 

91”.};

unis

,

my

:zﬁ‘d

; 533389:

,3“
\

.2?
V !

H5} 5 v“

‘v

..r

i t.

"1'37

33”.,

“I 1‘: a" a

.

'25?
H

A at.
9'5”.

1

mo

3‘
H “I“{J
«am-ma

 

_ “1.:
5

23:0? LIBRARY
Michigan State
University

This is to certify that the
dissertation entitled

THREE ESSAYS ON BEHAVIORAL FINANCE

presented by
Gabriele M. Lepori

has been accepted towards fulﬁllment
of the requirements for the

Ph.D. degree in ,f\ Economics

 

 

 

 

Major P'rotessor’s Signature
0 8 / / . QEES

Date

 

MSU is an afﬁnnative-action, equal-opportunity employer

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K'lPro’lAccsPres/CIRC/Dateoue mdd

THREE ESSAYS ON BEHAVIORAL FINANCE
By

Gabriele M. Lepori

A DISSERTATION

Submitted to
Michigan State University
in partial ﬁilﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Economics

2008

ABSTRACT

THREE ESSAYS ON BEHAVIORAL FINANCE

By
Gabriele M. Lepori

This dissertation consists of three empirical essays that explore whether and how
ﬁnancial decision-making is inﬂuenced by some psychological factors that do not have a
direct economic relevance. The hypotheses proposed here and the results obtained bring a
challenge to the accuracy of the paradigm that has been dominating the theory of asset
pricing, according to which investors’ decisions are exclusively shaped by economic
variables. Second, they represent a puzzle when interpreted in light of the Efﬁcient
Market Hypothesis.

The ﬁrst essay investigates the relationship between investor mood and
investment decisions. The evidence produced in the ﬁelds of psychology and behavioral
ﬁnance is employed to construct a hypothesis according to which daily changes in the
marginal investor’s emotional state, caused by changes in ambient air pollution, would be
expected to have an impact on aggregate demand in the stock market, ultimately affecting
equity returns in a predictable fashion. Such a behavioral hypothesis is tested by means of
a natural experiment allowed by an institutional change experienced by the Milan Stock
Exchange in the middle of the 1990’s. Interestingly, the empirical results do seem to

provide support to such a conjecture, and the “mood effect” detected here turns out to be

robust to the use of alternative air pollution proxies, estimation methods, and sets of
controls. It is also shown that, despite the predictions of the efﬁcient market theory, there
exist some unexploited proﬁtable trading strategies based on air pollution data.

The second essay extends the Seasonal Affective Disorder (SAD) hypothesis to
the foreign exchange market. According to medical evidence, seasonal changes in the
length of the day produce corresponding changes in human mood, which in turn is
conjectured to play a role in the process that governs investment decisions. Such a
framework appears to generate some neat predictions in terms of exchange rate
dynamics, which are then tested using weekly data on six currencies vis-a-vis the
Australian Dollar. The empirical analysis is conducted using an LSTR model, the
underlying idea being that the exchange rate moves between two regimes that arise out of
the divergence between the seasonal daylight cycles of the northern and southern
hemispheres. The ﬁndings, in this case, only provide weak evidence in support of the
hypothesis under investigation.

In the third essay, an empirical investigation is conducted to determine whether
investment decisions can be affected by superstition. After discussing the mechanisms
through which superstitions are conjectured to originate, the analysis focuses on a set of
superstitious beliefs that, according to folkloristics, can be expected to be held by large
groups of people. The behavioral hypothesis that arises out of this context is then tested
using daily data from the US. stock market, and the outcomes seem to be consistent with
the view that superstition can indeed affect investment choices. Moreover, it appears to
be possible to proﬁtably exploit some of the empirical regularities that superstitious

beliefs cause in the time series of equity returns.

ACKNOWLEDGMENTS

No researcher works in a vacuum. We are all influenced by others, and good
research takes a network of knowledge. This work was made possible by the
contributions that, directly or indirectly, came from many people.

First, I would like to thank Professor Richard T. Baillie for serving as my advisor
through the stages of my graduate school experience. His fruitful advice and kind support
are enormously appreciated. Indeed, the origins of this dissertation can be precisely
traced back to some early ideas that I developed when I was a student in his classes.

I am also highly indebted to Professors Geoffrey Booth, Raoul Minetti, and Emma
Iglesias, whose detailed comments and suggestions greatly improved the quality of the
analyses presented here. Similarly, I wish to thank Professor Michael Conlin for the
insights he gently shared with me about the mechanics of human decision-making.

Thanks to the former director of the economics graduate program, Professor
Thomas Jeitschko, for believing in my English proﬁciency, and to the current director,
Professor Paul Menchik, for giving me the opportunity to gain precious teaching
experience.

I am also happy to say that this long journey was made more enjoyable by the
encouragement that came on the one hand from my fellow doctoral students, especially
Makoto, Sanders, Simon, and Vincent, and on the other hand from the residents of Owen

Hall, particularly Alex, Chia-Yu, Harrison, Jay, Jorge, Lan, and Tatsuya.

iv

I want to express the deepest thank to my family, without which none of this
would have been achievable.

Last, I want to thank the advocates of orthodox ﬁnance and economics for the
research stimuli with which they have so far provided me. Heterodox approaches like the
one proposed here would probably lose much of their appeal, in my eyes, if there was not

a status quo to be challenged.

TABLE OF CONTENTS

LIST OF TABLES ................................................................................................. viii
LIST OF FIGURES .................................................................................................. x
CHAPTER 1

AIR POLLUTION, INVESTOR MOOD, AND FINANCIAL
CHOICES: EVIDENCE FROM THE MILAN STOCK

EXCHANGE ............................................................................................................... 1
1.1. Introduction ............................................................................................................ 2
1.2. Environment, Affective State, and Human Decisions ......................................... 6
1.3. Air Pollution: Taxonomy and Origins ............................................................... 10
1.4. Physiological and Psychological Reactions to Air Pollution ............................ 14

1.4. 1. Physiological Effects ...................................................................................... 14
1.4.1.1. Morbidity and Mortality .......................................................................... 14
1.4.1.2. Heart Rate ................................................................................................ 16
1.4.1.3. Threshold Concentrations ........................................................................ 17

1.4.2. Psychological Inﬂuence .................................................................................. 18
1.4.2.1. Risk Perception ........................................................................................ 20
1.4.2.2. Physical Health and Emotional State ....................................................... 22
1.4.2.3. Anxiety, Heart Rate, and Mood Misattribution ....................................... 23

1.5. Air Pollution, Mood Misattribution, and Investment Decisions ..................... 25
1.6. Data ....................................................................................................................... 28
1.6.1. Air Pollution Proxies ....................................................................................... 28
1.6.2. Equity Returns and Control Variables ............................................................ 32
1.7. Possible Issues of Endogeneity ............................................................................ 34
1.8. Empirical Examination ....................................................................................... 37

1.8.1. Particulate Matter and Equity Returns: a Natural Experiment ....................... 37

1.8.2. Basic Model .................................................................................................... 38

1.8.3. Controlling for Meteorological Factors .......................................................... 41

1.8.4. Controlling for Intemational Economic News ................................................ 43

1.8.5. Alternative Air Pollution Proxies .................................................................... 46

1.8.6. Further Robustness Checks ............................................................................. 48

1.8.7. Instrumental Variables Estimation .................................................................. 52

1.9. Air Pollutants, Mood, and Efﬁcient Market Theory ........................................ 56
1.10. Concluding Remarks ......................................................................................... 59

vi

CHAPTER 2
IS THERE A SAD FOREIGN EXCHANGE MARKET CYCLE?

EVIDENCE FROM AN LSTR MODEL .................................................... 74
2.1. Introduction: SAD Effect and Seasonal Stock Market Cycles ........................ 75
2.2. SAD and Exchange Rate Dynamics .................................................................... 77
2.3. Hypothesis Under Investigation ......................................................................... 80
2.4. An LSTR Model for the Exchange Rate ............................................................ 81

2.4.1. Transition Variable ......................................................................................... 82
2.4.2. Interpretation of the Parameters ...................................................................... 84
2.5. Data ....................................................................................................................... 87
2.6. Empirical Results ................................................................................................. 88
2.7. Concluding Remarks ........................................................................................... 90

CHAPTER 3

SUPERSTITION AND INVESTMENT DECISIONS ........................ 92
3.1. Introduction .......................................................................................................... 93
3.2. Origin of Superstitious Beliefs ............................................................................ 98
3.3. Superstitious Beliefs and Economic Decisions: Empirical Evidence ............ 101
3.4. Testable Hypothesis: Solar and Lunar Eclipses as Evil Presages ................. 103
3.5. The Physics of Eclipses ...................................................................................... 106
3.6. Environment, Emotional State, and Equity returns ....................................... 108
3.7. Data ..................................................................................................................... 110
3.8. Econometric Analysis ........................................................................................ 114

3.8.1. Total Eclipses and Magnitude of Equity Returns ......................................... 114
3.8.2. Total Eclipses and Direction of Equity Returns ........................................... 116
3.8.3. Multivariate Model ....................................................................................... 118
3.8.4. Controlling for Behavioral Factors ............................................................... 1 19
3.8.5. Controlling for Economic News ................................................................... 123
3.8.6. Direct vs. Indirect Visibility of the Phenomenon ......................................... 125
3.8.7. Controlling for Time—Varying Volatility ...................................................... 126

3.9. Superstitious Beliefs, Investment Decisions, and Efﬁcient Market Theory. 128
3.10. Concluding Remarks ....................................................................................... 130
REFERENCES ....................................................................................................... 138

vii

LIST OF TABLES

TABLE 1.1 SUMMARY STATISTICS - AIR POLLUTION AND ENVIRONMENTAL VARIABLES. 61
TABLE 1.2 SUMMARY STATISTICS — STOCK RETURNS ....................................................... 62

TABLE 1.3 AIR POLLUTION AND STOCK RETURN — A NATURAL EXPERIMENT (BASIC
MODEL) ..................................................................................................................... 63

TABLE 1.4 AIR POLLUTION AND STOCK RETURN — BASIC MODEL (ALTERNATIVE LAG
STRUCTURES) ............................................................................................................ 64

TABLE 1.5 AIR POLLUTION AND STOCK RETURN - CONTROLLING FOR METEOROLOGICAL
CONDITIONS ............................................................................................................... 65

TABLE 1.6 AIR POLLUTION AND STOCK RETURNS - CONTROLLING FOR INTERNATIONAL

ECONOMIC SHOCKS ................................................................................................... 66
TABLE 1.7 AIR POLLUTION AND STOCK RETURNS — ALTERNATIVE PROXY (NOX) ............ 67
TABLE 1.8 AIR POLLUTION AND STOCK RETURNS - ALTERNATIVE PROXY (SOZ) ............. 68
TABLE 1.9 AIR POLLUTION AND STOCK RETURNS - ALTERNATIVE SUB-PERIODS ............ 69
TABLE 1.10 AIR POLLUTION CROSS-CITY CORRELATION MATRIX .................................... 70

TABLE 1.11 AIR POLLUTION AND STOCK RETURNS — INSTRUMENTAL VARIABLES
ESTIMATION ............................................................................................................... 71

TABLE 1.12 A TEST OF THE EFFICIENT MARKET THEORY USING AN AIR POLLUTION INDEX

(GARCH MODEL) ...................................................................................................... 72
TABLE 1.13 A TEST OF THE EFFICIENT MARKET THEORY USING AN AIR POLLUTION INDEX

(LOGIT MODEL) ........................................................................................................ 73
TABLE 2.1 FORWARD PREMIUM ANOMALY REGRESSIONS .................................................. 91
TABLE 2.2 RESULTS FROM RESTRICTED LSTR UIP REGRESSIONS (C = 0) ........................ 91

viii

TABLE 3.1 SUMMARY STATISTICS OF US STOCK RETURNS .............................................. 132

TABLE 3.2 COMPARISON OF MEAN RETURNS .................................................................... 133
TABLE 3.3 SIGN TEST ....................................................................................................... 134
TABLE 3.4 REGRESSION ANALYSIS .................................................................................. 135
TABLE 3.5 CONTROLLING FOR BEHAVIORAL FACTORS ..................................................... 136
TABLE 3.6 CONTROLLING FOR ECONOMIC NEWS .............................................................. 137

ix

LIST OF FIGURES

FIGURE 1.1 DIRECT AND INDIRECT EFFECTS OF AIR POLLUTION ON PSYCHOLOGICAL WELL-
BEING. ........................................................................................................................ 18

FIGURE 1.2 DAILY AMBIENT CONCENTRATIONS (MG/M3) OF PARTICULATE MATTER (TSP) IN
MILAN. ...................................................................................................................... 29

FIGURE 1.3 MAIN SOURCES OF AIR POLLUTION EMISSIONS IN THE PROVINCE OF MILAN
(PERCENTAGE CONTRIBUTION). THE DATA ARE TAKEN FROM ARPA LOMBARDIA
(2001; 2002; 2003A; 2003B; 2003C; 2006). .............................................................. 31

FIGURE 1.4 DAILY EQUITY RETURNS OF THE ITALIAN STOCK MARKET (COMIT INDEX) ...... 32

FIGURE 2.1 DAILY HOURS OF NIGHT IN THE UNITED STATES AND AUSTRALIA (SEASONS
REFER TO THE NORTHERN HEMISPHERE). .................................................................... 77

FIGURE 2.2 FALL IN THE UNITED STATES. THE SAD EFFECT CAUSES THE US DOLLAR TO
APPRECIATE. .............................................................................................................. 79

FIGURE 2.3 DIFFERENCE IN THE DAILY NUMBER OF HOURS OF NIGHT BETWEEN UNITED
STATES AND AUSTRALIA (PANEL A) AND TRANSITION VARIABLE (PANEL B). NOTE:

SEASONS REFER TO THE NORTHERN HEMISPHERE. ..................................................... 83
FIGURE 3.1 EMPIRICAL FREQUENCY OF TOTAL ECLIPSES BY DAY OF THE WEEK. .............. 121
FIGURE 3.2 EMPIRICAL FREQUENCY OF TOTAL ECLIPSES BY CALENDAR MONTH .............. 122

Chapter 1
Air Pollution, Investor Mood, and Financial Choices:

Evidence from the Milan Stock Exchange

1.1. Introduction

There's so ”INC/I pollution in Ilzc air now that if i! “trawl/Or our lungs {licrc'd be no

place (0 put it all - Robert ()rlwn

It has long been suspected that some environmental stimuli have the power to
affect moods. For example, background music has long been in use in retail stores and
ofﬁces, allegedly to elicit certain desired behaviors or attitudes among Shoppers or
employees (Milliman, 1982; Bruner, 1990). Indeed, marketing researchers have been
investigating for decades how the store environment can be manipulated so as to
inﬂuence consumer purchases (Turley and Milliman, 2000). More recently, this kind of
effort has found a sound theoretical justiﬁcation in the psychological literature, whose
ﬁndings tend to support the belief that the feelings experienced at the time of making a
choice can bias the decision-making process through a mechanism of “mood
misattribution” (Schwartz and Clore, 1983; Schwartz, 1990; Mann, 1992).

Applying this same line of reasoning to the ﬁeld of ﬁnance is tempting for two
reasons. First, showing that investment decisions can be inﬂuenced by mood patterns
triggered by some environmental stressors would support the claims of behavioral
ﬁnance advocates, according to whom some psychological forces play a role in
determining the equilibrium prices that prevail in the ﬁnancial markets (Thaler, 1993).

Second, the computations required for making investment decisions are typically

complex, abstract, and involve risk, which are precisely the attributes that are believed to
induce people to rely more heavily on their emotions when making a choice (Forgas,
1995)

The present essay intends to give a contribution to this strand of literature by
investigating whether air pollution, through the adverse impact it is believed to have on
people’s physical and psychological Spheres, is able to exert a depressing inﬂuence on
investment decisions and, ultimately, on stock returns. Such a research question has been
stimulated by some promising ﬁndings which emerged in the ﬁeld of behavioral ﬁnance,
and the speciﬁc hypothesis that undergoes empirical testing has been strictly built upon
the evidence produced in the disciplines of medicine and psychology. As such, the results
can hardly fall under the criticism of data mining.

The empirical analysis is conducted using data from the Italian stock market].
Two are the main reasons that led to this choice. First, Milan, which hosts the
corresponding Stock Exchange, appears to be among the cities that have been suffering
the most from the air pollution phenomenon, at least within the industrialized countries
(The New York Times, 1989; International Herald Tribune, 2007). Second, the MSE
underwent an institutional change in the middle of the 1990’s, which allows for a natural
experiment that can shed light on the hypothesis under investigation (e.g., Pardo and
Valor, 2003). In particular, trading at the MSE switched from an open outcry system to a
computerized and decentralized one. If air pollution was an effective mood proxy and

mood had an inﬂuence on buying/selling decisions, then daily air pollution measured in

 

' The expressions “Italian stock market” and “Milan Stock Exchange” (MSE) will be used interchangeably
throughout the essay, as the city of Milan hosts the main (and now sole) Italian stock exchange.

the city of Milan, should exhibit a signiﬁcant relationship with stock returns only during
the era in which trading would physically take place on the ﬂoor of the MSE.

After controlling for well-known calendar anomalies, behavioral variables, and
international economic shocks, the empirical analysis produces results that are indeed
consistent with the hypothesis under scrutiny, for daily increases in air pollution levels
are estimated to have a signiﬁcant negative effect on stock returns during the sub-period
in which trading was centralized, whereas no effect is detected after the trading switch.
Furthermore, the lag structure of the air pollution effect appears to mimic the evidence
generated by medical and psychiatric studies, as current changes in air pollution
concentrations (proxied by particulate matter) are estimated to inﬂuence equity returns
concurrently and in the following few days.

The results turn out to be robust to the use of alternative air pollution proxies
(nitrogen oxides, sulﬁir dioxide), estimation frameworks (GARCH, OLS, Logit), and
sample choices.

A potential endogeneity problem arises due to the links that air pollution has with
economic activity on the one hand and social costs on the other hand. Though several
clues, as discussed later, suggest that the detected air pollution effect is not caused by
these links, an instrumental variables estimation is carried out in order to identify more
precisely the inﬂuence that pollution has on stock returns. More Speciﬁcally, three lagged
meteorological variables are employed as instruments: wind speed, atmospheric pressure,
and rainfall. Empirically, these variables show a high correlation with the endogenous
variable and moreover, it seems very reasonable to believe that they do not have any

systematic relationship with either current economic activity or equity returns. As a

result, I think that the empirical ﬁndings, which conﬁrm the outcomes of the previous
steps, do provide solid evidence of a causal impact of air pollution on investment
decisions and ultimately, on stock returns.

These results are remarkable on two grounds. First, they bring a challenge to the
paradigm that has been dominating economic analysis, according to which only purely
economic factors play a role in asset pricing. Second, they constitute an anomaly when
interpreted in light of the Efﬁcient Market Theory, as the study reveals the existence of
some unexploited proﬁtable trading strategies based on past (and public) air pollution
data.

Future research efforts may want to extend this framework to other ﬁnancial
markets, especially those that still operate through open outcry trading systems. The use
of experimental data seems quite problematic in this particular setting, but alternative
environmental stressors could be employed successfully maintaining a Similar approach.

The rest of the chapter is organized as follows. Section 1.2 discusses the
mechanisms through which investment decisions are conjectured to be affected by some
psychological factors, and comments on the empirical evidence produced so far. Section
1.3 gives on overview of the air pollution phenomenon and section 1.4 examines the
effects it has been found to exert on the human body and psyche. Section 1.5 puts
forward the behavioral hypothesis that will be tested, and section 1.6 describes the data.
Section 1.7 explores some potential sources of endogeneity, and the empirical analysis is
conducted in section 1.8, which also contains a battery of robustness checks. Section 1.9
investigates the implications that the ﬁndings have in light of the Efficient Market

Theory. Section 1.10 concludes.

1.2. Environment, Affective State, and Human Decisions

In the last ten years or so, ﬁnance researchers have begun to investigate a subject
that has been attracting marketing experts’ attention for decades. The question of whether
and how the store environment can be used to affect the time and money that consumers
spend in the store, obviously, has always been of interest to retailers. The concept of
“store atmospherics”, introduced by Kotler (1973—1974), refers precisely to “the effort to
design buying environments to produce Speciﬁc emotional effects in the buyer that
enhance his purchase probability”. From a theoretical viewpoint, this kind of endeavor
can ﬁnd validation, for example, in the work of Byme and Clore (1970), who maintain
that “affect elicited by a stimulus conditions behavior and attitudes toward other stimuli
merely associated with it”. In other words, the emotional state experienced at the time a
decision is being made is likely to condition the decision itself (Isen et al., 1978; Frijda,
1988; Forgas, 1995; Loewenstein, 2000), for emotions are believed to regulate thought
and inform judgment and cognitive evaluations (Damasio, 1994; Loewenstein et al.,
2001). What is intriguing here is the idea that mood works as a “source of information” to
individuals, and it inﬂuences their choices even when the source of the mood state does
not have anything to do with the decision being made, i.e. it is believed that a mechanism
of “mood misattribution” may be at work in people’s minds (Schwartz and Clore, 1983;

Schwartz, 1990).

Such a hypothesis can ﬁnd support, for instance, in several studies on consumer
behavior. Product choice, purchase intentions, behavior traits, and actual purchases have
all been shown to be partly affected by factors such as in-store music, ambient scent and
illumination (Milliman, 1982; McElrea and Standing, 1992; Areni and Kim, 1993; Gulas
and Bloch, 1995; North and Hargreaves, 1996; 1997; 1998; Summers and Hebert, 2001).
The existence of such links appears hard to justify unless one is willing to somewhat
loosen the deﬁnition of rational decision-making.

It seems therefore natural to wonder whether these fascinating ﬁndings can be
extended so as to encompass a broader set of human decisions and a wider deﬁnition of
environment. This is exactly where ﬁnance researchers come into play. Indeed, one of the
strands of literature spawned by the growth of behavioral ﬁnance, at the end of the
1990’s, has started to Show a keen interest in the relationship between investor mood and
investment decisions, where such a connection has been typically interpreted as being
mediated by the “mood misattribution” mechanismz. Broadly Speaking, behavioral
ﬁnance advocates claim that, contrary to the view of the paradigm that has dominated
economic analysis (labeled by Shiller (2006) as “neoclassical ﬁnance”), several
psychological factors play a role in the mental process that originates people’s investment
decisions and, as such, these same factors could be successfully incorporated into asset
pricing models (e.g., Kahneman and Tversky, 1979; Kahneman and Riepe, 1998). More
narrowly, the strand of literature previously mentioned has focused on trying to identify
some environmental variables that might act as mood proxies for large groups of

investors, the rationale being that changes in the environment may trigger mood changes

 

2 For an excellent review on the role of the mood misattribution framework applied to a behavioral ﬁnance
context, see Hirshleifer and Shumway (2003), and Lucey and Dowling (2005a).

and, ultimately, have an impact on investment decisions through, for example, the mood
rrrisattribution mechanism. The seminal contribution in this area can be traced back to
Saunders (1993), who, employing the percentage cloud cover in New York city as a
proxy for investor mood, observes that such an explanatory variable exhibits a signiﬁcant
relationship with the returns of three global indices of the US. stock market. These
results have been challenged by some subsequent investigations (Pme and Valor, 2003;
Keef and Roush, 2003; Tufan and Hamarat, 2004; Loughran and Schultz, 2004; Lucey
and Dowling, 2005a; Goetzmann and Zhu, 2005), yet some supporting evidence has also
been produced (Hirshleifer and Shumway, 2003; Chang et al., 2006).

Along similar lines, alternative environmental factors have been used as proxies in
an attempt to capture some detectible collective mood swing patterns. These include
temperature (Keef and Roush, 2003; Cao and Wei, 2005; Chang et al., 2006), humidity
(Pardo and Valor, 2003; Chang et al., 2006), rain and snow (Hirshleifer and Shumway,
2003) and wind (Keef and Roush, 2003). It should be said, however, that only a few of
these studies appear to have been rigorously constructed upon some Speciﬁc hypothesis
produced in the ﬁelds of medicine and/or psychology. One of the most intriguing
analyses in this area can be attributed to Kamstra et al. (2003), who empirically test
whether the well-known Seasonal Affective Disorder (SAD) phenomenon, caused by the
seasonal variation in the length of the day, is responsible for at least part of the seasonal
pattern that stock returns typically exhibit. Their ﬁndings are encouraging for behavioral
ﬁnance proponents, and, from a larger perspective, their work also suggests that, in order
to be considered solid, empirical evidence in this ﬁeld needs to be deeply rooted into

some underlying psychological/medical hypothesis.

This is precisely the goal of the present investigation. Particular care is paid in the
next few sections to the shaping of a behavioral hypothesis that is strictly built upon
medical and psychological ﬁndings and is subsequently tested empirically. Here the focus
will be on ambient air pollution, which, indeed, can be considered one of the most critical
environmental stressors to which individuals are exposed, and has been deemed
responsible for a broad Spectrum of physical and psychological effects on human beings.
Based on the evidence offered in the relevant literature, it will be possible to identify an
unequivocal hypothesis according to which daily increases in the level of air pollution are
believed to be associated with increased psychological distress (i.e., mood deterioration)
and, in turn, with reduced demand in the stock market. It is therefore conjectured that,
ceteris paribus, air pollution should have a negative marginal effect on equity returns.
The empirical analysis, carried out in the following sections, indeed seems to produce
results that are highly consistent with such an interpretation. The ﬁrst step now will
consist in presenting an overview of the air pollution phenomenon and of its

CODSCQUCI‘ICCS.

1.3. Air Pollution: Taxonomy and Origins

Air pollution is an umbrella term that refers to the presence of unhealthy particles
and gases in the atmosphere that may endanger the health of humans, animal life, plants,
and may even damage objects such as statues and buildings3. In this essay, most of the
analysis is centered upon one speciﬁc constituent of air pollution, i.e. Particulate Matter
(PM). Yet some attention is also devoted to the examination of the role played by

nitrogen oxides (NOX) and sulfur dioxide (S02). Several reasons have contributed to this

choice; ﬁrst, as the following sections will Show, most epidemiological, toxicological and
psychological studies have focused on these compounds, especially on PM. Second, daily
data about the levels of these pollutants in Milan and neighboring cities are readily
accessible and reliable. Third, though these three pollutants are somewhat related to each
other, as discussed later, their emission into the atmosphere originates from rather
different sources, which will allow us to address more convincingly some potential issues
of endogeneity.

PM consists of “a complex mixture of solid and liquid particles of organic and
inorganic substances suspended in the air” (WHO, 2003). These suspended particles may
vary in composition, dimension, and origin, and are generally categorized according to
their aerodynamic diameter. Total Suspended Particulates (TSP) include all airborne

particles. PMIO refers to particles with an aerodynamic diameter less than 10 pm

 

3 See Colls (2002) for an introduction to air pollution.

10

(micrometers). PM2_5 iS used for particles with a diameter less than 2.5 pm. PM can

either be directly emitted into the air (primary PM) or be produced secondarily

(secondary particulate) in the atmosphere from gaseous precursors, mainly S02, NOX,

ammonia, and non-methane volatile organic compounds (WHO, 2006). The largest
particles typically consist of wind-blown dust from agricultural processes, mining
operations, uncovered soil, and unpaved roads. Road dust caused by trafﬁc is also a
critical contributor to this category. Additional inputs come from mould spores, plant and
insect parts, and pollen grains. Smaller particles (diameter less than 2.5 pm) are largely
formed from gases, Whereas the smallest ones are formed by nucleation from heavy
metals, elemental carbon, organic carbon, sulfates, and nitrates (WHO, 2003). According
to WHO (2006) the major share of TSP emissions at the European level is estimated to
originate from “the combustion of solid fuels in small stoves in the residential and
commercial sectors, followed by industrial emissions from energy combustion and
manufacturing processes and from agricultural activities”.

Residence time in the atmosphere is a key factor in determining the travel distance
of air pollutants. The residence time of PM in the atmosphere may vary from 1-2 days to
4-6 days, depending mainly on the size and chemical composition of the particles (WHO,
2006). The larger particles are more easily deposited and normally travel less than 6
miles from their place of origination. Yet, dust storms may transport them for over 600
miles. Primary ﬁne particles can travel up to 1200-1800 miles, which implies that
pollutants emitted in one region can affect PM concentrations in adjacent regions and

even in neighboring countries.

11

The symbol NOx is used to represent the total concentration of nitric oxide (NO)
plus nitrogen dioxide (N02). NO reacts with the oxygen in air to form nitrogen dioxide
(Cotton et al., 1999). Although there are natural sources of NOx (e.g., volcanic action,

forest ﬁres), the combustion of fossil fuels is the major contributor in European urban
areas (WHO, 2003). Trafﬁc, in particular, represents the major anthropogenic source of

NOx, followed by stationary sources such aS power plants, domestic heating, furnaces
and boilers. As a result, NOx is a good indicator of automotive emissions, which also

makes it a proxy for other unmeasured pollutants emitted by vehicles (Samakovlis et al.,

2004). Additionally, NOx is “a precursor for a number of harmful secondary air

pollutants, including nitric acid, the nitrate part of secondary inorganic aerosols and photo
oxidants” (WHO, 2003).

Outdoor S02 is the main product from the combustion of sulfur compounds. The
most important natural sources of 802 are volcanoes, forest ﬁres, and oceans. As far as

anthropogenic emissions are concerned, they can be mainly tracked back to residential
heating, power plants, smelting of metals, paper manufacture and, residually, trafﬁc
(ARPA, 2003). Beyond being part of air pollution per se, sulfur dioxide and nitrogen
oxides also play a role in the acid rain phenomenon, as they mix with water vapor in the
atmosphere and then fall to earth as rain, snow and mist (McCormick, 1989).

Human beings are exposed to air pollution by inhalation and through contact with
the skin and the eyes (Oehme et al., 1996). Though people spend the majority of their
time indoors, WHO (2006) maintains that outdoor concentrations of air pollutants,
typically measured through monitoring networks, are representative of population

exposure.

12

Before moving to the existing evidence about the human response to air pollution,
it is important to emphasize that the investigations conducted in the ﬁelds of
epidemiology and toxicology have not been very successful yet at isolating the health
effects of individual pollutants, for toxins are normally present in the atmosphere in the
form of mixtures. The issue is further enhanced by the fact that combinations of different
air pollutants are conjectured to give rise to dynamic chemical reactions and generate
synergistic effects (Samakovlis et al., 2004). Additionally, any given chemical might
interfere with the absorption or detoxiﬁcation of some other substances (Oehme et al.,

1996). Throughout the essay, PM, NOx, and S02 will be merely regarded as proxies for

air pollution, keeping in mind that the impact they may be found to have on investment

decisions and, ultimately, on equity returns might originate from a mixture of pollutants.

13

1.4. Physiological and Psychological Reactions to Air Pollution

1.4.1. Physiological Effects

Air pollution has been the core of a massive number of neuro-toxicological and
epidemiological investigations4. Short-term and long-term exposures to ambient air
pollution have been connected with a wide variety of acute and chronic health effects,
respectively, ranging from slight irritation symptoms to restricted activity and to death
(American Thoracic Society, 2000; WHO, 2001). Though the most severe effects are
conﬁned to a relatively small portion of the population, a huge number of people is

believed to experience at least some minor symptoms and some form of physical distress.

1.4.1.1. Morbidity and Mortality

PM appears to be the most widely studied factor, and there is increasing evidence
pointing at PM as a dangerous contaminant per se (WHO, 2006). The research conducted
in this area has detected a broad spectrum of short-term effects, ranging from lung
inﬂammatory reactions and respiratory symptoms to increases in hospital admissions and
mortality rates (Ostro et al., 1995; Dockery and Pope, 1996; Pope, 1996; Cropper et al.,

1997; Daniels et a1, 2000; Chay and Greenstone, 2003; Peng et al., 2004; Peng et al.,

 

‘ See WHO (2003) for a detailed list of references.

14

2005; Maynard et al., 2007). Similarly, long-term exposure has been associated with rises
in cardiovascular diseases and reduced life expectancy (Ostro, 1994; Brunekreef, 1997;
Abbey ct al., 1999; Panyacosit, 2000; Pope et al., 2002). Focusing on morbidity rates, for
example, Ostro and Rothschild (1989), using US data from the Health Interview Survey,
maintain that there exists a Short-term positive association between ﬁne particulate and
“both minor restrictions in activity and respiratory conditions severe enough to result in
work loss and bed disability in adults”. Taking a closer look at Italian data, Michelozzi et
a1. (1998), using a sample covering the period 1992-1995, ﬁnd that a 10 pg/m3 increase
in TSP in Rome is estimated to produce a concurrent 0.4% increase in total daily
mortality. A comparable relationship (at lag 0 and 1) is detected by Biggeri et a1. (2001)

using daily data about PM") from eight Italian cities, and by Le Tertre et al. (2005) with

regard to the city of Milan.

The picture does not change substantially when it comes to NOX, which is also

believed to have relevant effects on morbidity and mortality rates, both in the short- and
in the long-term (Love et al., 1982; Sexton et al., 1983; Koenig et al., 1987; Folinsbee,
1992; Salome et al., 1996; Strand etal., 1997; Morgan etal., 1998; Blomberg etal., 1999;
Erbas and Hyndman, 2001; Samakovlis et al., 2004; Schildcrout et al., 2006). In their
study on Italian data, Michelozzi et a1. (1998) estimate that a 10 rig/m3 increase in N02
(at lag l and 2) produces a 0.3%-O.4% increase in total daily mortality. An analogous
representation emerges from the investigations on Italian data conducted by Biggeri et al.
(2001) and Fusco et al. (2001), the latter ﬁnding that total respiratory hospital admissions

are signiﬁcantly correlated with same-day levels of N02.

15

Moving to the third air pollutant under investigation, sulfur dioxide (302), there

exists an extensive body of research documenting its adverse inﬂuence on respiratory
symptoms, cardiovascular diseases, and premature mortality (Chinn et al., 1981;
Krzyzanowsli and Wojtymiak, 1982; Bates and Sizto, 1983; Schwartz et al., 1988;
Derriennic et al., 1989; Ponka, 1990; Schwartz et al., 1991; Ostro, 1994, Sunyer et al.,
1996; Wong et al., 2002; Low et al., 2006; Chen et al., 2007). Among the investigations

on Italian data, a Signiﬁcant association between daily S02 levels and some relevant

health outcomes has been found, for example, by Ciccone et al. (1995), Vigotti et al.

(1996), and Cadum et al. (1999).

1.4.1.2. Heart Rate

Among the direct effects of air pollution on the cardiovascular system, one turns
out to be particularly relevant in light of the current context. Numerous recent
investigations have documented the existence of a positive link between daily/home
rises in air pollution levels and augmented heart rate (Peters et al., 1999; Pope et al.,
1999; Dockery, 2001; Magari et al., 2001; Iagorio et al., 2003; Gong et al., 2004; Liu et
al., 2007). For example, Dockery et al. (1999) ﬁnd that “increased pulse rate, as well as
the odds of having a pulse rate 5 or 10 beats per minute above normal” are signiﬁcantly
associated with exposure to PM on the previous 1 to 5 days. Similarly, studying a sample
of middle aged healthy individuals living in a metropolitan area in France, Ruidavets et

al. (2005) claim that there exists a positive and contemporaneous connection between

16

daily levels of N02 and S02 and resting heart rate. Furthermore, the authors Show that

such an impact can be noticed even at low concentration levels of these pollutants.

These ﬁndings are critical here because heart rate changes, as we will see later,
are also associated with the onset of feelings of anxiety and fear (i.e., heightened anxiety
is linked to augmented heart rate). The fact that heart rate is connected both with air
pollution, on the one hand, and with affective arousal, on the other hand, might foster a
phenomenon of mood misattribution. Such a line of reasoning will be developed in depth

in section 1.4.2.3.

I . 4. 1. 3. Threshold Concentrations

A crucial question is whether there exists a threshold below which no effects of
air pollution on human health are to be anticipated. With regard to PM, the World Health
Organization conducted a review study with the speciﬁc purpose of giving an answer to
such a question, concluding that “most epidemiological studies on large populations have
been unable to identify a threshold concentration below which ambient PM has no effect
on mortality and morbidity. It is likely that within any large human population, there is a
wide range in susceptibility so that some subjects are at risk even at the low end of
current concentrations” (WHO, 2006). A similar rationale induced the World Health

Organization to state that “there is no evidence for a threshold for N02” (WHO, 2006).
More uncertainty surrounds the third pollutant under observation, $02. In the

2005 update of its air quality guidelines, the World Health Organization, after giving

emphasis to the ﬁnding that sulﬁrr dioxide “was signiﬁcantly associated with daily

17

mortality in 12 Canadian cities with an average concentration of only 5 jig/m3”, claims

that, if there were some Signiﬁcant S02 thresholds, they “would have to be very low”

(WHO, 2005).

1.4.2. Psychological Inﬂuence

Apart from its well documented biophysical effects, air pollution has also been

found to have an impact on human mental and emotional health (see Figure 1.1).

 

l Mass media HRisk perception

Environmental U
stressors

(air pollution) j Physical healthk/
Heart rate HMisattribution

Figure 1.1 Direct and indirect effects of air pollution on psychological well-being.

  

 

Psychological
discomfort
(lower mood,
anxiety, stress,

depression)

   

 

x
r

   

 

   

 

   

 

 

     

 

 

Meertens and Swaen (1997) even claim that the latter might often be of greater
importance to well-being than the former. Perceived effects, such as annoyance, have
been detected in numerous studies (e.g., Forsberg et al., 1997; Klaeboe et al., 2000;
Danuser, 2001; Rotko et al., 2002). Peper (1999) states that human beings exposed to
neurotoxins “may exhibit alterations in cognitive and affective functioning, and report a

wide range of subjective symptoms”, a point of view that is shared by numerous other

18

investigations (e.g., Weiss, 1983; Evans and Cohen, 1987; Schottenfeld, 1992). Among
the symptoms, feelings of fatigue, low mood and exhaustion have shown a signiﬁcant
association With air quality (Sagar et al., 2007). Indeed, air pollution can represent “a
major stressful stimulus to exposed persons and can lead to a variety of emotional, mental
and physical changes not only by direct toxic effects”, for an indirect cognitive mediation
can also play a role “in terms of a negative appraisal of pollutants” (Bullinger, 1990). In
the words of Lundberg (1996), environmental toxins can generate “symptoms compatible
with anxiety and depression, among them cognitive and behavioral changes”.

Bullinger (1989), using multivariate time-series analysis on German data,

estimates that daily increases in air pollution concentrations (S02) have a

contemporaneous and lagged negative effect (up to lag 4) on emotional well-being,
measured in terms of mood and perceived stress. Similarly, Evans et al. (1988),
investigating a sample of Los Angeles residents, ﬁnd a modest yet signiﬁcant relationship
between ambient photochemical oxidants and anxiety symptomss. Along the same lines,
Zeidner and Schechter (1988), and Chattopadhyay et al. (1995), using Israeli and Indian
cross-sectional data, respectively, claim that exposure to acute levels of ambient air
pollution is responsible for heightened levels of anxiety, depression, and tension.
According to the evidence uncovered by some epidemiological studies, the effects
of environmental toxins can also penetrate into the psychiatric sphere. More Speciﬁcally,

after controlling for meteorological variables and other confounders, PM, N02, and other

photochemical oxidants appear to have a positive association with the incidence of

 

5 The class of photochemical oxidants for the most part consists of ozone, nitrogen dioxide, and peroxyacyl
nitrates (WHO, 1979).

19

psychiatric emergencies (Strahilevitz et al., 1979; Briere et al., 1983; Rotton and Frey,

1984)

1.4.2.1. Risk Perception

Among the indirect effects that air pollution can exert on mental health, two
channels have received particular attention. The ﬁrst channel deals with the links between
beliefs, risk perception, and mental discomfort (Figure 1.1). Lazarus and Folkman (1984)
argue that the health consequences of an (environmental) stressor, such as a pollutant,
depend on the appraisal of the threat (perceived risk) and of the personal resources to deal
with it. Within such a framework, not only air pollution per se, but also people’s beliefs
about air pollution, through their inﬂuence on risk perception, are expected to have an
impact on human psychological well-being. In this regard, Lima (2004), studying a
sample of individuals living near an incinerator, ﬁnds that risk perception, together with
the interaction between risk perception and environmental annoyance created by air
pollution, “signiﬁcantly increase the prediction of psychological symptoms”, i.e., anxiety,
depression, and stress. Lercher et al. (1995), analyzing the relationship between perceived
air quality and psychosomatic symptoms in a sample of adults in Tyrol, Austria, argue
that poorly rated air quality is signiﬁcantly associated with low mood and feelings of
fatigue and exhaustion, though an analogous link could not be established with actual air
quality. Lima (2004) suggests that risk perception “ampliﬁes the effects of annoyance, as
it introduces a suggestion of danger to environmental changes”, in turn inducing people

to pay more attention to such changes as they can be perceived as potentially harmful.

20

Her empirical results Show that the “suspicion of threat”, per se, can produce heightened
levels of annoyance and, ultimately, more extensive symptoms of psychological
irritation.

These ﬁndings are supported by another strand Of literature that shows how the
release of “sham” airborne chemicals, totally innocuous per se, can trigger perceptions of
somatic change and illness through a process of symptom misinterpretation (MacGregor
and Fleming, 1996). It has been found, for instance, that symptom reporting is ampliﬁed
by the release of unfamiliar or unpleasant (innocuous) odors (Smith et al., 1978; Knasko,
1993; Dalton et al., 1997; Chen and Haviland-Jones, 1999; Dalton, 1999; Jones et al.,
2000), and even by the presence of “imaginary” pollutants/odors (O’Mahoney, 1978;
Knasko et al., 1990; Lange and Fleming, 2005).

In the context of the present investigation, such a line of reasoning is relevant
because not only individuals inhale neurotoxins from the environment in which they live,
for they also receive information from the mass media and their interpersonal networks
about air pollution daily ambient concentrations and the risks it spawns (Coleman, 1993;
Morton and Duck, 2001; Frewer et al., 2002; Wakeﬁeld and Elliott, 2003). In this sense,
the mass media may work as what Kasperson et al. (1988) call “risk ampliﬁcation
stations”. Indeed, it is reasonable to believe that media coverage of air pollution is
increasing in the levels of pollution itself: sudden and sharp rises in the ambient levels of
some neurotoxins are likely to cause increased volumes of information, the use of
“sensational” headlines, and more frequent debates which, ultimately, can be expected to
amplify risk perception among the public. If one believes in the framework proposed

above, the implication is that daily increases in air pollution concentrations can be

21

thought of as having both a direct negative impact on physical health and an indirect
negative inﬂuence, modulated by risk perception, in terms of psychological discomfort,

i.e., anxiety and stress.

1.4.2.2. Physical Health and Emotional State

The second indirect channel has do to with the interconnection existing between
physical and emotional health. In his theoretical work on mood, Morris (2000) asserts
that the fundamental function of the mood system is “to regulate goal-directed behavior
in such a way as to maintain a balance between the availability of goal-relevant resources
and the perceived level of demands for them”. In his interpretation, goal—relevant
resources include factors such as health, Skills, and money, Whereas demands for these
resources originate from goals that people set. The author then claims that mood changes
can be initiated by environmental events that modify people’s perceived demands and/or
resources, the implication being that “mood will deteriorate when available resources are
perceived as inadequate to meet active demands”. In his opinion, mood changes then
affect how individuals think, what they feel like doing, and how optimistic/pessimistic
they are (e.g., “when mood is bad, goal-directed activity seems less likely to succeed and
so we are more inclined to put off a task or lower our aspiration level”). When applied to
the current context, Morris’ theory suggests that events such as daily increases in air
pollution levels, by having a negative impact on physical health (i.e., a goal-relevant

resource), can indirectly provoke a deterioration of mood.

22

Such a view can ﬁnd support in numerous empirical studies (e.g., Aneshensel et
al., 1984; Livneh and Antonak, 1994; Maier and Watkins, 1998; Cohen et al., 1999;
Rasul et al., 2002; Baker, 2006). Eckenrode (1984), for instance, investigating the causes
of short-term changes in mood in a sample of women, concludes that the most important
direct determinants are concurrent daily stressors and physical symptoms, alongside
previous levels of psychological well-being. Pain and discomfort produced by the
symptoms of physical impairment are very likely to play a role in worsening mood (Rasul

etaL,2002)

1.4.2.3. Anxiety, Heart Rate, and Mood Misattribution

It is believed that exposure to physical and psychological stressors induces the
autonomic nervous system to produce physiological arousal (Levenson, 2003), and an
increased heart rate is characteristic of such a state of arousal (Appelhans and Luecken,
2006). Experiencing heightened autonomic activity, such as a heart rate acceleration, in
response to a situation or event “can serve as a somatic cue that one is encountering a
potentially difﬁcult or dangerous situation” (Hastings et al., 2007). An extensive body of
research has documented the existence of a positive correlation between anxiety and
cardiovascular activity, measured in terms of heart rate (Behnke et al., 1974; Behnke and
Beatty, 1981; Mezzacappa et al., 1997; Abel and Larkin, 1990; Calvo et al., 1996; Roth et
al., 1996; Breivik et al., 1998; Georgiades et al., 2000; McCarthy and Goffrn, 2004).
Analogously, pulse rate appears to reveal a good correspondence to self-reported fear

(Lang et al., 1970; Lang, 1971; Sartory et al., 1977).

23

As it has been previously argued, affective States (feelings or mood) contain
information that individuals employ for the assignment of meaning or relevance to the
situation. Nonetheless, if one embraces the mood misattribution theory, then what follows
is the View that human beings may often attribute arousal or feelings to the wrong sources
(e.g., Ross, 1977). In the present context, the mood misattribution framework may
operate as follows (see Figure 1.1): (1) when dealing with investment decisions, instead
of correctly attributing their increased anxiety and heart rate to increases in air pollution
levels, people may mistakenly ascribe them to an increased perceived riskiness of their
portfolio of assets and/or to an augmented perceived uncertainty about the economy’s
future prospects6 (e.g., Hirshleifer and Shumway, 2003; Kamstra et al., 2003; Lucey and
Dowling, 2005a); (2) if this was the case, then people’s affective state would play a role
in shaping their ﬁnancial decisions (i.e., increased anxiety and heart rate would likely
lead investors to shy away from risky stocks and prefer safer assets7). This is, of course, a
conjecture. The exact underlying mechanisms are left for investigation to others, such as

experimental psychologists.

 

6 Hirshleifer and Shumway (2003), Kamstra et al. (2003), and Lucey and Dowling (2005a), for example,
discuss a similar mechanism, though referring to different environmental factors.

7 This line of reasoning (i.e., the higher the level of anxiety, the lower the demand for risky stocks) is
suggested, for example, by Kamstra et al. (2000).

24

1.5. Air Pollution, Mood Misattribution, and Investment Decisions

As it Should be clear by now, the hypothesis advanced and tested in this essay
builds upon research from psychology and medicine which argue, on the one hand, that
people’s decisions are inﬂuenced by their mood and, on the other hand, that variations in
air pollution levels can affect people’s mood. Figure 1.1 summarizes the channels,
discussed in the previous sections, through which pollution is believed to exert an impact
on psychological well-being. What needs to be emphasized here is that all the causal
relationships discussed above point to the same direction, i.e., all of them suggest a
negative inﬂuence of poor air quality on emotional state. An extensive review of the
relevant literature has also shown that both contemporaneous and lagged (negative)
effects can possibly be produced. In the case of the latter, the time-series studies
examined here have revealed a lag structure that typically ranges between 1 and 5 days.

Applying the mood misattribution framework to the present setting appears to
yield an unambiguous testable hypothesis. The key idea in that framework is that “mood
tends to inform decisions even when the cause of the mood is unrelated to the decision
being made” (Lucey and Dowling, 2005a). Ceteris paribus, (1) if increases in the level of
air pollution cause market participants to experience lower mood, greater anxiety, and
augmented heart rate and (2) investors attribute such affective changes to the wrong
source (e.g., perceived riskiness of their portfolios, pessimism about future developments

of the economy), then (3) their investment decisions will turn out to be inﬂuenced by a

25

factor that, per se, is economically irrelevant. For instance, following the reasoning
hinted at by Kamstra et al. (2000), greater subjective anxiety could be expected to induce
agents to prefer safer assets over risky stocks. In turn, all else equal, this would lower
demand in the equity market, pushing down stock returns. In one word, if the previous
conjecture was valid, then in a reduced-form model one would expect to observe a
negative relationship between air pollution concentrations and stock returns. The goal of
the present investigation is precisely to empirically test such a linkage. Keeping in mind
the medical and psychological literature cited above, the data will be inspected for both
simultaneous and lagged effects.

It is useful to remark that, given the existence of limits to arbitrage (e.g., Barberis
and Thaler, 2001), for this air pollution effect to leave any trace of evidence in the pattern
of stock returns it may be sufficient that only a subset of agents suffer from the above-
mentioned mood ﬂuctuations (Lucey and Dowling, 2005a). A necessary condition,
however, is that agents do not realize that their decisions are driven by changes in their
moods (Mehra and Sah, 2002).

Interestingly, the lag structure potentially embedded in the air pollution effect also
generates an opportunity to put to test the Efﬁcient Market Theory, according to which
past data (e.g., lagged pollution levels) contain no useful information for predicting future
stock returns.

The Italian stock exchange, located in Milan, seems to represent an especially
suitable setting for performing these tests. First, as shown below, during the period under
observation air pollution reached high levels numerous times in the city of Milan and, as

such, its inhabitants are likely to have been appreciably affected by it. Second, the Milan

26

Stock Exchange moved from a ﬂoor trading system to a computerized and decentralized
one at the beginning of the 1990’s; this event offers the chance to make use of a natural

experiment that can give more strength to the empirical results.

27

1.6. Data

1.6.1. Air Pollution Proxies

Ambient air pollution concentrations are typically measured through a network of
monitoring stationss. Given the goal of the present study and the potential issues of
endogeneity discussed below, one would like to employ a measure of air pollution that is
highly representative of population exposure near the Milan Stock Exchange (where
traders were physically located) and, at the same time, is very unlikely to be
systematically associated with overall economic activity in Italy. In particular, ideally,
this measure should exhibit no systematic connection with the emissions coming from
businesses that are strictly related to companies listed on the Italian stock market. This
would make air pollution an exogenous factor when explaining the behavior of Italian
stock returns.

In the city of Milan, over the period under consideration, data about PM have

been recorded by three monitorsg. Given the characteristics of these monitors, their

 

8 According to the guidelines set by the Italian Environmental Protection Agency (APAT, 2004), monitors
are classiﬁed according to their type/purpose (Background, Industrial, Roadside), the area in which they are
located (Urban, Suburban, Rural), and the characteristics of that same area (Residential, Commercial,
Industrial, Agricultural).

9 One monitor (Roadside/Urban/Residential-Commercial) is located in Liguria Street, 1.9 miles South of
the MSE. A second monitor (Roadside/Urban/ResidentialCommercial) is located in De Vincenti Street, 2.2
miles North-West of the MSE. The third monitor (Baekground/Urban/Residential) is located in Juvara
Street, 1.9 miles North-East of the MSE. According to APAT (2004), roadside monitors are meant to
measure the air pollution mainly caused by trafﬁc emissions coming from neighboring streets; the
concentrations they record are representative of areas whose radius generally does not exceed a few
hundred yards. Industrial-type monitors are used to quantify the air pollution mainly emitted by close

28

proximity to the Milan Stock Exchange, and the purpose of the present analysis, only one
station has been judged as suitable for providing the data"). It is worth to stress that the
measurements made by such a monitor are representative of population exposure in a

small area that precisely encircles the MSE.

  

      

 

'T T-F'T ‘ 1 'ﬁ.—‘ '7 ‘T__ I

o _, -
Q '\ (Ir ’5 ‘3 ’\ % Q '\ (Ir '5 ft.) ’\
3% 9% 9% 3? '99 .83" '9’? 9% \‘ib '9}, P999 '99 v39 '99 ’99} ('9 N90103:,

Figure 1.2 Daily ambient concentrations (pg/m3) of particulate matter (TSP) in Milan.

Hourly data about PM, NOX, and S02 ambient concentrations, as measured by

such a monitoring station, from January 1, 1980 through May 19, 2006, have been

acquired from ARPA Lombardia”. Particulate matter is measured as TSP from

 

industrial sites; the concentrations they track are representative of areas whose radius is generally less than
100 yards. Background stations normally represent overall city-wide exposure more closely than do
roadside sites. In particular, background-urban-residential stations are employed to monitor air pollution
levels within large urban areas. The measurements they make are not directly inﬂuenced by nearby trafﬁc
or industrial activities, and are meant to track air pollution mainly generated within the urban area under
observation (APAT, 2004); signiﬁcant contributions might also come from outside that area, due to
pollutants being transported in the atmosphere. Typically, the ambient concentrations these monitors record
are representative of areas Whose radius does not exceed 3 miles.

’° This is the monitoring station located in Juvara Street.

” http://www.arpalombardia.it/qaria/default.asp. ARPA Lombardia is the Regional Environmental
Protection Agency that manages the network of monitoring stations in Lombardy, which is the Italian
region in which Milan is located.

29

01/01/1980 through 02/13/1998, and as PMIO thereafter”. All concentrations are

measured in terms of rig/m3. Daily average values have then been calculated using hourly
data from Sam through 6pm”.

Figure 1.2 allows a graphical inspection of the PM data”. Summary statistics are
displayed in Table 1.1. TSP data have also been collected from monitors located in
fourteen other cities within Lombardy”.

Figure 1.3 provides some details about the contribution that different sources

make to the emissions of PM, NOx, and S02 in the Province of Milan. As far as

particulate matter (TSP) is concerned, road transport (i.e., trafﬁc) is responsible for the
largest amount of emissions, followed by residential and commercial heating, and
combustion in manufacturing industry (ARPA Lombardia, 2001; 2002; 2003a; 2003b;
2003c; 2006). Even greater appears to be the role of trafﬁc in the release of nitrogen
oxides, whereas energy production and heating are the main factors to blame for the

emissions of sulﬁrr dioxide.

 

‘2 The conversion rate between these two measures is not necessarily stable over time; as such, it has been
chosen not to employ the PM”) data in the empirical analysis, thus reducing the sample period under

investigation.

'3 When more than three hourly observations were missing, a missing daily value was assigned.

14 The ﬁgure shows that quite a few observations are missing. Nevertheless, this Should not be regarded as a
matter of concern. First, the missing observations can be safely considered as purely random in the context
of the current analysis. Second, the period covered by the air pollution data appears to be quite extended,
the result being that a few thousand daily observations turn out to be available for estimation purposes.

'5 These include all the cities, within Lombardy, for which TSP data are available starting from, at least, the
beginning of 1991.

30

l ICombustion in energy and l
transformation industries l

800/0 I _._ . _ ._ . _. _ _._ ._ __ . __ _. _ .__ .. .__ .. _

o ,— l INon-industrial combustion '
70/0' — -—~ .__ , -- ~- ~ ~ A; I plants(heatingsystems)
60% I" e I" , 1

E Combustion in
I manufacturing industry

13 Road transport

 

 

 

machinery l

 

 

 

 

 

 

I I Other mobile sources and
I B Production processes

1 998 2000 2001 2003

Panel (a). Total Suspended Particulate.

 

 

 

 

 

 

 

 

    

2000
Panel (c). Sulfur dioxide (802).

Figure 1.3 Main sources of air pollution emissions in the Province of Milan (percentage contribution). The
data are taken from ARPA Lombardia (2001; 2002; 2003a; 2003b; 2003c; 2006).

31

1.6.2. Equity Returns and Control Variables

Stock returns for the Italian market have been computed using daily closing
values of three MSE global indices, i.e., MIB Storico, Comit, and Datastream Italy-
market, from January 2, 1980 through May 19, 2006”. Figure 1.4 allows a graphical
examination of the equity data, whereas the summary statistics are reported in Table 1.2.
Similarly, the daily returns of the German and US. markets have been calculated using

the Datastream Germany-market index and the S&P500 index, respectively.

10% W —- W - ,, , ,
5%11‘ , _. _______________________________________ '

0%;

 
 
     

 

-5% - ' .- H , ______

l
l

40% ,,,,,,,,,,,,,,,,,,,,,,,,,
l

45% i __ - _ .. . . . , - ................. I

.20% IWW ..W.W_._.__- WWW-

 

o "t ,9 fo co o '1. qt- to to o
‘15 ‘b ‘b «o to ca 9
,9 ,9 .9 .9 9 NO? .9 ,9 .9 9 0°

Figure 1.4 Daily equity returns of the Italian stock market (Comit index).

To control for well-known calendar anomalies, a Monday dummy has been

created, taking the value of 1 on Mondays and 0 otherwise (e.g., Gibbons and Hess, 1981;

 

'6 Adopting the standard approach, daily stock returns have been calculated as the logarithmic difference
between any two consecutive closing values of a given MSE index on trading day t and t-l, respectively.
All data have been collected from Datastream; the Datastream codes for the indices mentioned above are
ITMHIST, MILANBC, and TOTMKIT, respectively.

32

KO et al., 1997). Also, a Tax dummy has been constructed, being assigned the value of 1
over the ﬁrst seven days of January (e.g., Branch, 1977; Dyl, 1977).

The other control variables employed in the econometric analysis come from
disparate sources”. Daily data about temperature and rain in Milan have been obtained
from ARPA Lombardia‘SThe Seasonal Affective Disorder (SAD) effect has been
calculated according to the methodology proposed by Kamstra et al. (2003)”. Following
Krivelyova and Robotti (2003), a variable that captures temporary disturbances of the
Earth’s magnetosphere has been obtained from the National Geophysical Data Center in
Boulder, Colorado”. As the authors suggest that the inﬂuence of geomagnetic storms on
stock returns may be a lagged one, a dummy variable Geostorm has been constructed,
taking the value of 1 on the three days following the geomagnetic index being above 6.
Finally, to control for the alleged behavioral effects of the lunar cycle (Yuan et al., 2006),
data about the lunar phases, obtained from NASA, have been employed to create a
FuIlMoon dummy variable taking the value of 1 up to three days before and after each
Full Moon date and 0 otherwise”. Analogously, a NewMoon dummy has been
constructed, being assigned the value of 1 up to three days before and after each New

Moon date.

 

'7 These are the behavioral variables that are typically included in this stream of literature.

'8 http://www.arpalombardia.it/meteo/meteo.asp. Daily temperature values, in Celsius degrees, are
computed as the mean of maximum and minimum daily temperatures. Rain is measured in mm. Both time
series begin on January 2, 1989.

'9 More speciﬁcally, the SAD effect is assumed to be captured by two variables: a Fall dummy variable,
taking the value of 1 from September 21 through December 20 of each year and 0 otherwise, and a SAD
variable that measures the (normalized) number of hours of night in Milan. The SAD variable takes the
value of 0 from March 21 through September 20.

2° httpz/lwww.ngdc.noaa.gov/stp/GEOMAG/kp_ap.shtml. Unlike the authors, here it has been chosen to use
the C9 index as a proxy for geomagnetic activity. This index gives a qualitative estimate of the overall level
of geomagnetic activity according to a scale that ranges from O (quiet) to 9 (highly disturbed). The
estimation has also been carried out employing the Ap index used by the authors; no difference has been
detected in the results.

2' http://sunearth. gsfc.nasa. gov/eelipse/phase/phaseZOO I gmthtml.

33

1.7. Possible Issues of Endogeneity

Before turning to the econometric analysis, it seems useful to explore some
potential sources of endogeneity. Since overall air pollution, at the country level, is
clearly connected with a country’s economic activity (factor 1) and also generates some
social costs (factor 2) through the extemalities it creates, if one wanted to accurately
measure the alleged behavioral effect of pollution on stock retums, then these two factors
should be controlled for. Sadly, as the present study employs daily data, no good proxies
seem to exist for this purpose. Nevertheless, several approaches can be adopted to tackle
this problem.

Standard economic theory suggests that, on a global scale, both economic activity
and the social costs associated with air pollution are likely to exhibit a connection with
stock returns. As far as the ﬁrst factor is concerned, it seems the case that large amounts
of pollutants are emitted into the air through the use of materials and fuel inputs in the
industry and service sectors. Ignoring for a moment the elements that govern the
dispersion of pollutants in the atmosphere (e.g., meteorological phenomena), it seems
safe to assert that the air pollution levels recorded in a given area are associated with
economic activity in that area. For instance, if one focused upon an entire country, e.g.
Italy, then unexpected changes in overall economic activity in Italy would be assumed to

both generate unexpected changes in overall air pollution concentrations and have an

34

impact on the Italian stock marketzz. In particular, one would expect to observe a
(probably positive) contemporaneous relationship between unexpected positive changes
in air pollution levels (meaning augmented economic activity) and stock returns. In such
a setting, failing to statistically control for economic activity would cause the estimator of
the (alleged) pollution effect to be biased.

Secondly, as anticipated, airborne contaminants are also known for having a
negative feedback onto economic activity via the social costs (extemalities) they
generate. These costs take the form of premature mortality (WHO, 2006), reduced
productivity, work loss, and “restricted activity days” due to the health effects of
pollution (Ostro, 1983; Hausman et al., 1984; Ostro and Rothschild, 1989; Zuidema and
Nentjes, 1997; Ostblom and Samakovlis, 2004), and damage to the agricultural sector
(Henderson, 1996; Spash, 1997; Murphy et al., 1999). Hansen and Selte (1997), for
instance, ﬁnd that ambient concentrations of PM have a signiﬁcant relationship with
daily sick-leaves in Oslo, Norway. At the country level, computations made for some
European nations have revealed that the overall costs generated by air pollution are
currently high, ranging between 0.1% and 1% of GNP (OECD, 1994). As a result,
standard economic theory suggests that unexpected increases in a country’s air pollution
levels could have a negative impact on stock returns through the unexpected social costs
they produce. Which effect dominates (whether factor I or 2) is mainly an empirical
question, and is likely to depend on the particular situation considered.

What really matters here is that air pollution, unlike some other environmental

stressors analyzed in the behavioral ﬁnance literature (e.g., Saunders, 1993; Kamstra et

 

22 Alternatively, one could conjecture that, since many aspects of a country’s economic activity cannot be
observed on a daily basis (e.g. industrial production, energy consumption), traders use air pollution daily
data to infer information about overall economic activity.

35

al., 2003; Hirshleifer and Shumaway, 2003), cannot be treated a priori as a purely
exogenous factor when explaining the behavior of stock returns. Greater care is needed
here to make sure that the results do not suffer from endogeneity. To address this issue, as
partly anticipated, the empirical strategy will involve (l) a scrupulous geographical
selection of the air pollution measures, (2) the use of a natural experiment, and (3) an

instrumental variables estimation.

36

1.8. Empirical Examination

1.8.1. Particulate Matter and Equity Returns: a Natural Experiment

AS previously mentioned, the Italian stock market provides a favorable setting for
the purpose of testing whether investment decisions can be inﬂuenced by the mood
changes triggered by some environmental stressor. This is because the MSE experienced
an institutional change at the beginning of the 1990’s, the shift being from a ﬂoor trading
system to a computerized and decentralized one. The transformation process was over by
April 15, 1994, when no more shares were traded through the open outcry system23 .

This institutional revolution allows for a “natural experiment” (e.g., Pardo and
Valor, 2003) to be conducted as follows: (1) one needs to consider that, until April 15,
1994, traders were physically present on the trading ﬂoor located in Milan and, as such,
their exposure to air pollution was accurately tracked by the monitoring station employed
in the current study. If the behavioral hypothesis constructed in the previous sections was
correct, then, during this period, air pollution concentrations Should be expected to be a
good agent mood proxy and, thus, exhibit a negative relationship with Italian stock
returns. On the other hand, after April 15, 1994, Since traders began to operate remotely

from many other places, characterized by dissimilar environmental conditions, air

 

23 More Speciﬁcally, the transition process began on November 25, 1991, when 5 companies' shares started
to be traded through the computerized system. Another 5 companies’ shares were moved on 01/16/1992, 25
on 05/18/1992, 4 on 04/19/1993, 41 on 07/16/1993, 73 on 12/16/1993, 26 on 01/17/1994, and 225 on
04/15/1994 (Pia, 1997).

37

pollution levels in Milan should not be any longer a good proxy for agent mood; as such,
no association between pollution and Italian equity returns should be expected during this
second era“. Also, (2) if the relationship between pollution levels and stock returns was
Spurious (i.e., caused by an omitted variable, such as economic activity or the social costs
generated by air pollution), then one would expect to ﬁnd the same statistical evidence in
both sub-periods”.

As a result, observing a negative relationship between air pollution and Italian
stock returns during the ﬁrst era and a breaking of such a link in correspondence of the
institutional change mentioned above would provide solid evidence in support of the

hypothesis under investigation.

1.8.2. Basic Model

A careful examination of Figure 1.4 and Table 1.2 shows that the distributions of
the equity returns under scrutiny are affected by excess kurtosis and volatility clustering.
As such, a GARCH speciﬁcation has been deemed as the most appropriate to conduct the
empirical analysis (Engle, 1982; Bollerslev, 1986). In particular, the following model has
been estimated maximizing the Gaussian density function and using robust standard

errors (Huber, 1967; White, 1980):

 

2’ In fact, such a restriction can be loosened a bit as, undoubtedly, after the system switch some traders kept
operating from structures situated in Milan. As a result, for the argument to work it would be sufﬁcient that
the air pollution effect observed during the second era, if present, was “weaker” than during the ﬁrst one.
25 Unless, of course, the linkage between the omitted variable and Italian stock returns broke down at the
end of the ﬁrst era. This issue will be discussed later.

38

P
PM
’t = 63+ Zﬂj'i—j Hit—k PMt—k + YSADSADr + Viral/Fa”: +
i=1
+ yFuuMoonFullMoon, + yNewMoonNewMoon, + (1.1)
+ 7MondayM0’7dayt + 7TaxTaxt + 8t

8 =20" z ~iid(0,1)
t It I

2 2 2
0'1 = 5 ‘1' Wt—1+¢0'I—1

where r, represents the daily return of a given MSE index between time H and time t,
PM,_k measures the daily ambient concentration of particulate matter (TSP) in Milan at

time t-k, and the remaining variables have the meaning previously Speciﬁed. A ﬂexible
number of lagged returns has been added to each regression, when statistically
Signiﬁcant, to purge the time series of equity returns from any intrinsic autocorrelations.
All the coefﬁcient estimates, for the case of k equal to I, are reported in Table 1.3.
Table 1.4, instead, collects only the estimated marginal effects of the air pollution proxy
for values of k up to 5. Focusing on the ﬁrst sub-period, 1980-1994, the MIB index
provides strong evidence of a negative relationship between particulate matter and stock
returns, for the coefﬁcients on the proxy are negative and statistically different from zero,
at least at the 10% level, from lag 0 up to 4. For example, Table 1.3 suggests that a 10
ug/m3 increase in the level of TSP is estimated to reduce the following day’s equity
returns by 1 basis point. The evidence is somewhat weaker for the Comit index, given

that the air pollution proxy is statistically signiﬁcant at lag 1 (p-value 0.04) and partially

39

Signiﬁcant at lag 0 and 2. When considering the Datastream index, marginal Signiﬁcance
is reached only at lag l (p-value 0.13). It is worth to emphasize that all the coefﬁcients
exhibit the (negative) expected sign.

Based on the medical and psychological literature cited in the previous sections, it
is speculated that the effects of a prolonged exposure to high levels of air pollution might,

to some extent, add to each other. To test this hypothesis, 3 variable Pit/[,Mea" is

constructed and assigned a value equal to the mean of the PM concentrations observed
during the previous ﬁve days (i.e., from H through t-5). Model [1.1] has been estimated
again using such a pollution proxy and the relevant coefﬁcients are reported in the last
row of Table 1.4. The picture that emerges is highly supportive of the hypothesis under
investigation, especially when considering the MIB and Comit indices (the Sign of the
coefﬁcient on the pollution variable is negative, and standard statistical signiﬁcance is
achieved). Marginal evidence is also reached in the case of the third index (p-value 0.1 1).

Moving to the second sub-period (1994-1998), as conjectured, no evidence of a
signiﬁcant impact of particulate matter on stock returns is found at any lag (see Table
1.4). The coefficients generally Show a negative Sign, yet statistical signiﬁcance is never
even approached. As such, it can be concluded that these preliminary estimates are
consistent with the behavioral hypothesis under study, suggesting that further scrutiny
Should be granted.

Some attention must also be paid to the remaining explanatory variables in model
[1.1]. As Table 1.3 Shows, the estimates from both sub-periods corroborate the SAD
effect advanced by Kamstra et al. (2003), for the coefﬁcient on the SAD variable is

positive and the one on the Fall dummy is negative, as suggested by the authors. Solid

40

statistical signiﬁcance is reached across the three market indices. Same reasoning applies
to the Monday effect, which appears to be even stronger in the second sub-sample,
whereas the Tax dummy does not turn out to have any remarkable explicative power.
More difﬁcult to interpret is the role played by the lunar cycle, Since the estimates seem
to either contradict the ﬁndings of Yuan et al. (2006) and Dichev and James (2003) (i.e.,
the coefﬁcient on the NewMoon dummy is negative) or provide no evidence at all. It may
be worth to mention that Lucey and Dowling (2005b) found a similar “paradoxical”

result when studying Italian data.

1.8.3. Controlling for Meteorological Factors

Some recent studies in the ﬁeld of behavioral ﬁnance have Shown that certain
environmental variables, supposedly acting as proxies for investor mood, display an
association with stock returns. These include temperature (Keef and Roush, 2003; Cao
and Wei, 2005; Chang et al., 2006), rain (Lucey and Dowling, 2005b), and geomagnetic
storms (Krivelyova and Robotti, 2003). At the same time, a number of connections
between weather and air pollution concentrations have been established in the
meteorological literature (e.g., Campbell and Gipps, 1975; Mossetti et al., 2005). To
avoid the risk of obtaining biased estimators, it is therefore essential to add more controls

to model [1.1], which becomes:

41

P
PM M
r: = 0+ Zﬂj'i—j +/1MeanPMI ea" + YSADSADI + 7Fa11F011I+
j=1
+ yFuUMoonFullMoon, + yNewMoonNewMoon, + yMondayMonday, + (1.2)

Rain
'l' yTaxTaxt + llTempTempt + 7RainDt + 7 GeostormGeos tor mt + 8t

2 2 2
0'1 =5 +78r—1+¢0'r—1

where D/W” is a dummy that takes the value of 1 if a positive amount of rain has fallen
in Milan at time t, Temp, is the average temperature in Milan on day t, and Geostorm, is a

dummy that captures disturbances of the Earth’s magnetosphere. The remaining variables
are self-explanatory”.

The coefﬁcients obtained estimating model [1.2] are collected in Table 1.5. The
marginal effects of the air pollution proxy (average PM concentration between day t-l
and day t-5) are again negative and highly statistically signiﬁcant across the three equity
indices when the ﬁrst sub-period is analyzed”. In absolute value, the size of the pollution
effect turns out to be even greater than in model [1.1]: for example, focusing on the MIB
index, a 10 ug/m3 increase in the average PM level over the previous ﬁve days is

estimated to lower the following day’s Italian stock returns by 3 basis points. Such an

 

2" It should be noted here that, due to the time span covered by the available data for these weather series,
their inclusion into the regression equation generates a shrinkage of the sample data.
27 Model [1.2] has also been estimated using PM,-kas the pollution proxy. Across the three indices, and for

k ranging from 0 through 5, the coefﬁcient on such a variable has always turned out to be negative and
highly statistically signiﬁcant. As such, for brevity, only the results referring to PM,Mea" are reported in
Table 1.5.

42

impact appears to be economically signiﬁcant, as the mean and standard deviation of the

Pit/1,5460" variable, during the ﬁrst half of the 1990’s, were approximately 85 ug/m3 and

40 ug/m3, respectively.

As expected, the effect cannot be detected any longer after the transformation of
the Italian stock exchange in 1994. Qualitatively, the other explanatory variables exhibit
patterns similar to the ones reported in Table 1.3. No support is found in favor of the
conjecture that wants rain and geomagnetic activity to be good proxies for investor mood.
Different is the case of the temperature variable, for this factor exhibits a seasonal
behavior that tracks very closely the length of the day (i.e., the SAD variable). In many
occurrences, these last two controls turn out to be jointly statistically signiﬁcant yet
individually insigniﬁcant. It appears very challenging to empirically identify the

contribution that is individually attributable to each of them, if any.

1.8. 4. Controlling for International Economic News

Up to this point, no explicit economic/ﬁnancial factor has been employed in the
analysis. It is natural to wonder whether the results discussed in the previous sections are
robust to the inclusion of some variables that are able to capture the stream of economic
news ﬂowing into the ﬁnancial markets. Given the frequency of the data used here,
adopting a direct approach seems quite impractical. Instead, the strategy employed to
tackle this issue will consist in inserting into model [1.2] the returns of two international
stock market indices, i.e. the S&P500 and a global index of the German market. The goal

is to use these indices to indirectly capture (at least some of) the unexpected economic

43

shocks that occurred daily over the period covered by the present investigation. In
particular, the German index is expected to incorporate more closely European-Speciﬁc
Shocks, whereas the S&P500 is employed to capture global shocks”. Also, the inclusion
of these two international indices should purge the time series of Italian stock returns
from those seasonal patterns that are common to other international markets. The

following model has been ﬁtted by quasi-maximum likelihood:

P
Pl” A1
rt = 0+ 2.31%; +ﬂMeanPMr W + YSADSADI + rFallFallt +
l=l
+ 7FullMoonFullM00nt + 7NewMoonNEWM00nt + l/MondayMondaYI + “-3)

Rain
+ yTaxTaxt + 7TempTempt + 7RainDt + yGeostormGeOStormt +

+ 76ermanyGermanyt + 7USAUSAt—l + 8t
E, = 2,0', 2, ~ iid (0,1)

2 2 2
0'1 = 5 ‘1' 251—] + (901-1

where Germany, represents the returns of the German market and USA ,_ l alludes to the
returns of the US. market. The latter is included with a lag because the American

markets terminate their daily operations at least six hours after trading in Milan has

already come to an end, the implication being that the S&P500 index on day t

 

28 Obviously, Italian-speciﬁc news plays an important role in Shaping Italian stock returns. However, the
potential omitted-variable issue attributable to the occurrence of country-speciﬁc shocks is addressed in
some other sections of this essay, especially through the natural experiment examined in section 1.8.1 and
the instrumental variables estimation that will be discussed in section 1.8.7.

44

incorporates some economic news that is absorbed by the Italian equity indices on the
following trading day”.

The estimates are collected in Table 1.6, and clearly Show that a strong positive
correlation exists between the returns of the Italian stock market and the US. and
German returns. Yet, the evidence regarding the air pollution effect is left untouched. The
coefficients on the regressor under observation are once again negative across the three
indices, and the null hypothesis of no relationship with Italian equity returns can be
rejected at very high conﬁdence levels.

As it Should be expected, some control variables are now less statistically
signiﬁcant than they were in model [1.2], their temporal pattern being the same (or
similar) across the three stock markets whose returns are included in the regression
equation (e. g., Monday effect, lunar cycle, SAD effect).

Model [1.3] has also been estimated, in turn, by OLS (using Newey-West adjusted
standard errors) and adopting a LOGIT speciﬁcation”. In both cases the results are
qualitatively similar to the ones described above, and the evidence turns out to be highly
consistent with the behavioral hypothesis under scrutiny (i.e., the coefﬁcient on the air
pollution proxy is always negative and statistically Signiﬁcant when the ﬁrst sub-sample
is analyzed, whereas it is not distinguishable from zero when it comes to the second sub-
sample). As such, overall, the ﬁndings appear to be robust to the addition of alternative

sets of controls and to alternative estimation frameworks.

 

29 This is, of course, a basic speciﬁcation, as the goal here is not to fully describe the Spillover effects
among these three markets, which may be much more complex.
30 The estimates are not reported here for brevity; they are available from the author.

45

1.8.5. Alternative Air Pollution Proxies

As emphasized during the literature review, medical research has been hesitant in
attributing speciﬁc health effects to individual pollutants, for most epidemiological
studies have to deal with complex mixtures of environmental toxins whose speciﬁc (and
synergistic) effects are not yet completely understood. Therefore, it appears interesting to
test whether the air pollution effect, detected in the previous sections using PM as an
indicator, is robust to the use of alternative pollution proxies.

Second, daily data about ambient concentrations of nitrogen oxides and sulfur
dioxide in Milan are available for a longer time span than PM, their time series extending
up to 2006.

Third, employing a set of different proxies may help shed light on the potential
endogeneity generated by the existence of a link between economic activity and

pollution. As shown in section 1.6, the direct emissions of PM, NOX, and 802 produced
in the Province of Milan can be credited to rather dissimilar sets of sources. While the
major contributor to the emissions of TSP and NOx is road transport, residential heating
systems and power plants take the lion’s share when it comes to S02. Showing that air
pollution in Milan, independently of its sources, is systematically connected with Italian
stock returns is an important step towards proving that the results illustrated in the
previous sections are not spurious, and that the behavioral hypothesis tested in this essay

is indeed supported by the facts.

46

Following this line of reasoning, model [1.3] has been ﬁtted replacing PM with

either NOx or S02. The relevant coefﬁcients are contained in Table 1.7 and 1.831.

Focusing on the ﬁrst sub-sample (1989-1994), the lag structure exhibited by the air

pollution effect when NOx is used as a proxy is somewhat similar to the one discussed for

PM, though the statistical Signiﬁcance of the individual lags is more limited. The size of
the marginal effect is estimated to reach its maximum around lag 3 and 4, which is

reasonable if one considers that NOx also contributes to the formation of secondary PM
in the atmosphere, a process that may take some days. A variable NOW/"ea" has also

been constructed to test whether a protracted exposure to high levels of air pollution may

generate effects that add up to each other. It is assigned a value equal to the average NOx

ambient concentration between day t-l and day t—4. As shown in Table 1.7, the estimated
marginal effect of this variable conforms to the evidence collected so far, though in the
case of the MIB index the statistical relevance is only marginal. Quantitatively, a one

standard deviation increase in the mean level of NOx over the previous four days is

estimated to reduce the following day’s stock returns by approximately 8 basis points. As
such, the magnitude of this effect is comparable to what has been found using PM as
proxy. Once again, when the second sub-period (1994-2006) is examined, the air
pollution effect seems to disappear.

The picture that emerges when sulfur dioxide is employed as a proxy turns out to

be similar. When individual lags are used, the statistical evidence is quite solid for the

 

3' Only the marginal effects of the air pollution proxies are reported in these two tables. The patterns
characterizing the control variables are virtually the same as the ones shown in Table 1.6.

47

Datastream and MIB indices, in the latter case extending up to the third lag”. Based on
the same rationale stated above, a variable that measures the average ambient

concentration of S02 over the previous few days (three in this case) is constructed, and

its marginal effect on equity returns is assessed. The sign happens to be negative, as
expected, and the null hypothesis of no effect can be rejected, at least at the 10%
conﬁdence level, across the three market indices. Also, the size of the estimated pollution
effect is equivalent to the outcomes obtained using the other two proxies, for a one

standard deviation increase in the average 502 level, computed over the previous three

days, is expected to decrease stock returns by approximately 10 basis points.

1.8. 6. Further Robustness Checks

A possible critique to the natural experiment proposed here is that the institutional
change in the trading system of the Milan Stock Exchange might not be the only relevant
event that took place during the period analyzed. In other words, one might conjecture
that the air pollution effect seems to disappear after April 14, 1994 because the
correlation between stock returns and air pollution experienced an alteration for reasons
unrelated to the institutional change mentioned above. Examining Table 1.1 and Figure
1.2, it is indeed possible to notice that air pollution levels generally decreased between
the end of the 1980s and the end of the 1990s. Such an abatement seems to be attributable

to several factors: (1) the introduction of European emission standards that set

 

32 A critical point Should be highlight here, though. When analyzing the second sub-period ( 1 994-2006), the
coefﬁcient on the S02 variable turns out to be marginally statistically signiﬁcant at lag 2. Given the

considerable evidence previously discussed, it seems the case that this single piece can hardly substantiate a
decisive argument to refute the hypothesis under scrutiny.

48

progressively lower limits for exhaust emissions of new vehicles sold in EU. member
states, (2) technological improvements, (3) the renovation of residential and commercial
heating systems, (4) the moving of some industrial ﬁrms”.

One reasonable strategy to tackle the issues created by the (potential) existence of
unobserved (possibly relevant) structural breaks consists in reducing the length of the two
sub-periods employed in the natural experiment. As the two sub-samples shrink, the
probability that the disappearance of the air pollution effect after April 14, 1994 has been
caused by factors other than the transformation of the Milan Stock Exchange should
diminish accordingly. Based on this intuition, model [1.3] has been re-estimated using
alternative samples, and the results are reported in Table 1.9. Though the statistical
signiﬁcance‘of the effect is somewhat reduced when the ﬁrst sub-period shrinks, the
coefﬁcients remain negative and the null hypothesis of no relationship between air
pollution in Milan and Italian stock returns can always be rejected at standard conﬁdence
levels (at least in the case of the Comit and MIB indices). Similarly, reducing the second
sub-sample does not have any notable impact on the picture depicted thus far as, once
again, no evidence of a connection between air pollution and equity returns can be found.
Given these empirical patterns, it seems unlikely that some structural change, other than
the trading system switch at the MSE, is responsible for the observed breaking of the link
between the two variables of interest.

One might still wonder, though, whether the inability to detect an air pollution
effect after April 14, 1994 is due the fact that air pollution concentrations have been

lower, on average, during the second sub-period. In principle, lower levels of

 

3’3 http://Www.arpalombardia.it/qaria/doc_EvoluzioneAnni.asp. These factors are mentioned by ARPA
Lombardia on its web site, which was accessed on July 10, 2007.

49

environmental toxins may imply that the physical/psychological discomfort experienced
by people is not severe enough to affect their mood and their decision-making process. If
this was the case, then no trace evidence of an air pollution effect Should be found in the
time series of stock returns in the second era. Contrary to this argument, however, is the
medical literature cited in section 1.4.1.3, according to which no Speciﬁc threshold has
been identiﬁed below which the adverse inﬂuence of PM on the human body ceases to
work. In order to give an empirical answer to the problem, model [1.3] has also been re-

estirnated using the ﬁrst difference of the pollution proxy, as follows

P
r: = 0+ Zﬂer—j +ﬂ APMiMea" + YSADSADI + 7Fa/lFallt +
1:1
+ yFullMoonFullMoon, + yNewMoonNewMoon, + yMondayMonday, + (1.4)

Rain
+ 7TaxTaxt + 7TempTempt + 7RainDt

+ 7GermanyGermanyt + 7USAUSAt—l + 8!

+ yGeosmmGeostorm, +

2 2 2
0'1 = 5 + lit—1 + (1501—1

where APM,M€0” represents the change in the variable PMMW" between time H and

time t. The results, not reported here for brevity, are qualitatively very close to the ones
discussed for model [1.3]. AS expected, the coefﬁcient on the pollution proxy is negative

and statistically Signiﬁcant (at least at the 5% level) across the three market indices in the

50

ﬁrst sub-period (1989-1994), whereas no evidence is produced when the second sub-
period (1994-1998) is analyzed. These estimates, therefore, suggest that the air pollution
effect is at work even at low concentrations (i.e., concentration changes matter
independently of the air pollution level outstanding), which refutes the conjecture
according to which its disappearance, in the second era, would be attributable to air
quality improvements.

Yet another route is worth of exploration. In principle, not only stock returns but
also their variance might be inﬂuenced by the environmental stressor under scrutiny here.
To examine this potential further link, the pollution proxy has been inserted into the
conditional variance equation of model [1.3], giving origin to the following GARCH

model:

P
PM M.
’i = a+ Zﬂer—j +luMeanPMt “m '1' 7SADSADI + Viral/Fa”: ‘1'
Fl
+ yFuHMoonFullMoon, + yNewMoonNewMoon, + yMondayMonday, + (1.5)
BIRCH.”

+ 7TaxTaxt + 7TempTempt + 7 Rain + YGeostormGeOStomt +

+ l’GermanyGermanyt 'l' 7USAUSAt—l + 5t

2
0': = 5 +77PMIMean '1' 7572—1 '1' Wrz—l

51

where all the variables have the same meaning previously illustrated. When the estimates
for the MIB and Comit indices are evaluated, the null hypothesis of no connection
between air pollution and the variance of equity returns cannot be rejected at standard
conﬁdence levels (the p-values equal 0.39 and 0.17, respectively). Moving to the
Datastream index, a negative and statistically signiﬁcant marginal effect is Observed (p-
value 0.01), implying that increases in pollution concentrations appear to be associated
with reduced variability of returns. Though this individual ﬁnding, per se, seems to have
no obvious interpretation, taken together these outcomes suggest that the negative impact
that air pollution is estimated to have on stock returns cannot be attributed to the fact that

the former generates an increase in objective risk (i.e., in the variance of retums).

1.8. 7. Instrumental Variables Estimation

Up to this point, several methods/justiﬁcations have been used in an attempt to
Show that the air pollution proxies employed in the empirical analysis are not
endogenous: (1) the daily concentration data are taken from a sole monitor located in the
city of Milan, and are merely representative of human exposure around the Milan Stock
Exchange (few squared miles), which means they are very unlikely to represent a global
measure of air pollution in the Italian peninsula”; (2) a natural experiment, based on an
institutional change experienced by the Italian stock market, has shown that the air

pollution effect can only be detected over the period when traders would physically

 

3" Table 1.10 shows the pair-wise correlation between PM daily concentrations in Milan and in a set of
neighboring cities. The data demonstrate that air pollution patterns can be very dissimilar even when
considering any two cities in close proximity (few miles away from each other). This further proves how
the concentrations recorded by the monitor located in Milan are representative of a very small area and
cannot even be thought of as representing a provincial measure.

52

gather on the ﬂoor of the MSE; (3) the main source of PM in Milan is road trafﬁc, which
is unlikely to have an unambiguous and systematic link with overall Italian economic
activity; (4) the estimated air pollution effect exhibits a complex lag structure, implying
that current modiﬁcations in the ambient concentration of a given air pollutant will also
have an impact on stock returns in the near future; on the other hand, if both daily air
pollution levels and stock returns were driven by concurrent economic activity (i.e.,
pollution is endogenous), then only a contemporaneous relationship between pollution
and returns should be detected; (5) assuming for a moment that air pollution was indeed
an endogenous variable, standard ﬁnance theory suggests that only unexpected changes
in economic activity and, in turn, unexpected changes in daily air pollution levels, should
be correlated with equity returns; in other words, only the unexpected component of air
pollution should be useful in determining the direction of stock returns. Nevertheless, this
line of reasoning appears to be refuted by the evidence discussed so far (i.e., it is the
actual air pollution level in Milan, not its unexpected component, that exhibits a
systematic relationship with Italian equity returns).

This said, identifying a link of causality is always challenging, and all the feasible
techniques should be explored. The results proposed here would surely become more
solid if good instruments for air pollution were employed. Luckily, a careful examination
of the forces at work in this setting reveals that the instruments necessary to overcome the
potential endogeneity issue can be derived from a “natural” experiment. Indeed, this
appears to be one of the cases in which it is possible to exploit the forces through which
Nature has generated an environment somewhat similar to a randomized experiment. As

it is well understood, ambient air pollution is deeply affected by meteorological

53

conditions (e.g., Campbell and Gipps, 1975; Mossetti et al., 2005). AS such, a subset of
the latter can turn out to be helpful when employed as instruments. In particular, our
instruments will include: (1) wind speed, (2) atmospheric pressure, and (3) rainfall, all
measured in the city of Milan.

The intuition behind this identiﬁcation strategy is straightforward. While high
wind Speed, low pressure, and heavy rain in Milan all contribute to reduce air pollution
concentrations in that city (Campbell and Gipps, 1975; Mossetti et al., 2005), one can
hardly maintain that they also directly affect overall Italian economic activity. Similarly,
there seems to be no theoretical justiﬁcation for believing that lagged values of these
meteorological variables should have any systematic relationship with Italian stock
returns”. For these reasons, we believe that it is realistic to take these variables as

exogenous. The ﬁrst-stage regression equation is the following

P 3 3
PM, = a-l- Zﬂj’t—j + zlkWind,_k +Zﬂ'kPressure,_k +
1:1 k=2 k=2 (1 ' 6)
3
+ 2 Wk Rain,_ k + Controls + 8,
k=2

where Wind,,k measures wind Speed in Milan on day t-k, Pressure,,k captures the

atmospheric pressure in Milan on day t-k, and the remaining variables are self-

explanatory. The ﬁrst-stage estimates conﬁrm that these instruments have high

 

35 It has been chosen to use lagged values because, in principle, these environmental factors might work as
mood proxies and, thus, have a contemporaneous relationship with equity retums. However, the following
should be kept in mind: (I) no study has documented a link between atmospheric pressure and stock
returns; (2) the estimates produced by ﬁtting model [3] have shown that rainfall has no effect on returns;
(3) model [3] has also been estimated including contemporaneous wind speed as a regressor, yet no
marginal effect on equity returns has been found.

54

explanatory power for current PM, as across the three stock market indices it is possible
to reject the null hypothesis that their six coefﬁcients are jointly equal to zero at
conﬁdence levels far below 1%. As such, the weak instruments problem illustrated by
Bound et al. (1995) does not seem to apply here”.

Table 1.11 collects the second-stage estimates, which, once again, tell a consistent
story and conﬁrm what has been documented in the previous sections. When the ﬁrst
sub-sample is used, all three market indices display an air pollution effect that is both
statistically signiﬁcant and economically relevant. In particular, a one standard deviation
increase in the current PM ambient concentration is estimated to reduce stock returns by
approximately 0.2%. No statistical signiﬁcance is achieved, instead, when the second
sub-period is analyzed.

Taken as a whole, we believe that our empirical methodology provides robust
evidence of a causal effect of air pollution on Stock returns, which is consistent with the
hypothesis according to which some psychologically relevant (yet economically neutral)
factors play a role in the process that generates investment decisions. More Speciﬁcally,
one can speculate that this empirical regularity appears to be caused by what could be
deﬁned as a “mood effect”. Also, in line with the ﬁndings of Goetzmann and Zhu (2005),
the natural experiment described in section 1.8.1 effectively suggests that such a mood
effect mainly passes through the traders physically located in the city that hosts the Stock

Exchange.

 

36 The results are robust to the use of alternative lag structures.

55

1.9. Air Pollutants, Mood, and Efﬁcient Market Theory

The evidence produced by the present empirical analysis appears to be intriguing
on two grounds. First, it represents a conundrum in light of the paradigm that dominates
economic analysis, as it suggests that some forces, other than economic ones, contribute
to shape equilibrium prices in the ﬁnancial markets. Second, it poses a challenge to the
Efﬁcient Market Theory, as it insinuates that some publicly available past information
(i.e., air pollution levels) is helpﬁIl for predicting future stock returns.

Though the latter ﬁnding is appealing per se, in order to determine whether this
truly constitutes a “market anomaly” one needs to demonstrate that it is possible to
construct some proﬁtable trading strategy using air pollution data. Several routes could
be explored here. The easiest approach that comes to mind consists in creating an air
pollution index (API) that generates buying/selling signals when a given threshold
concentration is crossed. To keep things simple, the focus here will be on selling Signals,
though this framework could be straightforwardly extended to purchases as well. In

particular, an indicator is constructed according to the following speciﬁcation:

F .
1, if P.M,Me""2 180 ttg/m3 or Nofffa" 2 650 tlg/m3 or 502%“ 2 180 ttg/m3

AP], =ﬁ

 

KO’ otherwrse

56

where the variables have the same meaning previously illustrated, and the threshold
values have been chosen so that they are approximately equal to the sum of the mean and
standard deviation of the corresponding variables”. When the API index assumes value
1, it obviously indicates that air quality has been remarkably low over the previous few
days (high-pollution state), so that, ceteris paribus, a poor performance of the Italian
stock market should be expected”. In other words, a selling signal is generated. The
portfolio could then be bought back when the API index takes on again value 0 (low-
pollution state).

To verify the validity of this strategy, the following GARCH model has been

ﬁtted using data from the ﬁrst sub-sample (1989-1994):

P
r: = 0! + 2.5 jrt— j ”rt/AHA”: + 7sADSA-Dr + 7FallFallt +
j=1
+ 7FullMoonFullM00nt + 7NewM00nNeWM00nt + 7MondayMondayt + (17)

Rain
+ 7TaxTa-xt + 7 Temp Temp, + 7RainDt + yGeostormGeOStormt +

+ 70ermanyGermanyt + 7USAUSAt-I + 8t

2 2 2
0t = 5 + 752-] + (Mr—1

 

37 Mean and standard deviation have been computed using the relevant data from January 2, 1989 to April
14, 1994. It must be noticed that the results, discussed below, are robust to the choice of the threshold
values. Based on random experimentation, it can be safely said that the range within which each of the
three pollution proxies generates valuable signals is relatively large, so that a huge set of proﬁtable
combinations can be built.

38 It Should be noticed that the value that the API index will take on day t is already known on day t-/, as
the three pollution proxies on which the index is based are constructed using data from time t-I through t-5.

57

The estimates, reported in Table 1.12, indeed reveal that poor air quality, as
captured by the API index, is expected to signiﬁcantly reduce stock returns”. An
inspection of the three market indices suggests that, on average, the marginal effect on
returns equals -0.25%. As such, even if transaction costs are taken into account, the
empirical regularities exploited by this strategy seem to be strong enough to grant
(institutional) investors an excess return over a simple buy-and-hold strategy“).

In other words, given the existence of unexploited proﬁtable trading Strategies
based on air pollution data, the air pollution effect appears to qualify as a market
anomaly. Future research endeavors may be directed toward testing whether this violation
of the efﬁcient market theory can be detected in other stock markets that, nowadays, still

employ open outcry trading systems.

 

39 Regression equation [1.7] has also been estimated using a Logit speciﬁcation. The outcomes, reported in
Table 1.8, suggest that the probability that stock returns turn out to be positive is reduced by approximately
10% when the API index takes value 1. For comparison purposes, it can be said that such a value is greater
than the Monday effect estimated through the same framework.

40 For a discussion about the magnitude of transaction costs, see Krivclyova and Robotti (2003).

58

1.10. Concluding Remarks

This essay provides a contribution to the debate on behavioral fmance. On the one
hand, advocates of traditional asset pricing theories maintain that investors are able to
isolate the relevant economic variables, process them in a fully rational way, and
incorporate them into the price of stocks. On the other hand, detractors argue that
investment decisions, just like many other types of human decisions, are inﬂuenced by a
set of biases that are inherently part of the way people’s mental processes develop. More
speciﬁcally, psychologists have found evidence that the emotional state (i.e., mood)
experienced at the time of making a decision is able to affect the decision being made
even if the source Of the former is unrelated to the latter (so-called mood misattribution).
Following this line of reasoning, behavioral ﬁnance researchers have started to analyze
whether collective mood swings, triggered for example by some environmental factors,
can alter investment decisions and generate some detectible pattems in the series of
equity returns.

The present investigation has focused on air pollution and on its role as an
environmental stressor. Based on the evidence produced in the medical and psychological
ﬁelds, it has been conjectured that daily increases in the ambient concentration of air
pollution are able to deteriorate agents’ mood and induce them to decrease their demand
for risky stocks, ultimately lowering equity returns, ceteris paribus. Such a hypothesis

has been tested empirically using data from the Milan Stock Exchange, and exploiting a

59

natural experiment allowed by the occurrence of an institutional change in the way
trading was conducted in the Italian stock market. The evidence appears to be consistent
with the hypothesis under investigation, and supports the view that some psychologically-
relevant yet economically-neutral factors indeed play a role in shaping investment
decisions.

The ﬁndings also constitute a puzzle when interpreted in light of the Efﬁcient
Market Theory, as public past data about air pollution seem to represent valuable

information for predicting ﬁIture stock returns.

60

Table 1.1 Summary Statistics - Air Pollution and Environmental Variables

Tablel displays a number of summary statistics that describe the sample. For each variable, summary
statistics are reported for two sub-samples, the ﬁrst one (upper row) covering the period when trading at the
Milan Stock Exchange was centralized, and the second one (lower row) referring to the era in which
trading was computerized and decentralized. The variable described as C9 Index measures the intensity of
disturbances of the Earth’s magnetosphere (i.e. geomagnetic Storms), and ranges from 0 (quiet) to 9 (highly
disturbed). All the other variables have been measured daily by a monitor located in the city of Milan, Italy.

 

 

Variable Period Mean (81:21:35: Min Max Skcwness Kurtosis
PM 01/02/(129331036il/1994 122.25 80.55 6.28 679.86 1.77 7.69
3 -

(Pg/m ) WIS/182‘; 28:13/1998 56.32 20.21 5.13 148.77 .70 3.80
NOX 01/02/(‘2‘9fgojéslf’ '994 342.48 309.00 19.07 2592 2.45 10.62
(118/m3) 04/15/(1299gﬁgojégf/2006 204.31 166.31 28.57 1501 2.48 11.42
502 0“ 022131825032)“ 1994 141.69 187.44 0 1311 2.47 10.31
(Pg/m3) 04/‘5/(1gggéojéslf/2OO6 16.94 18.08 .44 163.36 2.10 9.15

0“ 02/ (11933-03651? 1994 14.07 7.78 -4.60 29.80 .07 1.89
Temp (C°)

04/15/(1393300363/2006 14.36 7.83 -4.80 32.00 .03 1.88

01/02/(1193395'03tjslf’1994 2.23 7.97 0 122.20 7.73 86.91
Rain (mm) 04/15/1994 05/19/2006

(3037' obs) 2.55 8.23 0 212 8.49 155.15

01/02/ (1 39:260th s1)4/1994 3.25 2.“ 0 9 24 201

C9 Index 04/15/1994-05/19/2006
(3059 obs) 2.57 2.12 0 9 .58 2.29

01/02/ 133223“ 1994 1.44 .57 .39 4.25 1.10 498
Wind (m/s)

04’ 1 5/ (13:33:41? 2006 1.66 .59 .20 6.09 1.61 8.42
Pressure OHM/(11911379703651? ‘994 1005.37 7.89 977.21 1027.28 -.07 3.31
hp -
( a) 04’ 15/(12993405651)9/ 2006 1001.55 7.35 969.58 1023.60 -.13 3.24

 

 

61

Table 1.2 Summary Statistics - Stock Returns
Table 11 displays a number of summary statistics that describe the time series of stock returns used in
the analysis. For each variable, summary statistics are reported for two sub-samples, the ﬁrst one (upper
row) covering the period when trading at the Milan Stock Exchange was centralized, and the second
one (lower row) referring to the era in which trading was computerized and decentralized. Comit,
Datastream Italy-market, and MIB are global indices of the Italian stock market. Datastream Germany-
market is a global index of the German stock market, and S&P500 is a global index of the US. stock
market. Equity retums are expressed in percentage points.

 

 

Variable Period Mean Standard Min Max Skewness Kurtosis
deviation
01’02/1980'04/14/1994 .064 1.41 -14.85 8.30 -.84 12.95
com“ index 04/1 5713959230053192006
(30590138) .026 1.19 -8.47 6.22 -.53 6.54
Datastream 01/02/1980'04/14/1994 .072 1.42 -9.84 9.18 -.37 9.09
Italy-market (3607 obs)
. 04/15/1994-05/19/2006
Index (3073 obs) .024 1.27 —7.79 6.90 -.18 5.94
01/03/1985‘04/14/1994 .056 1.31 -1031 6.88 -.62 8.96
MIB index (2285 obs)
04/1 5/1994-05/19/2006
(30560bs) .027 1.19 -8.48 6.22 -54 6.55
01/02/1980-04/14/1994
3&P500 (34700bs) .039 1.02 -22.89 8.71 -347 82.78
index 04/15/1994-05/19/2006
(29780bs) .036 1.08 -711 5.57 -.11 6.58
Datastream 01/02/‘980'04’14/1994 .033 0.96 -12—14 5.91 -1.12 17.86
Germany- (3550 obs)
market index 04’15/1994'05/19’2006 .023 1.18 -7.21 5.48 -.37 5.98
(3059 obs)

 

 

62

Table 1.3 Air Pollution and Stock Return - A Natural Experiment (Basic Model)

This table displays the results of estimating model [1.1] by quasi-maximum likelihood and using robust
standard errors. The left-hand side of the table contains the coefﬁcient estimates produced using the ﬁrst
sub-sample of data, covering the period when trading at the Milan Stock Exchange was centralized. The
right-hand side shows the estimates generated using the second sub-sample, which refers to the era in
which trading was computerized and decentralized. The air pollution proxy employed is the ambient
concentration of Particulate Matter (PM) at lag 1. Stock returns are expressed in percentage points. P-
values are in parenthesis below the corresponding coefﬁcients. One, two, and three asterisks denote
statistical Signiﬁcance at the ten, ﬁve, and one percent level, respectively.

 

Centralized Market Decentralized Market

 

 

Comit Datastream MIB Comit Datastream Ml B
01/02/1980— 01/02/1980- 01/03/1985- 04/15/1994- 04/15/1994- 04/15/1994—
04/14/1994 04/14/1994 (24/14/1994 02/13/1998 02/13/1998 2/13/1998
.136*" .132" 179*" .054 .020 .043
a (0.003) (0.011) (0.001) (0.630) (0.859) (0.695)
250*" 284*" 286*“ 134*" 145*"
P] (0.000) (0.000) (0.000) (0.002) (0.001)
ﬂ -.115"'" -.105** -.082"‘
2 (0.000) (0.032) (0.081)
,6 055* .067" 075*
3 (0.078) (0.043) (0.089)
pM -.00063** -.00052 -.00099*** -.00049 .00037 -.00034
ﬂI—I (0.027) (0.129) (0.008) (0.807) (0.853) (0.863)
.068m .064m .057* .087** .095** .088**
KW? (0.003) (0.007) (0.051) (0.019) (0.022) (0.022)
-.122* -.176"”'“" 4151* 4190* -.236"‘ -.189“
””0” (0.055) (0.005) (0.064) (0.077) (0.056) (0.082)
.009 -.008 -.039 .096 .1 10 .1 17
751114400" (0.870) (0.889) (0.547) (0.268) (0.249) (0.193)
-109" 4127" -.104* .013 .013 .016
77VewM00" (0.036) (0.017) (0.090) (0.893) (0.902) (0.857)
4131" -.091* -.113* -.218** -249" -.238"”"
7M0nday (0.016) (0.097) (0.085) (0.021) (0.017) (0.013)
-094 -.113 -.027 -.101 -.028 -.092
7704 (0.535) (0.517) (0.901) (0.709) (0.943) (0.736)
L” (L) 4399.1 4516.2 -2746.793 4421.9 -1463.5 4414.2
Obs 2730 2772 I743 920 923 917
Gaps 295 296 223 29 29 29

 

63

Table 1.4 Air Pollution and Stock Return — Basic Model (Alternative Lag Structures)

This table displays the results of estimating model [1.1] using alternative lag structures of the air pollution
proxy, Particulate Matter (PM). The model has been estimated by quasi-maximum likelihood and using
robust standard errors The left-hand side of the table contains the coefﬁcient estimates produced using the
ﬁrst sub-sample of data, covering the period when trading at the Milan Stock Exchange was centralized.
The right-hand side shows the estimates generated using the second sub-sample, which refers to the era in
which trading was computerized and decentralized. Each row refers to a different version of model [1.1],
where the difference arises from the type of lag applied to the pollution proxy. The ﬁrst row contains the
estimated coefﬁcient on the air pollution proxy at lag 0, whereas in the following rows alternative lags of
the pollution proxy are employed, from 1 through 5. In the last row, the pollution proxy employed is the
mean of the PM levels recorded between day t-5 and day t-I. Equity returns are expressed in percentage
points. P-values are in parenthesis below the corresponding coefﬁcients. The number of lagged returns
included in each estimated equation is Shown in brackets. One, two, and three asterisks denote statistical
signiﬁcance at the ten, ﬁve, and one percent level, respectively.

 

Centralized Market Decentralized Market

 

 

 

 

Comit Datastream MIB Comit Datastream MIB
01/02/1980— 01/02/1980- 01/03/1985- 04/75/7994- 04/75/1994— 04715/1994-
(24/74/1994 04/14/7994 0471477 994 02/73/1998 02/13/7 998 02/73/1998

-.00043 -.00035 -.00076** -.00009 .001 16 -.00027

#fM (0.124) (0.335) (0.033) (0.961) (0.598) (0.891)
[3] [3] [3] [3] [0] [3]

-.00063"”“ —.00052 -.00099"* —.00049 .00037 -.00034

”(C/I4 (0.027) (0.129) (0.008) (0.807) (0.853) (0.863)
[1] [3] [3] I31 [01 [3]

-.00049 -.00017 -.00061* -.00149 -.00138 -.00136

”if; (0.103) (0.591) (0.080) (0.419) (0.506) (0.461)
[3] [2] [31 [3] [0] [3]

—.00043 -.00026 -.00066* -.00029 .00086 -.00038

#57134 (0.169) (0.435) (0.080) (0.878) (0.689) (0.841)
[3] [31 [3] [3] [0] [3]

-.00036 .00012 -.00070* .00205 .00161 .00207

”£71: (0.246) (0.673) (0.074) (0.267) (0.423) (0.261)
[3] [3] [2] [31 [0] [3]

—.00035 -.00014 -.00041 -.00156 -.00132 -.00164

pf}; (0.249) (0.625) (0.289) (0.38) (0.471) (0.356)
[3] [3] [31 [31 [0] [3]

-.00068" -.00056 -.00087" -.00006 .00051 -.00009

”£1211" (0.032) (0.1 13) (0.021) (0.979) (0.852) (0.972)
[31 [31 [2] [3] [0] [3]

 

64

Table 1.5 Air Pollution and Stock Return - Controlling for Meteorological Conditions

This table displays the results of estimating model [1.2] by quasi-maximum likelihood and using robust
standard errors. The leftohand side of the table contains the coefﬁcient estimates produced using the ﬁrst
sub-sample of data, covering the period when trading at the Milan Stock Exchange was centralized. The
right-hand side shows the estimates generated using the second sub-sample, which refers to the era in
which trading was computerized and decentralized. The air pollution proxy employed is the mean level of
Particulate Matter (PM) recorded between day t-5 and day t-I. Compared to model [1.1], three weather
proxies are included as controls, i.e. temperature, rain, and geomagnetic stems. Stock returns are expressed
in percentage points. P-values are in parenthesis below the corresponding coefﬁcients. One, two, and three
asterisks denote statistical signiﬁcance at the ten, ﬁve, and one percent level, respectively.

 

Centralized Market

Decentralized Market

 

 

 

 

Comit Datastream MIB Comit Datastream MIB
01/02/1989- 01/02/1989- 0170277 989- 0475/1994 0405/7994 0477577994-
0477477994 0471477994 04/74/1994 0277377998 0277377 998 02/73/7998
.430" .443** .439" -005 .026 -001
a (0.01 1) (0.017) (0.010) (0.981) (0.909) (0.996)
295*" 267*" 306*” .155*** 154*"
'61 (0.000) (0.000) (0.000) (0.000) (0.000)
,6 4097* -112" -.080"' -.O81*
2 (0.053) (0.025) (0.086) (0.082)
pM -.00287*" -.00263*** -.00304*" -.00032 .00012 -.00034
#Mean (0.000) (0.001) (0.000) (0.899) (0.965) (0.894)
.066 .032 .076 .111" 109* .109"
7340 (0.172) (0.579) (0.137) (0.043) (0.072) (0.046)
4253‘" -.262*“ -.263*** -227" -.255* -.226**
77:0” (0.010) (0.011) (0.010) (0.044) (0.053) (0.045)
-157" -.116 -.l63"'* .142 .156 .142
”HI/M00" (0.042) (0.170) (0.040) (0.114) (0.108) (0.117)
-.169" -.151* -.I64* .023 .034 .023
mew/”00" (0.042) (0.077) (0.051) (0.804) (0.729) (0.803)
-.099 -.153* -.117 -.243"* -.267"”"* -.250***
7Monday (0.233) (0.064) (0.168) (0.008) (0.007) (0.007)
-.090 -.012 -.086 -.054 .060 -.049
7701‘ (0.759) (0.967) (0.772) (0.855) (0.901) (0.869)
-.007 -.008 -.006 .002 .001 .002
Hemp (0.329) (0.311) (0.393) (0.772) (0.885) (0.795)
. -.007 -.01 I -.009 -.027 -.077 -023
”tam (0.919) (0.889) (0.893) (0.727) (0.345) (0.766)
.033 .040 -.015 -.068 -212 -.076
7080mm (0.658) (0.622) (0.840) (0.701) (0.195) (0.665)
Ln (L) -1475.6 -15 19.5 -1476.3 -1458.1 -1506.6 -1456.4
Obs 975 984 961 951 955 948
Gaps 35 35 35 5 5 5

 

65

Table 1.6 Air Pollution and Stock Returns - Controlling for International Economic Shocks
This table displays the results of estimating model [1.3] using the ﬁrst sub-sample of data, which refers
to the period when trading at the Milan Stock Exchange was centralized. The model has been estimated
by quasi-maximum likelihood and using robust standard errors. The air pollution proxy employed is the
mean level of Particulate Matter (PM) recorded between day t-5 and day t-l. Compared to model [1.2],
two international stock market indices are included as controls, i.e. Datastream Germany-market and
S&P500. Stock returns are expressed in percentage points. P-values are in parenthesis below the
corresponding coefﬁcients. One, two, and three asterisks denote statistical signiﬁcance at the ten, ﬁve,
and one percent level, respectively.

 

Centralized Market

 

 

Comit Datastream MIB
0770277 989-04/14/1994 01/02/1989-04/14/1994 0770277 989-04/14/1 994
.356** 345* 364*
a (0.026) (0.051) (0.028)
.230*** .I85*** .239***
’6’ (0.000) (0.000) (0.000)
6 -.144*** -.071* -.154***
2 (0.000) (0.075) (0.000)
0 .077** .076**
3 (0.026) (0.034)
pM -.00251*** -.00291 *** -.00284***
#Meon (0.000) (0.000) (0.000)
.027 .036 .038
7540 (0.549) (0.448) (0.425)
-.163* -.230** -. 175*
7F0” (0.062) (0.019) (0.061)
-092 -.056 -079
”Pu/WOO" (0.244) (0.477) (0.323)
-. 109 —.1 1 1 -.094
mew/”00" (0.163) (0.151) (0.227)
-.061 -.091 -.O65
71140"de (0.425) (0.241) (0.401)
.025 .205 .021
”wt (0.935) (0.418) (0.946)
—.004 -.003 -003
”cm/2 (0.520) (0.660) (0.624)
. -.074 -.045 -.080
780m (0.293) (0.546) (0.260)
.01 1 .026 -.035
70805’0r'" (0.873) (0.739) (0.619)
.518*** .498*** .540***
70ermany (0.000) (0.000) (0.000)
.166*** .I77*** .168***
77/54 (0.000) (0.000) (0.000)
Ln (L) -1326.5 -1383.6 4323.1
Obs 952 960 937
Gaps 53 54 54

 

66

Table 1.7 Air Pollution and Stock Returns — Alternative Proxy (NO‘)
This table displays the results of estimating model [1.3] using Nitrogen Oxides (NOX), instead of

Particulate Matter (PM), as a proxy for air pollution. The model has been estimated by quasi-maximum
likelihood and using robust standard errors. The left-hand side of the table contains the marginal effects
of air pollution estimated using the ﬁrst sub-sample of data, which covers the period when trading at the
Milan Stock Exchange was centralized. The right-hand Side shows the estimates generated using the
second sub-sample, which refers to the era in which trading was computerized and decentralized. Each
row refers to a different version of model [1.3], where the difference arises from the type of lag applied
to the pollution proxy. The ﬁrst row contains the marginal effect of the air pollution proxy at lag 0,
whereas in the following rows alternative lags of the pollution proxy are employed, from 1 through 5. In
the last row, the pollution proxy employed is the mean of the NOx levels recorded between day t-4 and

day t-I. Equity returns are expressed in percentage points. P-values are in parenthesis below the
corresponding coefﬁcients. The number of lagged returns included in each estimated equation is shown
in brackets. One, two, and three asterisks denote statistical Signiﬁcance at the ten, ﬁve, and one percent
level, respectively.

 

Centralized Market Decentralized Market

 

 

Comit Datastream MIB Comit Datastream MIB
0770277 989- 0770277 989- 0770277 989- 0477577994- 0477 577994- 0477 577994-
0477477994 0477477994 0477477994 0577972006 0577972006 0577972006

-.00003 .00010 .00001 -.00001 -.00005 .00001

”(A/Ox (0.864) (0.562) (0.980) (0.980) (0.670) (0.997)
[21 [I] [2] [11 [0] [11

-.00008 —.00025 -.00010 .00001 .00003 .00001

#ﬁiox (0.651) (0.161) (0.588) (0.901) (0.791) (0.899)
[3] [ll [31 [ll [0] [1]

-.00021 -.00019 -.00022 .00007 -.00004 .00007

#1113" (0.166) (0.221) (0.174) (0.556) (0.706) (0.554)
[3] [ll [3] [1] [0] [1]

-.00023 -.00028* -.00025 .00002 -.00003 .00004

”ﬂex (0.140) (0.074) (0.1 13) (0.879) (0.787) (0.766)
[3] [ll [3] [ll [0] [1]

-.00031* -.00034** -.00030* .00005 .00001 .00006

#719): (0.050) (0.032) (0.074) (0.675) (0.908) (0.658)
[2] [I] [2] ll] [01 [11

—.00021 —.00021 -.00019 .00003 -0001 .00004

71,qu (0.132) (0.167) (0.212) (0.773) (0.273) (0.720)
[2] [11 [2] ll] [01 [1]

-.00033* -.00039** -.00030 .00003 -.00009 .00003

Ali/7%: (0.081) (0.034) (0.136) (0.855) (0.531) (0.833)
[3] 12] [3] [I] [01 Ill

 

67

Table 1.8 Air Pollution and Stock Returns — Alternative Proxy (802)
This table displays the results of estimating model [1.3] using Sulfur Dioxide (S02), instead of

Particulate Matter (PM), as a proxy for air pollution. The model has been estimated by quasi-maximum
likelihood and using robust standard errors. The left-hand side of the table contains the marginal effects
of air pollution estimated using the ﬁrst sub-sample of data, covering the period when trading at the
Milan Stock Exchange was centralized. The right-hand side shows the estimates generated using the
second sub-sample, which refers to the era in which trading was computerized and decentralized. Each
row refers to a different version of model [1.3], where the difference arises from the type of lag applied
to the pollution proxy. The ﬁrst row contains the marginal effect of the air pollution proxy at lag 0,
whereas in the following rows alternative lags of the pollution proxy are employed, from 1 through 5. In
the last row, the pollution proxy employed is the mean of the 502 levels recorded between day t-3 and

day t-l. Equity returns are expressed in percentage points. P-values are in parenthesis below the
corresponding coefﬁcients. The number of lagged returns included in each estimated equation is shown
in brackets. One, two, and three asterisks denote statistical signiﬁcance at the ten, ﬁve, and one percent
level, respectively.

 

Centralized Market

Decentralized Market

 

 

 

Comit Datastream MIB Comit Datastream MIB
0770277 989- 0770277 989- 0770277 989- 0477 577994— 0477577994- 0477 577994-
0477477994 0477477 994 0477477994 0577972006 0577 972006 0577972006
—.001 -.001* -.001 -.001 -.001 -.001
#1502 (0.176) (0.093) (0.082) (0.498) (0.248) (0.479)
[2] [2] [3] [1] [0] I I l
-.001 -.002** -.001 -.001 -.002 —.001
#15912 (0.172) (0.043) (0.084) (0.399) (0.130) (0.461)
[2] [2] [31 [1] [0] [1]
-.001 -.002** -.001 -.002* -.002* -.002*
#7532 (0.212) (0.01 I) (0.083) (0.080) (0.094) (0.090)
[2] [2] [3] [1] [ll [1]
—.001 -.001 —.001 -.00001 -0003 .0002
#15032 (0.123) (0.180) (0.074) (0.992) (0.812) (0.865)
[21 [1] [3] [ll [0] [I]
-001 -.001 -.001 .001 .001 .001
”(~32 (0.223) (0.293) (0.258) (0.317) (0.695) (0.320)
[2] [1] [3] [I] [01 Ill
-0003 -.001 -001 .001 -.001 .001
”(5:052 (0.630) (0.323) (0.338) (0.541) (0.667) (0.536)
[2] [I] [3] [1] [0] [I]
-.001* -.002** -.002* -001 -002 -.001
”Egg" (0.099) (0.022) (0.053) (0.330) (0.211) (0.401)
[2] [ll [3] [1] [01 [ll

 

68

Table 1.9 Air Pollution and Stock Returns - Alternative Sub-Periods
This table displays the results of estimating model [1.3] using smaller sub-samples. The model has been
estimated by quasi-maximum likelihood and using robust stande errors. The air pollution proxy
employed is the mean level of Particulate Matter (PM) recorded between day t-5 and day t-l. The left-hand
side of the table contains the marginal effects of the air pollution proxy estimated using alternative subsets
of the ﬁrst sub-sample of data, which covers the period when trading at the Milan Stock Exchange was
centralized; each row contains the air pollution effect estimated from a different subset of data. The right-
hand side Shows the marginal effects of the air pollution proxy estimated using alternative subsets of the
second sub-sample of data, which refers to the era in which trading was computerized and decentralized.
Equity returns are expressed in percentage points. P-values are in parenthesis below the corresponding
coefﬁcients. The number of lagged returns included in each estimated equation is Shown in brackets. One,
two, and three asterisks denote statistical signiﬁcance at the ten, ﬁve, and one percent level, respectively.

 

 

 

 

 

Centralized Market Decentralized Market
Comit Datastream MIB Comit Datastream MIB
-.003*"“" -.003"* -.003*** -.002 -.0002 -.002
01/02/1989- 04/15/1994-
04/14/1994 (0.000) (0.000) (0.000) 13/02/1998 (0.383) (0.922) (0.379)
[3] [21 [31 [ll [0] [I]
-.002*** -.003*** -.002"”'”" .001 .002 .001
01/02/1990- 04/15/1994-
7 7
04/14/1994 (0.003) (0.001) (0.004) 12/31/1996 (0.7-9) (0.566) (0.74-)
[31 [21 [31 [ll [0] [2]
-.002"‘ -.002* -.002* -.003 -.001 -.003
01/02/1991- 04/15/1994-
04/14/1994 (0.069) (0.084) (0.086) 12/29/1995 (0.535) (0.834) (0.517)
[21 [2] [2] [ll [0] [21
-.003"' -.002 -.003"‘
3;??31333- (0.061) (0.142) (0.068)
[2] [I] [21

 

69

Table 1.10 Air Pollution Cross-City Correlation Matrix
This table displays estimated cross-city air pollution correlation coefﬁcients for a sample of cities located in
Lombardy, the Italian region whose capital is Milan. The air pollution proxy employed in the computations
is Particulate Matter (PM). The ﬁrst column contains the names of the cities in which PM daily
concentrations have been measured by background monitors. The second column indicates the distance
between a given city and the monitoring station located in the city of Milan. The third column Speciﬁes the
province in which each given city is located.

 

 

 

 

 

Distance

City 1:13: Province Milan Pero Rho Agrate Villasanta lnzago Cassano
(km)

Milan - Milan 0

Pero 11 Milan 0.73 e

Rho 14.5 Milan 0.41 0.80 e

Agrate 15 Monza 0.43 0.63 0.63 e

Villasanta l6 Monza 0.36 0.64 0.69 0.67 e

lnzago 21 Milan 0.65 0.83 0.54 0.61 0.57 e

Cassano 23 Milan 0.56 0.84 0.68 0.59 0.51 0.86 o

Legnano 27 Milan 0.50 0.78 0.73 0.72 0.60 0.70 0.65

Nibionno 30 Lecco 0.72 0.78 0.51 0.57 0.50 0.76 0.73

Filago 31 Bergamo 0.69 0.75 0.45 0.55 0.52 0.73 0.67

Castano 34 Milan 0.70 0.83 0.60 0.53 0.61 0.74 0.63

Turbigo 37 Milan 0.59 0.72 0.69 0.48 0.44 0.73 0.62

Varese 48 Varese 0.51 0.35 0.13 0.09 0.08 0.40 0.24

Casnigo 60 Bergamo 0.64 0.70 0.31 0.49 0.41 0.74 0.64

Colico 72 Lecco 0.48 0.46 0.31 0.45 0.46 0.37 0.39
Distance

City 133:)“ Province Legnano Nibionno Filago Castano Turbigo Varese Casnigo
(km)

Milan - Milan

Pero 11 Milan

Rho 14.5 Milan

Agrate 15 Monza

Villasanta l6 Monza

lnzago 21 Milan

Cassano 23 Milan

Legnano 27 Milan 9

Nibionno 30 Lecco 0.60 o

Filago 3 l Bergamo 0.57 0.76 e

Castano 34 Milan 0.71 0.82 0.76 e

Turbigo 37 Milan 0.67 0.58 0.56 0.78 e

Varese 48 Varese 0.27 0.47 0.35 0.46 0.24 e

Casnigo 60 Bergamo 0.54 0.78 0.65 0.70 0.60 0.48 e

Colico 72 Lecco 0.42 0.40 0.57 0.54 0.39 0.32 0.40

 

70

Table 1.11 Air Pollution and Stock Returns -lnstrumenta1 Variables Estimation

This table reports the coefﬁcients and signiﬁcance levels from the instrumental variables estimation. The
left-hand side of the table contains the coefﬁcient estimates produced using the ﬁrst sub-sample of data,
covering the period when trading at the Milan Stock Exchange was centralized. The right-hand side shows
the estimates generated using the second sub-sample, which refers to the era in which trading was
computerized and decentralized. The air pollution proxy employed in the computations is the
contemporaneous concentration of Particulate Matter (PM). The instruments are wind Speed, atmospheric
pressure, and rainfall, all of which employed with a lag, as speciﬁed in model [1.6]. Equity returns are
expressed in percentage points. P-values based on robust standard errors are in parenthesis below the
corresponding coefﬁcients. One, two, and three asterisks denote Statistical signiﬁcance at the ten, ﬁve, and
one percent level, respectively.

 

Centralized Market Decentralized Market

 

 

 

Comit Datastream MIB Comit Datastream MIB

0770277 989- 0770277 989- 01/02/1989- 0477577994- 0477 577 994- 0477577994-

0477 477994 0477 4/7994 0477 477994 0277377998 0277 3/1998 0277377998

.377 .339 .366 .435 .559 .398

a (0.145) (0.208) (0.174) (0.275) (0.187) (0.324)
.226*** .147*** .226*** .102*** .099***

'61 (0.000) (0.001) (0.000) (0.006) (0.008)

0 -.121*** -.069* -.120***
2 (0.000) (0.077) (0.000)

FM -.0052** -.0043* -.0055** -.0096 -0102 -.0088
#7 (0.025) (0.081) (0.024) (0.163) (0.177) (0.211)
.105 .089 .127* .105* .116* 099*

7340 (0.104) (0.181) (0.073) (0.080) (0.098) (0.100)
-.329** -.299** -.353** -.092 -.157 -.087

77’0” (0.013) (0.015) (0.012) (0.393) (0.21 1) (0.417)
-.088 -097 -.076 .240*** .274*** .246***

7Fu11M00” (0.350) (0.331) (0.428) (0.006) (0.004) (0.005)
-029 -.020 -.019 .021 .031 .016

mew/1400" (0.751) (0.837) (0.841) (0.794) (0.736) (0.837)
-.101 -.178 -.124 -.235*** -.349*** -.243***

7Monday (0.365) (0.106) (0.293) (0.010) (0.001) (0.007)
-323 -.221 -354 -.019 -227 —.008

77046 (0.384) (0.582) (0.343) (0.948) (0.525) (0.978)
.0003 —.001 .002 .003 -0002 .003

”emit (0.967) (0.925) (0.792) (0.677) (0.983) (0.715)
. .037 .065 .027 —.I69* —.169* -.159*
780m (0.668) (0.462) (0.767) (0.069) (0.087) (0.089)
.109 .074 .071 -.130 -.282 -124

7090mm (0.250) (0.466) (0.469) (0.486) (0.122) (0.509)
.539*** .563*** .556*** .807*** .612*** .809***

VGermany (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
.170*** .153*** .180*** .051 -011 .050

”/54 (0.002) (0.005) (0.001) (0.339) (0.836) (0.347)
Obs 744 752 733 856 859 853

 

71

Table 1.12 A Test of the Efﬁcient Market Theory Using an Air Pollution Index (GARCH model)
This table displays the results of estimating model [1.7] using the ﬁrst sub-sample of data, which refers
to the period when trading at the Milan Stock Exchange was centralized. The model has been estimated
by quasi-maximum likelihood and using robust standard errors. The air pollution index (API) is a
binary variable constructed based on lagged concentrations of Particulate Matter (PM), Nitrogen Oxides
(NOx), and Sulfur Dioxide (S02). Loosely speaking, the API index takes value 1 when air quality has

been relatively poor during the past few days. According to the trading strategy discussed in section 1.9,
when this happens a selling signal is produced. Equity returns are expressed in percentage points. P-
values are in parenthesis below the corresponding coefﬁcients. One, two, and three asterisks denote
statistical signiﬁcance at the ten, ﬁve, and one percent level, respectively.

 

Centralized Market

 

 

Comit Datastream MIB
0770277 989-04/14/1994 0770277 989-04/14/1994 0770277 989-04/14/1 994
.122 .078 .102
a (0.401) (0.607) (0.493)
.234*** .182*** .241 ***
'31 (0.000) (0.000) (0.000)
4139*** 4148***
ﬂZ (0.000) (0.000)
ﬂ .076" .076"
3 (0.028) (0.034)
4222** 4289*** 4245**
”4P1 (0.027) (0.003) (0.019)
.047 .056 .060
7540 (0.318) (0.249) (0.222)
-.135 -.18l"‘ -143
7Fall (0.1 13) (0.062) (0.1 10)
4083 4040 4080
mil/MOO" (0.289) (0.606) (0.317)
4106 4101 4098
”WW/1400" (0.183) (0.185) (0.219)
4066 4099 4074
Monday (0.384) (0.201) (0.334)
.050 .266 .056
7765: (0,365) (0.230) (0.853)
-003 -.001 -002
7789272 (0.653) (0.861) (0.795)
. 4078 4050 4084
78w" (0.266) (0.500) (0.242)
4009 4004 4049
7090mm (0.888) (0.957) (0.493)
.505*** .473*** .522***
7Germany (0.000) (0.000) (0.000)
.I6l"”MI .1764"!!! _]7lxrnu
7084 (0,000) (0.000) (0.000)
Ln (L) 4341.5 4400.9 4338.8
Obs 963 971 948
Gaps 54 55 55

 

72

Table 1.13 A Test of the Efﬁcient Market Theory Using an Air Pollution Index (LOGIT model)
This table displays the results of estimating equation [1.7] using a LOGIT model. The coefﬁcients that
appear in the table are marginal effects, i.e. they Show the estimated impact that each variable has on the
probability of observing a positive return. The estimation is based on the ﬁrst sub-sample of data, which
refers to the period when trading at the Milan Stock Exchange was centralized. The air pollution index
(API) is a binary variable constructed based on lagged concentrations of Particulate Matter (PM),
Nitrogen Oxides (NOX), and Sulfur Dioxide (802). Loosely speaking, the API index takes value I

when air quality has been relatively poor during the past few days. According to the trading strategy
discussed in section 1.9, when this happens a selling Signal is produced. Equity returns are expressed in
percentage points. P-values are in parenthesis below the corresponding coefficients. One, two, and three
asterisks denote statistical signiﬁcance at the ten, ﬁve, and one percent level, respectively.

 

Centralized Market

 

 

Comit Datastream MIB
0770277 989-04/14/1994 0770277 989-04/14/1 994 01/02/1989-04/14/1994

.150*** .100*** .158***

'31 (0.000) (0.004) (0.000)

-.092* -.109** -.118"

HAP! (0,032) (0.035) (0026)
.032 .022 .040

7540 (0.227) (0.408) (0.143)

-.083* -.063 -.099**

779677 (0.085) (0.183) (0.042)
, -.058 -.061 -.048

”ll/WOO" (0.185) (0.158) (0.285)
-.048 -.108 -.049

mew/"00" (0.257) (0.009) (0.254)

-.089** -.052 -.083"‘

7Monday (0.035) (0.219) (0.054)
.079 .161 .068

mu: (0.602) (0.245) (0.664)
—.001 -.001 -.001

mm); (0,703) (0.774) (0.753)
-.005 -.017 -007

980"? (0.900) (0.650) (0.853)
-.006 -002 -009

I'Geoslorm (0.892) (0.969) (0.833)

.155*** .166*** .154***

7Germany (0.000) (0.000) (0.000)

.118*** .104*** .127***

l’USA (0,000) (0.000) (0.000)

Ln (L) -596.8 -605.3 -583.9
Obs 963 971 948

 

73

Chapter 2
Is There a SAD Foreign Exchange Market Cycle?

Evidence From an LSTR Model

74

2.1. Introduction: SAD Effect and Seasonal Stock Market Cycles

The empirical literature on asset pricing and investor mood has recently
discovered an interesting link between stock returns and the so-called Seasonal Affective
Disorder (SAD) effect. According to Kamstra et al. (2003), there exists a stock market
cycle caused by the interaction between modiﬁcations in people’s medical/psychological
conditions and their decision-making process. The authors maintain that investors’ degree
of risk aversion can be affected by a set of psychological factors related to the mood
realm (i.e. depression and anxiety), which in turn can be altered by some environmental
cues. The speciﬁc variable considered in their analysis is the seasonal daylight cycle,
which is believed to be responsible for the winter blues phenomenon and its more severe
manifestation, i.e. SAD.

The framework advanced by Kamstra et al. suggests that, as investor mood
progressively deteriorates during the fall (due to a diminished number of hours of
daylight), market participants become increasingly risk averse and tend to shy away from
risky stocks, ceteris paribus. In the winter, as the number of hours of daylight gradually
increases, investor mood gets back to normal and, during the adjustment process, the
market participants experience a decrease in their degree of risk aversion, which in turn
leads them to be more aggressive in terms of portfolio choices. AS such, all else equal,
the demand for risky stocks is expected to increase during the winter, until the SAD

effect eventually disappears, presumably at the onset of the Spring.

75

In the time series of stock returns of several major equity markets the authors ﬁnd
some empirical regularities which indeed appear to be consistent with the previous
interpretation, i.e. stock returns are lower than average in the fall and higher than average
in the winter, their magnitude being correlated with the seasonal variation in the length of
the night. Based on their ﬁndings it seems possible to construct a set of trading strategies
that take advantage of the differences in the daylight cycles that characterize countries
located at different latitudes, ultimately earning an “excess return” over a simple global
diversiﬁcation strategy.

The rest of the chapter is organized as follows. Section 2.2 extends the SAD
framework to the foreign exchange market. Section 2.3 puts forward the testable
hypothesis that emerges, and section 2.4 describes the model that is employed in the
empirical investigation. Section 2.5 illustrates the data and results of the econometric

analysis are discussed in section 2.6. Section 2.7 concludes.

76

2.2. SAD and Exchange Rate Dynamics

To my knowledge, the analysis contained in this chapter represents the ﬁrst
attempt to extend the SAD framework to the foreign exchange market. It seems a natural
step though, as the ﬁrst question that comes to mind is whether the empirical regularities
that the SAD effect generates in several countries’ stock markets can spill over into the

dynamics of those same countries’ exchange rates.

Spring Summer

 

United States

# hours of night

 

Fall Winter

 

 

 

Fall Winter

Australia

  

I hours of night

Spring Summer

 

 

 

Figure 2.] Daily hours of night in the United States and Australia
(seasons refer to the northern hemisphere).

77

In particular, since the seasonal daylight cycles in the northern and southern
hemispheres are out of sync, one would wonder if their corresponding SAD cycles could
produce a periodic pattern in the rate of exchange between the currencies of any two
countries located in opposite hemispheres.

The story goes as follows. Here we focus for simplicity on the exchange rate
between US. dollars and Australian dollars, where the latter acts as the numeraire. First,
as shown in Figure 2.1, the length of the night reaches its peak in Sydney six months after
it has done so in New York. Second, according to the medical literature discussed by
Kamstra et al. (2003), SAD is assumed to strike in the fall and fade away in the winter,
whereas it is believed that in the spring and in the summer people are not affected at all
by such a medical condition.

Combining these two facts together yields an interesting outcome in terms of
exchange rate behavior, if one believes in the SAD hypothesis proposed by Kamstra et al.
(2003). When Americans enter the fall, SAD hits the United States, whereas no change
takes place in Australia. As such, Americans are assumed to become increasingly risk
averse and shy away from risky stocks. The original conjecture advanced in this essay is
that American investors not only reduce their demand for domestic stocks, ceteris
paribus, but also decrease their demand for any foreign stocks, among which Australian
stocks. If this is the case, then in the foreign exchange market one should observe a
decrease in the supply of US. dollars (Americans now need fewer Australian dollars to
buy their desired amount of Australian stocks), which would be expected to cause an
increase in the equilibrium value of the exchange rate (i.e. an appreciation of the US

dollar), all else equal. Such an adjustment process is depicted in Figure 2.2.

78

As the season changes and Americans enter the winter, SAD fades away in the
United States and, again, no relevant event occurs in Australia (which is now in the
summer). It follows that Americans progressively move back to exhibiting their standard
degree of risk aversion, increasing their holdings of risky stocks during the adjustment
process, all else equal. As in the previous case, my conjecture is that their increased
demand for stocks not only falls onto domestic equity but also on foreign equity, among
which Australian stocks. If this intuition is correct, then in the foreign exchange market
one should observe an expansion in the supply of US. dollars, ceteris paribus, and
ultimately a fall in the exchange rate (i.e. a depreciation of the. US. dollar).

Analogous reasoning applies when Americans enter the spring and the summer, so

that Australians enter the fall and the winter, respectively.

AUD/U SD

 

 

USD

 

 

 

Figure 2.2 Fall in the United States. The SAD effect causes the US dollar to appreciate.

79

2.3. Hypothesis Under Investigation

Overall, such a framework puts forward an unambiguous testable hypothesis. The
exchange rate between US. dollars and Australian dollars is expected to exhibit a
seasonal pattern: the US. dollar should appreciate in the summer and in the fall and
depreciate in the winter and in the spring, all else equal.

More generally, the rate of exchange between the currencies of any two countries
located in opposite hemispheres should exhibit such a behavior, whereas this pattern
should not appear when one considers the currencies of any two countries located in the
same hemisphere (or at least at similar latitudes). Also, the appreciation/depreciation
process should be a function of the daylight cycle and be more acute for countries that are

further apart in terms of latitude.

80

2.4. An LSTR Model for the Exchange Rate

One possible way to tackle the issues raised by the hypothesis discussed in the
previous section is to assume that the exchange rate seasonally moves between two
distinct regimes. In one regime (which covers summer and fall), due to the seasonal
pattern of the SAD effect, the currency of the country in the northern hemisphere is
expected to experience an excess demand and therefore appreciate against the currency of
the country in the southern hemisphere. In the other regime (winter and spring), the
timing of the SAD cycle is such that the same currency is expected to experience an
excess supply and thus depreciate.

One may also conjecture that there is no abrupt shift between the two regimes
discussed above. Instead, the exchange rate may smoothly move from one to the other,
the transition process being governed by the difference between the seasonal daylight
cycles in the two hemispheres.

Following such a line of reasoning, a Logistic Smooth Transition Regression
(LSTR) model seems the most appropriate to conduct the empirical investigation. More

precisely, the following model will be employed in the econometric analysis

ASH] 21a] +761(ft ’Stll‘ [1 —G(z,,}’,c)]+[a'2 +762lfr —St)]' G(:,,}’,C)+u,+j (2-1)

81

where As” I is the rate of appreciation (depreciation) of a currency, s, is the logarithm of

the spot exchange rate (measured in terms of units of a southern hemisphere’s country’s

currency per one unit of a northern hemisphere’s country’s currency), f, is the forward

exchange rate, and u,+ , is the disturbance term. G(.) is a logistic transition function

G(z,,y,c)= 1+exp — withy>0 (2.2)

where z, is the transition variable, 0'2, is the standard deviation of 2,, yis the transition
Speed parameter, and c is the threshold between the two regimes.

The forward premium is included in the regression equation mainly as a control
variable. As it will be argued later, from a theoretical viewpoint it is not obvious whether
its impact on the dependent variable Should be different across the two regimes. Nor is it
obvious whether the SAD framework highlighted here imposes some other kind of

restrictions on ,6, and ,62. The next section will discuss in more detail how the transition

variable has been constructed.

2.4.1. Transition Variable

As previously mentioned, the transition process between the two regimes is likely

to be associated with the seasonal daylight patterns in the two hemispheres. An initial

inspection of Figure 2.1 may lead one to think that the daily difference between the

82

length of the night in the two countries under observation represents a suitable transition

variable.

Nevertheless, as Figure 2.3 reveals, the variable just described follows a pattern

that is not entirely consistent with the Shift between the summer-fall regime and the

winter-Spring regime. More accurately, the daily difference in the length of the night

between, for example, United States and Australia reaches an annual peak at the end of

the northern hemisphere fall and a trough at the end of the northern hemisphere Spring.

Spﬁng

Summer

 

-3 ‘ Fall

 

Winter

Winter

 

Spﬁng

 

 

Fall

 

(a)

Winter

Winter

 

-3 ‘ Fall

 

 

 

Summer

 

 

 

 

(b)

Figure 2.3 Difference in the daily number of hours of night between United States and Australia (panel a)
and transition variable (panel b). Note: seasons refer to the northern hemisphere.

This drawback can be easily ﬁxed by shifting the whole time series of such a

variable three months forward, so that it reaches an annual maximum at the end of the

northern hemisphere summer (i.e., in the middle of the summer-fall regime) and a yearly

83

 

minimum at the end of the northern hemisphere winter (i.e., in the middle of the winter-
spring regime), as Shown in Figure 2.3. The resulting variable is employed in model [2.1]

as the transition variable 2,.

2. 4. 2. Interpretation of the Parameters

Once the transition variable has been determined, it becomes relatively
straightforward to clarify the meaning of model [2.1] and identify the predictions
generated by the SAD framework in terms of restrictions on the parameters of the model.
The transition function reaches a maximum when the transition variable reaches its yearly
maximum at the end of the northern hemisphere summer, so that the summer-fall regime
represents the upper regime in model [2.1], whereas the winter-spring regime represents
the lower regime.

As a result, if the Uncovered Interest Rate Parity (UIP) condition held (i.e. (1:0

and [i=1 in the following model),

AS744 =4 +75 (f7 “Sr)+ur+1 (2.3)

then one would expect a, to be negative and a2 to be positive in model [2.1], given that

the currency of a country in the southern hemisphere is expected to appreciate against the
currency of a country in the northern hemisphere during the (northern) winter-spring

period and depreciate during the (northern) summer-fall period, that is

84

a] <0 , (22 >0 (2-4)

Given the overwhelming evidence about the empirical failure of the UIP (e.g.
Maynard and Phillips, 2001; Baillie and Kilic, 2006), it might also be possible to make
the previous prediction less stringent.

On the other hand, it appears to be more complex determining whether the SAD

framework also yields some neat predictions in terms of the parameters ,8, and ,8). At this

point of the analysis, I do not see any particular reason for the slope parameter to be
consistently different across the two seasonal regimes. Nevertheless, an interesting
implication could be derived if the data sample at hand was divided into two sub-
samples, one containing those observations for which the forward premium is positive
and the other one containing the observations for which the forward premium is negative.

The parameters may differ systematically according to the following rule

If (f, - 8,) > 0, then ,6] < ,62 (2.5)

<0. then .37 >332

The reasoning is the following. Consider again the example about the rate of
exchange between US. dollars and Australian dollars. When the forward premium is
positive, the Australian dollar is expected to depreciate and such a rate of depreciation
might be augmented when the event happens to take place in the upper regime (northern
summer-fall), because that is the regime in which the SAD effect is already forcing the

Australian dollar towards a depreciation. If the event took place in the lower regime

85

(northern winter-spring), instead, the SAD effect would be driving the Australian dollar
in the opposite direction, thus lowering the extent of its depreciation. As a result, one may

conjecture that, given a positive forward premium, ,6) should be smaller than ,62.

Analogous reasoning applies when the forward premium is negative, the difference being
that in such a scenario the inequality sign would be reversed.

Though mentioned here, this hypothesis is not tested directly. Instead, it is left for
future investigations that may have the beneﬁt of having to deal with larger samples. As
the next section will show, the present work is based on a relatively small sample of

currencies and covers a relatively short time span.

86

2.5. Data

The data set used in the empirical analysis contains weekly data on spot exchange
rates for the currencies of the US. Dollar (UD), Euro (EU), UK Pound (UK), Swiss
Franc (SF), Japanese Yen (JY), and Canadian Dollar (CD) vis-a-vis the Australian Dollar
(AD). This data set has been constructed using the daily data provided by the Federal
Reserve Bank of St. Louis (noon buying rates in New York city for cable transfers
payable in foreign currencies), and covers the period 0170271998 through 0471172007“.

Weekly data on LIBOR rates for the above currencies have been obtained from
the British Bankers’ Association and have then been employed to construct the time
series of forward exchange rates for the these same currencies”. A simple Visual
inspection of the data, not reported here, reveals that the UIP does not hold, on average,
for the currencies considered here.

The transition variable has been constructed for each exchange rate and for the
relevant time period starting from the corresponding countries’ daylight cycles, and
according to the formulas presented by Kamstra et al. (2003) for the SAD effect. In
particular, the difference in the length of the night for each pair of countries has been
adjusted, as discussed in section 2.4.1, to obtain the ﬁnal transition variable used in the

regressions.

 

4' The only exception is the Euro currency, for which the available time series starts at the beginning of
January, 1999.
‘2 http://www.bba.org.ukfbba/j sp/polopoly. j sp?d=l 03.

87

2.6. Empirical Results

The results from estimating model [2.3] by maximum likelihood are collected in
Table 2.1. Excluding the case of the Japanese Yen, the coefﬁcient ,6 is always estimated
to be negative, even though statistical Signiﬁcance is never achieved. The time series
analyzed here, therefore, seem to exhibit the typical properties documented in the
literature about the forward premium puzzle.

The estimated coefficients for the LSTR in model [2.1] are reported in Table 2.2.
Overall, the results do not provide strong evidence is support of the hypothesis under

investigation. For the Euro, UK Pound, Japanese Yen, a, is indeed estimated to be
negative, whereas a2 is estimated to be positive, but the corresponding standard errors are

relatively high, and one cannot conﬁdently reject the null hypothesis that these two
coefﬁcients are jointly equal to zero. In the case of the US. Dollar, Swiss Franc and
Canadian Dollar, the ﬁrst parameter is estimated to be positive and in two instances it
turns out to be even greater than the second one; yet, once again, the null hypothesis that
the two parameters do not statistically differ from zero cannot be rejected.

In terms of transition between regimes, the best results have been obtained for the
Swiss Franc, UK Pound and Japanese Yen, as their estimated transition functions appear
to be satisfactorily smooth. Euro, Canadian Dollar and US. Dollar, on the other hand,
present an extremely abrupt switch between regimes, which raises serious doubts about

the adequacy of the model employed here.

88

It Should be kept in mind that the evidence provided here is based on a sample that
contains a single southern hemisphere’s country’s currency, i.e. the Australian Dollar.
Future investigations should extend the data set and include at least the currencies of New
Zealand and South Aﬁica. Also, weekly data are likely to incorporate a relevant amount
of noise, which suggests that future analyses should also be based on lower frequency

data.

89

2.7. Concluding Remarks

This analysis contained in this chapter represents a ﬁrst attempt to extend the
SAD framework (developed by Kamstra et al., 2003) to the foreign exchange market.
From a theoretical perspective, the SAD theory appears to generate some neat and
testable predictions in terms of exchange rate dynamics. In particular, during the
(northern hemisphere) winter and spring, the currency of any country located in the
northern hemisphere is anticipated to experience an excess supply and depreciate against
the currency of any country located in the southern hemisphere. The opposite is expected
to happen during the (northern hemisphere) summer and fall.

A Logistic Smooth Transition Regression model has been chosen to investigate
such a phenomenon, the underlying idea being that the two time periods discussed above
(winter-spring and summer-fall) constitute tow separate regimes through which the
exchange rate smoothly moves while being affected by the differences between the
daylight cycles of the two countries considered.

The empirical results, overall, appear weak and do not provide solid evidence in
support of the initial hypothesis. Such an empirical failure might have been caused, at
least in part, by the short time span covered by the available data (a little more than four
hundred weekly observations) and by the inclusion into the sample of only one southern

hemisphere’s country’s currency.

90

Table 2.1 Forward premium anomaly regressions

HCC standard errors are in parenthesis below the corresponding parameter estimates. For the US Dollar the
sample has been reduced to the more recent three years of observations“.

As,+1=a +ﬂ (fr ‘Sr)+“t+1

 

 

UD EU _ UK SF JY CD
a 0.001 0.003 0.0002 0.005 -0004 0.0001
(0.002) (0.003) (0.001) (0.004) (0.009) (0.001)
-2595 -6.681 -0014 -6.754 3.877 -0154
(5.219) (4.923) (4.458) (4.744) (9.102) (3.294)
Obs. 142 274 434 434 432 434

 

Table 2.2 Results from Restricted LSTR UIP regressions (c = 0)

HCC standard errors are in parenthesis below the corresponding parameter estimates. For the US Dollar the
sample has been reduced to the more recent three years of observations (see footnote 43). The threshold
parameter c is set to zero in all regressions.

ASr+1=1a1‘l'761(fr _St)]'[1_G(Zt’l/’C)]+[a2 +ﬂzlfr ‘Stll'Glzril’aCl'l'uiH

where G(z,, 7,6): (1+ expl— 7(2) - C)/ 02,)

I

 

 

UD EU UK SF JY CD

a, 0.001 -0001 -00004 0.00003 -0021 0.002
(0.003) (0.004) (0.001) (0.004) (0.012) (0.002)

131 4.153 -0.762 -0.456 4.219 20.103 -3.978
(7.507) (7.318) (5.821) (5.526) (12.404) (4.967)

a, 0.0003 0.007 0.001 0.01 1 0.016 -0002
(0.003) (0.003) (0.001) (0.007) (0.015) (0.002)

132 4.254 -12.580 0.216 42.462 -15.289 3.604
(7.071) (6.431) (6.893) (8.459) (14.501) (4.328)
7 81.93 211.13 18.68 11.65 23.93 527.85
(19.45) (33.51) (13.32) (4.05) (4.63) (15.70)

Obs. 142 274 434 434 432 434

 

 

43 When using the whole US Dollar sample, the transition function would not converge, so the sample Size

was repeatedly reduced until a reasonable estimate was obtained.

91

Chapter 3

Superstition and Investment Decisions

92

3.1. Introduction

As mentioned in the previous chapter, the research carried out in the ﬁeld of
behavioral ﬁnance has discovered a rich collection of cognitive and emotional biases that
affect investors’ decision—making process (e.g. Shleifer, 1999; Hirshleifer, 2001).

This chapter aims at contributing to such a stream of ﬁnancial literature by
exploring whether and how investment decisions can be inﬂuenced by superstitious
beliefs. Numerous investigations conducted in the social sciences, in particular in the
ﬁelds of psychology and sociology, have produced some evidence in support of the view
that superstition can inﬂuence individuals’ attitudes and choices in everyday life, and
they have also revealed that even the most important human decisions are not immune to
the power of superstitious beliefs (e.g. Kaku, 1974; Campbell, 1996; Ankerberg and
Weldon, 1999; Torgler, 2003). Yet, very few empirical examinations have been carried
out in such an area by economists and ﬁnance researchers. Few recent examples include
the articles of Dichev and Janes (2003) and Yuan et al. (2006) on the alleged link
between equity returns and lunar phases.

One may conjecture that such a dearth of investigations has been caused by (1) the
assumption that people tend to behave rationally, (2) the view that superstition is not
worth of scientiﬁc scrutiny, and (3) the view that, even if some people’s decisions were
affected by superstitious beliefs, this would not be likely to produce any detectible impact

on relevant aggregate economic/ﬁnancial variables. In response to the ﬁrst and second

93

concerns, one should consider that superstition may produce real effects in the same way
that self-fulﬁlling prophecies do. As an example, the reader may think about the well
known fact that in many western countries the number 13 is considered an unlucky
number and building owners will sometimes deliberately skip a ﬂoor so numbered“.
Assuming for simplicity that building owners are fully rational, their choice is likely to
have been determined by the worry that their business might be negatively affected by the
existence of superstitious tenants or customers. If people believe that other people hold
superstitious beliefs, then, just like in the case of self-fulﬁlling prophecies, superstition
may indeed produce real consequences. In turn, as long as superstitious beliefs are able to
spur real effects in the realm of human decision-making, then it is my opinion that they
Should become the object of scientiﬁc investigation. As for the third argument, it is
certainly true that a large set of superstitious beliefs (e. g. avoiding walking under ladders,
throwing salt over one’s Shoulder) is unlikely to have any signiﬁcant impact on relevant
macro variables. Yet, this fact alone, should not preclude research to be conducted in this
area. Testing whether superstition can exert an inﬂuence on interesting ﬁnancial variables
is possible provided that one identiﬁes a superstitious belief that can inﬂuence the
attitudes of large groups of investors at the same time and in a predictable fashion. If the
decisions of the marginal investor are affected by such a factor, then a detectible
aggregate effect Should follow.

In order to achieve such a goal, the present investigation focuses on the
superstitious beliefs allegedly attached to total eclipses, which are natural phenomena that

can be observed approximately at the same time by millions of individuals on earth.

 

4" According to a Spokesman for Otis Elevator Company, the world’s largest manufacturer of elevators,
around 85% of the buildings in the world do not feature a named 13th ﬂoor (Perkins, 2002).

94

Folklorists have provided solid evidence that, for thousands of years, solar and lunar
eclipses have been frequently interpreted as bad presages in many societies around the
planet. More interestingly, they have also documented that such beliefs are still somewhat
rooted in many modern cultures“). The occurrence of eclipses has been (wrongly)
commonly taken as a bad omen, that is a sign that would precede the occurrence of
horrible events such as famines, calamities, and outbreaks of lethal diseases. The
existence of such an alleged link has been disproved by modern science and, as such, one
can safely conclude that eclipses are a purely exogenous factor in the context of the
present analysis. Total eclipses could produce an actual effect on some macro
socioeconomic variables only if people attached some superstitious beliefs to eclipses
themselves.

Reformulating this View in the context of ﬁnancial markets, one would expect
that, (I) if agents were entirely rational and asset prices efﬁciently incorporated all
relevant information (null hypothesis), then the occurrence of total eclipses would have
no impact whatsoever on equity returns. Similarly, (2) if investors held superstitious
beliefs yet the stock market was ﬁrlly efﬁcient, then one would not expect stock returns to
follow any abnormal pattern in correspondence of an eclipse; after all, eclipses are
perfectly predictable events and, therefore, asset prices should already incorporate their
future occurrence. Only if superstition was at work and, concomitantly, the stock market
was not entirely efﬁcient, then one would expect to observe a (negative) relationship
between the occurrence of total eclipses and stock returns. Consequently, the present

study appears to be appealing on two grounds: if the null hypothesis of no relationship

 

’5 See Goldman (1940), Greenlee (1945), Allison (1948), Seibold (1967). Hand and Talley (I984), Castro
(I995), Vojcik (1996), and Milan (2000).

95

between eclipses and equity returns was rejected, that would suggest that (1) some
psychological factors (i.e., superstitious beliefs) play a role in determining investors’
decisions and (2) ﬁnancial markets conceal some inefﬁciencies that could be proﬁtably
exploited by an informed agent.

The investigation conducted here is interlinked with a growing body of literature
that analyzes how some economically neutral yet psychologically relevant factors can
affect investors’ behavior. The effects of mood on asset prices have been extensively
documented (e.g. Saunder, 1993; Hirshleifer and Shumway, 2003; Kamstra et al., 2003;
Krivelyova and Robotti, 2003; Cao and Wei, 2005). Interestingly, some environmental
variables that are believed to be capable of triggering mood changes (e.g. sunshine, cloud
cover, temperature, geomagnetic ﬁeld) may also exhibit a link with the occurrence of
total solar eclipses, so it turns out to be necessary to control for those factors in order to
demonstrate that the impact of total eclipses on equity returns, if any, is not due to some
already known cause. Analogously, as total lunar eclipses only take place during full
moon periods, one needs to make sure that the eclipse effect is not generated by the
lunar-cycle effect identiﬁed by Dichev and Janes (2003) and Yuan et al. (2006).

The empirical examination, conducted using a large sample of daily data on three
major US stock market indices, shows that the occurrence of total eclipses is indeed
associated with lower-than-average returns, consistently with the view that superstitious
beliefs contribute to shape investors’ decisions and, concurrently, the Efﬁcient Market
Theory does not hold strictly. Not surprisingly, when the analysis focuses on the total
eclipses visible in north America, the eclipse effect turns out to be even stronger. It is also

Shown that such an effect, which is both statistically Signiﬁcant and economically

96

relevant, can be used to construct some proﬁtable trading strategies and, consequently, it
seems to qualify as a market anomaly.

What makes the analysis of superstition particularly appealing from a social
scientist’s perspective, and what distinguishes it from the literature on investor mood, is
that the former refers to a purely self-ﬁrlﬁlling and self-reinforcing phenomenon.
Explaining what triggers people’s actions is the core of any social science, and it is in my
opinion that any force that can drive human behavior and is capable of producing tangible
effects in multiple spheres of human life should be granted thorough investigation.

The rest of the chapter is organized as follows. Section 3.1 describes the
mechanisms through which superstitions are conjectured to emerge. Section 3.2 reviews
the empirical analyses that have been conducted in such an area in the ﬁelds of
economics and ﬁnance, and section 3.4 puts forward the superstition hypothesis that is
tested in the present work. Section 3.5 illustrates the mechanics of the phenomenon under
investigation (i.e. eclipses) and section 3.6 discusses some of the controls that need to be
employed. Section 3.7 describes the data, whereas the empirical analysis is contained in
section 3.8. Section 3.9 discusses the implications that the fmdings have in light of the

Efﬁcient Market Theory. Section 3.10 provides some concluding remarks.

97

3.2. Origin of Superstitious Beliefs

The term “superstition” is recurrently used in common speech and is typically
employed in a pejorative sense. In most instances, its meaning is merely assumed rather
than deﬁned. Numerous deﬁnitions have been proposed in the ﬁelds of psychology and
sociology. Considering the goal of the present analysis, it has been chosen to focus on
one particular deﬁnition, that is the one advanced by Jarvis (1980), according to whom
superstition is “an attitude individually held by people which relate their existence to the
general order of the cosmos but which is not based upon empirical evidence nor
incorporated within the institutionalized belief systems of a society as deﬁned by leading
representatives of these systems at any given time.”

Jarvis also offers a classiﬁcation of superstitions, which may be sorted into
taboos, spells, and omens. Omens are signs that believers consider as prophecies and may
have to do with bad luck, good luck, and death. Taboos involve prohibitions whose
violation may give origin to a punishment. Lastly, spells are measures employed to
promote good luck or get around bad luck.

Needless to say, sociologists have long been puzzled by the survival of
superstition in modern industrialized societies. In the nineteenth century, they anticipated
that superstition “would gradually fade away in the face of that rationalism and
empiricism which was presumed to be characteristic of modern capitalist society”

(Campbell, 1996). Yet, several investigations Show that the levels of superstition have not

98

decreased substantially during the twentieth century. Campbell (1996), based on several
surveys undertaken in Britain between the 1950s and the 1980s, illustrates that a large
portion of the population is consistently willing to depict themselves as superstitious, and
a considerable fraction even admits to “engaging in some superstitious practices”.
Similarly, Ankerberg and Weldon (1999), surveying the literature on the effects of
astrology on human behavior, uncover evidence that the inﬂuence of the former on the
latter has clearly intensiﬁed in recent decades. Torgler (2003), analyzing a sample of data
from seventeen countries, claims that citizens of most countries display a fairly high level
of superstition in terms of beliefs in horoscope, good luck charms, and fortune-tellers. In
1990, Gallup polls showed that 25% of the US. population believed in astrology (Gallup,
1990) and, based on the results of two other surveys (Gallup, 1978; Stark and Bainbridge,
1985), it seems that this ﬁgure has remained relatively steady in recent years“.

Up to now, most of the theories proposed to rationalize superstition have been
given either an anthropological or a psychological foundation. According to Skinner
(1948), superstitious beliefs develop through an incorrect learning process. More
speciﬁcally, individuals may tend to draw erroneous conclusions when they are unable to
precisely identify cause-and-effect relationships. Superstitions typically arise following a
recurrent pattern: a person (group of people) has to deal with uncertainty about a future

relevant event or outcome (factor 1); before the event takes place or the outcome is

 

‘6 A simple search conducted using any Internet search engine reveals that there exist literally hundreds of
web sites that deliver trading Signals based on the tools of astrology and the horoscope. Lo et al. (2000)
state that “it has been argued that the difference between fundamental analysis and technical analysis is not
unlike the difference between astronomy and astrology”. Here it is claimed that “ﬁnancial astrology” itself
appears to have been following the footsteps of technical analysis. Testing the predictions of the former
does not seem to be substantially different from testing the predictions of the latter, as the mechanics of
both ﬁelds are heavily based on self-fulﬁlling expectations. While the latter discipline has received
considerable academic scrutiny, the same cannot be said for the former. In a broad sense. together with the
studies of Dichev and Janes (2003) and Yuan et al. (2006) on the lunar cycle, the present work may be
considered as a ﬁrst effort in this direction.

99

uncovered, something fortuitous happens (factor 2); the accidental event is then
interpreted as cause, even if, in fact, there is no relationship at all between factor 1 and
factor 2. Subsequent fortuitous reiterations of this accidental association of events then
act as a positive reinforcer.

Anthropologist Malinowski (I948) maintains that superstitions cannot be labeled
as pointless primitive beliefs, as they are psychologically needed for existence and fulﬁll
some essential functions within a culture. His “theory of the gap” asserts that superstition
ﬁlls the void of the unlmown and reduces anxiety. Scheibe and Sarbin (1965) summarize
these ideas claiming that “where he lacks truth and needs it, (man) makes it up,
somehow”. Cohen at al. (1959) ﬁnd that people are more likely to distort facts or logic
rather than abandon their superstitions.

AS highlighted by Henslin (1967), the social context plays a key role in the
survival and diffusion of superstitions. Individuals’ behavior originates in a social
framework and is reacted to by signiﬁcant others who serve as reinforces. Behaviors are
then transmitted or taught to new members and, as a result, become part of a culture.
Scheibe and Sarbin (1965) summarize this argument asserting that “it is as if each culture
presents its participants with a rule book which states standard beliefs about what leads to
what”.

Interestingly, even the (supposedly) most important human decisions (e.g.
reproduction, health) do not appear to be immune to superstition (e.g. Kaku, 1974;
Stevenson, 2000; Do and Phung, 2006). It seems therefore natural to wonder whether
some superstitious beliefs extend to economic decisions. The next section discusses the

evidence that has been uncovered in this area.

100

3.3. Superstitious Beliefs and Economic Decisions: Empirical Evidence

In the last two decades, relatively few studies have been carried out in the ﬁelds of
economics and ﬁnance on the relationship between superstition and decision-making.
Woo and Kwok (I994), analyzing license plate auctions in Hong Kong, discover that
plates featuring the number 8 (a sign of good luck in Cantonese-speaking countries) are
sold for statistically higher prices than plates featuring the number 4 (a Sign of bad luck).

Kolb and Rodriguez (1987), employing a sample of US. daily returns from the
CRSP value- and equally weighted indices ﬁ'om 1962 through 1985, ﬁnd that mean
returns for Friday the Thirteenths are signiﬁcantly lower than those of other Fridays,
which is what one would expect according to the “Friday the Thirteenth” superstition".

Brown at al. (2002) investigate the effect of superstition on price clustering in
some Asia-Paciﬁc ﬁnancial markets (Australia, Hong Kong, Indonesia, the Philippines,
Singapore and Taiwan). In the Chinese culture some numbers are considered highly
inauspicious (e.g. number 4), whereas others (e.g. number 8) are considered highly
auspicious (e.g. number 8). The authors uncover some evidence in support of the View
that Chinese superstitions play a role in shaping the number preferences of traders in
Hong Kong, but the evidence is weak for the remaining ﬁve countries.

Since foreign investors’ trading might eliminate the traces of any culture-Speciﬁc

superstition from the time series of equity returns, Brown and Mitchell (2004) focus on

 

’7 Some subsequent studies carried out by other authors, using different samples. have not lent support to
such a ﬁnding, though. See Dyl and Maberly (1988), Chamberlain et al. (1991), and Coutts (1999).

101

trading at the Shanghai and Shenzhen stock exchanges, between 1990 and 2002. A strict
segmentation characterized these two markets over much of the period under
consideration, as domestic investors were only allowed to trade A-shares, whereas
foreign investors were only allowed to trade B-shares. The authors claim that the prices
of A-shares are statistically more likely to end in 8 than in 4, while such a pattern is much
weaker for B-shares, consistently with the Chinese superstition story.

A valuable lesson taught by these investigations is that, as the degree of
integration among international ﬁnancial markets has considerably increased during the
recent decades, in order to test the inﬂuence of superstition on investment decisions one

Should isolate some beliefs that are shared by many cultures around the world.

102

3.4. Testable Hypothesis: Solar and Lunar Eclipses as Evil Presages

The folklore literature provides extensive evidence about the widespread
interpretation of total eclipses as bad omens. In Hindu mythology, it is believed that some
demons “biting” the sun or the moon are responsible for the occurrence of eclipses,
which are deemed as ominous events (Dickens, 1996). Such a symbolism has not
disappeared, and many Indians Still think that some catastrophe will occur“. In Thailand,
devotees burn black joss sticks and hand over offerings in the days that precede an
eclipse, and when it takes place they clash gongs and explode ﬁrecrackers to scare the
demon that is believed to be “biting “ the sun or the moon. Likewise, in Cambodia,
during solar eclipses thousands of people Shoot with handguns and riﬂes to frighten the
moon and “help the sun escape”. In China, solar eclipses were taken as bad omens and
people used to think that the emperor might die when those phenomena occurred. In
Japan, people used to think that eclipses were symptoms of a disease that hit the sun and
that protection was needed if one wanted to stay away from such a disease. Milan (2000)
asserts that, even in our time, some Japanese people will cover their wells to prevent their
water from being infected by the disease hitting the sun. Eskimos’ beliefs are analogous
to the Japanese’s and some Eskimos will turn over utensils to prevent them from being
contaminated. One of the two key books of orthodox Judaism, the Talmud (Sukkah 29a),

declares that “An eclipse of the sun is a bad omen for the world”. Goldman (1940)

 

48 See, for example, Asiaweek — October, 20 I995.

103

provides evidence that the Cubeo Indians, a group Of natives that live in southeastern
Colombia, rush out of their houses and scream at the beginning of an eclipse. When one
of them bears an answer to her shouts, that is taken as a presage that he or She will soon
pass away after suffering from the same disease that hit the sun. In many rural areas of
Mexico it is still believed that, when a solar eclipse takes place, pregnant women are at
risk of having a baby with a cleft lip (Castro, 1995).

Moving the focus to the United States, Vojcik (1996) claims that the dropping of
atomic bombs in Japan at the end of World War II has initiated an “era of nuclear
apocalypticism that has ﬂourished in American religious culture, folklore and popular
culture”. Wilson (1977) and Boyer (1992) cite numerous surveys that indicate how
apocalyptic speculations are pervasive at a popular level, and assert that millions of
Americans at present hold beliefs about an approaching destruction of the world.
According to a 1994 survey conducted for US. News & World Report, many people think
the Bible should be taken literally when it narrates about a Battle of Armageddon (44%)
and a ﬁnal Judgment Day (60%)”. In this regard, the book of Revelation, which has been
one of the main sources of inspiration for most contemporary Christian prophecy,
contains a clear reference to what has been typically considered a total solar eclipse.
More Speciﬁcally, when describing the opening of the seven seals that precedes the end
of the world, the book of Revelation (6: 12) reads: “And I saw when he opened the sixth
seal, and there was a great earthquake; and the sun became black as sackcloth of hair”.
Hand (1961-1964), in his massive collection on popular beliefs and superstitions,

includes eclipses among the signs of an imminent doomsday. Similarly, Hand and Talley

 

49 US. News & World Report, 19 December 1994, p. 64.

104

(1984), in a similar compilation, list eclipses as one of the signs that precede the end of
the world.

Taken as a whole, there appears to exist a large body of evidence in support of the
View that total eclipses, for centuries or even millennia, have been interpreted as bad
omens in numerous societies around the world. Contemporary investigations also seem to
indicate that such superstitious beliefs have not been completely abandoned, though their
sphere of inﬂuence is probably restricted and only a relatively small fraction of the
population is likely to be affected.

Interestingly, the stock market features at least two of the characteristics that
Campbell (1996) considers as essential for the development and survival of superstitions:
(l) outcomes are very relevant (losing money is extremely different from winning
money) and (2), the participants are aware of their lack of control over the outcomes
themselves (i.e., equity returns seem to be randomly distributed and no theory seems to
work when investors are called upon to predict future returns). As a result, ﬁnancial
markets appear to provide an ideal setting for testing whether superstitious beliefs
contribute to shape human choices.

Given the evidence discussed in this section, it appears that a very speciﬁc
hypothesis can be put forward and subsequently tested empirically. More speciﬁcally, the
investigations conducted in the ﬁeld of folkloristics about total eclipses suggest that, if
the marginal investor attached superstitious beliefs to such natural phenomena, then their
occurrence should be anticipated to worsen her expectations about the future
developments of the economy, in turn causing a reduced demand for stocks and,

ultimately, having a negative impact on equity returns.

105

3.5. The Physics of Eclipses

Total solar eclipses take place when the moon passes in front of the sun and
obscures it entirely. Overall, the duration of such natural phenomena is limited to a
couple of hours, at most, and totality generally lasts for less than seven minutes.
Empirically, eclipses have been found to have a temporary inﬂuence on numerous
geophysical variables and also on plants, animals’ behavior and human beings’ health.
Kotrappa et al. (1981) claim that solar eclipses are associated with increases in relative
humidity and conspicuous falls in ground level temperature”. Keshavan et al. (1981), and
Strestik and Jaroslav (1999) document a corresponding variation in the Earth’s
geomagnetic ﬁeld.

Solar eclipses have also been found to alter the circadian rhythm of plants
(Haberle et al., 2001), and animals (Wojtusiak et al., 1975a; 1975b; 1976), yet, to the
best of my knowledge, no evidence has been produced on their ability to inﬂuence the
human circadian rhythm“.

Hundreds of investigations have documented one speciﬁc effect of solar eclipses

on human, i.e. solar retinopathy. This is a medical condition that is caused by

 

5 0 Some of the locations monitored by the authors experienced a drop in temperature up to about 10° C.

5 ' A circadian rhythm is a type of “biological clock” that presides over numerous physiological processes of
plants, animals and even humans. It partly determined by some external cues, such as temperature and
sunlight. For a literature review on the role of circadian rhythms, see Paranjpe and Sharma (2005). Clock
shifting at the beginning and end of daylight saving time has been shown to produce some effects on sleep
patterns and, in turn, on human behavior (Kamstra et al., 2000). On the other hand, solar eclipses are
transitory events and totality lasts no more than few minutes. As a result, it seems safe to assume, as a
working hypothesis, that solar eclipses do not produce any shift in people’s biological clocks that, in turn,
might be responsible for changes in their behavior.

106

unprotected or inappropriate sun gazing or exposure, whose risks have been known since
at least the time of Plato (Shuttleworth and Galloway, 2002). Staring at the sun with a
naked eye or without a proper ﬁlter, even during an eclipse, may cause serious damages,
and such harmful effects may have contributed, over the centuries, to develop and
reinforce the superstitious beliefs attached to solar eclipses. Recent surveys have revealed
that there is still a great deal of uncertainty among the general public about how to view
an occultation of the sun in safety (Shuttleworth and Galloway, 2002), which may

suggest that, for many people, the physics of solar eclipses, to a certain extent, remains a

mystery.

107

3.6. Environment, Emotional State, and Equity returns

Over the last ﬁfteen years, behavioral ﬁnance researchers have accumulated
considerable evidence about the effects that alterations in the Earth and space
environments can exert on investor mood and, ultimately, on investment decisions and
equity returns. Taken as a whole, this body of research casts doubt on the view that
economic agents are entirely rational and calls for the inclusion of behavioral variables
into asset pricing models.

Saunders (1993) can be credited with making the seminal contribution in this area.
In particular, he ﬁnds that the weather in New York (percentage daily cloud cover) is
highly correlated with the returns of three major indices of the US. stock market, and
suggests that sunshine inﬂuences people’s mood, which in turn affect the demand for
risky stocks and, ultimately, stock returns (the better investor mood, the higher equity
returns). Along similar lines have moved several subsequent investigations (Kramer and
Runde, 1997; Hirshleifer and Shumway, 2003; Pardo and Valor, 2003; Goetzmann and
Zhu, 2003; Loughran and Schultz, 2004).

Adopting an analogous approach, Cao and Wei (2005), using a sample of data
from eight industrialized countries, ﬁnd that stock returns are negatively correlated with
the temperature prevailing in the cities that host the corresponding stock exchanges.

Kamstra et al. (2003) assert that much of the seasonal pattern found in stock

returns is attributable to the so-called seasonal affective disorder (SAD). After discussing

108

the evidence produced in the ﬁelds of medicine and psychology that shows a positive
relationship between the length of the night and seasonal depression, as well as between
depression and risk aversion, the authors examine whether the seasonal variation in the
length of the day (measured at different latitudes) is correlated with equity returns. Their
empirical ﬁndings appear to be consistent with the existence of an SAD effect, and
suggest that such a mood effect is both statistically and economically signiﬁcant.
Krivelyova and Robotti (2003), drawing upon a body of psychiatric research that
documents how changes in the geomagnetic ﬁeld can inﬂuence people’s mood, test
whether stock returns exhibit an abnormal pattern during geomagnetic storms, and ﬁnd
some support for their hypothesis. Controlling for other well-known market anomalies,
the authors show that the occurrence of geomagnetic storms produces a statistically
signiﬁcant negative (lagged) inﬂuence on equity returns in nine industrialized countries.
Finally, Dichev and Janes (2003) and Yuan et al. (2006) uncover some evidence
in support of the view that the lunar cycle exerts an inﬂuence on people’s behavior, in
general, and on investment decisions, in particular. Their ﬁndings seem to suggest that
stock returns are higher-than-average during new moon periods and lower-than-average

during full moon periods.

109

3.7. Data

Here the analysis is centered upon the US. market and covers three broad Stock
market indices: S&P500, Dow Jones Industrial Average and Nasdaq Composite. Data
about equity returns have been collected from Yahoo! Finance”. Daily log returns,
adjusted for dividends, have then been computed from January 2, 1930 through January
27, 2006 for the S&P500 and Dow Jones indices. In the case of the Nasdaq index, the
sample covers the period February 8, 1971 through January 27, 2006. Table 3.1 allows a
graphical inspection of the data and shows that the time series of stock returns under
investigation suffer from negative skewness and excess kurtosis.

Table 3.1 also include summary statistics for two sub-samples of the S&P500 and
Dow Jones indices: January 2, 1946 through January 27, 2006 and January 2, 1970
through January 27, 2006. Three reasons have led to such a choice. First, as shown in
some previous studies, during the Great depression equity returns were exceedingly
volatile compared to the subsequent decades, and such a pattern might Obscure the role
played by the phenomenon under scrutiny (Schwert, 1989). Second, as pointed out in
section 3.2, superstitious beliefs appear to have revived in the American culture following
World War H; as a result, and also in order to leave out the events associated with the
war, January 2, 1946 has been chosen as the starting date for the ﬁrst sub—period.

Moreover, in order to obtain results that are comparable across the three market indices in

 

52 http://ﬁnance.yahoo.com/.

110

terms of time span, a second sub-sample has been constructed whose starting date is
January 2, 1970. Naturally, such a method can also be interpreted as a form of robustness
check, as a case for a superstition effect can only be made if the empirical evidence is
relatively stable over time.

Data about solar and lunar eclipses have been collected ﬁ'om NASA”. Only total
eclipses have been considered, and their date has been detemrined according to Universal
Time“. Considering the whole sample, total solar eclipses seem to be visible somewhere
on Earth every seventeen months, on average (53 observations overall).

NASA also provides information about the geographic region in which each total
eclipse is visible. Initially, given the small number of observations, the analysis is
conducted using all total eclipses, whereas, at a subsequent stage, attention is restricted to
those eclipses visible in north America. Newspapers and other news providers around the
world generally document the occurrence of a total solar eclipses visible somewhere on
Earth, but given that stock markets are not fully integrated, one would expect eclipses
visible in the US. to have a stronger impact on the domestic stock market”. As this
process of news circulation takes some time (newspapers giving accounts of this event
are normally issued on the subsequent day), one should think about the possibility that
the effect of total eclipses on the equity market, if any, may be a lagged one.

Total lunar eclipses occur somewhat more often than solar ones, and can also be

seen from a much. larger area on Earth. Considering the whole sample, 64 such

 

53 http://sunearth.gsfc.nasa.gov/eclipse.html.

5‘ The underlying idea is that total eclipses are Special in that there is a powerful symbolic meaning
attached to them. There exist two other kinds of solar eclipses: annular and hybrid. Lunar eclipses can also
be of penumbral type. A preliminary empirical analysis of eclipses other than total, not reported here for
brevity, has suggested that they do not seem to exhibit any systematic relationship with equity returns.

55 See, for example, Shuttleworth and Galloway (2002).

111

occurrences have been recorded. As such, total lunar eclipses are visible somewhere on
Earth every fourteen months, on average. When solar and lunar eclipses are combined,
the frequency increases to one occurrence every eight months, on average.

Data about the Earth’s geomagnetic ﬁeld have been collected from the National
Geophysical Data Center in Boulder, Colorado. Unlike Krivelyova and Robotti (2003),
who measure geomagnetic activity through the Ap index, the present analysis employs
the aa index“. The only difference between the two indices originates from the fact that
data coming from different panels of observatories are used in their construction, and no
signiﬁcant difference in the empirical results should be expected. The time series
employed here spans from January 2, 1940 through December 31, 1998.

Data about temperature and cloud cover in New York have been obtained from
the National Climatic Data Center in Asheville, North Carolina”. The former extends
over the period January 2, 1964 through January 27, 2006, whereas the latter spans from
January 4, 1965 to April 30, 1996. The SAD variable (daily number of hours of night in
New York city) has been calculated according to the methodology suggested by Kamstra
et al. (2003).

Finally, following Yuan et al. (2006), in order to capture the alleged effects of the
lunar cycle it has been chosen to construct two dummy variables: a full—moon durruny,

taking the value of I up to three days before and after each full moon date and 0

 

5" http://www.ngdc.noaa.gov/stp/GEOMAG/aastar.shtml. The choice has been motivated by data
availability reasons.

57 http://Iwf.ncdc.noaa.gov/oa/ncdc.html. In particular, the data have been provided by the JFK airport
station.

112

otherwise, and a new-moon durruny, taking the value of 1 up to three days before and

after each new moon date. Data about the lunar phases have been obtained from NASA”.

 

53 http://sunearth.gsfc.nasa.g0v/eclipse/phase/phasecat.html.

113

3.8. Econometric Analysis

A preliminary investigation, conducted using the empirical methodology that will
be illustrated in this section, has produced no evidence of a contemporaneous relationship
between total eclipses and stock market retumssg. AS such, the rest of the analysis will
focus on the detection of a lagged effect. From a theoretical viewpoint, this may be
justiﬁed by the fact that a large portion of the market participants get to know about the
occurrence of an eclipse from the mass media, and typically with a time lag that is a
natural outcome of the process of information diffusion (e.g. newspapers). A lagged
eclipse durmny has been created that takes the value of l on the trading day that
immediately follows a total eclipse and zero otherwise. As it will be shown, the outcomes
of the tests employed here do seem to support the hypothesis that the occurrence of

eclipses produces a negative lagged impact on investment decisions“).

3.8.1. Total Eclipses and Magnitude of Equity Returns

The analysis starts off with a simple t-test aimed at verifying if equity returns are

lower than average on the trading days that immediately follow the occurrence of total

 

59 The results are not reported here for brevity.

6° The same analysis has been carried out using alternative lags for such a dummy variable. No statistical
evidence has been found in this regard. Analogously, no anomalous behavior of stock returns has been
detected on the days that immediately preceded total eclipses.

114

eclipses“. Table 3.2 displays the outcomes of this test and reveals that, on average, daily
equity returns are negative when they are preceded by a total eclipse. Such a result arises
in correspondence to every market index and sample period examined. On the other hand,
daily stock returns are estimated to be positive, on average, during “ordinary” trading
days (i.e., days that do not follow total eclipses). When the whole sample is considered,
for the Dow and S&P500 indices the difference between the two means is approximately
0.2%, and is statistically signiﬁcant at standard conﬁdence levels (p-value below 0.10).

The divergence is even larger when the focus is moved to the intermediate
sample. For the two above-mentioned indices, stock returns are lower (roughly by 0.25%)
on the trading days that follow total eclipses than on ordinary days, the difference being
statistically signiﬁcant at least at the 5% level.

The most recent sub-sample generates similar results. Across the three market
indices, equity returns are consistently lower after a total eclipse has taken place than on
regular days, the effect being both economically and statistically signiﬁcant (p-value
below 0.01). On average, the difference is about 0.45%.

Taken as a whole, these estimates tell a consistent story and provide some initial
support to the hypothesis that equity returns negatively react to the occurrence of solar
and lunar eclipses. Such an outcome is compatible with the view that (at least some)
investors hold superstitious beliefs about these natural phenomena and take such beliefs
into account (either consciously or subconsciously) when they are called upon to make
ﬁnancial decisions. Interestingly, the table also suggests that the magnitude of the alleged

superstition effect seems to have grown in the last decades.

 

6' More speciﬁcally, stock returns are regressed on a constant and a lagged eclipse dummy, and Newey-
West heteroskedasticity and serial correlation robust standard errors are calculated (max 5 lags). The t-test
determines whether the coefﬁcient on the eclipse dummy is Statistically different from zero.

115

Another possible explanation is that, even if the marginal investor is fully rational,
she may assume that a large fraction of the population holds superstitious beliefs and, in
turn, may plan her investment decisions accordingly in an attempt to take advantage of
the price changes caused by superstitious investors (self—fulﬁlling prophecy). In this
regard, the empirical regularity documented and discussed above suggests that rational

investors, if existing at all, cannot completely eliminate the alleged superstition effect.

3 .8.2. Total Eclipses and Direction of Equity Returns

The previous section tried to verify if stock returns tend to be lower than average
when they are preceded by a total eclipse. In this section, instead, the goal is to determine
if also the direction of equity returns is affected to some extent by the occurrence of total
eclipses.

In order to perform such a task a sign test can be employed, which determines if
the direction of a market index change aﬁer a speciﬁc event (i.e. a total eclipse) is
systematically different from the direction that one would be anticipating simply based on
the unconditional distribution of returns. In other words, the objective is to verify whether
there is a recurrent pattern in the stock return “data generating process” according to
which returns tend to be not only lower after an eclipse than on regular days but also
negative.

The unconditional probability of observing a negative daily return has been
estimated from the data themselves for each sub-sample and stock market index, as

shown in Table 3.3. The null hypothesis assumed for testing purposes is that the number

116

of negative returns actually observed in each sub-sample after a total eclipse is not
different from the number that one would anticipate to observe when drawing a random
sample. The binomial distribution is then employed to calculate the probability of
observing a number of negative returns greater or equal to the one actually observed in
the sample. It follows that such a probability may then be used as a one tail p-value for
the purpose of determining whether to reject or not to reject the null hypothesis.

Table 3.3 displays the outcomes of such a test. Depending on the market index
and sub-period analyzed, the estimated unconditional probability of observing a negative
return (fourth column) ranges between 43.8% and 48.4%. When the whole sample is
examined, for the Dow and S&P500 indices there is some mild evidence that equity
returns are more likely to be negative on those trading days that directly follow the
occurrence of total eclipses (p-value below 0.09).

The results are somewhat less clear-cut when the intermediate sub-sample is
evaluated, though there is still some marginal evidence in favor of the view that total
eclipses seem to alter the direction of the following trading day’s equity returns in the
expected way.

Lastly, when the most recent sub-sample is assessed, all three indices reveal that,
after an eclipse, a negative return is much more likely to be observed than a positive one,
the corresponding p-values being less than or equal to 0.01. Depending on the market
index considered, the probability that equity returns turn out to be negative following a
total eclipse ranges between 64% and 67%, whereas the same probability ranges between

43.8% and 48.5% on ordinary trading days.

117

Once again, overall, the ﬁndings appear to be consistent with the view that the
occurrence of total eclipses causes an abnormal behavior of stock returns, and such a
departure is consistent with what one would anticipate based on the superstition story
discussed in the previous sections. Taken as a whole, the tests carried out in the previous
and current section suggest that, after a total eclipse, not only stock returns tend to be
lower than they would be otherwise, but they even tend to be negative. Up to this stage of
the analysis no control variable has been employed. The next step will consist in
verifying whether the empirical regularities documented here can be explained in light of
some standard economic variables or based on some psychological factors advanced in

the behavioral ﬁnance literature.
3. 8. 3. Multivariate Model

The initial goal is to purge the time series of equity returns from any residual
autocorrelations. This is done by estimating the following equation by OLS and

performing statistical inference using Newey-West adjusted standard errors (max 5 lags)

P
t‘, = (I + Z ﬂt—krt—k +lchIipseEdipset—l +81 (3'1)
k=l

where r, is the daily stock return between time H and time t, and Eclipse,_, is a lagged

dummy variable that takes the value of 1 when a total eclipse has taken place at time t-I.

Lagged returns up to p periods are added to the model when statistically Signiﬁcant.

118

Table 3.4 displays the estimated coefﬁcients and corresponding p-values. The
index returns included in the present sample seem to exhibit the typical autocorrelation
structure that has been detected in the ﬁnance literature, for the coefﬁcients on the lagged
returns are statistically signiﬁcant up to the second lag. When evaluating the whole

sample, the coefﬁcient pecupse turns out to be negative and statistically different from

zero for both the S&P500 and Dow Jones indices. The point estimate of the eclipse effect
is approximately -0.2%. Analogous results are obtained when the intermediate sample
(1946-2006) is assessed. In this case the marginal effect turns out to be both slightly more
economically relevant (-0.23%) and statistically Signiﬁcant. lastly, when the focus is
moved to the most recent sub-sample (1970-2006), the eclipse effect still clearly emerges
across the three market indices, and its size appears to be particularly large (-0.4%).
Additionally, the null hypothesis of no relationship with the dependent variable can be

rejected at remarkably high conﬁdence levels.

3 .8. 4. Controlling for Behavioral Factors

In principle, the empirical regularity documented in the previous sections might
be caused by other factors. AS a result, it is critical to control for all the variables that are
inﬂuenced by the physics of eclipses and, at the same time, have been shown to have a
relationship with stock returns.

Cao and Wei (2004) have documented the existence of a link between equity
returns and temperature and, at the same time, Kotroppa et al. (1981) have shown that

total solar eclipses can inﬂuence the Earth’s ground temperature. As a result, a variable

119

that measures the daily temperature in New York city will be added to the regression
equation. Based on the same rationale, one also needs to control for the SAD effect, the
percentage cloud cover in New York, and Earth’s geomagnetic activity.

Dichev and Janes (2003) and Yuan et al. (2006) claim that stock returns are
generally higher during new-moon periods than ﬁrll-moon periods. Since solar eclipses
can only occur during a new moon, whereas lunar eclipses can only take place during a
full moon, one will also want to control for the lunar cycle. This is done by adding the
full moon and new moon dummies, previously discussed, to the regression equation.

The intersection of the time series of all the variables involved generates a sample
that covers the period ﬁ'om January 4, 1965 through April 30, 1996 for the S&P500 and
Dow Jones indices, and from February 8, 1971 through April 30, 1996 for the Nasdaq
index. The following model has been ﬁtted by OLS and using Newey-West adjusted

standard errors (max 5 lags):

p
"t = a + Z ﬂt—k’t—k +HEclipseEClipset—l + 7MondayMondayt +
k=l

+ ASADSAD, + AhmemmreTemperature, + zlaoudCloud, + (3-2)
+ Aceomagne,,-6Geomagnetic,_l + dNewmoonNewmoon, +

+ 5Fu11m00nF ullmoon, + 8,

where Temperature, measures the daily temperature (Fahrenheit degrees) in New York,
Cloud, measures Sky cloudiness in New York”, and Geomagnetic” is a lagged dummy

that takes the value of I when the aa geomagnetic index is greater than 60 at time t-I.

 

62 Sky cover is measured in tenths, where 0 means less than 10% coverage and 10 means 100% coverage.

120

Newmoon, is a dummy that takes the value of I up to three days before and after each
new-moon day, and Fullmoon, is a dummy that takes the value of I up to three days
before and after each full-moon day. SAD, captures the number of hours of night in New
York and has been adjusted so as to incorporate the occurrence of total solar eclipsesé'3

It is natural to wonder whether the occurrence of total eclipses follows any
particular pattern in terms of day of the week or calendar month“. Figure 3.1 classiﬁes
the total eclipses that occurred in the whole sample by day of the week, and reveals that
no recurrent pattern seems to exist. Nevertheless, with the purpose of getting more
precise estimates, a Monday dummy has been added to model [3.2], taking the value of 1

when the trading day t is a Monday and zero otherwise.

14%
312% l
E
g 10%
3' ., I
I: ﬁle I
6% I
4%
2%
0%

066‘ 95$
V‘p «‘30 fobo «“oéi

*1 A A

6° 8° 06°
0 S

Q 66‘" 9°

Figure 3.1 Empirical frequency of total eclipses by day of the week.

 

(’3 One hour of night has been added to the SAD variable on each day in which a total solar eclipse
occurred, except in the summer and Spring when the value of the SAD variable remains equal to zero. This
is a very conservative choice as totality typically lasts for very few minutes, so that the actual amount of
daylight is not signiﬁcantly reduced. Estimation has also been conducted without adjusting the SAD
variable. The results are practically identical.

6" The periodicity and recurrence of eclipses is governed by the so-called Saros cycle, a period of
approximately 6,585.3 days (18 years 11 days 8 hours). Any two eclipses separated by one Saros cycle
Share very Similar geometries (Espenak, I987).

121

Similarly, Figure 3.2 sorts all total eclipses by month. Based on such a
categorization, total eclipses do not appear to occur evenly throughout the year. Instead,
they seem relatively more likely to take place in October, November and June. For this
reason, model [3.2] has been estimated both with and without a whole set of monthly
dummies. The outcomes are practically the same in both cases and, as such, only the

estimates associated with the more parsimonious model are reported and discussed here.

 

14%
12%4 --------------------------- ~
10%4 ------
8% ___________ ......

6% «e

Frequency

4% «—

2% «»

 

 

 

0%;

dae~se.<~°,see:ee
56$ (8’90) ‘9 Y9 ‘35 I 1'99 $49 05530 Ads (96‘
9

Figure 3.2 Empirical frequency of total eclipses by calendar month.

Table 3.5 displays the empirical estimates, which are very similar across the three
market indices and suggest that the superstition effect previously documented does not
seem attributable to other well-known market anomalies that may have a connection with
total eclipses. In particular, equity returns following a total eclipse are estimated to be
approximately 0.35% lower than they would be otherwise, all else equal. As a result, the
coefﬁcient on the eclipse dummy appears to be not only highly statistically signiﬁcant (p-

value below 0.01) but also economically relevant.

122

AS for the other explanatory variables, only the Monday dummy and the
percentage sky cover appear to be signiﬁcantly correlated with equity returns, and the
signs of such effects are coherent with the typical results documented in the literature. It
is worth to mention that the SAD variable turns out to be individually statistically
insigniﬁcant, but it is jointly signiﬁcant with the temperature variable. Since these two
variables follow a very similar seasonal pattern, it seems hard to isolate the contribution

that each of them gives, if any, to determining the behavior of stock returns.

3.8. 5. Controlling for Economic News

In principle, one would also want to check whether the empirical regularity
previously discussed is merely attributable to some economic shocks that took place
concurrently with (or the day after) total eclipses rather than to these natural phenomena
themselves. Such a question can be answered accurately only if one is able to identify the
time series of unexpected economic shocks that affected the stock market. In turn, in
order to isolate the unexpected components of a sequence of economic shocks, one would
need a good measure of investors’ expectations“. Since such a strategy seems quite
impractical, here it has been chosen to adopt an indirect approach and take advantage of
the investors’ expectations that are embedded in the term Structure of interest rates.

Previous research has shown that the Yield Curve, in particular the yield Spread

between some long-term and short-term Treasury bonds, incorporates some information

 

65 For a review of the issues associated with such a strategy, see Rigobon and Sack (2006).

123

useful for predicting inﬂation, economic activity, and future short-term interest rates“.
The underlying idea is that the shape of the Yield Curve changes in response to changing
expectations about the future dynamics of interest rates, and changes in expectations are
caused by the arrival of new pieces of information on the market.

In the present context, the yield spread can be used to capture unexpected
economic shocks to the extent that such shocks are incorporated into the Shape of the
Yield Curve itself. More speciﬁcally, the spread between the interest rates on the ten-year
US Treasury bond and the three-month Treasury bill is added to the regression equation
[2]. Furthermore, contemporaneous and lagged changes in such a Spread are added to the

model if statistically signiﬁcant, so that the regression equation becomes:

P
rt = a + 2 ﬂt—k’t—k +HEcIipseEClipset—l + 7MondayM0ndayt +
k=l
+ 715 A DSAD, + MemPemMeTemperature, + ACIoudCloud, +

+ ’lGeomagneticGeomagnetict-1 + §Newmoon Newmoon, '1' (3-3)
+ §Fu11m00nF ullmoon, + ”SpreadSpread, + 17,ASP readASpread, +

+ ngpreadASpread,_1 + 8,

where Spread, is the yield spread discussed above, ASpread, is the change in such a
spread between time t-I and time t, and ASpread_ I is its change between time t-2 and
time t-I.

The estimated coefﬁcients are collected in Table 3.6 and Show that the yield

spread indeed seems to contain some useﬁrl information about the economic activity. In

 

(’6 See, for example, Estrella and Hardouvelis (1991), Plosser and Rouwenhorst (I994), Estrella and
Mishkin ( I997), Dueker (I997), Estrella and Mishkin (1998), and Hamilton and Kim (2002).

124

particular, it exhibits a positive signiﬁcant correlation with equity returns. Furthermore,
lagged changes in the yield spread also appear to incorporate some valuable information
in terms of asset pricing.

If one accepts as true the View that the arrival of new relevant economic
information is reﬂected into the Shape of the yield curve, then the ﬁndings are once again
consistent with the hypothesis that superstition plays a role in investment decisions.
Indeed the coefﬁcients on the eclipse dummy are highly statistically signiﬁcant and
economically relevant across the three market indices. Put in another way, based on the
previous tests, there appears to be no reason to conclude that the abnormal behavior of
stock returns following total eclipses is caused by the market reaction to the arrival of

new economic information.

3 .8. 6. Direct vs. Indirect Visibility of the Phenomenon

So far, the empirical analysis has been conducted including all the total eclipses
that were visible somewhere on Earth during the sample period under consideration. One
may argue though that, if the superstition story contains some truth, then the size of the
eclipse effect should be larger when the sample is cut down so as to include only those
total eclipses that were visible in north America and, therefore, could directly affect US.
investors.

Once such an adjustment is made, the number of total eclipses recorded in the
relevant sample period (1965-1996) Shrinks from 49 to 29. Indeed, when models [3.2]

and [3] are re-estimated considering only the eclipses visible in north America, the

125

statistical Signiﬁcance of the eclipse dummy stays practically unchanged whereas the
magnitude of its coefﬁcient increases in absolute value. More precisely, the estimates,
which are very Similar across the three market indices and the two regression models,
suggest that when a total eclipse is visible in north America, equity returns in the US.
market are estimated to fall by approximately 0.45% on the following trading day, all else
equal. Once again, these outcomes seem to be hard to reconcile with fully rational asset

pricing, and are in line with the superstition hypothesis under observation.
3.8. 7. Controlling for T ime- Varying Volatility

The stock market index returns analyzed here appear to be characterized by
volatility clustering and excess kurtosis. In order to deal directly with such issues a
GARCH framework may be employed to jointly model the mean and variance of returns
(Engle, 1982). More speciﬁcally, the following GARCH (1,1) model has been estimated

by quasi-maximum likelihood and using the Huber-White estimator of variance:

p
rt = a + Z ﬂt-krt—k +HEclipseEC1ipset—l + yMondayMondayt +
k=l
+ 2.5 A DSAD, + MemperamreTemperature, + 71C)0udCloud, +

+ liceomagnen-cGeomagnetic,_1 + 6Newmoon Newmoon, + (3.4)
+ dpullmoonFullmoon, + ”SpreadSpread, + UPSPreadASpread, +

+ 77:51.1) readASpread,_l + 8,

e, =2,0', z, ~iid (0,1)

126

2 2 2
07 = 5 1' 757—1 +¢07—1

The results, not reported here for brevity, are comparable to the ones discussed in
the previous sections. When the GARCH model is used, the superstition effect is
estimated to be slightly smaller than in model [33]“. Such an effect is still Statistically
signiﬁcant and economically relevant. As for the remaining variables, their patterns are
analogous to the ones previously described.

Overall, it seems safe to conclude that the evidence in support of the superstition
story has proven to be robust to the use of alternative sets of controls and estimation
techniques. It is natural to wonder whether such an empirical regularity can be exploited
to construct some proﬁtable trading strategy, which would make it a market anomaly.

This is precisely the goal of the next section.

 

”7 When all eclipses are evaluated, the superstition effect ranges from -0.25% to —0.3%, depending on the
particular market index. Similarly, when the sample includes only the total eclipses that were visible in
north America, the superstition effect ranges from —0.32% to —0.38%.

127

 

 

 

3.9. Superstitious Beliefs, Investment Decisions, and Efﬁcient Market Theory

It is worth to stress here that, as mentioned earlier, total eclipses are perfectly
predictable events. As a result, if the stock market was entirely efﬁcient, the occurrence
of total eclipses Should have no effect whatsoever on equity returns even if investors
attached superstitious beliefs to such phenomena and were affected by those same beliefs
when making ﬁnancial decisions“.

Yet, there seems to exist the opportunity to exploit the empirical regularities
associated with total eclipses. If, at the beginning of 1970, an investor had bought a
portfolio of stocks that replicated the S&P500, for example, and subsequently adopted a
Simple buy-and-hold strategy, holding her portfolio until the beginning of 2006, then she
would have earned an average annual return of 7.59%“. If, instead, such an investor had
implemented an active trading strategy based on the occurrence of total eclipses (i.e., sell
when a total eclipse takes place and buy back at the end of the following trading day),
then she would have earned an average annual “excess return” of approximately 0.6%.

Considering that total eclipses took place every eight months, on average, over the
sample period, it seems safe to assume that a trading strategy based on the superstition

effect would have allowed investors to gain an excess retum over a passive strategy even

 

6” NASA has already calculated all the total eclipses that will be visible somewhere on Earth within the
next two thousand years.

69 In this section daily stock returns are computed as the change in a given stock market index divided by
the previous trading day‘s closing value. No loss of generality should be anticipated as a result of this
choice.

128

if transaction costs are taken into account"). The excess return could be even larger if
agents were allowed to sell short and/or trade derivatives. Analogous results can also be
obtained by applying the same trading strategy to the Dow Jones and Nasdaq indices.
Taken as a whole, these ﬁndings suggest that the superstition effect not only takes
the form of a recurrent pattern in the time series of equity returns, but truly seems to
represent a violation of the efficient market hypothesis, as its own existence and
magnitude reveals the existence of some unexploited proﬁt opportunities. In a larger
perspective, it can be concluded that the empirical results seem hard to reconcile with
fully rational asset pricing and lend support to the view that investment decisions can be

inﬂuenced by some purely psychological factors (e. g. superstitious beliefs).

 

7” For a discussion about the magnitude of transaction costs, see Krivelyova and Robotti (2003).

129

3.10. Concluding Remarks

This chapter has focused on the link that seems to exist between the occurrence of
total eclipses and US. equity returns. The analysis has been motivated by the hypothesis
that human decision-making can be inﬂuenced by superstitious beliefs, which ﬁnds
support in several empirical investigations conducted in the ﬁelds of sociology and
psychology.

More speciﬁcally, the work is aimed at verifying whether the superstitious beliefs
attached to total eclipses, as documented by folklorists, can inﬂuence investors
buying/selling decisions in the same way they have affected human behavior in many
societies for centuries. Since the evidence produced in the ﬁeld of folkloristics suggests
that these natural phenomena have been commonly interpreted as evil omens, it is
conjectured that their occurrence might be taken as a presage of ﬁrture poor performances
of the economy, in turn inducing investors to shy away from Stocks and, ultimately,
decreasing stock returns, all else equal.

Such a hypothesis has then been put to test using daily data from the US. stock
market covering a relatively long period of time. The results have revealed that the
occurrence of total eclipses is associated with an abnormal behavior of equity returns. In
particular, after controlling for a set of behavioral and economic variables, stock returns
appear to be lower than average in the aftermath of total eclipses, which is consistent with

the superstition story proposed here.

130

 

Since eclipses are perfectly predictable events, the existence of such unexploited
empirical regularities also questions the validity of the efﬁcient market theory.
Additionally, in a larger perspective, these ﬁndings challenge the view that investors are

entirely rational and that only purely economic factors are relevant for asset pricing.

131

Table 3.1 Summary statistics of US stock returns

This table displays summary statistics of daily log returns on three US stock market indices: S&P500,
Dow Jones Industrial Average and Nasdaq. Returns are in percentage points.

 

Index Mean Standard Min Max Skewness Kurtosis
Period Devratron

S&P500
01/02/1930-01/27/2006 .0204 1.1140 -22.899 15.366 -.347 22.679
(20014 obs)

S&P500
01/02/1946-01/27/2006 .0282 .9011 -22.899 8.708 4.307 35.792
(15237 obs)

S&P500
01/02/1970-01/27/2006 .0289 .9972 -22.899 8.708 -l.398 37.652
(9108 obs)

Dow Jones
01/02/1930-01/27/2006 .0198 1.1134 -25.631 14.27 -.543 29.347
(19104 obs)

Dow Jones
01/02/1946-01/27/2006 .0267 .9096 -25.631 9.666 -1 .633 50.116
(15109 obs)

Dow Jones
01/02/1970-01/27/2006 .0286 1.0263 -25.631 9.666 -I .752 49.990
(9108 obs)

Nasdaq
02/08/1971-01/27/2006 .0355 1.1946 -12.043 13.254 -.313 13.774
(8826 obs)

 

132

Table 3.2 Comparison of mean returns

This table displays daily mean returns on trading days following a total eclipse (second column) and on
“regular” days (third column). The number of observations is in parenthesis. The last column contains
the t-statistics of a difference of means test (p-values are in brackets). More precisely, returns are
regressed on a constant and a one-day lagged eclipse dummy, and Newey-West heteroskedasticity and
serial correlation robust standard errors are computed (max 5 lags). The test veriﬁes whether the
coefﬁcient on the eclipse dummy is statistically different from zero. Returns are in percentage points.
One, two and three asterisks denote statistical signiﬁcance at the 10%, 5% and 1% level, respectively.

 

Day after Regular days

 

Index total eclrpse Difference t-test
Period
Mean Mean

S&P500 -.208 .022 _0 186* -I .82
01/02/1930-01/27/2006 (117) (19896) ' [0.07]
S&P500 -.063 -.004 O 059 -0.25
01/02/1930-12/31/1945 (23) (4752) - ' [0.80]
S&P500 .130 .026 0 103 0.55
01/02/1946-12/31/1969 (38) (6090) ' [0.59]
S&P500 -.249 .029 0 220“ -l .93
01/02/1946-01/27/2006 (94) (15143) ' ' [0.05]
S&P500 -.470 .031 0 439*" -3.29
01/02/1970-01/27/2006 (56) (9052) ' ' ' [0.01]
Dow Jones -.225 .021 -0 204" -2.03
01/02/1930-01/27/2006 (I 17) (18986) ' [0.04]
Dow Jones -.l 18 -.005 113 -0.40
01/02/1930-12/31/1945 (23) (3971) -' [0.68]
Dow Jones .080 .023 0573 0.34
01/02/1946-12/31/1969 (38) (5963) ' [0.74]
Dow Jones -.255 .028 0 227" -2. l 7
01/02/1946-01/27/2006 (94) (15015) - ' [0.03]
Dow Jones -.451 .031 _0 420*" -3.30
01/02/1970-01/27/2006 (56) (9052) ' [0.01]
Nasdaq -.433 .038 _0 395,," -2.51
02/08/1971-01/27/2006 (55) (8770) ' [0.01]

 

133

Table 3.3 Sign test

This table displays the results of a Sign test on the direction of stock returns following a total eclipse.
The second column contains the number of eclipses that occurred in each sub-sample (trials). The
third column Shows the total number of negative returns observed (successes). The fourth column
Shows the estimated unconditional probability of observing a success (negative return) in each
particular sub—sample for a given index. The number of observations is in parenthesis. The last
column contains the probability of observing a number of negative returns greater or equal to the
number actually observed, based on the binomial distribution. Note: observations containing neither
positive nor negative returns have been dropped. One, two and three asterisks denote statistical

Signiﬁcance at the 10%, 5% and 1% level, respectively.

 

Number of

 

Index Number of negative Probability of One-tail
. total negative
Period . returns after p-value
eclipses . return
eclipses
S&P500 .. 0.472
01/02/1930-01/27/2006 1 17 63 (19637) “089
S&P500 0.4833
0170271930-1273171945 23 12 (457]) 0.435
S&P500 0.4580
017027194642/3171969 38 14 (5985) 0399
S&P500 .. 0.469
01/02/1946-01/27/2006 94 5' (15066) ”092
S&P500 m 0.477
01/02/1970-01/27/2006 56 37 (9081) 0004
Dow Jones * 0.479
01/02/1930-01/27/2006 ‘ ‘7 64 (19019) 0083
Dow Jones 0.4838
,

017027193042731/1945 23 '3 (3979) 0:783
Dow Jones 0.4681
01/02/1946-12/31/1969 33 '5 (5971 ) 0357
Dow Jones 0.478
01/02/1946-01/27/2006 94 51 (15040) 0.125
Dow Jones 1, I, 0.484
01/02/1970-01/27/2006 56 36 (9069) 0'01 2
Nasdaq H, 0.438
02/08/1971-01/27/2006 55 37 (8789) 000‘

 

134

Table 3.4 Regression analysis

This table contains the estimated coefﬁcients of regression model [3.1]. P-values
computed using Newey-West adjusted standard errors (max 5 lags) are in parenthesis.
The number of observations is displayed in the sixth column. The coefﬁcients are in
percentage points. One, two and three asterisks denote statistical Signiﬁcance at the 10%.
5% and 1% level, respectively.

 

 

Index

Period 0' ﬂt-I ﬂI-z rchlipse Obs.
S&P500 .021*** 049*" -024 -. 192* 20012
01/02/1930-01/27/2006 (0.008) (0.000) (0.1 12) (0.060)
5&p500 .028*** .085*** -.047*** -.231** 15237
01/02/1946-01/27/2006 (0000) (0.000) (0.004) (0.042)
3&p500 .031*** .069*** -032 -.443*** 9108
01/02/1970-01/27/2006 (0.003) (0.000) (0.149) (0001)
Dow 10",, .021** 027* -.207** 19103
01/02/1930-01/27/2006 (001 1) (0.070) (0040)
Dow Jones .02711Itt .076**! “.047‘mI “.235** 15109
01/02/1946-01/27/2006 (0.000) (0.000) (0.01 1) (0.025)
Dow 10”,, .031 *** .057*** -.039* —.422*** 9108
01/02/1970-01/27/2006 (0-004) (0-000) (0099) (0.001)
Nasdaq .034TTT .102*TT -.4I4’.‘T‘I 8825
02/08/1971-01/27/2006 (0.007) (0.000) (0.007)

 

135

Table 3.5 Controlling for behavioral factors

This table contains the estimated coefﬁcients of regression model [3.2]. P-values computed using
Newey-West adjusted standard errors (max 5 lags) are in parenthesis. The coefﬁcients are in percentage
points. One, two and three asterisks denote statistical signiﬁcance at the 10%, 5% and 1% level,

 

 

respectively.
Index S&P500 Dow Jones Nasdaq
Pcriod 01/04/l965-04/30/l 996 01/04/1965-04/30/1996 02/08/ I 97 l -04/30/l 996
088* _096* . 182"”
a (0.073) (0.063) (0.000)
.1242":- .096*** 288“”
787-7 (0000) (0.000) (0-000)
-.368*** 2392“” -.367""MI
lchlipse (0003) (0.002) (0-003)
-_131*** -.107*** -.280""
I’Monday (0.000) (0.001) (0-000)
71 .013 .014 .013
SAD (0.244) (0.269) (0.279)
A -.001 -.001 -.001
Temperature (0830) (0.595) (0.219)
,1 -.009** -.009*** -.01 1*"
Cloud (0,012) (0.008) (0001)
A _ .064 .074 .064
Geomagnetic (0,164) (0.1 17) (0. 182)
d .020 .032 .007
Newmoon (0384) (0.203) (0-765)
0- .011 .014 -.004
Fullmoon (0,648) (0.583) (0.852)
Observations 7886 7886 6372
Total eclipses 49 49 40

in the sample

 

136

Table 3.6 Controlling for economic news

This table contains the estimated coefﬁcients of regression model [3.3]. P-values computed using
Newey-West adjusted standard errors (max 5 lags) are in parenthesis. The coefﬁcients are in percentage
points. One, two and three asterisks denote statistical signiﬁcance at the 10%, 5% and 1% level,

respectively.

 

 

Index S&P500 Dow Jones Nasdaq
Period 01/04/1965-04/30/l 996 01/04/1965-04/30/1996 02/08/1971-04/30/1996
.081 .082 .l75***
6' (0.108) (0.122) (0.001)
I3 .123*** 096*” 284*“
t-I (0,000) (0.000) (0.000)
. -.348*** -.376*** -.359***
(“Eclipse (0.005) (0.003) (0.003)
-,139*** -,123*** -.291***
I’Monday (0.000) (0.000) (0.000)
) .009 .01 l .009
~SAD (0,422) (0.393) (0464)
’3 -.001 -.001 -.001
~Temperature (0.531) (0,439) (0.1 19)
I -.009** -.009** -.01 1*"
"Cloud (0,017) (0.012) (0-001)
71 _ .053 .065 .061
Geomagnetic (0,25 5) (0.175) (0.215)
(5 .018 .033 .006
New-moon ((1436) (0,200) (0.808)
0- .012 .017 -.006
F ullmoon (0.622) (0,494) (0.798)
_ .017** .018** .016*
’7 YieldSpread (0_034) (0.029) (0-057)
ASPread -.746*** -.698*** -.477***
777 (0.001) (0.003) (0.002)
ASpread 376*" 458“” 323*
r—l (0.006) (0.001) (0.057)
Observations 7536 7592 6122
Total eclipses 49 49 40

in the sample

 

I37

REFERENCES

Abbey D. E., N. Nishino, W. F. McDonnell, R. J. Burchette, S. F. Knutsen, W. Lawrence
Beeson, J. X. Yang (1999). Long-terrn inhalable particles and other pollutants

related to mortality of non- smokers, American Journal of Respiratory and
Critical Care Medicine, 159, 373—382.

Abel J. L. and K. T. Laan (1990). Anticipation of Performance among Musicians:
Physiological Arousal, Conﬁdence, and State-Anxiety, Psychology of Music, 18,
171-182.

Allison L. (1948). Folk beliefs regarding weather in southeastern Illinois, Journal of
American Folklore 61, 68-70.

American Thoracic Society (2000). What constitutes an adverse health effect of air

pollution? American Journal of Respiratory and Critical Care Medicine, 161,
665—673.

Aneshensel C. S., R. R. Frerichs, and G. J. Huba (1984). Depression and physical illness:
A multiwave, nonrecursive causal model, Journal of Health and Social Behavior,
25 (4), 350-371.

Ankerberg J., and J. Weldon (1999). Encyclopedia of Cults and New Religions, Harvest
House.

APAT — Agenzia per la Protezione dell’Ambiente e per i servizi Tecnici (2004). Linee
Guida Per La Predisposizione Delle Reti Di Monitoraggio Della Qualita Dell’aria
In Italia, Roma.

Appelhans B. M. and L. J. Luecken (2006). Heart Rate Variability as an Index of
Regulated Emotional Responding, Review of General Psychology, 10 (3), 229-
240.

Areni C. S. and D. Kim (1993). The inﬂuence of background music on shopping
behavior: Classical versus top-forty music in a wine store, Advances in Consumer

Research, 20, 336-340.

ARPA Lombardia - Regione Lombardia (2001). Rapporto sulla qualita dell’aria di
Milano e provincia: Anno 2001.

138

ARPA Lombardia — Regione Lombardia (2002). Rapporto sulla qualita dell’aria di
Milano e provincia: Anno 2002.

ARPA Lombardia - Regione Lombardia (2003a). Rapporto sulla qualita dell’aria di
Milano e provincia: Anno 2003.

ARPA Lombardia - Regione Lombardia (2003b). Rapporto sullo stato dell’ambiente in
Lombardia — segnali ambientali, Regione Lombardia.

ARPA Lombardia - Regione Lombardia (2003c), INEMAR, Inventario Emissioni in
Atmosfera: emissioni in regione Lombardia nell'anno 2001, ARPA Lombardia
Settore Aria, Regione Lombardia DG Qualita dell'Ambiente.

ARPA Lombardia - Regione Lombardia (2006), INEMAR, Inventario Emissioni in
Atmosfera: emissioni in regione Lombardia nell'anno 2003. Dati ﬁnali, ARPA
Lombardia Settore Aria, Regione Lombardia DG Qualita dell'Ambiente.

Baillie, R. T., and R. Kilic (2006).Do asymmetric and nonlinear adjustments explain the
forward premium anomaly?, Journal of International Money and Finance, 22, 25-
47.

Baker S. R. (2006). Towards an idiothetic understanding of the role of social problem
solving in daily event, mood, and health experiences: A prospective daily diary
approach, British Journal of Health Psychology, 11 (3), 513-531.

Barberis N. and R. Thaler (2001). A survey of behavioral ﬁnance. Working Paper,
University of Chicago.

Bates D. V. and R. Sizto (1983). Relationship between air pollutant levels and hospital
admissions in southern Ontario, Canadian Journal of Public Health, 74, 117-122.

Behnke R. R. and M. J. Beatty (1981). A cognitive-physiological model of Speech
anxiety, Communication Monographs, 48, 158-163.

Behnke R. R., L. W. Carlile, and D. Lamb (1974). A psychophysiological study of state
and trait anxiety in public speaking, Central States Speech Journal, 25, 249-253.

Biggeri A., P. Bellini, and B. Terracini (2001). Meta-analysis of the Italian studies on
short-term effects of air pollution, Epidemiologia e Prevenzione, 25 (2), 1-71.

Blomberg, A., M. T. Krishna, R. Helleday, M. Soderberg, M. C. Ledin, F. J. Kelly, A. J.
Frew, S. T. Holgate, and T. Sandstrom (1999). Persistent airway inﬂammation but
accommodated antioxidant and lung ﬁrnction responses after repeated daily

exposure to nitrogen dioxide, American Journal of Respiratory and Critical Care
Medicine, 159, 536—543.

139

Bollerslev T. (1986). Generalized autoregressive conditional heteroskedasticity, Journal
of Econometrics, 31, 307-328.

Bound J., D. A. Jaeger, and R. M. Baker (1995). Problems with Instrumental Variables:
Estimation When the Correlation between the Instruments and the Endogenous
Explanatory Variable is Weak, Journal of the American Statistical Association,
90, 443-450.

Boyer P. (1992). When time shall be no more: prophecy belief in modern American
culture, Cambridge: Belknap Press of Harvard University Press.

Branch B. (1977). A tax loss trading rule, Journal of business, 50, 198-207.

Breivik G., W. T. Roth, and P. E. Jorgensen, (1998). Personality, psychological states and
heart rate in novice and expert parachutists, Personality and Individual
Diﬂerences, 25, 365-380.

Briere J., Downes A., and J. Spensley (1983). Summer in the city: Urban weather
conditions and psychiatric emergency-room visits, Journal of Abnormal
Psychology, 92 (1), 77-80.

Brown R, A. Chua, and J. Mitchell (2002). The inﬂuence of cultural factors on price
clustering: evidence from Asian-Paciﬁc stock markets, Paciﬁc-Basin Finance
Journal 10, 307-332.

Brown R, and J. Mitchell (2004). Culture and stock price clustering: evidence from the
people’s Republic of China, Working paper, University of Michigan.

Brunekreef B. (1997). Air pollution and life expectancy: is there a relation? Occupational
and Environmental Medicine, 54, 781—784.

Bruner G. C. (1990). Music, mood, and marketing, Journal of Marketing, 54 (4), 94-104.

Bullinger M. (1989). Psychological effects of air pollution on healthy residents—A time-
series approach, Journal of Environmental Psychology, 9 (2), 103-118.

Bullinger M. (1990). Environmental stress: effects of air pollution on mood,
neuropsychological function and physical state. In: Puglisi-Allegra S. and A.
Oliverio (1990). Psychobiology of stress, NATO ASI Series vol. 54, Kluwer
Academic Publishers, Dordrecht.

Byme, D. and G. L. A. Clore (1970). A reinforcement model of evaluative responses,
Personality, 1, 103-1 28.

140

Cadum E., G. Rossi, D. Mirabelli, M. A. Vigotti, P. Natale, L. Albano, G. Marchi, V. Di
Meo, R. Cristofani, and G. Costa (1999). Air pollution and daily mortality in
Turin, 1991-1996, Epidemiologia e Prevenzione, 23 (4), 268-76.

Calvo M. G., A. Szabo, And J. Capafons (1996). Anxiety and heart rate under
psychological stress : The effects of exercise-training, Anxiety, Stress, and
Coping, 9(4), 321-337.

Campbell C. (1996). Half-belief and the paradox of ritual instrumental activism: a theory
of modern superstition, British Journal of Sociology 47, 151-166.

Campbell N. A. and J. Gipps (1975). The Inﬂuence of Meteorological Conditions on Air
Pollution, Australian Science Teacher Journal, 21 (2), 67-73.

Cao M., and J. Wei (2004). Stock market returns: a note on temperature anomaly,
Journal of Banking and Finance 29, 1559- l 573.

Castro R. (1995). The logic of a traditional health belief: solar eclipse and pregnancy in
Ocuituco, Mexico, Salud Publica de Mexico 37, 329-338.

Chamberlain T. W., C. S. Cheung, and C. C. Kwan (1991). The Friday the thirteenth
effect: myth or reality?, Quarterly Journal of Economics and Business 30, 111-
117.

Chang T., C. Nieh, M. Yang, and T. Yang (2006). Are stock market returns related to the
weather effects? Empirical evidence from Taiwan, Physica A, 364, 343-354.

Chattopadhyay P. K., B. Som, and P. Mukhopadhyay (1995). Air Pollution And Health
Hazards In Human Subjects: Physiological And Self-Report Indices, Journal of
Environmental Psychology, 15, 327—33 1.

Chay K. Y. and M. Greenstone (2003). The impact of air pollution on infant mortality:
evidence from geographic variation in pollution shocks induced by a recession,
Quarterly Journal of Economics, 1 l8 (3) , 1 121-1 167.

Chen D. and J. Haviland-Jones (1999). Rapid mood change and human Odors, Physiology
and Behavior., 68 (1-2), 241-50.

Chen T. M., J. Gokhale, S. Shofer, W. G. Kuschner (2007). Outdoor air pollution:
nitrogen dioxide, sulfur dioxide, and carbon monoxide health effects, The
American Journal of the Medical Sciences, 333 (4), 249-256.

Chinn S., V. Florey, I. G. Baldwin, and M. Gorgol (1981). The relation of mortality in

England and Wales 1969-73 to measurements of air pollution, Journal of
Epidemiology and Community Health, 35, 174-179.

141

Ciccone G., F. Faggiano, and P. Falasca (1995). 802 air pollution and hospital
admissions in Ravenna: a case-control study, Epidemiologia e Prevenzione, 19
(62), 99-104.

Cohen A. R., J. V. Brehm, and D. Latane (1959). Choices of strategy and voluntary
exposure to information, Journal of Personality 27, 63-73.

Cohen S., G. A. Kaplan, and J. T. Salonen (1999). The Role of Psychological
Characteristics in the Relation Between Socioeconomic Status and Perceived
Health, Journal of Applied Social Psychologi, 29 (3), 445-468.

Coleman C. (1993). The Inﬂuence of Mass Media and Interpersonal Communication on
Societal and Personal Risk Judgments, Communication Research, 20, 611 — 628.

Colls J. (2002). Air pollution, London, New York: Spon Press.

Cotton F. A., G. Wilkinson, C.A. Murillo, M. Bochmann (1999). Advanced Inorganic
Chemistry, 6th ed. Wiley-Interscience, New York, 1999

Coutts J. A. (1999). Friday the thirteenth and the ﬁnancial times industrial ordinary
shares index, Applied Economics Letters 6, 35-37.

Cropper M. L., N. B. Simon, A. Alberini, and P. K. Sharma (1997). The health effects of
air pollution in Dehli, India, Polici Research Working Paper No. 1860, The World
Bank.

Dalton P. (1999). Cognitive inﬂuences on health symptoms from acute chemical
exposure, Health Psychology, 18, 579-590.

Dalton P., C. J. Wysocki, M. J. Brody, and H. J. Lawley (1997). The inﬂuence of
cognitive bias on the perceived odor, irritation, and health symptoms from

chemical exposure, International Archives of Occupational and Environmental
Health, 69, 407-417.

Damasio A. (1994). Descartes’ Error: Emotion, Reason, and the Human Brain. New
York: Putnam.

Daniels M. J., Dominici F., Samet J. M. and S. L. Zeger (2000). Estimating Particulate
Matter-Mortality Dose-Response Curves and Threshold Levels: An Analysis of
Daily Time-Series for the 20 Largest US Cities, American Journal of
Epidemiology, 152, 397 - 406.

Danuser B. (2001). Candidate physiological measures of annoyance from airborne
chemicals, Chemical Senses, 26 (3), 333-337.

142

Derriennic F., S. Richardson, A. Mollie, and J. Lellouch (1989). Short-term effects of
sulphur dioxide pollution on mortality in two French cities, International Journal
of Epidemiology, 18, 86-197.

Dichev I. D., and T. D. Janes (2003). Lunar cycle effects in stock returns, Journal of
Private Equity 6, 8-29.

Dickens C. R. (1996). The Indian eclipse, 1995, Journal of the British Astronomical
Association 106, 6-7.

Do Q., and T. D. Phung (2006). Superstition, family planning, and human development,
World Bank Policy Research working paper No. 4001.

Dockery D. W. (2001). Epidemiologic Evidence of Cardiovascular Effects of Particulate
Air Pollution, Environmental Health Perspectives, 109 (4), 483-486.

Dockery D. W. and C. A. Pope HI (1996). Acute respiratory effects of particulate air
pollution,Annual Review of Public Health, 15, 107-132.

Dockery D. W., C. A. Pope, R. E. Kanner, G. Martin Villegas, and J. Schwartz (1999).
Daily changes in oxygen saturation and pulse rate associated with particulate air
pollution and barometric pressure, Research Report (Health Effects Institute), 83,
1-19.

Dueker M. J. (1997). Strengthening the Case for the Yield Curve as a Predictor of
US. Recessions, Federal Reserve Bank of St. Louis Review 79, 41-51.

Dyl E. (1977). Capital gain taxation and year-end stock market behavior, Journal of
Finance, 32, 165-175.

Dyl E. A., and E. D. Maberly (1988). The anomaly that isn’t there: a comment on Friday
the thirteenth, Journal of Finance 18,1285-1286.

Engle R. (1982). Autoregressive Conditional Heteroscedasticity with Estimates of
Variance of United Kingdom Inﬂation, Econometrica 50, 987-1008.

Erbas B. and R. J. Hyndman (2001). Statistical Methodological Issues in Studies of Air
Pollution And Respiratory Disease, Working Paper 6/2001, Department Of
Econometrics And Business Statistics, Monash university, Australia.

Espenak F. (1987). Fifty Year Canon of Solar Eclipses: 1986—2035, NASA RP-l 178,
Sky Publishing Corp., Cambridge.

Estrella A., and F. S. Mishkin (1997). The Predictive Power of the Term Structure of

Interest Rates in Europe and the United States: Implications for the European
Central Bank, European Economic Review 41, 1375—1401.

143

Estrella A., and F. S. Mishkin (1998). Predicting U.S. Recessions: Financial Variables as
Leading Indicators. Review of Economics and Statistics 80, 45—61.

Estrella A., and G. A. Hardouvelis (1991). The Term Structure as a Predictor of Real
Economic Activity, Journal of Finance 46, 555—576.

Evans G. W. and S. Cohen (1987). Environmental stress. In: Stokols D. and I. Altman,
eds. Handbook of environmental psychology. New York: Krieger, 571-609.

Evans G. W., S. D. Colome, and D. F. Shearer (1988). Psychological reactions to air
pollution, Environmental Research, 45 (1), 1-15.

Field A. P. and H. Schorah (2007). The verbal information pathway to fear and heart rate
changes in children, Journal of Child Psychology and Psychiatry, OnlineEarly
Article.

Folinsbee, L. J. (1992). Does nitrogen dioxide exposure increase airway responsiveness?,
Toxicology and Industrial Health, 8, 273—283.

Forgas J. P. (1995). Mood and judgment: The affect infusion model (Aim), Psychological
Bulletin, 117, 39—66.

Forsberg B., N. Stjemberg, and S. Wall (1997). People can detect poor air quality well
below guideline concentrations: A prevalence study of annoyance reactions and
air pollution from traffic, Occupational and Environmental Medicine, 54 (l), 44-
48.

Frewer L. J ., S. Miles, and R. Marsh (2002). The Media and Genetically Modiﬁed Foods:
evidence in Support of Social Ampliﬁcation of Risk, Risk Analysis, 22, 701-711.

Frijda N. (1988). The laws of emotion, Cognition and Emotion, l, 235— 258.

Fusco D., F. Forastiere, P. Michelozzi, T. Spadea, B. Ostro, M. Area, and C. A. Perucci
(2001). Air pollution and hospital admissions for respiratory conditions in Rome,
Italy, European Respiratory Journal, 17 (6), 1 143-50.

Gallup (1978). Surprising number of Americans believe in paranormal phenomena. In
The Gallup Poll (release, June 15). Princeton, New Jersey.

Gallup (1990). Belief in psychic and paranormal phenomena widespread among
Americans. In The Gallup Poll (News Service, August 6). Princeton, New Jersey.

Georgiades A., A. Sherwood, E. C. D. Gullette, M. A. Babyak, A. Hinderliter, R. Waugh,
D. Tweedy, L. Craighead, R. Bloomer, and J. A. Blumenthal (2000). Effects of
Exercise and Weight Loss on Mental Stress—Induced Cardiovascular Responses in
Individuals With High Blood Pressure, Hypertension, 36, 171-176.

144

Gibbons M. R. and P. Hess (1981). Day of the Week Effects and Asset Returns, The
Journal of Business, 54 (4), 579-596.

Goetzmann W. N. and N. Zhu (2005). Rain or Shine: Where is the Weather Effect?,
European Financial Management, 11 (5), 559-578.

Goldman I. (1940). Cosmological beliefs of the Cubeo Indians, Journal of American
Folklore 53, 242-247.

Gong H., W. S. Linn, S. L.Terrell, K. W. Clark, M. D. Geller, K. R. Anderson, W. E.
Cascio, and C. Sioutas (2004). Altered heart-rate variability in asthmatic and

healthy volunteers exposed to concentrated ambient coarse particles, Inhalation
Toxicology, 16 (6-7), 335-43.

Greenlee R. F. (1945). Folktales of the Florida Seminole, Journal of American Folklore
58, 138-144.

Gulas C. S. and P. H. Bloch (1995). Right under our noses: Ambient scent and consumer
responses, Journal of Business and Psychology, 10, 87-98.

Haberle K-H., I. Reiter, K. Patzner, C. Heyne, and R. Matyssek (2001). Switching the
light off: a break in photosynthesis and sap ﬂow of forest trees under total solar
eclipse, Meteorologische Zeitschrift 10, 201-206.

Hamilton J. D., and D. H. Kim (2002). A reexamination of the predictability of
economic activity using the yield spread, Journal of Money, Credit, and
Banking 34, 340-360.

Hand W. D. (1961-1964), Popular beliefs and superstitions from North Carolina. Vol. 6-7
of The Frank C. Brown collection of North Carolina folklore, Durham, NC: Duke
University Press.

Hand W. D., and J. E. Talley (1984), Popular beliefs and superstitions from Utah,
collection Anthon S. Cannon and others, Salt lake city: University of Utah Press.

Hansen A. C. and H. K. Selte (1997). Air Pollution and Sick-leaves — is there a
Connection? A Case Study using Air Pollution Data from Oslo, Discussion Papers
No. 197, Statistics Norway.

Hastings P. D., C. Zahn-Waxler and B. A. Usher (2007). Cardiovascular and affective
responses to social stress in adolescents with internalizing and externalizing
problems, International Journal of Behavioral Development, 31, 77-87.

Hausmann J.A., B. Ostro, and DA. Wise (1984). Air Pollution and Lost Work, Working
Paper No. 1263, National Bureau of Economic Research, Cambridge.

145

Henderson J. V. (1996). Effects of air quality regulations, American Economic Review,
86 (4), 789-813.

Henslin M. J. (1967). Craps and magic, American Journal of Sociology 73, 316-330.

Hirshleifer D. (2001). Investor psychology and asset pricing, Journal of Finance 56,
1533-1597.

Hirshleifer D., and T. Shumway (2003). Good day sunshine: stock returns and the
weather, Journal of Finance 58, 1009-1032.

Huber P. J. (1967). The behavior of maximum likelihood estimates under non-standard
conditions, Proceeding of the Fifth Berkeley Symposium on Mathematical
Statistics and Probability l: , 221-233.

International Herald Tribune (2007). Made in Italy: fashion, food, Fiat, pollution.
Published: June 11, 2007.

Isen A. M., T. E. Shalker, M. Clark, and L. Karp (1978). Affect, accessibility of material
in memory, and behavior: A cognitive loop?, Journal of Personality and Social
Psychology, 36, 1—12.

Jarvis P. (1980). Toward a social understanding of superstition, Social Compass 27, 285-
295.

Jones T. F., A. S. Craig, D. Hoy, E. W. Gunter, D. L. Ashley, D. B. Barr, J. W. Brock,
and W. Schaffner (2000). Mass psychogenic illness attributed to toxic exposure at
a high school, New England Journal of Medicine, 342, 96-100.

Kahneman D. and M. W. Riepe (1998). Aspects of Investor Psychology: Beliefs,
Preferences, and Biases Investment Advisors Should Know About, Journal of
Portfolio Management, 24 (4), 52-66.

Kahneman D., and A. Tversky (1979). Prospect theory: an analysis of decision under
risk, Econometrica 37, 263-292.

Kaku K. (1974). Increased induced abortion rate in 1966, an aspect of a Japanese folk
superstition, Annals of Human Biology 2, 1 l l-115.

Kamstra M. J., L. A. Kramer, and M. D. Levi (2000). Losing sleep at the market: The
daylight-savings anomaly, American Economic Review, 90 (4), 1005—1011.

Kamstra M. J., L. A. Kramer, and M. D. Levi (2003). Winter Blues: A sad stock market
cycle, American Economic Review, 93 (1), 324—343.

146

Kasperson R. E., O. Renn, P. Slovic, H. S. Brown, J. Emel, R. Goble, J. X. Kasperson,
and S. Ratick (1988). The Social Ampliﬁcation of Risk: A Conceptual
Framework, Risk Analysis, 8 (2), 177-187.

Keef S. P. and M. L. Roush (2003). The weather and stock returns in New Zealand,
Quarterly Journal of Business Economics, 41 (1/2), 61-79.

Keshavan M. S., B. N. Gangadhar, R. U. Gautam, V. B. Ajit, and R. L. Kapur (1981).
Convulsive threshold in humans and rats and magnetic ﬁeld changes: observations
during total solar eclipse, Neuroscience Letters 22, 205-208.

Klaeboe R., M. Kolbenstvedt, J. Clench-Aas,, and A. Bartonova, (2000). Oslo trafﬁc
study—part 1: an integrated approach to assess the combined effects of noise and
air pollution on annoyance, Atmospheric Environment, 34 (27), 4727—4736.

Knasko S. C. (1993). Performance, mood, and health during exposure to intermittent
odors, Archives of Environmental Health, 48 (5), 305-8.

Knasko S. C., A. N. Gilbert, and J. Sabini, (1990). Emotional state, physical well-being,
and performance in the presence of feigned ambient odor, Journal of Applied
Social Psychology, 20, 1345-1357.

Ko W., L. Yurning, and J. Erickson (1997). A New Look at the Monday Effect, The
Journal of Finance, 52 (5), 2171-2186.

Koenig J. Q., D. S. Covert, S. G. Marshall, G. Van Belle, W. E. Pierson (1987). The
effects of ozone and nitrogen dioxide on pulmonary ﬁmction in healthy and in
asthmatic adolescents. The American review of respiratory disease, 136, 1152—
1157.

Kolb R. W., and R. J. Rodriguez (1987). Friday the thirteenth: “Part VII” — A note,
Journal ofFinance 17, 1385-1387.

Kotler P. (1973-1974). Atmospherics as a marketing tool, Journal of Retailing, 49, 48-64.

Kotrappa P., Y. S. Mayya, S. K. Dua, and P. C. Gupta (1981). Increase in ground level
concentration of radon during a solar eclipse, Health Physics. 40, 559-560.

Kramer W., and R. Runde (1997). Stocks and the weather: an exercise in data mining or
yet another capital market anomaly?, Empirical Economics 1 1, 63 7-64 1.

Krivelyova A. and C. Robotti (2003). Playing the ﬁeld: Geomagnetic storms and

international stock markets. Working Paper No. 2003-5b, Federal Reserve Bank
of Atlanta.

147

Krzyzanowsli M. and B. Wojtymiak (1982). Ten-year Mortality in Sample of an Adult
Population in Relation to Air Pollution, Journal of Epidemiology and Community
Health, 36, 262-268.

Lagorio S., F. Forastiere, R. Pistelli, I. Iavarone, V. Fano, R. A. Incalzi, S. Basso, R. T.
Benedetto, A. M. Della Corte, L. Fuso, C. Maiolo, S. Sammarro, M. Serra, S.
Spadaro, L. M. Tramaglino, G. Cattani, G. Stacchini, A. Marconi, G. Ziemacki,
and B. Ostro (2003). Air pollution and cardiac and respiratory function in three
groups of patients, Annali dell 'Istituto Superiore di Sanita, 39 (3), 395-404.

Lang P. J., B. G. Melamed, and J. Hart (1970). A psychophysiological analysis of fear
modiﬁcation using an automated desensitization procedure, Journal of Abnormal
Psychology, 76, 220-234.

Lang, P.J. (1971). The application of psychophysiological methods to the study of
psychotherapy and behavior modiﬁcation. In A. E. Bergin & S.L. Garﬁeld (Eds),
Handbook of psychotherapy and behavior change (pp. 75—125). New York: John
Wiley & Sons.

Lange L. J. and R. Fleming (2005). Cognitive Inﬂuences on the Perception of Somatic
Change During a Feigned Chemical Release, Journal of Applied Social
Psychology, 35 (3), 463-486.

Le Tertre A., J. Schwartz, and G. Touloumi (2005). Empirical Bayes and adjusted
estimates approach to estimating the relation of mortality to exposure of PM”),

Risk Analysis, 25 (3), 711-718.

Levenson R. W. (2003). Blood, sweat, and fears: The autonomic architecture of emotion,
Annals of the New York Academy of Sciences, 1000, 348—366.

Ling L., T. D. Ruddy, M. Dalipaj, M. Szyszkowicz, H. You, R. Poon, A. Wheeler, and R.
Dales (2007). Inﬂuence of Personal Exposure to Particulate Air Pollution on
Cardiovascular Physiology and Biomarkers of Inﬂammation and Oxidative Stress

in Subjects With Diabetes, Journal of Occupational and Environmental Medicine,
49 (3), 258-265.

Livneh H. and R. A. Antonak (1994). Psychosocial reactions to disability: a review and

critique of the literature. Critical Reviews in Physical Rehabilitation Medicine, 6,
1-100.

Lo A., H. Mamaysky, and J. Wang (2000). Foundations of Technical Analysis:
Computational Algorithms, Statistical Inference, and Empirical Implementation,
NBER Working paper No. 7613.

Loewenstein G. (2000). Emotions in economic theory and economic behavior, American
Economic Review, 65, 426—432.

148

 

Loewenstein G., E. U. Weber, C. K. Hsee, and N. Welch (2001). Risk as feelings,
Psychological Bulletin, 127, 267-286.

Loughran T., and P. Schultz (2004). Weather, stock returns and the impact of localized
trading behavior, Journal of Financial and Quantitative Analysis 39, 343-364.

Love G. J., S. P. Lan, C. M. Shy, and W. B. Riggan (1982). Acute respiratory illness in
families exposed to nitrogen dioxide ambient air pollution in Chattanooga,
Tennessee, Archives of Environmental Health, 37 (2), 75-80.

Low R. B., L. Bielory, A. I. Qureshi, V. Dunn, D. F. Stuhhniller, and D. A. Dickey
(2006). The relation of stroke admissions to recent weather, airborne allergens, air

pollution, seasons, upper respiratory infections, and asthma incidence, September
11, 2001, and day of the week, Stroke, 37 (4), 951-7.

Lucey B. M. and M. Dowling (2005a). The role of feelings in investor decision-making,
Journal of Economic Surveys, 19 (2), 211-237.

Lucey B. M. and M. Dowling (2005b). Weather, Biorhythms, beliefs and stock returns —
Some preliminary Irish evidence, Review of Financial Analysis, 14 (3), 337-355.

MacGregor D. G. and R. Fleming (1996). Risk perception and symptom reporting, Risk
Analysis, 16, 773-783.

Magari S. R., R. Hauser, J. Schwartz, P. L. Williams, T. J. Smith, and D. C. Christiani,
(2001). Association of Heart Rate Variability With Occupational and
Environmental Exposure to Particulate Air Pollution, Circulation, 104, 986-998.

Maier S. F. and L. R. Watkins (1998). Cytokines for Psychologists: Implications of
Bidirectional Immune-to-Brain Communication for Understanding Behavior,
Mood, and Cognition, Psychological Review, 105 (1), 83-107.

Malinowski B. (1948). Magic, science and religion, Garden City, New Jersey: Doubleday
and Co.

Mann L. (1992). Stress, affect, and risk taking In J. Frank Yates, ed.: Risk-taking
Behavior (Wiley, Chichester).

Maynard D., B. A. Coull, A. Gryparis, and J. Schwartz (2007). Mortality risk associated
with short-term exposure to trafﬁc particles and sulfates, Environmental Health
Perspectives, 115 (S), 751-755.

Maynard, A., and P. Phillips (2001). Rethinking an old empirical puzzle; the forward
premium paradox, Journal of Applied Econometrics, 16 (6), 671-708.

149

 

McCarthy J. and R. Goffm (2004). Measuring job interview anxiety: beyond weak knees
and sweaty palms, Personnel Psychology, 57, 607-637.

McCormick J. (1989). Acid Earth: the global threat of acid pollution, 2nd edition, London:
Earthscan.

McElrea H. and L. Standing (1992). Fast music causes fast drinking, Perceptual and
Motor Skills, 75, 362-370.

Mehra R. and R. Sah (2002). Mood ﬂuctuations, projection bias and volatility of equity
prices, Journal of Economic Dynamics and Control, 26, 869—887.

Mezzacappa E., R. E. Tremblay, D. Kindlon, J. P. Saul, L. Arseneault, J. Seguin, R. O.
Pihl, and F. Earls (1997). Anxiety, antisocial behavior, and heart rate regulation in
adolescent males, Journal of Child Psychology and Psychiatry, 38 (4), 457-69.

Michelozzi P., Forastiere F., Fusco D., Perucci C. A., Ostro B., Ancona C. and G. Pallotti
(1998). Air pollution and daily mortality in Rome, Italy, Occupational
Environmental Medicine, 55 (9), 605-10.

Milan W. (2000). Fear and Awe: eclipses through the ages, http://www.space.com/, 18
January 2000.

Milliman R. E. (1982). Using background music to affect the behavior of supermarket
shoppers, Journal of Marketing, 46, 86-91.

Morgan G., Corbett S. and J. Wlodarcyzk (1998). Air pollution and daily mortality in
Sydney, Australia, 1989-1993, American Journal of Public Health, 88, 759-764.

Morris W. M. (2000). Some thoughts about mood and its regulation, Psychological
Inquiry, ll (3), 200-202.

Morton T. A. and J. M. Duck (2001). Communication and Health Beliefs: Mass and
Interpersonal Inﬂuences on Perceptions of Risk to Self and Others,
Communication Research, 28 (5), 602 -626.

Mossetti S., S. P. Angius, and E. Angelino (2005). Assessing the impact of particulate

matter sources in the Milan urban area, International Journal of Environment and
Pollution, 24 (1-4), 247-259.

Murphy J. J., M. A. Delucchi, D. R. McCubbin, and H. J. Kim (1999). The cost of crop
damage caused by ozone air pollution from motor vehicles, Journal of

Environmental Management, 55, 273-289.

North A. C. and D. J. Hargreaves (1996). The effects of music on responses to a dining
area, Journal of Environmental Psychology, 16, 55-64.

150

North A. C. and D. J. Hargreaves (1997). Music and consumer behaviour. In D. J.
Hargreaves and A. C. North (Eds), The socialpsychology of music (pp. 268-
289). Oxford, UK: Oxford University Press.

North A. C. and D. J. Hargreaves (1998). The Effect of Music on Atmosphere and
Purchase Intentions in a Cafeteria, Journal of Applied Social Psychology, 28 (24),
2254-2273.

O’Mahoney M. (1978). Smell illusions and suggestion: reports of smells contingent on
tones played on radio and television, Chemical Senses, 3, 183-189.

OECD (1994). Intemalising the Social Cost of Transport, ECMT, Paris.

Ostblom G. and E. Samakovlis (2004). Costs of Climate Policy when Pollution Affects
Health and Labour Productivity: A General Equilibrium Analysis Applied to
Sweden, Working Paper No. 93, The National Institute of Economic Research,
Stockholm.

Ostro B. (1983). The Effects of Air Pollution on Work Loss and Morbidity, Journal of
Environmental Economics and Management, 10, 371-382.

Ostro B. D. (1994). Estimating the health effects of air pollutants: a method with an
application to Jakarta, Policy Research Working Paper No. 1301, The World
Bank.

Ostro B. D. and S. Rothschild (1989). Air pollution and acute respiratory morbidity: An
observational study of multiple pollutants, Environmental Research , 50 (2), 238-
247.

Ostro B. D., J. M. Sanchez, C. Aranda, and G. S. Eskeland (1995). Air pollution and
mortality: results from Santiago, Chile, Policy Research Working Paper No. 1453,
The World Bank.

Panyacosit L. (2000). A review of particulate matter and health: focus on developing
countries, Interim Report IR-00-005, International Institute for Applied Systems
Analysis, Austria.

Paranjpe D. A., and V. K. Sharma (2005). Evolution of temporal order in living
organisms, Journal of Circadian Rhythms 3, 1-13.

Pardo A., and E. Valor (2003). Spanish stock returns: where is the weather effect?,
European Financial Management 9, 1 17-126.

Peng R. D., F. Dominici, and T. A. Louis (2005). Model choice in time series studies of

air pollution and mortality, Department of Biostatistics Working Paper No. 55,
John Hopkins University.

151

Peng R. D., F. Dominici, R. Pastor-Barriuso, S. L. Zeger, and J. M. Samet (2004).
Seasonal analyses of air pollution and mortality in 100 US. cities, Department of
Biostatistics Working Paper No. 41 , John Hopkins University.

Perkins B. (2002). Bottom Line Conjures Up Realty's Fear Of 13, Realty Times,
09/ 13/2002, retrieved on 04/14/2008.

Peters A., S. Perz, A. Dﬁring, J. Stieber, W. Koenig, and H.-E. Wichmann (1999).
Increases in Heart Rate during an Air Pollution Episode, American Journal of
Epidemiology, 150 (10), 1094-1098.

Pia P. (1997). Il mercato azionario italiano, G. Giappichelli Editore, Torino.

Plosser C. I., and K G. Rouwenhorst (1994). International Term Structures and Real
Economic Growth, Journal of Monetary Economics, 33, 133—155.

Ponka A. (1990). Absenteeism and respiratory disease among children and adults in
Helsinki in relation to low-level air pollution and temperature, Environmental
Research, 52, 3446.

Pope C. A. III (1996). Adverse health effects of air pollutants in a nonsmoking
population, Toxicology, 111 (1-3), 49-155.

Pope C. A. III, R. L. Verrier, E. G. Lovett, A. C. Larson, M. E. Raizenne, R. E. Kanner, J.
Schwartz, G. M. Villegas, D. R. Gold, and D. W. Dockery (1999). Heart rate
variability associated with particulate air pollution, American Heart Journal, 138
(5), 890-899.

Pope C. A. III, R. T. Burnett, M. J. Thun, E. E. Calle, D. Krewski, K. Ito, and G. D.
Thurston (2002). Lung cancer, cardiopulmonary mortality, and long-term
exposure to ﬁne particulate air pollution, Journal of the American Medical
Association, 287, 1132—1 141.

Potter S. (1999). Nonlinear time series modeling: an introduction, Journal of Economic
Surveys, 13 (5), 505-528.

Rasul F., S. A. Stansfeld, C. L. Hart, C. Gillis, and G. D. Smith (2002). Common mental
disorder and physical illness in the Renfrew and Paisley (MIDSPAN) study,
Journal of Psychosomatic Research, 53, 1163—1170.

Rigobon R., and B. Sack (2006). Noisy macroeconomic announcements, monetary
policy, and asset prices, Working paper, Sloan School of Management, MIT.

Ross L. (1977). The intuitive psychologist and his shortcomings: Distortions in the

attribution process. In L. Berkowitz, ed.: Advances in Experimental Social
Psychology (Academic Press, New York, NY).

152

 

Roth W. T., G. B. Breivik, P. T. Jorgensen, and S. Hoﬁnann (1996). Activation in novice
and expert parachutists while jumping, Psychophysiology, 33, 63-77.

Rotko T., L. Oglesby , N. Kunzli. , P. Carrer , M. J. Nieuwenhuijsen, and M. Jantunen
(2002). Determinants of perceived air pollution annoyance and association
between annoyance scores and air pollution (PM2.5, N02) concentrations in the
European EXPOLIS study, Atmospheric Environment, 36, 4593—4602.

Rotton J. and J. Frey (1984). Psychological costs of air pollution: Atmospheric
conditions, seasonal trends, and psychiatric emergencies, Population &
Environment, 7 (1), 3-16.

Ruidavets J. B., S. Cassadou, M. Coumot, V. Bataille, M. Meybeck, and J. Ferrieres
(2005). Increased resting heart rate with pollutants in a population based study,
Journal of Epidemiology and Community Health, 59, 685-693.

Sagar A., M Bhattacharya, and V. Joon (2007). A Comparative Study of Air Pollution-
Related Morbidity Among Exposed Population of Delhi, Indian Journal of
Community Medicine, 32 (4), 268-273.

Salome C. M., N. J. Brown, G. B. Marks, A. J. Woolcock, G. M. Johnson, P. C.
Nancarrow, S. Quigley, and J. Tiong (1996). Effect of nitrogen dioxide and other
combustion products on asthmatic subjects in a home-like environment, European
Respiratory Journal, 9, 910—918.

Samakovlis E., Huhtala A., Bellander T. and M. Svartengren (2004). Air Quality and
Morbidity: Concentration-Response Relationships for Sweden, Working Paper
No. 87, National Institute of Economic Research.

Sartory G., S. Rachman, and S. Grey (1977). Investigation of relation between reported
fear and heart-rate, Behaviour Research and Therapy, 15, 435—438.

Saunders E. M. (1993). Stock prices and wall street weather, American Economic
Review, 83 (5), 1337-1345.

Scheibe E. Karl, and Theodore R. Sarbin (1965). Towards a Theoretical
Conceptualisation of Superstition, The British Journal for the Philosophy of
Science 16, 143-158.

Schildcrout J. S., L. Sheppard, T. Lumley, J. C. Slaughter, J. Q. Koenig, and G. G.
Shapiro (2006). Ambient Air Pollution and Asthma Exacerbations in Children: An
Eight-City Analysis, American Journal of Epidemiology, 164 (6), 505-517.

Schottenfeld R. S. (1992). Psychologic sequelae of chemical and hazardous materials

exposures. In: Sullivan J. B. and G. R. Krieger, eds. Hazardous materials
toxicology. Baltimore: Williams & Wilkins, 463-70.

153

Schwartz J., D. Wypij, D. Dockery, J. Ware, S. Zeger, J. Spengler, and B. Ferris Jr.
(1991). Daily diaries of respiratory symptoms and air pollution: Methodological
issues and results, Environmental Health Perspectives, 90, 181-187.

Schwartz J ., V. Hasselblad, and H. Pitcher (1988). Air pollution and morbidity: A further
analysis of the Los Angeles student nurses data, Journal of the Air Pollution
Control Association, 38, 158-162.

Schwarz N. (1990). Feelings as information: Informational and motivational functions of
affective states. In E. T. Higgins (ed.), Handbook of Motivation and Cognition,
Vol. 2. New York: Guildford Press, 527—561.

Schwarz N. and G. L. Clore (1983). Mood, misattribution, and judgments of well-being:
Informative and directive functions of affective states, Journal of Personality and
Social Psychology, 45, 513—523.

Schwert G. W. (1989). Why does stock market volatility change over time?, Journal of
Finance, 44, 1115-1153.

Seibold D. (1967). Moon lore from Nogales, Arizona, Western Folklore 26, 122-123.

Sexton K., R. Letz, and J. D. Spengler (1983). Estimating human exposure to nitrogen
dioxide: an indoor/outdoor modeling approach, Environmental Research, 32,
151—166.

Shiller R. J. (2006). Tools for ﬁnancial innovation: neoclassical versus behavioral
ﬁnance, The Financial Review, 41 , 1-8.

Shleifer A. (1999). Inefﬁcient markets: an introduction to behavioral ﬁnance, Oxford
university press.

Shuttleworth G. N., and P. H. Galloway (2002). Analysis of the United Kingdom solar
eclipse public health campaign 1999, Clinical and Experimental Ophthalmology
30, 308-312.

Skinner B. F. (1948). Superstition in the pigeon, Journal of Experimental Psychology 38,
168-172.

Smith M. J., M. J. Colligan, and J. J. Hurrell (1978). Three incidents of industrial mass
psychogenic illness: A preliminary report, Journal of Occupational Medicine, 20,
399-400.

Spash C. (1997). Assessing the economic beneﬁts to agriculture from air pollution
control, Journal of Economic Surveys, 11 (1), 47-70.

154

 

Stark R., and W. Bainbridge (1985). The future of religion. Berkeley: University of
California Press.

Stevenson M. D. (2000). The effect of superstition on the day of discharge from
maternity units in Northern Ireland: «a Saturday ﬂit is a short sit», Journal of
Obstetrics & Gynaecology 20, 139-142.

Strahilevitz M., A. Strahilevitz, and J. E. Miller (1979). Air pollutants and the admission
rate of psychiatric patients, American Journal of Psychiatry, 136 (2), 205-7.

Strand V., P. Salomonsson, J. Lundahl, and G. Bylin (1996). Immediate and delayed
effects of nitrogen dioxide exposure at an ambient level on bronchial
responsiveness to histamine in subjects with asthma. European Respiratory
Journal, 9, 733—740.

Strestik J. (1999). The response of the 11 August 1999 total solar eclipse in the
geomagnetic ﬁeld, Earth Moon & Planets 85, 561-566.

Summers T. A. and P. R. Hebert (2001). Shedding some light on store atrnospherics:
inﬂuence of illumination on consumer behavior, Journal of Business Research, 54
(2), 145-150.

Sunyer J ., J. Castellsague, M. Saez, A. Tobias, and J. M. Anto (1996). Air pollution and
mortality in Barcelona, Journal of Epidemiologt and Community Health, 50, 76-
80.

Thaler R. H. (1993). Advances in behavioral ﬁnance, Princeton University Press.

The New York Times (1989). Pollution: the risks for travelers. Published: August 6,
1989.

Torgler B. (2003). It is about believing: superstition and religiosity, CREMA working
paper No. 2003- 10.

Tufan E. and B. Hamarat (2004). Do Cloudy Days Affect Stock Exchange Returns:
Evidence from Istanbul Stock Exchange, Journal of Naval Science and
Engineering, 2 (1), 117-126.

Turley L. W. and R. E. Milliman (2000). Atmospheric Effects on Shopping Behavior: A
Review of the Experimental Evidence, Journal of Business Research, 49, 193—
211.

Vigotti M.A., G. Rossi, L. Bisanti, A. Zanobetti, and J. Schwartz (1996). Short Term

Effects of Urban Air Pollution on Respiratory Health in Milan, Italy, 1980-89,
Journal of Epidemiology & Community Health, 50, 71-75.

155

 

Wakeﬁeld S. E. and S. J. Elliott (2003). Constructing the News: The Role of Local
Newspapers in Environmental Risk Communication, The Professional
Geographer, 55, 218-228.

Weiss B. (1983). Behavioral toxicology and environmental health science: opportunity
and challenge for psychology, American Psychologist, 38, 1174-1187.

White H. (1980). A heteroskedastic-consistent covariance matrix estimator and a direct
test of heteroskedasticity, Econometrica, 48, 817-838.

Wilson D. (1977). Armageddon now! The premillenarian response to Russia and Israel
since 1917, Grand Rapids, MI: Baker book house.

Wojcik D. (1996). Embracing doomsday: faith, fatalism and apocalyptic beliefs in the
nuclear age, Western Folklore 55, 297-330.

Wojtusiak R. J., and Zoﬁa Majlert (1975a). Reactions of animals during the partial solar
eclipses in Poland in 1959, 1966, 1969 and 1971, Folio Biologica 23, 267-278.

Wojtusiak R. J., and Zoﬁa Majlert (1975b). Effect of partial eclipse of the sun on 15th
February 1961 on animal behavior, F olia Biologica 23, 245-266.

Wojtusiak R. J., and Zoﬁa Majlert (1976). Ethological and ecological analysis of animal
behavior during an eclipse of the sun, F olia Biologica 24, 191-211.

Wong T. W., W. S. Tam, T. S. Yu, and A. H. Wong (2002). Associations between daily
mortalities from respiratory and cardiovascular diseases and air pollution in Hong
Kong, China, Occupational and Environmental Medicine, 59 (1), 30-5.

Woo C. K., and R. H. F. Kwok (1994). Vanity, superstition and auction price, Economics
Letters, 44, 389-395.

WORLD HEALTH ORGANIZATION - Regional Office for Europe (2001).
Quantiﬁcation of the Health Effects of Exposure to Air Pollution, Report of a
WHO Working Group, Bilthoven, Netherlands, 20-22 November 2000.

WORLD HEALTH ORGANIZATION - Regional Ofﬁce for Europe (2006). Health risks
of particulate matter from long-range transboundary air pollution, Joint WHO /
Convention Task Force on the Health Aspects of Air Pollution, European Centre
for Environment and Health, Bonn Ofﬁce.

WORLD HEALTH ORGANIZATION (1979). International program on chemical safety,
Environmental health criteria 7, Photochemical oxidants, Geneva.

WORLD HEALTH ORGANIZATION (2003). Health aspects of air pollution with
particulate matter, ozone and nitrogen dioxide. Report on a WHO working group.

156

 

WHO, Bonn, Germany, EUR/03/5042688, http://www.euro.who.int/document/
/e79097.pdf.

WORLD HEALTH ORGANIZATION (2005). Air quality guidelines global update
2005, Report on a working group meeting, Bonn, Germany, 18-20 October 2005.

Yuan K., L. Zheng, and Q. Zhu (2006). Are investors moonstruck? Lunar phases and
stock returns, Journal of Empirical Finance 13, 1-23.

Zeidner M. and M. Schechter (1988). Psychological responses to air pollution: some
personality and demographic correlates, Journal of Environmental Psychology, 8,
191—208.

Zuidema T. and A. Nentjes (1997). Health Damage of Air Pollution: An Estimate of a

Dose-Response Relationship for the Netherlands, Environmental and Resource
Economics, 9, 291-308.

157

 

IIIIIIIIIIIIIIIIIIIIIIIIIIIIII

WWWWMWW