. fiwisu

v A . u z
“W .
a; T251. ...u
. z .
. fi%..

 

 

 

.Il. - L-UI.IL.>" 11":2.‘ <0l’tQ.‘.fVI-Hl§vfu.fi'r.f

 

 

This is to certify that the
dissertation entitled

Household Consumption and Labor Supply Response to
Economic Shocks in Russia

presented by

Ren Mu

has been accepted towards fulfillment
of the requirements for the

Doctoral degree in Economics

 

 

 

 

SJ? 7 0200 6%

Date

MSU is an Affirmative ActiorVEqual Opportunity Institution

 

PLACE lN RETURN Box to remove this checkout from your record.

TO AVOID FINES return on or before dat

e due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE

DATE DUE
will”? 20 210er

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJClRCIDatoDuopes-sz

HOUSEHOLD CONSUMPTION AND LABOR SUPPLY RESPONSE TO
ECONOMIC SHOCKS IN RUSSIA

By

Ren Mu

A DISSERTATION

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

DOCTOR OF PHILOSOPHY
Department of Economics

2004

ABSTRACT

HOUSEHOLD CONSUMPTION AND LABOR SUPPLY RESPONSE TO
ECONOMIC SHOCKS IN RUSSIA

By

Ren Mu

This dissertation consists of three chapters. The first two chapters are empirical
studies on household consumption and labor supply response to economic shocks,
using the data from the Russian Longitudinal Monitory Survey (RLMS). The third
chapter studies the econometric estimation methods in an unevenly spaced panel
data.

Chapter 1. ”Risk, Consumption, Wealth and Human Capital: Evidence from
Russia”. This paper investigates Russian households’ consumption response to
the income shocks. In particular, this paper examines the effects of education and
wealth on the ability of the households to smooth their consumption. A random
coefficient model of average treatment effect that allows for endogenous variables is
implemented in the estimation. After correcting for the possible sample attrition
by the inverse probability-weighting method, this paper finds that consumption
is only partially protected from idiosyncratic shocks. The analysis also provides
evidence that households in the wealthier group can smooth their consumption
better. In addition, education of the household members in the high asset group
increases their consumption smoothing ability while no education effect is detected

in the low asset group.

Chapter 2. ”Multiple Job Holdings As a Way to Smooth Consumption: Labor
Response to Wage Arrears Among Russian Couples”. This paper tests the hy-
pothesis that labor supply in the form of multiple job holdings was effective in the
attempts of households to maintain consumption when their income was declining
during the Russian economic transition period. The theoretical framework shows
that the possibility of holding secondary job(s) for an individual increases with
the possibility of getting wage arrears in their primary job. The paper applies
the ”Chamberlain” approach to a dynamic probit model of the decision to hold
multiple jobs. It finds that both husbands and wives are more likely to take sec-
ondary informal jobs when they have wage arrears shocks in their primary jobs.
The combined regression results from the reduced form and structural estimation
of households consumption suggest that the insignificant effect of wage arrears on
consumption reflect, to certain extent, the adjustment of secondary job holdings.

Chapter 3. ”Estimation With or Without Straight Exogeneity Assumption in
Unevenly Spaced Panel Data”. This paper studies the econometric estimation
methods for unevenly spaced data, which is very common in the survey data con-
ducted in developing countries. Classic minimum distance, and one-step GMM
estimators are used to impose the non-linear parameter restrictions in dynamic
models. The paper also shows that these two methods also can be applied to

estimations without strict exogeneity assumption.

To Xu and Chengcheng

ACKNOWLEDGEMENTS

I complete this dissertation with a keen awareness of and deepest gratitude for
the help and support from many people.

I owe a tremendous debt of gratitude to John Strauss, my mentor and major
professor. Over the past few years, Professor Strauss has played a substantial role
in my academic development and training as an economist. He has always been
generous with his time and his knowledge when it comes to my endless questions
at different stages of my research. He has never been short of encouragement at
times when I had doubts and worries. His guidance has made my graduate study
a rewarding experience, one that I can look back upon with fond memories. Most
importantly, he has set a model as a scholar, that I will always try to live up to
throughout my career.

I am also especially thankful to Professor Jeffrey Wooldridge. His art of teach—
ing has convinced me of the beauty of applied econometrics. Throughout the
undertaking of this dissertation, his valuable suggestions and insights made my
task much more interesting and easier. His great books in econometrics will con-
tinue to influence my way of thinking about and carrying out economics empirical
analysis.

I also owe thanks to Professor John Giles and Professor Jeff Biddle. Dr. Giles
stimulated my thinking about many of the ideas presented in this dissertation.
Equally importantly, I am grateful that he successfully persuaded me five years

ago that E. Lansing is a lot better place than Seattle. Dr. Biddle has carefully

and critically read different versions of my dissertation drafts. He has offered a
number of helpful comments and suggestions at each time of his reading.

I am also indebted to Professor Jeffrey Riedinger, who has given me valuable
supports in many ways. I enjoyed very much and learned a lot from working for
him as a research assistant in the past two years.

I would also like to acknowledge a number of professors who have influenced my
academic development, including Jack Meyer, David Neumark, John Gooddeeris,
Jay Choi, Steven Haider and James Stapleton. I am also thankful for the friendship
and encouragement from Professor Christine Amsler.

Many incredible fellow graduate students provide the warmth and support
throughout my study at MSU. Zhehui, Yanyan, Lebo, Pond, Olena, and Firman
are the best friends and colleagues that a person could ever have.

Finally, I would like to give a special thanks to my family. My father has always
encouraged me to do my best. My sister has never failed to give me loving advices
on different issues. My mother is a trustworthy source of strength and wisdom for
me. Her help with my baby made it possible for me to focus on my dissertation
and finish it. My husband, Xu, has been enthusiastic and effective in solving all my
computer-related problems, besides the fact that he is a good listener with great
patience and love. Our son, Chengcheng, joined us in our life journey six months
ago, and has since motivated me and brought us tremendous joy. In a concrete

way, this study is theirs.

vi

TABLE OF CONTENTS

LIST OF TABLES ix
LIST OF FIGURES xii
1 Risk, Consumption, Wealth and Human Capital: Evidence From
Russia 1
1.1 Introduction ............................... 1
1.2 Theoretical Framework and Empirical Implications ......... 6
1.2.1 A Full Consumption Insurance Model ............ 6
1.2.2 Empirical Implications and Test ................ 9
1.3 Data and Descriptive Statistics .................... 12
1.4 Sample Attrition: Evidence and Correction .............. 16
1.5 Endogenous Income Change ...................... 20
1.6 Results .................................. 23
1.7 Conclusions ............................... 29
BIBLIOGRAPHY .............................. 44
APPENDIX .................................. 47
2 Multiple Job Holdings As a Way to Smooth Consumption: Labor
Response to Wage Arrears Among Russian Couples 60
2.1 Introduction ............................... 60
2.2 Wage Arrears In Russia ........................ 65
2.2.1 The Causes of Wage Arrears .................. 65
2.2.2 The Distributions of Wage Arrears on the Primary Jobs . . 68
2.3 Theoretical Framework and Empirical Issues ............. 72
2.3.1 Theoretical Framework .................... 72
2.3.2 Empirical Specification ..................... 80
2.4 Data and Summary Statistics ..................... 86
2.4.1 Sample Creation ........................ 86
2.4.2 Summary Statistics ....................... 89
2.5 Empirical Results ............................ 92
2.5.1 Labor Supply Response to Wage Arrears—Secondary Job
Holdings and Job Changes ................... 92
2.5.2 Effects of Wage Arrears on Household Consumption ..... 95
2.6 Conclusion ................................ 97
BIBLIOGRAPHY .............................. 124
APPENDIX .................................. 127
3 Estimation With or Without Strict Exogeneity Assumption in
Unequally Spaced Panel Data 128
3.1 Introduction ............................... 128

vii

3.2 Estimation Under Strict Exogeneity Condition ............ 131

3.2.1 Static Model ........................... 131
3.2.2 Dynamic Model with Lagged Explanatory Variables ..... 132
3.3 Estimation Without Strict Exogeneity Condition ........... 141
3.3.1 Static Model ........................... 141
3.3.2 Dynamic Model ......................... 142
3.4 Conclusion ................................ 146
BIBLIOGRAPHY .............................. 148

viii

List of Tables

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

1.10

1.11

1.12

1.13

Household Real Income and Consumption Change Over Years and
Regions ................................. 31

Descriptive Statistics .......................... 32

Coefficient of Variation (CV) in Income and Consumption (per capita) 33

Descriptive Statistics in 1995 by Attrition .............. 34
Attrition Probability during 1995—2000 ................ 35
Predicting the Change of Income ................... 37
Consumption Smoothing for the Whole Sample ........... 38
Consumption Smoothing for the Low Asset Group .......... 40
Consumption Smoothing for the High Asset Group ......... 42
First Stage Regression ......................... 47

Regression Results With vs Without 1994 Data— Low Asset Group 49
Regression Results With vs Without 1994 Data— High Asset Group 51

Consumption Smoothing for Low Asset Group (Without Education
Interaction) ............................... 53

ix

1.14

1.15

1.16

1.17

2.1

2.2

2.3

2.4

2.5

2.6

2.7

2.8

2.9

2.10

Consumption Smoothing for High Asset Group (Without Education
Interaction) ............................... 55

Consumption Smoothing for Pooled Asset Group (With Education
Interaction; With Attrition Correction) ................ 57

Consumption Smoothing for Pooled Asset Group (With Asset In-
teraction; With Attrition Correction) ................. 58

Consumption Smoothing for Pooled Asset Group(With Education / Assets

Interaction; With Attrition Correction) ................ 59
Incidence and Level of Wage Arrears ................. 99
Wage Arrears by Occupation (before/ after 1998) ........... 104
Distribution of Employment States of Husband and Wife ...... 107
Individual and Household Characteristics by Wage Arrears ..... 108

Random Probit Estimation of Labor Response (Holding Secondary
Job) to First Time Wage Arrears ................... 110

Random Probit Estimation of Labor Response (Holding Secondary
Job) to Wage Arrears .......................... 112

Random Probit Estimation of Labor Response (Holding incidental
work) to First Time Wage Arrears ................... 114

Random Probit Estimation of Labor Response (Holding incidental
work) to Wage Arrears ......................... 116

Random Probit Estimation of Labor Response (Change of Job) to
First Wage Arrears ........................... 118

Random Probit Estimation of Labor Response (Change of Job) to
Wage Arrears .............................. 120

2.11 Consumption Regression (First Time Arrears) ............ 122
2.12 Consumption Regression (Wage Arrears) ............... 123

2.13 First Stage Regression of Multiple Job Holdings ........... 127

xi

List of Figures

2.1

2.2

2.3

2.4

2.5

2.6

2.7

2.8

Currently Owed Wage by Age (Men) ................. 100
Currently Owed Wage by Age (Women) ............... 100
Currently Owed Wage by Education (Men) .............. 101
Currently Owed Wage by Education (Women) ............ 101
Currently Owed Wage by Tenure (Men) ............... 102
Currently Owed Wage by Tenure (Women) .............. 102
Currently Owed Wage by Wage (Men) ................ 103
Currently Owed Wage by Tenure (Women) .............. 103

xii

Chapter 1

Risk, Consumption, Wealth and
Human Capital: Evidence From
Russia

1 . 1 Introduction

Several years into the economic and social transition in Russia, many households
have been in an environment of considerable economic uncertainty. The risk of
wage nonpayment to working men and women is pervasive and the incidence of
pension denial to retired people is high (Lokshin and Ravallion, 2000; Stillman
2001, Jensen and Richter 2002).1 In 1998, when the Russian government aban-
doned its defense of a strong ruble exchange rate against the dollar and defaulted

on the government domestic debt, one of the most serious financial crises hit the

 

1Based on Russian Longitudinal Monitory Survey (RLMS), the wage nonpayment rate in
1994, 1995, 1996, 1998 and 2000 is 38.8%,40.4%, 57%, 59% and 27% respectively. The rate of
pension nonpayment in these years is 3.3%,8.4%, 33.3%, 15.1% and 2.7% respectively.

economy.2 Two years after that economic shock, there has been substantial in-

crease in incomes, but the economic wellbeing of the households has not been fully

recovered (Mroz, Henderson and Popkin 2002).

Using the data from the Russian Longitudinal Monitory Survey (RLMS), we can
clearly see several features in the fluctuations of the households’ income and con-
sumption over the years (Table 1.1). Russian households have experienced several
years of income decline since 1994.3 In November 1998, three months after the
financial crisis, households income per capita dropped by 60% compared to 1994.
In 2000, income per capita increased by 24% from 1998 and consumption increased
by 19%, but consumption still remains the second lowest since 1994. Moreover,

this trend holds true even after geographic variations are controlled (Column 2, 4

of Table 1.1).

This paper asks a simple question: in such a tumultuous transition period, how
well have Russian households been dealing with their income shocks? Particu-
larly, this paper investigates the consumption smoothing ability of the households.
Previous research on consumption smoothing in Russia has found limited con-
sumption smoothing among households (Stillman 2001). Most of the empirical

tests for perfect consumption smoothing in developing countries find real and sig-

 

2 Official Russian statistics estimate the Russian real GDP contracted 4.6% in 1998. The
inflation rate, using the consumer price index, hit 84.4% and the interest rate on treasury bill
rose from 27.8% in May 1998 to 135.3% in August 1998. Unemployment rate had reached close
to 12% having increased from 7.0% in 1997. See Mroz and Popkin (1999), Cooper (1999).

3The panel data in RLMS has covered 1994, 1995, 1996 1998 and 2000. Russian GDP grew
slightly only in 1997 (0.4%), the first case since 1992. See Lokshin and Ravallion (2000)

nificant consumption smoothing, that is, for many households consumption does
not track household income particularly well regardless of the lack of efficient fi-
nancial infrastructure and well-functioning social security, but the studies also re—
ject the existence of full consumption insurance (Townsend 1995, Morduch 1995).
At the same time, empirical studies have also found that households with differ-
ent characteristics respond differently to income shocks. Wealthy households or
households with more assets such as land are found to be better insured against
income shocks (Townsend 1994, Morduch 1995, Jalan and Ravallion 1999). This
paper looks at the potential effect of education on household consumption smooth—
ing. The households with high—education members might, for instance, be able to
better obtain information and plan their expenditure, thus shielding consumption
from short-run fluctuations in individual income. This education effect is particu-
larly interesting is because it provides one piece of empirical evidence to test the
hypothesis that the ability of to deal successfully with economic disequilibria is
enhanced by education.4 Schultz (1975) argues that educated individuals adapt
more easily as economic circumstances change, using assets more efficiently, ob-
taining better credit arrangements and exploiting new income opportunities more
quickly. Under this hypothesis, we expect to see that households with higher lev-

els of human capital would have more allocative efficiency and thus do better in

 

4Welch (1970)stressed that the role of education in production may directly contribute to
physical product, which is the “worker effect ”of education. On the other hand, increased
education may enhance a worker’s ability to acquire and decode information about costs and
productive characteristics of the other inputs. This is “allocative effect ”of education. Schultz
(1975) contributes the enhanced ability of dealing with disequilibrium to the “allocative effect of
”education.

smoothing their consumption against income shocks.5

The identification of this effect is complicated because education enhances the
earning and thus the wealth of the households, and the wealth in turn enhances
the consumption smoothing ability of the households. To separate the education
effect, which might be over and above the wealth effect, I stratify the sample into
two wealth groups. Further, in each wealth group, the impact of income change
on consumption is allowed to depend on the maximum education level of adult
household members. Due to the endogeneity of income change and the fact that the
coefficient of income change on consumption change may differ across households
with differing human capital, I implement a two-step method using a random
coefficient model following Wooldridge (2002b). At the first step, income changes
of the households are predicted by exogenous shocks. The second step involves
using the predicted income change and the predicted income change interacted
with education variable as instrumental variables to estimate the main effect of

income change and the education effect on consumption smoothing.

The empirical work in this paper uses data from phase two of the Russian Lon-
gitudinal Monitoring Survey (RLMS), for the years 1994, 1995, 1996, 1998 and

2000 (Rounds IV—VIII).6 This survey is designed as a repeated sample of house-

 

5 Glewwe and Hall (1998) found that households with better educated heads are less vulner-
able to macro-economic shocks in Peru. The vulnerability is measured by change in per capita
consumption.

6 The regression analysis doesn’t include the 1994 data because the information of years of
education of the household members is not available.

hold dwellings, not of each household itself. Evidence is found that the households
move out and thus attrition from the survey are intrinsically different from the
households which remain in the survey. To address this problem, a predicted
weight associated with attrition probability is assigned to each household. This
inverse probability weighting scheme corrects sample attrition and leads to consis-

tent estimators (Wooldridge 2002a; Wooldridge 2002c )

The paper proceeds as follows. Section 2 reviews the theory of full consumption
insurance and its testable implications. Section 3 summarizes the data used to
test the hypothesis. The evidence of a possible attrition problem is presented in
Section 4 and the inverse probability weighting to correct for attrition is explained.
Section 5 discusses the endogeneity of income change and the implementation of
the random coefficient model of average treatment effect to estimate the main
effect of income change, as well as the interaction of income change and education.
Empirical results are discussed in section 6. We find that income changes matter to
consumption change, so there is no perfect consumption smoothing. We also find
evidence supporting the hypothesis that education of household members enhances
the consumption smoothing in the high wealth group but not in the low wealth

group. Section 7 concludes the paper.

1.2 Theoretical Framework and Empirical Impli-
cations

1.2.1 A Full Consumption Insurance Model

The full consumption insurance outcome can be obtained through a risk shar-
ing model cast in the setting of a social planner, where the planner maximizes
the sum of weighted utilities of individuals subject to an aggregated resource
constraint (Mace 1991; Cochrane 1991; Townsend 1994; McCarthy 1995).7 As-
sume a single-good economy 8with N households and each of them lasts for T
periods. Household i has state contingent time separable utility U (Cf-ASH, 6%)),
where Cit is the consumption per capita of household in the state of T at time
t ; sTt is the state of the world at time t, 7' = 1,2, ...3; and 6ft is taste shifters
of the household 2'. Let 7r(s.rt) denote the subjective prediction about state 7'
at time t with 2::1W(31t) = 1, where 7r(sTt) captures the expectation about
the uncertainty. So the life time discounted expected utility of household 2' is:
U1: = 2:le 2:21 mn(sfi)U(Cft(sTt),6it), where p,- indicates the time dis-
count rate of household 2' and 6ft is the taste shifters of household i. If the house-
holds are to pool their resources together and insure each other against idiosyn-

cratic shocks, this is equivalent to a social planner maximizing a weighted sum of

 

7Such optimal risk sharing allocation can also be achieved as a competitive equilibrium in a
decentralized economy with complete contingent markets (Arrow 1964; Townsend 1994; Deaton
1997)

8The full consumption insurance implications continue to hold for the multiple good economy
(Mace 1991).

household utilities subject to an overall constraint at each time and in each state
of the world. Then the Pareto-optimal consumption allocations can be derived

from the planning problem

N S T
1 . .
Max}: “’2' Z Z fi”(srt)U(Cit(Srti 5ft» (1-1)
i=1 T=1t=1( +p‘)
N . .
s.t. Z@,(57t,6;t)=CA(sTt) 7:1,2,...s;t=1,2,...r. (1.2)
i=1

where CA(sTt) is the total amount of the consumption good available at each
time t in each state 7; w,- is the weight assigned to household 2' with w,- > 0
and 2,1111),- = 1 for i = 1, 2, ...N. The control variable for the social planner is

it, the consumption of household 2' at each time given the state of the world.
Take the derivative with respect to Cit: the first-order conditions for the problem

maximizing (1.1) subject to (1.2) are:

1 . .
wi7f(37t)WUI(Cft(STta63ml) = /\(Srt) (1.3)
where /\(Srt) is the Lagrangian multiplier at time t in the state T. Equation (1.3)
says that in the optimal resource allocation in a given state of the world, the
weighted marginal utilities are equalized across individual households. When the

state of T happens, the ca: — post counterpart of (1.3) without any uncertainty is

mmv’rdwi» = At (1.4)

To remove the fixed effect of each household associated with 11),, divide (1.4) at

time t + 1 by (1.4) and we can get

U,( f+1(f+1)) 1 =At+1
U'(Cg(5;)) 1+p,- A,

 

(1.5)

The right-hand side of equation (1.5) consists only of aggregate variables and it
is the same for all households. This equation has the full consumption insurance
implication because it says that the growth of discounted marginal utility across
all the households in community should be equal and be a function of growth in

A, which is a function of growth in total resources available to the community.

This model does not imply the reference group of consumption smoothing. The
“community ”here can be a village (Towsend 1994), extended households (Al-
tonji,Hayashi and Kotlikoff 1992), ethnic lines (Grimard 1997), and asset groups
or stratifications of households according to their assets level(McCarthy 1995; J alan
and Ravallion 1999). The reference group used in this paper is asset groups within

each primary survey unit (PSU).9

 

9A PSU is identified using the variable “site ”contained in the RLMS and the table made
available by the Carolina Population Center. The number of households in each PSU varies from
37 to 172.

1.2.2 Empirical Implications and Test

To derive a testable form of (1.5), suppose the utility function is a power utility 10

. . . 1 .

water» = €$P(U5i);(ci)a (1.6)
Strict concavity requires a < 1.The households have the same constant relative
risk aversion, (1 — 0). So

I ' 22 2t -
U(C_‘2) =e$pa(6t+1-6t)(cli)(l_0) (1.7)

U’(Cf+1) C“
Substituting (1.7) into (1.5) and taking the logarithms, we will get the following

equation as the test for perfect insurance hypothesis:

. . 1 ' -
log Ctz+1 — log Ctz = :[Uwf-H _ 6f)

/\t 1
+log(1 +Pil+108(—/\f’)l (1-8)
Conditional on the change in taste shifters ((52+1 — 5%) and the time discount rate
of pi, equation (1.8) implies that in the optimal risk allocation, the consumption
growth of each household 2' depend only on the aggregate resource availability vari—
ables through At and At“, but doesn’t depend on the individual income growth

log Yi — log Yi or initial assets level of the household Ai. Using the above equa-
t+1 t 1

tion to test for perfect consumption insurance , we make the following assumptions.

 

10The specification of power utility directly gives the relationship between growth rate of
individual consumption and the aggregate variables. (Mace 1991)

First, the change of taste shifter 6:: is a function of the change of household char-
acteristics X; such as household size and household composition as well as the
household characteristics in the initial period. Discount rate p,- is a function of
household head characteristics H: such as age and sex. If the full consumption
insurance hypothesis is true within community r, we can include a set of com-
munity, year and the interactions of year and community dummies (Dr,Dt and
Drt) to control for the community average consumption. 11 Then the testable

econometric model of equation (1.8) takes the following form

A log C2 = a0 + alA log Yti + 012/42, + a3AX§ + O4Hi + OSDt + OGDr + 07D” + 5i
(1.9)

If the full consumption insurance exists, then 011 = 0 and (12:0.

The test based on (1.9) is under the null hypothesis that there is optimal con-
sumption allocations within a complete risk sharing system. But because of moral
hazard and incomplete information, such risk sharing systems, even in the village
economy, may not be viable regardless of its advantage(Deaton, 1997). In a more
realistic setting, the risk sharing among individuals can be obtained through in-
formal social networks, the formal credit market, financial markets, and insurance
markets. But because incomes covary, it may be that the group is inefficient in
producing insurance especially if the shock is large, as was the one in Russia dur-

ing this period. Maybe also participation in the risk-sharing system may incur

 

11Using a set of dummies is preferred to using the changes in community mean consumption.
(Deaton (1992), Jalan and Ravallion (1994))

10

costs and thus not every household is equally likely to participate in this system
and thus is not equally insured against idiosyncratic income shocks. For example,
liquidity constrained households may not have access to credit market. Likewise,
information constrained households are unlikely to participate in the financial mar-
ket or even insurance markets if the market entrance imposes information costs.
But information costs are lower for better educated people if education helps peo-
ple to obtain, process and use information (Welch 1970). Then we should expect
that households whose members have high education are more likely to be covered
by the risk-sharing systems than households with less-well educated people. This
suggests that the coefficient on the income depends on the education level of the
household members. If true, we can add in an interaction term of income change
and maximum education level of the household members (E‘) into the equation
(1.9) and expect to see a negative sign on this interaction term. I use demeaned
maximum education (Ef — E) in the interaction term. In this specification, the
coefficient of the change in income can be interpreted as the average treatment ef-
fect of education on consumption smoothing, that is , the consumption smoothing

of the households whose members have average education level.

But one problem with this procedure is that the coefficient on the interaction term
also captures the wealth effect on consumption smoothing since households with
high—educated peOple have higher income earning ability. One possible way to

separate this wealth effect of education from the allocative effect of education is to

11

stratify the data by assets. Thus, we estimate the following equation for two asset

groups separately:

A log CE = a + AA log Y." + M log Y." x (Ei — E) + BaE" + mg + @3sz +

+ fieHi + ant + aDr + all)... + e: (1.10)

1.3 Data and Descriptive Statistics

The data used for this paper come from the Russian Longitudinal Monitory Survey
(RLMS), which was conducted by the Population Center at the University of North
Carolina. The RLMS is a household-based survey designed to measure the effects of
Russian reform on the economic well-being of households and individuals starting
from 1992. Beginning in 1994, RLMS was designed to provide a longitudinal
study of populations of dwelling units. At each round, the RLMS interview was
completed with the household and its members in the original sample dwelling unit.
This sampling plan did not call for households to be followed if they moved from
the sample dwelling unit. Consequently, the RLMS is not a true panel design. But
unique household and individual identification numbers exist in the data and can
be used as a link to form a pure panel of the households who remain in the original
dwelling unit over time. The detailed information on household-level income and

expenditures can then be used to study the economic welfare of households in a

12

dynamic setting. Data on individual wage earnings, pension earnings, complete
years of education and employment status are also provided. The data used for
analysis are for the years 1994-1996, 1998 and 2000. In the regression analysis
containing the interaction term of income change and years of education, we drop
the 1994 data because it does not have the information about years of education

of the household members although it has information of level of schooling.12

The value of assets is the estimated worth of the non-financial households assets.13

It is used as a proxy for household wealth. The households then are divided into
two groups according to their asset levels in 1994. Total household consumption of
the household is the sum of expenditures on food consumption and non-food ex-
penditures on clothing, fuel, transportation, repair service, laundry, postal service,
medical service, marriage/funeral service, rent, child support, schooling, sanato—
rium, travel and clubs. The growth of consumption is calculated as the logarithmic
difference in consumption. Total household income is the constructed income in
RLMS, which is the sum of income from the workplace; fuel subsidies; child sup-
port; pensions; asset income and transfer payment from government, relatives or
friends. But some components of total income may serve to smooth consumption.

For example, transfers (both private and public) may increase after the shocks.

 

12Regression including the 1994 data is presented in the Appendix. 1994 data contains only
the categorical level of education of the household members, so we include three education level
dummies in that regression.

13The assets include (1) Refrigerator; (2) Homer; (3) Washing Machine; (4) Black & White
TV; (5) Color TV; (6) VCR; (7) Car or Tmck; (8) Motorcycle; (9) 'Il‘actor; (10) Garden Cottage;
(11) Dacha or Other House; (12) Other Apartment. Household survey respondent was asked to
estimate the current value of their assets based on the age of assets. The missing value is replaced
by the reported area-age specific asset.

13

Including endogenous components in the income measure will result in the volatil-
ity of income shock being systematically understated (Rosenzweig, 1988; Strauss
and Thomas, 1995). Because of this concern, we construct a variable we call net
income, which doesn’t include the endogenous components such as unemployment
benefits, transfers from relatives , friends, church, mosque, foreign and interna-

tional organizations; and incomes from sales and rental of assets.

Table 1.2 provides an economic and demographic profile of the sample of 1412
households with 673 households in the low asset group and 739 in the high asset
group. Total income is 40.3% higher than net income for the low asset group; for
the high asset group, it is 30.21% higher than net income. This large proportion of
endogenous components of income indicate that households smooth their income
by public and private transfer or assets selling. This justifies our concern over the
endogeneity of income change. Thus, we use net income we defined above as a

measure of household income in the regression analysis.

Households on average have 3.72 members. The number of senior people in the
low asset group is 0.786, significantly higher than 0.605 in the high asset group.
But the number of children in the high asset group is 0.906, higher than 0.673 in
the low asset group. These differences in the household composition between the
two assets groups may be due to the fact that households with more senior people
have older assets whose estimated values are usually not as high as the assets in

the households composed of relatively younger people.

14

Descriptive statistics of the incidence of households that report unemployment,
wage arrears and pension arrears are presented in Panel B of Table 1.2. More
households in the high asset group report unemployment and wage arrears, but
more households in low asset group report pension arrears. They correspond, not
surprisingly, to the age differentials between the two groups. Panel C of Table 1.2
presents the proportion of households which experienced fresh spells of unemploy-
ment, wage arrears or pension arrears in the month of the survey. For example,
about 8.1% of the households have reported at least one household member who
was newly unemployed and 8.8% of the households have wage arrears for one month
or less but have not experienced wage arrears in the previous year. More house-
holds in the low asset group experienced a pension shock than in the high asset
group. This may simply be due to the fact that there are more seniors in the

households in low assets group.

Table 1.3 presents information of the within-household coefficient of variation (CV)
of total income, net income and consumption of the households in the analysis
sample. The CV at the household level is calculated as the within-household
standard deviation divided by the within-household mean. The CV at the group
level is the average of CVs of the households in that group. The CV of total
income for both assets groups is more than 0.5. Without controlling for endogenous
income smoothing, as we expect, the CV for net income is higher and is over 0.6

for both groups. On average, consumption has tighter distribution than the total

15

income. The Kruskal-Wallis rank test confirms that CV of consumption for the
lower assets group is significantly larger than that for the high asset group. This
is not surprising since we expect that households with more assets experience less
volatile changes in their consumption because they can draw on their assets to
smooth their consumption. The simple correlation between consumption and net
income is also lower for the high asset group although the magnitude is still as big
as 0.42. It means that consumption follows income and therefore the ability to
smooth consumption is very limited for households in both groups, if we assume

that households prefer smooth consumption.

1.4 Sample Attrition: Evidence and Correction

Among 3548 households included in the 1995 survey, 2214 of them remain in the
survey for four rounds, but 1334 of them dropped out during periods of before the
2000 survey. We call these households leavers. Are the households who remain
in the survey (stayers) significantly different from the leavers? Table 1.4 presents
the summary statistics by attrition using 1995 data. On average, the stayers
are poorer, have a less educated household head, larger household size and more
senior members. Fewer stayers live in Moscow and St. Petersburg, rather more of
them live in non—urban areas. All these differences between the two samples are
statistically significant at 5%. Apparently, the households in our analysis sample,

namely the stayers, are not a random sample from the original sample in 1995.

16

Using the stayers in the analysis without correcting for any potential attrition bias
may result in overestimating the consumption response to income change, since the
stayers are poorer and may face more credit constraint because disproportionably

more of them live in non-urban areas.

There are two other variables in the data which are closely related to household
attrition but may not be correlated with consumption change. They are the assess-
ments of the interviewer about the level of cooperation of the household respondent
with the survey enumerators. That is, at the end of the survey, the interviewer was
asked to assess the respondent’s attitude toward the interview as well as the re-
spondent’s behavior during the interview. The attitude was listed as “friendly and
interested”; “not particularly interested ”;“impatient and worried ”; and “hostile
” The behavior includes “comfortable”; “occasionally nervous ”; and “nervous
”. On a scale of 1 to 4 with 1 representing the highest degree of cooperation, the
mean attitude of the movers is 1.28, while the mean attitude of the stayers is 1.23.
The difference is small in magnitude, but statistically significant at 5% level. It

means that the households which were more willing to cooperate in the interview

were more likely to stay in the survey.

Significant differences between the stayers and the movers suggest that estimates
that don’t correct for potential attrition appropriately might be inconsistent. We
use the inverse probability weighting (IPW) method (Wooldridge 2002a)to correct

potential bias from sample attrition.

17

Under the key assumption that sample attrition is ignorable with respect to the
consumption growth conditional on the observables in the attrition equation”,
the IPW gives more weight to the households which are less likely to stay in the
sample to make the analysis sample more representative of the original sample.
The IPW procedure involves two stages of estimation. In the first stage, at time
t (t=2, 3, 4) an attrition probit model is estimated, restricting attention to those
households still in the sample at t — 1. Out of this sample, some are lost to
attrition at time t, and some are not. Let aids“ = 1[:r,-t6t+v,-t]|s,-t_1 = 1, s,t_2 =
1, ..., 3,1 = 1), where t=2, 3, 4 and Uitlmz't ~ Normal (0,1), be the probit estimation
of the conditional probability of household 2' to stay in the survey at time t. These
predicted conditional probabilities can not be used directly in the IPW procedure
because the sample at each time period is not representative of the population
that was originally sampled at t=1. However, Wooldridge (2002a) shows that the
joint probabilities calculated from these predicted conditional probabilities can
be used in the IPW procedure and lead to consistent estimators. The predicted
conditional probabilities 7i“ (t=2, 3, 4) are used in the second stage to calculate
the joint probability that the households stay in the survey for two, three and four
rounds. We denote the unconditional probabilities by 15,2, 13,3, 152-4 respectively.
We ignore the initial condition and assume that all the households have the same
probability of being in the original sample. Then 15,1(sfl = 1) = 7}“ = 1. So the

joint probability, 1312(82'1 = 1,812 = 1) = fin X 5B2, 1523(82‘1 = 1; 8r2 = 1:513 = 1) =

 

14In our case, this means that the selection is not related to the idiosyncratic errors in regression
1.9. Our test confirms this assumption and the coefficient of the lead selection indicator in a
fixed effect regression of 1.9 is not significant with a t-statistic being -1.01.

18

fin >< 7E2 X fr2'3; 152210911 = 1,822 = 1,823 = 1,3,4 = 1) = fin X 7h? x 7m X 7m.
According to the estimated probability pit, each household 2' at time t is assigned

a WBIght wit = l/fiit-

Table 1.5 presents the probit estimation of the conditional probabilities of being
in the survey for the years 1996, 1998 and 2000. Conditional on the household
demographic characteristics, geographic locations as well as respondent’s level of
cooperation in the survey, we find that income is no longer significant in explaining
probabilities of being in the survey. The age of household head affects the prob-
ability in an inverse U shape pattern. For example, ceteris paribus, households
with household head at age 50 in 1998 are the most likely to remain in the survey
in 2000. The more senior people a household has, the more likely that it is going
to remain in the survey. Bigger households are more likely to stay in the survey
continuously. Households in the Moscow and St. Petersburg region are more likely
to drop out compared to any other region covered by the surveys. The level of
cooperation of the household respondent in the oral survey is also positively as-
sociated with staying 15 For example, the least cooperative households in 1996
and 1998 are more likely to drop from later surveys. And if the respondent was
nervous during 1995 or 1996 survey, then the household is more likely to remain in

the survey in 1996 and 1998. The predicted probabilities (frit) from these probit

 

15Heckman’s solution requires that there be at least one exogenous variable affecting selection
that does not appear in the structure equation. By reasonable assumption, the attitude and
behavior variables which measure the level of respondent’s cooperation may well serve as such
identifying variables. But in the IPW method, no such identifying variables are needed.

19

estimation are then used to calculate the joint probability for each year (p,,)and

each household is assigned a weight equal to the reciprocal of the joint probability

(wit = 1/152‘tl-

1.5 Endogenous Income Change

Net income defined in section 3 is used as household income in our analysis in
order to reduce the potential bias from endogenous income smoothing. But two
empirical problems still remain. The first problem is the measurement error in
income. The measurement error in income will lead to an attenuation bias in the
estimate of the coefficient in income growth and thus we will be less likely to reject
the complete consumption insurance hypothesis. The second problem is that net
income change may still be endogenous. For example, unobservables that affect
income change may also affect consumption change. A credit constraint level that
the household faces may affect income change as well as the precautionary savings
of the households, thus consumption of the households. Also income change may
be correlated with unobserved preference shifts (Cochrane 1991). If income change
leads to higher preference for leisure and if leisure/ consumption are substitutes,
then the estimate of the coefficient of income change on consumption change will
be biased downward. In the case that leisure and consumption are complements,

we might overestimate the response of the consumption change to income change.

20

In our model, the coefficient of income change on consumption change is also
allowed to differ across the households with different human capital. We implement
a new method following a random coefficient model of average treatment effect
with endogenous variables (Wooldridge 2002b) to estimate equation (1.10). The
traditional 2SLS estimator uses interactions of exogenous variables with (Ef — E)
as instrumental variables for log I”; x (E — E). This method may result in too
many overidentifying restrictions and the 2SLS estimator may have poor finite
sample properties (Wooldridge 2002b). The new method involves two steps. First,
the change of income is predicted from the estimation of a linear reduced form
by regressing income change on exogenous variables. Call this predicted income

A A

change A log Yti' The second step is to use the predicted income change (A log Yt’)

, the interaction of predicted income change with education(A log Y: x E‘) as

instrumental variables for A log Y," and Alog Y," x (Ei — E) in equation (1.10).

We use two sets of exogenous variables to predict the change of income. First
the change of total household wage earnings is used as a predictor for the change
in the household income 16. The total wage earnings is constructed separately
from individual questionnaires. Each adult in the household reports their total
labor earnings in the individual questionnaires. These individual responses can be
added up to the household level and form a separate measure of household income.

This measure of income also contains measurement error but this error may be

 

16Stillman(2001) used aggregated individual total income as instrumental variable for house-
hold income.

21

uncorrelated with the measurement error in the household total income reported
by the household respondent. Household income comes from a separate household
questionnaire and the respondent is asked to estimate total household income.
Thus the change in this aggregated wage earning can be used as an instrumental
variable to purge the measurement error from the total income. But the change
of total wage earning can’t account for the potential endogeneity associated with
household income change. As an alternative to correct for the endogeneity of
income change and also to capture the unexpected nature of idiosyncratic income
shocks, a second set of exogenous variables are used. They are: whether there is
a household member who is newly owed wage (for no more than one month) 17,
whether there is a household member who is newly unemployed, whether there are
retired household members who are newly owed pensions, and an interaction of
pension arrears with a dummy variable for the year 1996, which was the year of

the pension crisis.

Table 1.6 presents the estimation of income change in the first step. We can
see that the change of wage income obtained from individual questionnaires are
significant both statistically and economically in predicating household income

change. When we use the shocks of unemployment, wage arrears and pension

 

17The following survey questions are used to construct this variable. First, “At the present
time, does your place of work owe you any money which for some reason they didn’t pay you on
time? ”. Second, “For how many months has your enterprise not paid this money to you? ”. If
the answer to the first question is “Yes ”and the answer to the second question is “one month ”,
I take it as “owed wage for no more than one month ”. The “newly owed wage for no more than
one month ”means that no household member is owed wage in the previous period but there is
at least one household member is owed wage for no more than one month in this period. The
variables, “newly unemployed ”and “newly owed pension ”, are defined in the same way.

22

arrears as instruments, we obtain the expected signs. They are jointly significant
in low and the high asset groups and the p — values for the F — statistics are
0.002 and 0 respectively. Individually, three out of four predicting variables are

significant for the low asset group and all of them are significant for high asset

group.

1 .6 Results

The estimated results of equation (1.10) for low and high asset groups are presented
in Table 1.7 and 1.8 respectively. Columns (1) to (3) contain the estimation results
without sample attrition correction. Columns (4) to (6) contain the estimation
results with sample attrition correction. The first stage regression using predicted
income change and the interaction of predicted income change and education as
instrument variables for income change and the interaction of income change and

demeaned education are presented in Table 1.10 in the Appendix.

Comparing the OLS estimator and the IV estimators for each of the two asset
group, we can see that the coefficient of income change is significantly larger in
the IV results when we use wage income change to predict the household income
change (1V1). It confirms our concern of measurement error in the income variable.
When we use the shocks of unemployment, wage arrears and pension arrears as

predictors of income change (1V2), we find that the magnitude of coefficient lies

23

in between the OLS and 1V1. It might be the case that 1V1 overestimates the
consumption response to income change because it doesn’t correct for common
unobservables between income change and consumption change. Sample attrition
correction leads to smaller estimators of the coefficient on income change for both
OLS and two IV estimations in the low asset group. In the high asset group,
the coefficient of income change is bigger in IV2 after corrected for attrition. But
the differences between the coefficients in the regressions with and without sample
attrition correction are generally very small in magnitude. We are going to focus on

the IV2 regression results with sample attrition correction in the following analysis.

Neither the low asset nor the high asset group can completely smooth their con-
sumption. The coefficient on the income change means how consumption change
responds to income change for the household whose member has average education
level 18. For example, in the low asset group (Column 6 in Table 1.8, if income
change 10%, consumption will change 1.74% for the household whose member has
11 years of education . In the high asset group (Column 6 in Table 1.9), 10%
change in income will lead to 0.95% change in consumption for the household with
11 years of education. Education effect on consumption smoothing is found in high
asset group but not in the low asset group. In the high asset group, the coefficient
on the interaction term of income change and education is -0.0386, significantly

different from 0 at 10% significant level. It means that if the household has a

 

18 The average years of education is 11 years for the low asset group and 13 years for the high
asset group.

24

university graduate 19, for example , then a 10% change in income only causes a
0.58% change in consumption. The result is puzzling because we might expect that
people in the low asset group would draw on their human capital more in times
of income shock but the wealth effect in the high asset group might outweigh any
education effect. What we find here, however, is that education and wealth are
complements and that human capital can help people to better use their wealth in

smoothing consumption against income shock.

Assets are significant in helping the households in the low asset group to smooth
their consumption and not surprisingly, the asset effect is not significant in the high
asset group. As we can see in the descriptive statistics (Table 1.2), net income per
capita in the high asset group is about 32.8% higher than that in the low asset
group. At the time of income shock, the households in the the high asset group
do not necessarily need to sell or rent out their assets, but for the low asset group,
drawing on their assets may be one of the main options to cope with the shock. For
the low asset group, drawing down the assets by 10% can increase the consumption

by 0.39%.

The coefficient on the number of children in the initial period has a different sign
in the two groups and both of them are significant. It seems that children can help,
in some way, the households in the low asset group to smooth consumption. One

explanation for this result may be related to means-testing schemes in child benefits

 

19University graduate on average has 16.03 years of education in the sample.

25

in Russia. 20 The child benefits that low income households receive may serve to
help the households to smooth their consumption. For the high asset group, the
presence of children in the first period doesn’t have any consumption smoothing
effect. Instead, if there are more children in the first period, consumption will

increase, possibly to meet the growth of the children.

For households in both groups, the change in the number of senior people has a
negative and significant effect on consumption change. One more senior person in
the household is related to a 0.37% and 0.16% decrease in consumption for the low
asset and high asset groups respectively. It may just be due to the fact that seniors
do not consume as much as younger people. Alternatively, living arrangements with
a senior person might be related to whether the pension payment is fulfilled. If
a household’s members choose to live with seniors when they have pensions, then
we can see a negative sign in the change of number of old people in the households
because the households with more old people will have better means to cope with
income shocks. But the causation could go other way from smoothing consumption
to attracting seniors. If this is the case, then the coefficient on the change of senior
people is not consistently estimated in this specification. Although we don’t focus
on the endogenous living arrangement here, caution should be taken in interpreting

this result.

 

20In 1991, the government introduced a special child support benefit to be paid to all families
with children. In 1995 some regional governments decided to limit the payments of child support
benefits to only the poor families. In 1997-1998, a federal law was adopted, which limited the
payments to only the families with per capita income below the regional subsistence level. See
Denisova, Kolenikov and Yudaeva (2000) on the child benefits policy in Russia.

26

The coefficients on the year dummies measure how consumption changes during
1996-1998 and 1998-2000 compared with the 1995-1996 period. For the households
in low asset group, consumption decreased by 18.8% during 1996-1998 compared
with the 1995-1996 period. For the households in high asset group, consumption
decreased by 27.2% . These differences are not statistically significant. During
the 1998-200 period, consumption increased by 43.7% for the low asset group
and 24.4% for the high asset group, compared with 1995-1996 period. These are
statistically different at 10% significance level21. This means that consumption
recovered more during the 1998-2000 period for the low asset group than for the
high asset group. This result is consistent with the findings that there is a de-
cline in inequality during the 1998-2000 period and much of this decline is due to

government transfers and pension payments (Mroz, Henderson and Popkin 2001).

To check whether dropping the 1994 data will lead to different estimation results,
we include the 1994 data in the regression without the interaction term of income
change and years of education (since 1994 data doesn’t contain information on
years of education). Instead we include three dummies to identify the four cate-
gorical education groups. The results are presented in Table 1.12 and Table 1.13
in the appendix. The inverse probability weights are calculated in the same way as
Section 4 but using 1994 as the first period. Almost all the previous results hold
in these regression with 1994 data. The coefficient of income change in the high

asset group (Column 6 Table 1.12) is 0.1451 lower than 0.1679 (Column 6 Table

 

21The p — value is 0.053.

27

1.8) when 1994 data is not included. The regressions in Table 1.12 don’t count
for the differential consumption smoothing across households with different human
capital while the coefficient of 0.1679 in Table 1.8 is the consumption response to
income change for household whose members have average education level. The
coefficient of income change in the low asset group (Collum 6 Table 1.11) doesn’t
differ significantly from that in Column 6 of Table 1.7 when we exclude 1994 data
(0.1657 vs. 0.1608). This is not surprising. Since no education effect is found
in the low asset group, without including interaction term of income change and

education in Table 1.11 shouldn’t change the coefficient on income change.

Different regressions without education interaction term are also reported in Table
1.13 and Table 1.14. Without controlling for education effect, the coefficients of
income change increase in both of the assets group. It is intriguing, especially for
low asset group, where we do not find education effect on consumption smoothing.
When we pool the asset group together (Table 1.15), the education effect is not
significant. As we observe earlier, that the education effect exists in high asset
group only, thus it might be the case that such effect is not dominant in the whole
sample. The results of the regressions for the whole sample, but with income change
interacting with asset group dummy variable is reported in Table 1.16. And the
results for including the interaction terms for both asset group and education are
included in Table 1.17. Those results yield insignificant coefficients on income

change variable. Such results are likely to be caused by multicollinearity, due to

28

the interaction terms.

1 .7 Conclusions

Using the Russian Longitudinal Monitory Survey (RLMS), we examine the differ-
ences of consumption insurance between two groups of households based on their
initial asset level. In the econometric model, we corrected the potential attrition
bias with the inverse probability weighting method and we use the random coeffi-
cient model for average treatment effect to estimate the endogenous income change
and education effect. We reject the full consumption smoothing hypothesis in both
wealthy and less wealthy group. We find that the education in the high asset group
can increase the consumption smoothing while no education effect is detected in
the low asset group. We also find that poor households may rely on their assets
to smooth their consumption while assets effect is not significant in consumption
equation for wealthy households. The number of children in the initial period has
different impact on consumption smoothing for the poor and wealthy households.
The change of policy in child support may cause this difference. The number of
senior people in the households also helps to smooth consumption. And wealthy
households recovered slowly from the financial shock. Several of the results needs
more detailed work. For example, we found that income shocks have but small im-
pact on households’ consumption. A further study of how the Russian households

managed to smooth their consumption would be necessary to understand how the

29

households adjust to different shocks in the transitional economy. Multiple job

holdings and living arrangement would be some topics interesting to look at.

30

Table 1.1: Household Real Income and Consumption Change Over Years and
Regions

 

 

 

Dependent Variable Income Consumption
(log) per capita per capita
1 2 1 2
Year 1995 -0.314 -0.253 -0.176 -0.140
(0.019)**(3) (0.041)“ (0.019)“ (0.039)“
Year 1996 -0.476 -0.388 -0.351 -0.355
(0.025)“ (0.048)“ (0.022)“ (0.041)“r
Year 1998 -0.580 -0.615 -0.718 -0.741
(0.021)“ (0.046)“r (0.021)“ (0.040)“
Year 2000 -0.388 -0.375 -0.573 -0.554
(0.021)“ (0.042)“ (0.021)“ (0.041)“
Moscow and 0.466 0.411
St. Petersburg (0.075)“ (0.078)“
Northern and 0.141 0.120
North Western (0.079)* (0.078)
Volga-Vaytski and -0.079 -0.141
Volga Basin (0.051) (0.052)“
North Caucasian -0.127 0.002
(0.070)* (0.056)
Ural 0.012 -0.037
(0.055) (0.058)
Western Siberian 0.069 0.106
(0.069) (0.074)
Eastern Siberian and 0.253 0.080
Far Eastern (0.072)“ (0.068)
F-statistics for 207.39“ 49.23“ 380.98“ 111.13“
Year Dummies
F -statistics for 3.42“ 2.21“
Region/ Year Dummies
Number of Obs 8928 8928 8928 8928
R-squared 0.044 0.079 0.091 0.115

 

 

(1) The omitted region is Central and Central Black Earth.

(2) Regressions with region dummies also include interactions between region dummies
and year dummies.

(3) Standard errors (robust to correlation of residuals within households and
heteroscedasticity)in parentheses

(4) ”indicates significance at 5% level and” 10% significant level.

31

Table 1.2: Descriptive Statistics

 

 

 

ALL Low Assets High Assets

 

Panel A: Household Characteristics

Net Income per Capita 1181.370 889.796 1436.404
(1786.130) (1098.702) (2187.700)
Total Income per Capita 1417.600 1248.402 1870.340
(1652.770) (1535.417) (2425.912)
Consumption per Capita 1578.230 1141.678 1658.942
(2079.450) (1278.420) (1888.691)
Total Assets 4794.190 813.749 8768.132
(9139.080) (450.440) (11623.610)
Household Size 3.716 3.471 3.929
(1.516) (1.536) (1.467)
Number of Seniors 0.689 0.786 0.605
(0.792) (0.776) (0.796)
Number of Children 0.7981 0.673 0.906
(0.987) (1.022) (0.942)
Age of Houshold Head 47.391 49.828 45.225
(14.825) (16.293) (13.011)
Gender of Houshold Head 0.776 0.663 0.876
(0.417) (0.428) (0.330)
Years of Education of 10.897 9.908 11.769
Household Head (3.746) (3.814) (3.558)
Maximum Education 12.038 10.818 13.113
of Household Members (3.724) (3.919) (3.176)
Panel B: Incidence of Unemployment, Wage Arrears and Pension Arrears
Unemployment 0.120 0.111 0.127
(0.325) (0.314) (0.333)
Wage Arrears 0.415 0.353 0.471
(0.493) (0.478) (0.499)
Pension Arrears 0.098 0.115 0.082
(0.297) (0.319) (0.274)
Panel C: Change of Unemployment,Wage Arrears and Pension Arrears
Change of Unemployment 0.081 0.071 0.089
(0.272) (0.257) (0.285)
Wage Arrears for 0.088 0.065 0.108
One Month or Less (0.284) (0.247) (0.311)
Change of Pension Arrears 0.076 0.089 0.065
(0.265) (0.284) (0.246)
Number of households 1412 673 739

 

(1) Net income is defined as total income not of public transfers,private transfers,
unemployment benefits and incomes from sales and rental of assets.
(2) Standard deviations in parentheses.

32

Table 1.3: Coefficient of Variation (CV) in Income and Consumption (per capita)

 

 

All Low Assets High Assets
CV for Total Income 0.562 0.549 0.575
(0287)“) (0.281) (0.292)
p — value for Kruskal-Wallis test 0.037
CV for Net Incomem 0.632 0.614 0.649
(0.430) (0.444) (0.416)
p — value for Kruskal-Wallis test 0.0001
CV for Consumption 0.625 0.650 0.602
(0.305) (0.324) (0.285)
p — value for Kruskal-Wallis test 0.0001
Correlation of Net Income 0.4412 0.4909 0.4192
and Total Consumption
Number of Households 1412 673 739

 

(1) Standard deviations in the parentheses.

(2) Net income is defined as the total income net of public transfer, private transfer;

unemployment benefits and incomes from sales and rental of assets.

33

Table 1.4: Descriptive Statistics in 1995 by Attrition

 

 

 

Stayers Leavers p—value
(1) (2) (1)-(2)-

Net Income per capita 1793.554 2555.519

(38.002) (3906.432) 0.000
Consumption per capita 1264.867 1749.451

(33.343) (68.078) 0.000
Years of Education 10.899 11.212
of Household Head (0.079) (0.114) 0.024
Gender of Household 1.224 1.282
Head (0.008) (0.012) 0.0001
Moscow & St.Petersburg 0.051 0.156

(0.004) (0.009) 0.000
Household Size 3.715 3.211

(0.032) (0.038) 0.000
Age of Household Head 47.170 46.947

(0.315) (0.470) 0.693
Number of Seniors 0.690 0.537

(0.017) (0.186) 0.000
Urban 0.607 0.729

(0.010) (0.0116) 0.000
Attidude of Respond 1.231 1.280

(0.011) (0.021) 0.035
Behavior of Respond 1.186 1.211

(0.010) (0.018) 0.236
Number of Obs. 2214 1334

 

(1) Stayers are households which stayed all 4 rounds since 1995 survey.

Leavers are households which were in 1995 survey but dropped out either

in 1996,1998 or in 2000 survey.

(2) Attitude=1 if household survey respondent is friendly and interested;

Attitude=2 if respond is not particularly interested; Attitude=3 if respondent
is impatient and worried; Attitude-=4 if respondent is hostile.
(3) Behevior=1 if respondent is comfortable during the interview; Behavior=2 if

respondent is occasionally nervous; Behavior=3 if respondent is nervous.

(4) Standard errors in parentheses.

(5)p — value is for testing the null hypothesis that the variable mean is not

different across the two samples.

34

Table 1.5: Attrition Probability during 1995-2000

 

Dependent Variable:

1=Stay; 0=Leave

 

 

(96l95) (98l96,95) (2000|98,96,95)
Net Income per capita 0.034 0.016 0.026
(log) (0.022) (0.019) (0.0167)
Education of -0.001 0.006 0.0067
Household Head (0.010) (0.009) (0.008)
Gender of Household -0.055 -0.064 -0.056
Head (0.081) (0.069) (0.058)
Age of Household 0.045 0.039 0.050
Head (0.012)“ (0.010)“ (0.086)“r
Age2 of Household -00003 -00003 000%
Head (0.0001)“ (0.0001)”"’r (0.0001)**
Number of Seniors 0.134 0.181 0.172
(0.067)“( (0.056)“ (0.047)“
Household Size 0.087 0.075 0.076
(0.030)“ (0.026)“ (0.022)“
Moscow and -0.405 -0.367 -0.577
St. Petersburg (0.137)“ (0.127)“ (0.111)“
Northern and 0.037 0.035 0.0167
North Western (0.104) (0.092) (0.079)
Volga-Vaytski and -0.266 -0.353 -0.200
Volga Basin (0.120)“( (0.105)“ (0.093)“r
North Caucasian -0.101 -0.094 -0.002
(0.102) (0.091) (0.082)
Ural -0.136 -0.161 -0.119
(0.118) (0.106) (0.093)
Western Siberian ~0.216 -0.335 -0.318
(0.119)* (0.105)“ (0.095)“r

35

Table 1.5 (cont’d).

Eastern Siberian and

Far Eastern
Attitude2

Attitude3

Attitude4

Behavior2

Behavior3

Number of obs

(1)The omitted region is Central and Central Black Earth.

0325
(0.077)“
-0244
(0.074)“
-0.156
(0.185)
-0.446
(0.440)
-0.018
(0.084)
0.635
(0.239)v
2314

-0.312
(0.067)“
-0173
(0.059)“
-0249
(0.148)*
-0512
(0.291)*
-0014
(0.071)
0.309
(0.167)*
2129

0337
(0.060)“
-0197
(0.049)“
—0.314
(0.125)“
-0444
(0.239)*
-0101
(0.056)*
0.101
(0.133)
2263

(2)Standard errors (robust to correlation of residuals within households and

heteroscedasticity)in parentheses

(3)**indicates significance at 5% level and"r 10% significant level.

(4)Attitudc2=not particularly interested; Attitude3=impatient,worried;Attitude4=
hostile. The omitted attitude category is Attitude1=friendly interested.

Behavior2=occasionally nervous;Behavior3=nervous.The omitted behavior category

is Behaviorlzfelt comfortable.

36

62.83 8.5 23 E 98: 30895: 23 .«o 898w was 8.853 owe some 60th

8:3 23 E efim 30855: was 9838 582:3 mo Long: 85 “on? 20880: 98 82:8 EBEEo no 838::

85 E owfifio £8382: 20895: :38 Mo :03858 859585 mo 88» ”8538.» mosmteuoeemso 308803 AS

.62: eoaaocmcwmmoag 9:85:me x. was 623 vocaofiammmam undocmcmmw .5.

885883 E Amfiozmdvoomouoaes 98 86895: 5:33 3226?: «o comes—oboe 3 8530.3 88.8 Eavfifim A:

 

 

2mm 2mm 88 88 one. 8% Bo Lo 85:2
8% 8% 8% 8% 8% 8% 8:935 Eo%\:o%em
8% 8% 8% 8% 8% 8% m88< ESE Eonemsom
8% 8% 8% 8% 8% 8% 852:3 mosmtmuogennv 208803
”833213, 3:80
good mood assessed as e33 1 a
482.3 3.383 482.3 82 33> x
83. SE- was- 2325. nosed Lo mango
438.3 3.323 482.3
83. $2- 3.3. 352% season“ a0 3:20
3.883 44:33 3.833
awed- 83¢. and. satisfies: 3 355
3.333 2:3 $383 3.3 5 £52 25
$8. wood- 83. as 235. 333
822m
3.68.3 3.883 3.333 3:322: 5a 8:330
and $2. «So 4885 3.33 .8 3:20
3:80 88¢. aw;— QsoLU 88¢. 304 299% 2055

8885 wo owqeno 833.3% B89889

 

Emma? 4359305 882: 55 8283on mac

@885 mo emceao 05 35028.?“ 64 833.

37

 

 

.383 4383 438.3 883 0.83 0.8.3 3.8 E
oooo 83o oooo 8oo oooo «moo 2: 2 32m 222:8
488.3 488.3 3.883 883 4:83 4883 oesa 32 2: 5
oooo- oooo- 8o.o- mooo- 38o- 2.88. 228 8 82:2
883 883 883 :83 883 883 238.8 E 2: a
88. oooo- ooooo mooo mooo oooo 22280 co 382:2
38.3 98.3 .383 0.83 883 88.3
388. oooo- 288. 283 mooo- :oo. 28 22325: a 3820
.1883 488.3 .1383 4883 4:83 4883 228m
ago. oooo- oooo- oooo- 83. ~88. Lo 23222 2: 2 3820
883 383 883 883 0.83 88.3 o2380
oooo- Koo- oooo- wooo- oooo- mooo- 8 82:2 2: 2 32220
883 883 :83 :83 883 :83
oooo Boo 8oo oooo mooo goo 22$. :28.
883 883 883 $83 88.3 883 222232 22388 0
ooooo ooooo 8oo.o- oooo oooo mooo 8:825 2:252
4:83 4883 .2883 4883 3.883 .4383
wodo ammo mooo 82o Goo mooo 3820 2:83
NE E 80 NZ 0: 80

 

00300800 203552 55>

20500800 sosgo< 305:5

 

8090 008958000

”833.83 0808000

 

macaw 205$ 05 000 wEfiooEm 003082800 Kg 0305

38

0w0000 000005 20:00:00 00.0 00520.00, 0000005002

000 00 0000000 008000 E $0.000 500000 000 005 0008 00 00000.8 0mm? B0000080300000 00 09800 0000 NB .0w0000
0000000 20000000 000 000000.505 00 0000000300000 B0232: 50¢ 0000050 00805 0w03 030... «0 0w0000 0000 73 A8

._0>0~ 0000050300002 0* 20.0 .96. 0000050wmm§m 00.0020me .11

”0000500000 5 30030000000030: 28 020000000 02me 02.002000 .00 005200.000 00 00050: 000000 000285 AS

 

oooo
m®>
1883
008
1:83
80o-
:83
oooo-
10ooo3
ooooo
1820.8
88.

ooov
m®>
1883
go
infimodv
oooo-
:83
088.
£883
ooooo
.1883
33.

oooo
w®>
:Awmodv
oooo
1883
oooo-
883
mooo-
£883
888
1883
888.

oooo ooov oooo .30 00 508:2
00> 00> 00> 00000000 000>\000_w0m
1883 1883 .1883
oooo Homo Boo oooo 08>
.1383 1883 1883
888- 088. oooo- 82 08>
883 883 $83
mooo- oooo- oooo- 0080 22380 00 8080
£883 £883 £883
moooo ooooo o8o.o 0.80 220250 8 no?
1083 1383 .1383
088. 888. 88. 080 220800 00 a?

3.283 5 some

39

 

$88.8 $88.8 $88.8 $88.8 $08.8 $Coo8 :25: E 2:
Eoo oooo Boo Eoo oooo too a mum 223:0:
$88.8 $88.8 $88.8 $28.8 £858 $338 :25 :2 a
oodo- ado- who- So- wooo- So- acaom o: 5:52
$38.8 $388 $0058 $83.8 $38.8 $88.8 82$ :3 a
oooo- oooo- oooo- oooo- :ooo- Hooo- 5:230 o: 5:52
33.8 63.8 $3.8 $3.8 $3.8 $3.8 mam
Boo- Hooo goo- oooo- Soo Soo- 22880 a @0830
$28.8 $68.8 $88.8 $Aooo8 $38.8 Loooov 325m 0:
oooo- oooo- oooo- 20o- go- Homo- 35:2 5 owqfio
88.8 888 3.88 88.8 88.8 38.8 5:230 o:
:oo- 83. Nooo- Eoo- wooo- goo- 28:2 8 083:0
$888 38.8 $288 88.8 Aoooov $28.8
o:o.o- oooo- goo- oooo- mooo- oooo- 332 :33
$8.8 $8.8 $8.8 $8.8 38.8 388 38:52 228:3: o:
oooo- oooo- :ooo- Hooo- soo- soo- 808:3: 8:552
:38 :58 $8.8 $8.8 88.8 8.88 83:32.: 8:852
oooo- zoo- Hooo- oooo- :oo- oooo- 8:858 x Magoo 2:85
$508 $658 $0.88 $§o8 $38.8 $888
$3 oooo Hooo ooao homo oooo 09:20 8:85
NE E 20 :2 :0 £0

 

005000000 00§50< 53>

005000000 005538 00055,

 

098:0 003083000

”0303.3/ 0000:0000

 

0:000 00mm< 304 05 000 wEfiooEm 005955000

no: 05:9

40

0&0000 0000000 3000000: 000 00300000, 0000000005

00: 00 00.00000 020000 00 0m0.0:0 500000 000 0003 00000 00 0000000 0003 5000008300000 00 0w0000 0000 Na .0w0000
0000000 2000000: 000 00000000000 00 0000000000000 1003305 00000 0000300 0000000 0w03 0300 .00 0&0000 0000 Ha A3
000,00 0000050900053 0 000 000,00 00000£0w_0§m 0000E0m00 030

”00000000000 5 33006000000230: 000 03000000: 00303 00003000 00 000300000 00 0005000 000000 0000000m AS

 

ooom
0%
$38.8
mono
ihmwodv
mooo-
38.8
33o-
$0ooo8
moooo
3oo.8
Boo-

omoo
0%
$38.8
mono
00890.8
33.
3.8.8
goo-
$:ooo.8
moooo
«820.8
:oo-

oooo
00>
$38.8
:02
$3088
o:m.o-
38.8
Boo-
:ooo.8
0ooo.o
3oo.8
zoo-

oooo
00>
$388
$2
$388
33.
38.8
oooo-
$:ooo.8
moooo
38.8
Boo-

oooo
00>
$38.8
goo
38.8
:Eo-
32.8
Boo-
:ooo8
Hoooo
388
go.

omom 30 0: 088:2
00> 000000000 000>\000w0m
$38.8
goo oooo 08>
$338
08o- 002 08>
38.8 080
mooo- 0:903:00 0: 02080
$3808 080
moooo 2:60:00 0: 0000..
$388 080
Boo- 20:00:00 0: 00¢.

30:88 0.0 03:0

41

 

38.8 38.8 38.8 38.8 38.8 38.8 08.50 :8 8:
08.: 08.: 3:: 08:. :83- 08.8 8 :80 2:88:00
38.8 38.8 33.8 33.8 38.8 38.8 88:0 E :0
:83- 08:- 08.8 08.: 08.: 08.3- 808:0 0: 088:2
$38.8 $38.8 $388 38.8 $38.8 38.8 08.80 :8 :0
38.: 38.: 88.: 80.: 00.: 03.: 8.8080 0: 088:2
38.8 $38.8 $38.8 $32.8 $38.8 $388 800
08:- 03:- 03:- 0.3:- 03:- 03.8 8:88:00 8 00880
$38.8 $38.8 $38.8 $38.8 $388 $3.38 808:0 0:
303- 03:- 000.3. 3.3- 02.3- 80.8 888:2 8 000880
33.8 38.8 38.8 38.8 38.8 338 8:880 0:
08.8 083. 08:- :83- 38:- 03:- 808:2 8 @0880
35.8 35.8 38.8 38.8 38.8 388
2:3 08.: :0: 3:3 3:: ms: 882 08:8
388 38.8 38.8 38.8 38.8 38.8 888:2 8:88:00 0:
8:3- 08:- 0:83- 38:- 88.8 88.: 8:88:00 8:808:
$38.8 $38.8 $38.8 3.8.8 $38.8 $38.8 8:08:00 8:80:82
32.8 08.8 20:- :33- 8.0:- 08.3- 888800 x 00:20 .888
$08.8 $33.8 $38.8 $38.8 $38.8 $38.8
0.2.: :00 08.0 02.: 80.: 003.0 00:20 088::
«>0 :0 0.00 «>0 :0 0.00

 

005000000 005000300. 55>

GOSUWEOO goswpua< uSOSfiB

 

0&0000 00500000000

0EE00> 000000000

 

00000 000.00. 00me 0% 00m wEfiooEm 0059000000

3.0 03:8

42

0w0000 0000000 20000000: 00.0 00300000, 00000000000

00: 00 0000000 0000000 00 00000000 0030000 000 005 00000 00 0000000 0w03 .000000000000000 00 0900000 800 NE .0w00:0

0000000 000000000: 000 00000000000 00 0000000000000 0000003000 00000 0000030 0000000 0w03 00000 .00 0900000 0000 02 A3
.00>0_ 000005030003 00 000 000000 00000000300000: 000000030 0&0

0000000000000 00 30050000000030; 000 03000000: 000:3 30000000 00 0030000000 00 00050-5 000000 0000005 3v

 

0000
00>
$300.8
000.0
$00008
0000-
300.8
0000-
00000.8
0000.0
$300.8
000.0-

0000
0%
$300.8
000.0
$3008
0000-
300.8
000.0-
$000008
00000
$3008
000.0-

0000
00>
$300.8
000.0
$200.8
000.0-
300.8
000.0-
$000008
0000.0
$300.8
000.0-

0000
00>
$3008
0000
$300.8
000.0-
300.8
000.0-
$000008
0000.0
300.8
000.0-

0000
w®>
«#Gvodv
0000
$3008
0000-
300.8
000.0
0000.8
0000.0
$300.8
000.0-

0000
00>
$300.8
0000
$3008
0000.
0000.8
000.0-
$000008
0000.0
1320.8
000.0-

000 00 0000000 Z
000000005 000>\000w0m

ooom 000>

wag 000>
0003

20000000: .00 000000

0.80

200000000: 00 N0W0.

000m

0000000003 00 0m<

00.880 0.0 030-0

43

BIBLIOGRAPHY

Altonji, J. G., F. Hayashi, and L. J. Kotlikoff (1992), “Is the Extended Family
Altruistically Linked? Direct Tests Using Micro Data,” The American Economic
Review Vol. 82, No. 5: 1177-1198.

Arrow, K. (1964), “The Role of Securities in the Optimal Allocation of Risk Bear-
ing, ”Review of Economic Studies 31: 91-96.

Cochrane, J. H. (1991), “A Simple Test of Consumption Insurance,” The Journal
of Political Economy Vol. 99, Issue 5: 957-976.

Deaton, A. (1997), The Analysis of Household Surveys Baltimore: The Johns Hop-
kins University Press.

Deaton, A. (1992), Understanding Consumption Oxford, Clarendon Press.

Dynarski, S., and J. Gruber (1997), “Can Families Smooth Variable Earnings?,
”Booking Papers on Economic Activity Vol. 1997, Issue 1: 299-284.

Glewwe, P., and G. Hall (1998), “Are Some Groups More Vulnerable to Macroeco-
nomic Shocks Than Others? Hypothesis Tests Based on Panel Data from Peru,” Journal
of Development Economics Vol. 56: 181-206.

Grimard, F. (1997),“Household Consumption Smoothing through Ethnic Ties: Ev-
idence from Cote D’Ivoire,” Journal of Development Economics v53, n2, 391-422.

Jalan, J ., and M. Ravallion ( 1999), “Are the Poor Less Well Insured? Evidence on
Vulnerability to Income Risk in Rural China,” Journal of Development Economics
Vol.58: 61—81.

Jensen, R., and K. Richter (2002), “Social Security, Income Volatility and Health:
Evidence from The Russian Pension Crisis,” Working Paper.

Lokshin, M., and M. Ravallion (2000), “Welfare Impacts of the 1998 Financial Cri-
sis in Russia and the Response of the Public Safety Net,” Economics of Transition.
Vol. 8(2), 269-295.

44

Mace, B. J.(1991), “Full Insurance in the Presence of Aggregate Uncertainty,” Journal
of Political Economy v99, n5: 928-56

McCarthy, J. (1995), “Imperfect Insurance and Differing Propensities to Consume
Across Households,” Journal of Monetary Economics Vol. 36: 301-327.

Michael, R. (1973), “Education in Nonmarket Production,” The Journal of Political
Economics Vol.81, Issue 2, Part 1: 306-327.

Morduch, J. (1995), “Income Smoothing and Consumption Smoothing,” Journal of
Economic Perspective Vol. 9, Number 3: 103-114.

Mroz. T. and B. Popkin (1995),“Poverty and the Economic Transition in the
Russian Federation,”. Economic Development and Cultural Change Vol. 44: 1-31.

Mroz, T., D.Mancini, and B. Popkin (1999), “Monitoring Economic Conditions in
Russian Federation: The Russia Longitudinal Monitoring Survey 1992-1998,” Report
submitted to the US. Agency for International Development. Carolina Population
Center, University of North Carolina at Chapel Hill, North Carolina.

Mroz, T., L. Henderson, and B. Popkin (2001), “Monitoring Economic Condi-
tions in Russian Federation: The Russia Longitudinal Monitoring Survey 1992-
2000,” Report submitted to the US. Agency for International Development. Car-
olina Population Center, University of North Carolina at Chapel Hill, North Car-
olina.

Paxson, C. (1992), “Using Weather Variability to Estimate the Response of Savings
to 'Dansitory Income in Thailand,”American Economic Review 82 no. 1: 15-33.

Rosenzweig, M. R. (1988), “Risk, Implicit Contracts and the Family in Rural Areas
of Low-Income Countries,” The Economic Journal Issue 393: 1148-1170.

Schultz, T. W. (1975), “The Value of the Ability to Deal with Disequilibria, ” Jour-
nal of Economic Literature, Vol 13, Issue 3: 827-846.

Stillman, S. (2001), “The Response of Consumption in Russian Households to
Economic Shocks ”Rand Working Paper.

45

Strauss, J., and D. Thomas (1995), “Human Resources: Empirical Modeling of
Household and Family Decisions,” Handbook of Development Economics Vol. III A
Elsevier Science B.V.

Townsend, R. M. (1994), “Risk and Insurance in Village India, ”Econometrica Vol.
62, No. 3: 539-591.

Townsend, R. M. (1995), “Consumption Insurance: An Evaluation of Risk-Bearing

Systems in Low-Income Economies, ”Journal of Economic Perspectives Vol. 9, No.
3: 83-102.

Welch, F. (1970), “Education in Production, ”The Journal of Political Economy
Vol. 78, Issue 1: 35-39.

Wooldridge, J. M. (2002a), “Inverse Probability Weighted M—Estimation for Sam-
ple Selection, Attrition and Stratification,” Portuguese Economic Journal 1:117-
139.

Wooldridge, J. M. (2002b), “Further Results on Instrumental Variables Estimation
of Average 'IYeatment Effects In The Correlated Random Coefficient Model,” mimeo,
Department of Economics, Michigan State University.

Wooldridge, J. M. (2002c), “Econometric Analysis of Cross Section and Panel Data
”MIT press: Cambridge, MA.

46

APPENDIX

Table 1.10: First Stage Regression

 

Dependent Variable:

Income Change

Income Change
x Demeaned
Maximum Education

 

 

IV1 IV2 IV1 IV2
Panel A: Whole Sample
Predicted Income Change 1 1.000

(0.088)“
Predicted Income Change 2 1.000

(0.138)“

Control Variables:
Household Characteristics Yes Yes Yes Yes
Household Total Assets Yes Yes Yes Yes
Region/ Year Dummy Yes Yes Yes Yes
Number of Obs 4236 4236 4236 4236
Panel B: Low Asset Group
Predicted Income Change 1 0.327 -11.914

(0.293) (1.559)“
Predicted Income Change 1 0.058 1.027
X Maximum Education (0.024)“r (0.121)“
Predicted Income Change 2 0.539 -8.306

(0.281)* (1.192)“

Predicted Income Change 2 0.043 0.766
x Maximum Education (0.038) (0.121)“
Control Variables:
Household Characteristics Yes Yes Yes Yes
Household Total Assets Yes Yes Yes Yes
Region / Year Dummy Yes Yes Yes Yes
Number of Obs 2020 2020 2020 2020

47

Table 1.10 (cont’d).

Panel C: High Asset Group

Predicted Income Change 1

Predicted Income Change 1
x Maximum Education
Predicted Income Change 2

Predicted Income Change 2
x Maximum Education
Control Variables:
Household Characteristics
Household Total Assets
Region/ Year Dummy
Number of Obs

1.752
(0.358)“
-0053
(0.025)“

Yes
Yes
Yes
2216

1.165
(0.277)“
-0.014
(0.023)

Yes
Yes
Yes
2216

-3.054
(2.453)
0.313

(0195)

Yes
Yes
Yes
2216

-5.778
(0.962)“
0.553
(0.090)“

Yes
Yes
Yes
2216

(1) Predicted Income Change 1 is fitted value of income change using change of wage

earnings obtained from individual survey as predictor. Predicted income change 2

is fitted value of income change using unemployment shock, wage arrears shock and

pension shock as predictors. (2) Standard errors (robust to correlation of residuals

within households and heteroscedasticity) in parentheses *‘k significant 5%

significance level and * significant 10%significance level. (3) Household characteristics

variables: years of maximum education of the household members; change in the

number of children; seniors and household size; the number of children, seniors and

household size in the first period; age, age squared and gender of the household

head in the first period.

48

 

.1333 3.333 $333 :33 :33 333 85a E 5
33- 33- 33- N33- 333- 333. eoeom 3 2882
$333 .1333 1.333 83.3 $3.3 2.3.3 333 E a
33- :33- 333- :33- 333- 333- 8330 3 285:2
$3.3 33.3 83.3 83.3 33.3 :33 33
33.3.. 333 333- 33.3- 233. 333- 22383 a @330
1.83.3 £333 #333 1.223 162.3 1823 333
333- 33.3- 333- 333- 33- 32- 3 338:2 5 @325
63.3 33.3 33.3 63.3 33.3 63.3 5330 3
333- 333- 333- 333- N33- 333- 2352 a .336
83.3 $3.3 33.3 33.3 83.3 33.3 5330 .3. a
33.3 333 333 33 33 :33 853:3 8:852
33.3 33.3 83.3 33.3 33.3 93.3 E330 33 a
333 33.3 333 333- 333- 53- 85323.. 5:852
33.3 $3.3 $3.3 33.3 :33 33.3 3.5330 33 a
:33- 333- 333- 33.3. 333- 33.3- 3335 asses:
33.3 33.3 .333 .333 :33 .833
333- :33- 33- 333- 333- 33.3- eon-.3. 33.
.333 #333 £333 1.333 13:33 .1833
333 33 333 33 :33 333 3:30 3885
NE E So 3: E 30

 

33 3 39:5,

QED vm 55>

own-EU 2053:5380 33> .anQ

 

3:80 uomm< 304 [13ng «0mm: 305:5 m> 33>? 3:33“ :ommmoawom ”SA 2nt

49

0:530 «0:0 56:03:: 3 ~0>0~ £59m 400500 3:05:09, mm _0>0_ EBB .Am0wgw oHHV

. :0330300 53::0000 130:0»: mm _0>0_ ::000m 383% 3 VV .8333? 3:30:88 3:0:0w 30:0: 3 _0>0_ :0530200 3:583

$53233 3:053:05 05 mm 350:3 :0_m:0: E 0w§fio ”5:9: 0:0 :35 0:05 0: 0.80:3 0mm? 65:32:53: .«0 0w:-m;0 80: NE

038:0 0:805 20:03.0: :8 3:063:05 00 0305050030 33:03:05 80b 8:850 0:505 0mm? :30... mo 0w:0;0 000: CC A8

.~0>0_ 00::0m3w6332 3:05:30 3 :90 ER: 00:85:30R0m. 3:005:30 I

0000558: 5 3205038002030: 98 3:050:03 :23? £03300: “0 :0530300 0... 30:20.; 20:0 :8::3m 2V

 

 

3333
8>
$:3.3
333
«.waodv
3:3-

833
333-
$:33.3
3333
Lmoodv
333-
$633
333

3333
33>
$333
333
33.3
333-

C33
333-
$:333.3
3333
$833
:33-
$:3.3
333

3333
33>
$833
333
$833.33
333-

$333
333-
$:333.3
3333
*Amoodv
333-
$333
333

3333
33>
$:3.3
333
$83.33
:33-
$333
333.
63.3
:33-
:333
333
333.3
333-
83.3
333

3333
00>
$333
333
$833
333-
$333
33.
33.3
333
:3333
3333
33.3
333-
33.3
333

3333
m®>
$333
333
$:3.3
333-
633
3:3-
333
:33-
:3333
3333
333.3
333-
:33
333

30 “0 30:8: Z
00:88:: 30>\:0_w0m

ooom 30>
wag 50>

82 330%

30m

20:03:02 “0 30:50
30m

20:03:03 “0 ~0w<
@303

20:03:03 30 0m<
vow-Sm 33H 05

5 03m 20:03:0m

.2583 2.: 233-

50

 

$8.8 $8.8 $8.8 $8.8 38.8 38.8 825 E 2:
8o.o- 83. ~88. zoo- zoo- Boo- a 38 223.2%
$8.8 88.8 88.8 88.8 €88 $8.8 Bram E 5
oooo oooo 8o.o- o8.o o8.o oooo 158% 8 38:2
1288 188.8 188.8 188.8 188.8 188.8 Beam 2 E
888 oooo Goo wooo goo 88o 9:220 8 58:52
1888 188.8 188.8 1688 1:88 £288 38
goo- goo- goo- go- Eoo- Eoo- 2238: a 855
188.8 188.8 188.8 18988 18898 13888 958% o0
838. oooo- «So- oodo- 2:8. 83. 38:2 a @8526
$88 88.8 88.8 :88 88.8 38.8 5:230 o0
888. M88. mooo- M88. 88. 988. 38:2 5 8:30
38.8 $8.8 $8.8 88.8 88.8 88.8 boosoo £8 a
oooo- oooo- 8o.o «.88 ovoo goo 838:5 8:852
88.8 :88 38.8 88.8 88.8 88.8 8828 Ho 8
goo- moo- 8o.o- «.88 «88 $88 oossoom 8:852
88.8 88.8 88.8 88.8 $8.8 88.8 30380 EN a
888. go. goo- oooo oooo goo 5:828 8:852
:88 88.8 88.8 88.8 88.8 88.8
Eoo Eoo o8.o oooo oooo 8o.o 33$. 38
188.8 188.8 188.8 1888 1:88 188.8
8:8 83 oooo oodo oEo wooo 8526 0885
82 So So NZ go So

 

83 8 82:3

33 8 55>

 

@95on sowoaasmcoo ”.83 .onQ

 

@280 ommm< :wi 1885 $2 820355 m> 55> mfizmom commmmowom ”NH; 2an

51

.264: 98 3853:: mm ~92 :35.“ 48:8 3:058? E 732 BEE A8688 SUV
. 583:3 DEESmm 855w 8 6%: 388m .Awowfiw 3 VV 558:8 5:503 885% 323 mm ER: 538:? ambamv
838:8, «5:5,:me 2: mm 989:: 588: E mwfifio ”5:2: 28 :8: v.88 0: 38b: emu? 5553382: We vwfifio mom: Na
6885 2:85 20585: :8 “=88:me ma 38::2826 833365 Eob 8:830 2:85 am?» 138 go wwfifio 8m: H>~ A3
:96: 8585:8989 €85:me 1 :5 ~86: 8:8cfiwmmo8m “:85:me fr

$8558: 5 Amfiofimdeomohmumn v5 mEonmmso: :3me flag—Ema: mo :052280 3 38508 $0th @8285 AC

 

 

 

 

S8 88 88 S8 98 28 80 8 38oz

m®> m®> m®> wm> mm.» m®> @285va Hmm>\flomw.®m
1:88 1888 188.8 168.8 1:88 1:88

:88 o8.o v88 83 :88 88o oooo 5%
188.8 188.8 1:88 1:88 1:88 1:88

88o- mono- 888- 82- oooo- :88. 82 50>
88.8 88.8 :88

~88- oooo- «88 o8: 5%

:88 :88 :88 :88 88.8 :88 :8:

88.8 83 8.8 83 8o.o 83. 2288: 8 “880

188.8 188.8 188.8 88.8 88.8 888 So:

888 8oo.o 888 888 8oo.o 888 2238: 8 031

188.8 188.8 188.8 88.8 88.8 88.8 88:

Eoo- Boo- o8.o- ooo.o- oooo- Boo- 22:80: 8 a?

8.888 2.: 2%?

52

 

188.8 188.8 1:88 1:88 1888 1:88 :28: 8:
:88 888 888 888 888 8.88 2: a 88 88888:
1:88 1:88 1:88 1:88 1:88 1:88 :28: 8: 2: 5
888- 888- 888- 888- 88- 888- 8888 8 8:882
1:88 1:88 138.8 88.8 :88 1:88 8:8: 8: 2: a
888- 888- 8o.o- 888- 888- 888- 8880 8 8:882
:88 88.8 88.8 :88 88.8 88.8
888- 88. :88- 888- 888- 888. 88 88888: a 8880
188.8 1:28 1:88 1:888 1888 182.8 8888
888- 888. 888- 888- 888- 888- 8 8:882 2: a 88:0
88.8 88.8 :88 :88 :88 :88 888:0
888- 888- 888- 88. 888- 888- 8 8882 2: 8 88:0
1:88 :88 1:88 :88 88.8 :88
:88- 88- 888- 888- 888- 88. :88 88:-
:88 $8.8 :88 888.8 :88 :88 888882 88888: oo
88 888 88 888 88 888 8:88: 888882
1:28 1888 1:88 1:28 1888 1:88
8:8 888 888 88 :88 888 88:0 8885
82 :2 8:0 82 E So

 

8:058:00 :o§:S< 5:5

5:38:00 BUSES-8: 305:5

 

mwSEO 8858558580

”83883 8:288:95

 

CEBU-8:85 983833. gofimgv 95:0 8088:. 304 :8 wEEooEm 85:83:00 ”24 @388

53

8888
m®>
18888
888.8
18888
888.8-
«vaodv
8888-
1888.8
8888.8
1888
888.8-

8888
w®>
18888
888
1:88.88
8888-
18888
88.8-
1888.8
8888.8
1:88
888.8-

.9588 8:8 38-83:; mm 3?: 530m .3083. Enos-80> mm .832 8.838 .Awmfiflw SUV
. 85:89:88. E88888 88.85% £ ER: 853% A883» 3 VV coma-8:888 88.88.8888 355% 3285 2. ~92 88382888 888593
@5833 30538.85 23 md 8.88:8 :06ch E mwsdgo ”5:9: 28 58:8 808 o: 8888 .8me 385588383: .80 $530 mom: Na
.8955 2:85 83:88:08 5.8 8888:me m8 Padanoflmmsc 88:88:85 508 885830 2:85 “Wm-83 833 we mmfifio 8m: :3 A8
.832 won-8088:3882 8:865:me .8 8:8 _m>w_ mogocammfim 8:85:me f8
mwmmficoamm E Qflofimaggggg 8:8 £83535: 5585 E83868 mo .8883.on 08 $8505 8.88.8 8.8855 CV

8888
88>
18888
888.8
iAvadV
8888-
1:888
8888-
1888.8
8888.8
1888
8888.

8888
88>
18888
888.8
1888
88.8-
8888
88.8-
1888.8
8888.8
:Afifiodv
888.8-

.880 mo 888: Z

828859 88o>\50_wom

ooom “88>

wag 58>

883-“ 8680850: mo .8850
888: 83888:on mo mmw<

88mm anwmsom mo ow<

8,888 8: 888,8

54

 

$8.8 $8.8 $8.3 coooo 38.8 33.8 35 E
oooo oooo oooo Hooo- Hooo- oooo- 2: a 3% Eofimoom
$8.8 $8.8 88.8 63.8 smog 38.8 cocoa o2 2: 5
oooo- Hooo- oooo- mooo- oooo- oooo- 228m o0 5952
Looooo Looooo Looooo $8.8 Looooo Loooov 35 E 2: 5
oooo oooo ovoo oooo oooo ovoo 5620 E soaoz
88.8 Loooov Loooov Looooo flooov Looooo
«moo- Boo- oooo- goo- Soo- oooo- mam Eoogoom a omoooo
r838 Ltooo :Eoov tfloooov $38.3 .9383 228m
$8- 83. ES. $3. $3. oodo- o0 352 2: a mango
$8.3 Cooov :oooo sooov Cooov Cooov 8.220
Boo- Soo- Hooo- :oo- :oo- oooo- oo soaoz 2: a voodoo
aoooo Afioov $8.8 $5.8 Loooov Afiooo
Soo Soo oooo oooo oooo oooo 3&2 Eco
$8.8 $8.8 cooov $8.3 $8.8 38.3 28%: ooooomoom oo
Hooo- Hooo- ooooo- ooooo- ooooo- ooooo- :osoosom 8:852
$38.3 rcoooo $88.8 $35.8 :Coooo 135.3
go oooo Soo moao oofio oooo mango 2:85
NE 5 £0 NE So So

 

2050250 coSEo< 55>

:oSomtoO ”8353‘ 393:5

 

wwSfiO cosqfigaoo

”wamEdN/ unowcoaoa

 

Acosowoofi: ~568st osofigc anew oomm< swam .5“ wEfiooEm 5:385:00 ”EA 2nt

55

 

.9690 0:0 36003:: mm _0>0~ 5.59m 30:00 1503009, $ _0>0_ EBB .Ammvflw 3H8

. 00:00:00 0000:0000 1.200% E ER: 008% $0an 3 VV 00300300 38000000 350% 3303 mm 62: 00300300 uBEAS
mflnflg 0:08:59: 05 00 802.8 020000 E 030:0 50:08 000 :2: 0008 0: 9800.3 0w?» #:0050383: .80 030:0 89 NB
.mwzafi 08005 20:09.0; 00w 3083505 no 0035530000 2033?: Bob 000850 08005 0mg, :38 80 098:0 wow: 35 Amy
40%: 00:00aEwmmvooH 0:005:me « 0:.» ER: mocaocawmwfixon 0:03:36 I

83500an 5 30803000080080: 0:0 0205000: 555 015208 “0 003200000 00 00000.8 80:0 vowvnwum A3

 

SE SE oao Boo Boo oao .30 oo soaoz
m®> mm> mm.» m®> mo> m®> momaafla udm>\fimomwmwm
$33.8 $33.8 $338 $338 $33.8 $33.8
20o oao oooo :oo oao mooo oooo 5% %
$28.8 $38.8 $28.8 $38.8 $33.8 $23.8
zoo- Homo- mono- go. go. oooo- oooo 3%
38.8 33.8 38.8 38.8 38.8 38.8
oooo- mooo- oooo- oooo- oooo- zoo- 3% 220200 3 350

£888 L888 L888 $:ooo.8 $:ooo.8 $:ooo.8

ooooo ooooo ooooo ooooo ooooo ooooo 2% 223.8: oo $3.
$38.8 $38.8 $388 38.8 $38.8 $38.8

Boo- Boo- Boo- zoo- zoo- 38o- Ham 2302550 e?

.3088 o: 2330.

 

Table 1.15: Consumption Smoothing for Pooled Asset Group (With Education
Interaction; With Attrition Correction)

 

Dependent Variables:

Consumption Change

 

 

OLS IV1 IV2

Income Change 0.062 0.257 0.169
(0.020)“ (0.044)“r (0.058)“r

Income Changex Demeaned -0.007 -0.031 -0.024
Maximum Education (0.005) (0.014)“ (0.021)
Maximum Education -0.0002 -0.0003 -0.0004
Of Household Members (0.003) (0.003) (0.003)
Total Assets -0.003 -0.001 -0.002
(0.008) (0.008) (0.008)

Change in the Number of -0.035 -0.035 -0.036
Children (0.042) (0.045) (0.043)
Change in the Number of -0.239 -0.214 -0.229
Seniors (0.057)“ (0.059)W (0.059)“
Change in Household Size —0.044 -0.034 -0.037
(0.027)* (0.027) (0.027)

Number of Children -0.0002 -0.004 -0.003
In the lst Period (0.019) (0.020) (0.019)
Number of Seniors —0.062 -0.059 -0.062
In the lst Period (0.030)“ (0.030)* (0.031)“
Household Size in the 0.025 0.026 0.026
lst Period (0.015)* (0.015)* (0.015)*
Age of Household Head -0.015 -0.016 -0.016
(0.006)“ (0.006)“ (0.006)“

Age2 of Household Head 0.0002 0.0002 0.0002
(0.0001)Hr (0.0001)** (0.0001)**

Gender of Household Head -0.028 -0.028 -0.029
(0.026) (0.027) (0.027)

Year 1998 -0.300 -0.229 -0.261
(0.048)“ (0.050)“ (0.051)“

Year 2000 0.302 0.271 0.285
(0.036)“ (0.039)“ (0.038)”

Regioin/Year Dummies Yes 4395 Yes
Number of Obs. 4236 4236 4236

 

See notes in Table 1.14

57

Table 1.16: Consumption Smoothing for Pooled Asset Group (With Asset Inter-
action; With Attrition Correction)

 

Dependent Variables:

Consumption Change

 

 

OLS IV1 IV2

Income Change 0.006 0.511 0.342
(0.075) (0.170)“ (0.262)

Income Changex Asset Group 0.006 -0.036 -0.025
(0.009) (0.020)* (0.033)

Maximum Education 0.0001 -0.0001 -0.0002
Of Household Members (0.003) (0.003) (0.003)
Total Assets -0.003 -0.0013 -0.001
(0.008) (0.008) (0.008)

Change in the Number of -0.032 -0.034 -0.035
Children (0.042) (0.045) (0.043)
Change in the Number of —0.237 -0.211 -0.226
Seniors (0.057)“ (0.059)“ (0.059)“
Change in Household Size -0.045 -0.038 -0.040
(0.027)* (0.027) (0.027)

Number of Children -0.005 -0.005 -0.004
In the lst Period (0.019) (0.020) (0.020)
Number of Seniors -0.061 —0.058 -0.061
In the lst Period (0.030)“ (0.030)* (0.030)“r
Household Size in the 0.025 0.025 0.025
lst Period (0.015)* (0.015)* (0.015)*
Age of Household Head -0.015 -0.016 -0.016
(0.006)“ (0.006)“ (0.006)”

Age2 of Household Head 0.0002 0.0002 0.0002
(00001)“r (0.0001)** (00001)“

Gender of Household Head -0.026 -0.025 -0.027
(0.026) (0.027) (0.027)

Year 1998 -0.302 -0.231 -0.263
(0.048)“ (0.051)“ (0.052)“

Year 2000 0.303 0.270 0.285
(0.036)“ (0.039)W (0.038)“

Regioin / Year Dummies Yes Yes Yes
Number of Obs. 4236 4236 4236

 

See notes in Table 1.14

58

Table 1.17:

Correction)

Consumption Smoothing for Pooled Asset Group(With Educa-
tion/ Assets Interaction; With Attrition

 

Dependent Variables:

Consumption Change

 

 

OLS IV1 IV2

Income Change -0.097 0.367 0.239
(0.097) (0.176)Mr (0.272)

Income Change x Asset Group 0.064 0.022 —0.007
(0.060) (0.116) (0.205)

Income Changex Demeaned 0.015 -0.008 -0.022
Maximum Education (0.032) (0.062) (0.124)
Income Changex Asset Group -0.003 —0.003 —0.0002
x Demeaned Maximum Education (0.004) (0.008) (0.015)
Maximum Education -0.0004 -0.0004 -0.0004
Of Household Members (0.003) (0.003) (0.003)
Total Assets -0.003 -0.0005 -0.002
(0.008) (0.008) (0.008)

Change in the Number of -0.033 -0.039 -0.038
Children (0.042) (0.045) (0.044)
Change in the Number of -0.234 -0.215 -0.231
Seniors (0.058)“ (0.059)“ (0.060)“r
Change in Household Size -0.045 -0.035 -0.037
(0.027)* (0.027) (0.027)

Number of Children 0.001 -0.005 -0.003
In the lst Period (0.019) (0.020) (0.020)
Number of Seniors -0.061 -0.060 -0.063
In the 1st Period (0.030)“ (0.030)“ (0.031)“r
Household Size in the 0.025 0.026 0.026
lst Period (0.015)* (0.015)* (0.015)*
Age of Household Head -0.015 -0.017 -0.016
(0.006)“ (0.006)“r (0.006)“

Age2 of Household Head 0.0002 0.0002 0.0002
(00001)“r (0.0001)** (0.0001)**

Gender of Household Head -0.026 -0.029 -0.030
(0.026) (0.027) (0.027)

Year 1998 —0.299 -0.228 -0.262
(0.048)“ (0.050)“ (0.051)“

Year 2000 0.302 0.271 0.285
(0.036)“ (0.039)“ (0.038)W

F —test of Interaction Terms 3.02 2.00 0.71
p — value of F statistics 0.0286 0.112 0.548
Regioin/Year Dummies Yes Yes Yes
Number of Obs. 4236 4236 4236

 

See notes in Table 1.14

59

Chapter 2

Multiple Job Holdings As a Way
to Smooth Consumption: Labor
Response to Wage Arrears
Among Russian Couples

2.1 Introduction

During ten years of economic and social transition, Russia has witnessed a weak
correlation between employment change and output change: the declines in total
employment were substantially below the steep declines in GDP 1. Instead of
mass layoffs, wage arrears, unpaid administrative leave and short-time work were
common practices adopted by the employers as adjustment mechanisms. Among
these practices, wage arrears were the dominant form of labor market adjustment

and the main source of insecurity for Russian workers (Lehmann, Wadsworth and

 

1Between 1990 and 1998, GDP fell by about 45%, while aggregate employment declined by
about 16%. OECD (2001)

60

Acquisti (1999)). At the end of 1998, nominal wage arrears reached 77 billion
roubles, which was equivalent of 200 percent of the monthly wage bill. After three
years of recovery from the 1998 financial crisis, by the end of 2001, the stock of
wage arrears was reduced to 29.9 billion roubles but was still higher than the level
of 17.5 billion roubles in 1995.2 Wage nonpayment was also pervasive in terms
of the percentage of working people affected: about 64% of working age men and
women were owed wages in 1998 and this number was still over 25% in 2001(Mroz,

Osmolovskii and Popkin, 2002).

Various studies have shown that workers respond actively in one way or another
to wage arrears. For example, using nationally representative household survey
data together with matched firm-individual data, Earle and Sabirianova (2002)
found that wage arrears have a positive impact on workers’ quits in the regions
where wage arrears level are less than average. Lehmann, Wadsworth and Acquisti
(1999) also found that wage arrears positively affect the incidence of job-to-job
movement and such effect is strengthened by the viability of the outside labor
market. Foley(1997) looked at the individual pattern of transitions between labor
market states from 1992 to 1996 in a pooled cross section data set and found that
wage arrears had no effect on propensity to switch jobs; however, wage arrears
increased the probability of taking on additional work (Foley 1997b). Desai and
Idson (2001) detected worker-initiated turnover among workers subjected to wage

nonpayment. Moreover, they found that although wage arrears had no effect on

 

2Russian Economic T‘rends,June 1999,vol.8, no.4 ; December 2002, vol 11, no.4

61

the tendency of workers to hold more than one formal job, wage arrears did have

distinct labor supply effects with respect to less formal supplemental work.

Given that labor markets are flexible, any labor response, particularly an increase
in household labor supply in secondary jobs, might mitigate the negative impact
of wage arrears in the primary job on household consumption. It has been well
known that labor supply is an alternative to dissaving, asset decumulation, or
increasing debt in the attempts of households to maintain consumption in the
face of declining income (Mincer 1962). Various studies have focused on the labor
market responses, such as hours worked, self-employment and household labor
supply to economic shocks 3. Empirical evidence also shows that shift of labor
from farm to off-farm employment can explain the observed lack of correlation
between consumption and idiosyncratic crop shocks in rural India (Kochar 1999).
To test the hypothesis that in Russian economy the labor response helps to mitigate
the impact of wage arrears on household consumption, we focus on multiple job
holdings by household members. 4 Applying a test proposed by earlier research
(Kochar 1999), our reduced-form regression of consumption on wage arrears reveals

no significant effect of wage arrears in the primary job on household consumption. 5

 

3Smith, Thomas, Frankenberg, Beegle, and Teruel (2002) on Indonesia; McKenzie (2003a) on
Mexico and Mckenzie (2003b)on Argentina.

4The impact of wage arrears on total working hours is ambiguous because it might reflect
job changes. Furthermore, changing primary jobs need not reflect the desire of the household to
smooth their consumption because with well-functioned labor market, job-specific shocks would
result in a shift to new jobs, even if households have access to insurance market (Kochar 1999).

5This result is consistent with the result in our previous findings that the consumption of
Russian households was relatively well protected from shocks such as wage arrears, pension
arrears and unemployment.

62

Conditional on multiple job holdings, however, wage arrears have a negative effect
on consumption. This result confirms that the ability to smooth consumption in
times of wage arrears reflects, in some part, multiple job holdings of the male and

female household heads.

The experience of wage arrears by one individual can trigger the change of worker’s
labor allocation from one job to another job or from single job holding to multiple
job holding, as documented in the previous literature. At the same time, it may
also affect the consumption-leisure decision within his / her household and change
the labor supply of other household members. Previous theoretical analysis high-
lights the implications of income shocks on spousal labor supply in a household
life-cycle model in which individual’s labor supply increases when exogenous shocks
experienced by their spouse causes a transitory reduction in their household in-
come.6 In the case of wage arrears, if individual worker responds to their own
wage arrears by increasing labor supply, then the cross effect of wage nonpayment
on spouse labor supply might be subdued. In our estimation of household labor
supply responses to wage arrears shocks, we analyze how individual labor supply
responds to wage arrears experienced by the spouse. Unlike previous studies of
“the added worker effect ”, which focus on the wife’s labor supply response to the
husband’s labor market shock, we examine the labor supply response of both wife

and husband to the wage nonpayment experienced by their spouse. It is necessary

 

6The effect of a husband’s job loss on the labor supply of his wife is known as the “added
worker ”effect (Mincer 1962; Heckman and MaCurdy 1980; Lundberg 1985; Stephens 2001; ect.)

63

in the context of the Russian labor market to include both women and men in the
analysis because the labor force participation rate of women is comparable to that

of men.7

In our estimation of multiple job holdings as well as job-to-job movement, we apply
the “Chamberlain ”approach to a dynamic probit model. This model explicitly
allows individual fixed effects to be correlated with time varying variables including
the incidence of wage nonpayment. It is especially useful because the unobserved
time-invariant individual characteristics may very well be correlated with both the
labor supply decisions and the experience of wage arrears. We also allow the effect
of first time wage arrears on the labor supply to be different from wage arrears
in general. Such difference is important because it allows us to learn how people

respond to repeated shocks such as wage arrears.

We find that both husband and wife are more likely to take secondary informal
job when they have wage arrears shocks in their primary jobs. Moreover, they
are more likely to change their primary job after they experience wage arrears in
their primary jobs. In the reduced form regression of household’s consumption
on wage arrears, we find that the household consumption is very well protected
from wage arrears shock. But the results in the structural equation shows that

household consumption falls with the wage arrears shocks experienced by the wife.

 

7Labor force participation rate of women was 83.3%; 84.1%, 82.3%, 80.7% 78.7% and 78.9%
in year 1994, 1995, 1996, 1998, 2000 and 2001 respectively. The labor force participation rate of
men in the same period was 87.9%, 88.2%, 86.6%, 84.4%, 85.9% and 86.1%. (Mroz, Henderson,
Bontch-Osmolovskii and Popkin, 2001)

64

The results suggest that overall insignificant effect of wage arrears on consumption

reflects adjustment of secondary job holdings.

The remainder of this paper proceeds as follows. Section 2 reviews the literatures
on the causes of wage nonpayment in Russia and describe the features of wage
arrears with descriptive statistics from RLMS. Section 3 presents the theoretical
model of household labor supply with uncertainty and discusses the empirical
specification strategy. Section 4 discusses the dataset used in the analysis. Section

5 presents the central results. Section 6 concludes.

2.2 Wage Arrears In Russia

2.2.1 The Causes of Wage Arrears

Wage arrears in Russia are not an independent and isolated phenomenon. They
were rather a part of more general and pervasive payment arrears, which included
nonpayment from government to its suppliers in industries; nonpayment of tax
from enterprise to government and nonpayment from enterprise to enterprise. All
these arrears were to some extent inter-correlated in a vicious circle (Gimpelson
and Lippoldt 2001; Ivanova and Wyplosz 1999). The causes of wage arrears and
nonpayment in general have been the focus of many economic studies. At a macro
level, tight monetary policy in the mid-19905 targeted at inflation is often cited

as one possible root of nonpayment, because it induced economic depression and

65

cash shortages. Two features of fiscal policies during this period, high tax rates
and budgetary cuts are also believed to have caused the wage arrears. High tax
rates created incentives for firms to hide cash, while the budget cuts resulting
from the budget deficit targets specified by the International Monetary Fund by
the end of 1995 caused persistent underfinancing of the army as well as other public
sectors and generated mass late wages and pensions. Absence of regulatory control
also contributed to the widespread wage arrears, because it left firms managers
with unchecked powers to manage money and unscrupulous managers were not
held responsible for delaying or withholding wage payment to workers (Earle and

Sabirianova 2002; Desai and Idson 2001;Gimpelson and Lippoldt 2001).

In terms of firms’ decision regarding wage arrears at micro level, one interpretation
is that wage arrears reflects an implicit contract between firms and employees.
According to this theory, the firm proposes a contract that includes a low monetary
wage and access to social service. The most productive workers leave the firm and
concentrate in the most productive firms; the less adequate workers remain in their
initial firm. In this framework, wage arrears can be viewed as an element of implicit
contract between firms and less productive workers (Grosfeld et. 2001). Another
View treats arrears as the outcome of generally poor economic performance of firms.
Unprofitable firms lacking cash and run up arrears as a way for them to cut costs
and stay afloat (Lehmann, Wadsworth and Acquisti 1999). This response may be

especially attractive when firms face the political and bureaucratic obstacles to

66

layoffs, such as costly regulations and approvals that have to be met. Or managers
may not want to create conflicts with local governments unwilling to allow large
open unemployment (Commander et al. 1996). Wage arrears could also be a device
used by managers to extract subsidies from the government, especially by firms
with close ties to federal or local governments or those with greater bargaining

power (Alfandari and Schaffer 1996).

Given the decisions of firms with respect to wage arrears, Russian workers seem
to have been very tolerant of such practices. Fear of unemployment might ex-
plain why workers accept deteriorating employment conditions. It is particularly
aggravated by the low level of unemployment benefits, which was also subject to
arrears(Clarke 1998). At the same time, keeping an “old” job could provide at least
some in-kind compensation, particularly enterprise-related social benefits, which
include housing, nursery, health care and recreation benefits, and added about 5
percent to total labor costs in general 8. The larger the fringe benefit portion of
compensation was, the stronger the incentives for workers to accept a backlog in
wages (Gimpelson and Lippoldt 2001).Weak union is also cited as a factor leading
to more wage arrears (Connor 1995). More importantly, when other firms in the
region also pay late, the employees of a late-paying firm become less likely to quit,

reduce effort or to strike. Such a self-propagation feature of the wage-arrear prac-

 

8The composition of fringe benefits in the Russian system is different than in the US. system.
The fringe benefits in the US. include payments to private retirement systems as well as to
life insurance, health benefits and other agreed-upon plans. Legally required payments such as
Social Security and unemployment insurance contributions are normally not counted as the fringe
benefits. Fringe benefits in the US. accounted for about 9.2%-l6.1% of total compensation in
the 19805 (Woodbury 1983).

67

tice makes it difficult for workers not to accept wage arrears (Earle and Sabirianova

2000).

2.2.2 The Distributions of Wage Arrears on the Primary
Jobs

Official information on wage arrears in Russia is limited to aggregate levels of
cumulative overdue wage debts. We use Russian Longitudinal Monitory Survey
(RLMS) (1994-1996, 1998, 2000, 2001) micro level data to describe the magnitude,
persistence and the distribution patterns of wage arrears on the primary jobs of
the respondents. The incidence of wage arrears could be measured separately by
two survey questions. The first one is “At the present time, does your primary
place of work owe you any money which for some reasons they didn’t pay you on
time?” This question measures the incidence of all wage arrears, which in Table 1
we call “owed wage ”. Since wage payment is very uncertain and workers got paid
periodically, we also use the second question to measure the “flow ”of wage arrears.
The respondent was asked “Tell me please, at your primary place of work in the
last 30 days, did you get a sum of money as wages, bonuses, benefits, revenues,
profits ? ”This question measures the incidence of wage arrears in the survey

month and we name it “working without payment at current month ”.

As we can see from Table 2.6, wage arrears have been prevalent over the years

since early in the 1990’s. About 40% of respondents reported that they were owed

68

overdue wage at the survey dates in 1994 and 1995, with higher proportion for men
than for women. The wage nonpayment rate rose to about 60% in 1996 and 64%
at the time of financial crisis in 1998. During the period in which the economy
was recovering from the financial crisis after 1998 , the rate of wage nonpayment
dropped to 33.2% for men and 26% for women in 2000. It further decreased to
26.3% for men and 21.7% for women in 2001. Under the concept of wage arrears on
the current month (3rd and 4th row of Table 2.6), the rate of wage nonpayment is
lower than the rate measured by “whether currently owed money by the employer
”, because people who have wage arrears might choose to quit working or change
jobs. Moreover, workers irregularly got repayment of back wages. The magnitude
of arrears can be measured by “months of wages not paid ”. The average months
of wage nonpayment was about 2.8 months in 1994 and rose to 6.0 months for
men and 4.7 months for women in 1998. In 2000 and 2001, even though the
incidence of wage arrears was much lower than previous years, the magnitude of
wage nonpayment was not getting much better, with an average of 3 months in

2001.

Variation of wage nonpayment across a number of individual characteristics was
also large. Figure 1.1-Figure 4.2 depict the patterns of wage arrears by age, ed-
ucation, tenure and relative monthly wage of individuals. The age effect (Figure
2.1 and 2.2) changes over the years, but for both men and women, young work-

ers (age 18-29) are less likely to have wage arrears than older workers. Arrears

69

are generally negatively related to the level of education completed, especially for
male workers (Figure 2.3 and 2.4). More educated workers are likely to have more
alternative job prospects and thus less likely to continue working without payment
at primary jobs. Workers with primary or less education were the most vulnerable
to wage arrears, especially during the period of financial crisis in 1998. Tenure is
generally positively related to the incidence of wage nonpayment (Figure 2.5 and
2.6). It might be explained by the fact that workers with long tenure might have
less outside opportunity because of their specialized skills. New employees with
tenure less than 1 year are far less likely to have wage arrears. The mobility cost
for a new employee might be lower thus they are more likely to change job after
nonpayment. Another explanation might be the fact that they were exposed to
less incidence of wage arrears because of their short tenure with the employees.
Figure 2.7 and 2.8 shows the relation between incidence of wage arrears and rela-
tive real monthly earnings 9. They imply that people in the lower earning quartile
have a higher incidence of wage arrears. The causality of wage arrears and wage
earnings can go either way. Wage nonpayment may affect people at the bottom of
wage distribution most if the managerial decision is to allocate wage arrears to the
lower-paid, less-skilled employees (Desai and Idson 2001). On the other hand, wage
arrears have negative effect on people’s earning and thus people who have wage

nonpayment may be more likely to be in the bottom of the earning distribution.

 

9The survey doesn’t have the information on actual wage, instead it asked the respondent
“How much money in the last 30 days did you receive from your workplace after taxes? ”.

70

Table 2.6 lists the wage arrears by occupation. In order to have enough observation
numbers in each occupation-gender-year cell, we look at the distribution of wage
nonpayment at the aggregate level, particularly before and after 1998. Consistent
with the data in Table 2.6, there is a substantial drop in wage arrears in most of
the occupations after 1998. Within each period there exists considerable variance,
with rates ranging from 76.2% for men who are “agricultural and fishery laborers
”to 14.8% for “models, salespersons and demonstrators ”. The rate of wage non-
payment is high in the semi-skilled job such as “machine operators and assemblers
”; or “drivers and mobile-plant operators ”. It is relatively low among profession-
als like “general managers ”. This is consistent with the wage arrears distribution
among different education categories displayed in Figure 2.1 and 2.2. Some fields
such as “teaching professionals ”and “physical mathematical , engineering science
professionals ”also experienced a high rate of wage arrears regardless of the high-
level education of the employees. This phenomenon might be explained the fact
that these jobs would fall under the budget sectors and government retrenchment
in the late 1990’s increased the rate of wage nonpayment in these areas. Within
each occupation, women are not necessarily less likely to get wage arrears although

in general women have less incidence of wage arrears as showed in Table 2.6.

71

2.3 Theoretical Framework and Empirical Issues

2.3.1 Theoretical Framework

This section examines how one might expect the likelihood of taking multiple job
to vary with the experience of wage arrears in the primary job. Further, we explore

the implication of multiple job holdings on household consumption decision.

Households’ Choice of Multiple Job Holdings

We begin by considering a simple theoretical model of a household’s choice of
leisure and consumption. An analysis of households’ choice of multiple job holdings
involves the specification of a model under uncertainty whereby observed choice
depends on households’ wage payment in primary jobs, thus the realization of the
wage arrears shock. Assume households exist until time T and their utility function
depends on household total consumption C as well as the leisure of male and female
household head. The leisure is represented by discrete choice variable im and if,
each of which stands for male and female member’s choice on multiple job holding.
Let the wage of male and female household head in primary job be WMm and
WMf. The realization of their wage arrears shock in the primary job is represented
by P3,! and PIM’ And we denote the nonpecuniary benefits that are associated

with having the primary job by B. If male or female household head chooses

to take multiple jobs (im = lor if = 1), we assume they can get wage payment

72

W5", and st at the non-primary job(s) without risk. The households’ objective
function at any point of the life cycle is then described by the maximization of the

expected utility subject to the budget constraint:

1

T
MaxUt = EtZWU(CT’lmT’lth) (2.1)
T=t

At = (1+ 7‘)At—1 — Ct—l + WMmt1(PMmt=1)+ WMft1(PMft = 1) +

+ Wsmtl(imt =1)+ WSft1(ift=1)+ B

CT = AT (2.2)

where 1(.) is the indicator function of the event between brackets—for instance,

“male household head taking multiple jobs period t ”is 1(imt = 1) = 1

Because PMmt = 1 and PM ft = 1 refer to the male and female household head
getting full wage payment at primary job in period t, we can write the probability

of full wage payment as the following:
PT(P1Wmt = 1) = Et[1(PMmt = 1)] = Pmt; (pmt 6 [0,1]) (2.3)
and
Pr<PMft =1) = Et[1(PMft = 1)] = pft; (pf. e [0.11) (2.4)

In order to explicitly derive how households’ choice of taking multiple jobs and the
wage arrears at the primary job are correlated, we take two-period case (T - 1,
T) as an example. In this finite-horizon discrete case, if the solution exists, this

maximization problem is equivalent to solving the Bellman (1957) equation. We

73

denote the value function of household h by V” and the discount factor by p, the

Bellman equation is written as:

Vh(AT—1) = )viaxU<CT_1.z'mT_1.z'fT-1) + pETMaxUIZT + W). 9(PTT- — 1)+
W’}%(Pf”%- — )+ Wsmtm'm. = 1) + W file ft = 1). z'mT, 2'le

(2.5)

where ZT = (1 + r)AT_1 — CT_1 + B. By the iterated expectation, we have:

 

ETU[ZT + W,,,Ti(P,n’T= 1+) wa1(P fT— _ )+ WSmtl(imT = 1) +
WSfT1(ifT =1))imT,ile

= ETEpU[ZT + W,’,’,"T(P,’,’,’T= )+ WTT(PTT— _ 1) + W5 mT1(imT—- — 1)+
WSfT1(ifT = llaimTvilePmTvaTl

= PmTPfTEtUlZT + W117iT+ +WfT + W SimT1( l—mr — 1) +

WSfTWfT = 1), z"1nT)"’le + l1 - PmTlll - PleETUlZT +

WSmT1(imT = 1) + WSfT1(ifT =1),imT,ile+PmT[1- PleF/I‘UlZT +
WnT + WSmTl(z'mT- — 1) + WSle(sz-—1),imT,2fT]+
Plel - PmTlETUlZT + Wj’f + WSmT1(imT = 1) + WSfT1(ifT = 1), imTvile

(2.6)

We treat the leisure of male and female household members equally in the house-
hold utility function. And also we assume that leisure is a normal good. Let
U1: U(C,0,0); U2 = U(C, 1,0) = U(C, 0,1) and U3 = U(C,1,1). The assump—

tion implies, given the consumption level, household’s utility will have the following

74

relationship :

Assumption 1. U1 < U2 < U3 (2.7)

Another assumption we make in this model is based on Blundell, Magnac and
Meghir (1997)’s result that if leisure is normal good, there will exist a reservation
asset level; if assets are below the reservation level then the probability of accepting
the job offer will be close to 1. Consequently, a lower level of wealth increases the
probability of taking secondary jobs. Based on this result, we can assume that
the utility gain from taking secondary jobs, is biggest when both of the household
members get wage arrears in their primary jobs. At the. same time, the utility gain
from taking secondary jobs is smallest when none of the households head have

wage arrears in their primary jobs:

Assumption 2. 32 st VZ < 2,
AUT < AUT < AUT or AUT < AUT < AUT

and AU’T < AUT’ < AUT or AUT < AUT’!’ < AUT’ (2.8)

where AUT = U[ZT + W333, + W}? + ng + WfST,1,1] — U[ZT + W35. + wf‘T +
Wis , 1,0] (i = m, f) , which is the household utility gain from both male and
female household head taking secondary job(s) over one of them taking secondary
job(s) when both of them get wage payment in their primary jobs. Accordingly,
AUT = U[ZT + ng + wa,1,1] - U[ZT + WTST, 1, 0], which is the utility gain

from both male and female household head taking secondary job(s) over one of

them taking secondary job(s) when both of them get wage arrears in their primary

75

 

jobs. And AUT = U[ZT + W,1,‘,4T + wflT + wa,1,1]- U[ZT + wng + WET, 1, 0],
which is utility gain from both male and female household head taking secondary
job(s) over one of them taking a secondary job when the female gets wage arrears
in her primary job. AUT’, accordingly, is the utility gain in this case when the male
gets wage arrears. Similarly, AU} = U [ZT + err’iIT + W}; + Wis , 1,0] - U [ZT +
Wig + Wfi'TIw,0,0] and is the household utility gain from one of the household
members taking a secondary job(s) over none of the household members taking
a secondary job when both of the households get wage payment in their primary
jobs. AU” = U[ZT + Wig», 1,0] — U[ZT, 0, 0]; and it is utility gain from one of the
household head taking secondary job(s) over none of the household heads taking
secondary job(s) when both of the households get wage arrears on their primary
jobs. Lastly, AU’” = U[ZT+WTI:§~+W’T"T‘, 1, 0]—U[ZT+WT’,’14T,0,0] and it stands for
the household utility gain from one of the household members taking a secondary
job compared to none of the household heads taking secondary job(s) when the
III!

female household head gets wage arrears in their primary jobs. And AUT is the

utility gain in this case when the male gets wage arrears.

There are three possible cases: both of the household members taking secondary
job(s); one of the household members taking secondary job(s) and none of the
household members taking secondary job(s). Both of the household members take
secondary job(s) if and only if their expected utility is higher than when only one

of them chooses to take secondary job(s). The difference in the expected utility

76

between two households members taking secondary job(s) and only one household
is:
AEUT = PmTPfTAUII‘ + (1 - pmr)(1- pr)AU;f~ + me(1_ prMUi’: +
+m<1 — PmrlAUf" (2.9)

The comparative statics of AEUT with respect of the probability of wage arrears

can be written as:

 

 

BAEU
a T = pr(AU11‘— AUf)+(1— pr)(AU% — AUT) (2.10)

me

BAEU
aprT = pmr(AU:’r - AUrf) + (1 - me)(AU§}~ — AUT) (2.11)

Following assumption 2 that AUT’w < AUT’T < AU% and AUT’T < AUT3~ < AU12~, we

know that 8A5; < 0. This shows that both male and female household head are

2

more likely to take secondary job when the probability of getting wage payment

in the primary job is lower.

Similarly, one of the household head will choose to take secondary job if and only if
their expected utility is higher than when none of them chooses to take secondary
job(s). This difference in expected utility between one household head taking

secondary job(s) and none of the household head taking secondary job(s) is:

AEUT’ : meprAUT + (1 — melll _ pr)AU§C + me<1 — prlAU¥I +

+1?fo1 - PmirlAUf” (212)

77

The relationship between the likelihood for one household head to take the sec-

ondary job and the probability of wage arrears can be shown as the following:

8A E U T, I rm III II
T".
and
I
93—521 = me(AUT — AUT”) + (1 — me)(AUT” — AUT’) < 0 (2.14)
fT

which means that one of the household head is more likely to take secondary job(s)

if the probability of getting wage payment in the primary job is lower.

The model above identifies the relationship between the likelihood of taking sec-
ondary jobs and the probability of getting wage arrears, which reflects the house-
hold’s expectation of wage payment. We cannot explicitly estimate such expecta-
tions on wage arrears because the firm-level information necessary for such esti-
mation is not available. Instead we can use the real incidence of wage arrears and

thus estimate the labor supply response to the realization of wage arrears shocks.

Households Consumption and Labor Supply

To further analyze the effects of wage nonpayment and secondary job holdings on
households’ consumption, we take the first order condition of the maximization

problem outlined in equation (2.1) and (2.2), we can have:

U(Ct,imtiift) = /\t(1+ 7‘) (2-15)

78

A /. 1

1+1“

= E A 2.16
1+p t( t+1) ( )

 

At

According to MaCurdy (1985), the realization of past variables, the forecast errors
at each period as well as the change of expectations about future wage enter
the decisions of households as shock to At. Households form expectations and
update their expectations about future variables including the possibility that one
or both of the household heads will get wage nonpayment. The marginal utility
of wealth at time t is a function of initial assets, the expectation of future wages,
the interest rate, the rate of time preference and other unobservables. Condition
(2.16) describes the time path of the change in A: households under uncertainty
set their savings so that the expectation of next period’s marginal utility of wealth

is updated by the realized forecast errors.

For a given household, wage arrears in time t brings a low wage realization in time
t, moreover it also changes the value of A. Its impact on A depends on how early the
household learns of the wage arrears. If the household has information about wage
arrears in advance, then A increases before wage arrears actually occur. In this case,
wage arrears at time t will not affect the consumption at time t in the structural
model specified by equation (2.15). If wage arrears in the primary job come as a
shock to the household at time t, the impact of such a shock will be embodied in
the change of A in equation (2.15). The wage effect of wage nonpayment affects
household labor supply only at time t, but the wealth effect of wage nonpayment,

which comes from the reduction in expected lifetime wealth, will lead to an increase

79

in household labor supply once they know of wage nonpayment 10. But in both
of these cases, if the household uses secondary jobs to buffer the consumption
against wage arrears in their primary jobs, then the overall impact of wage arrears
on consumption will be low as reflected in the “reduced-form ”regression specified
by equation (2.15) where the labor supply decision is not included. On the other
hand, conditioning on secondary job holdings should yield a significant negative

effect of wage arrears in primary job consumption.

2.3.2 Empirical Specification

Analyzing the Impact of Wage Arrears in Primary Job on Secondary
Job Holdings

Estimation of secondary job holdings requires a specification of wage arrears shock
as shown by Pmt and pft in equation (2.9) and (2.12). Also the estimation of
consumption equation (equation (2.15)) requires the specification of the time-
variant marginal utility of wealth for household 12: Am. Following the approach of

MaCurdy ( 1985), log Aht can be expressed as:

t
log At = log A0 + e + Z 5t (2.17)

T=0

where 5t represents the forecast errors at each time t. e is a household fixed
effect. Since changes in log At reflect both the realization of past variables and the

changes in the expectation of future wealth, we include dummy variables Smt and

 

10For the discussion of cross-wage effect and wealth effect in the family life-cycle labor supply
model with uncertainty, see Blundell (1999) and Stephens (2002)

80

5' ft to capture the changes in log At which are correlated with wage nonpayment
shocks experienced by male and female household members at time t. We use two
different measures of wage arrears “shocks ” for Smt and S ft- One is “the first-time
wage arrears ”, which is defined to be a dummy variable equal to 1 if wage arrears
happen for the first time between the survey years. The other is “the general wage
arrears ”, which is also a dummy variable but equal to 1 whenever a wage arrears
occurs. The impact of “the first-time wage arrears ”and “the general wage arrears
”might be different because of the difference in households’ ability to predict the
occurrence of wage arrears. For example, if “the first-time wage arrears ”comes as
a shock to households, then pmt and p ft have low values at time t. By equations
(2.9) and (2.12), we know that households will be less likely to take secondary
job(s) in the case of “the first-time wage arrears”. Similarly, households update
their information on wage payment in the primary jobs and are more likely to take

secondary job(s) in the case of “the general wage arrears ”.

Another issue in estimating secondary job(s) holding is that the realizations of
wage rates in the primary job and secondary job W,%, W}: , W3, and ngt might
be correlated with the shocks at local labor markets such as wage nonpayment,
unemployment, short-time work and other forms of shocks, which in turn are corre-
lated with individual’s decision on secondary job holding. Moreover the change in

wage rates may be also correlated with the wealth shock across the families. Thus

including these wage variables in the regression may pick up the heterogeneity in

81

the local market shock and wealth losses. This would lead to biased estimates of

the wage-effect.

The objectives of this paper, however, do not require an explicit estimation of wage
effect and so we accordingly circumvent the problems by estimating a reduced-
form regression that omits wages but includes the exogenous determinants of wage
rates. These are individual characteristics X mt and X ft such as age and education;
community labor market characteristics such as the presence of employment service
center; closed government factory and the presence of bank in the community;
the interacted terms in individual characteristics with community labor market
characteristics Bet; and region, year and region / year dummies (DT; Dt and Drt) 11.
The employment service centers were established to help people find work, conduct
retraining in new specialization and pay unemployment benefits. The presence of
employment service center in the population center would signal the existence of a
relative accessibility of information on the labor market. The presence of a closed
government factory, on the other hand, captures the demand side factor of local
labor market. We also include a dummy variable on whether there is a bank in the

community because of it might explain some differences in the likelihood of credit

constraints.

Given the specification of log At, Wmt and W ft explained above, the estimation

 

llRLMS provides extensive infrastructure information on 160 population centers, which can
be aggregated to 38 state level primary sampling units (PSUs). The “community ”refers to
population centers, while the “region ”refers to PSUs.

82

of the effect of wage arrears on the probability of husband / wife taking secondary
job(s)12 can be specified. In order to account for the potential correlation between
regressors and unobserved individual fixed heterogeneity, we implement Mundlak’s
(1978) random effects probit model 13. In this model, we allow unobserved individ-
ual fixed effect (uf, Um ) to be explicitly correlated with the mean of time-variant
regressors. Following Mundlak (1978), the fixed effect of individual 2' (2' = m if

male household head; 2' = f if female household head) can be expressed as:
vi = 10 + ¢1Xm + 111250;} + 11132 + $45; + (1)53". + 02' (i = m. f) (218)

where the conditional distribution of 0.,- follows normal distribution, that is:
aT|(th,Xft,Zt,Sft, Smt) ~Normal(0, 0%). Assume that the estimating equation

of secondary job or incidental work holding, in linearized form, is:

P2"; = 71Xit + 72Xit + 7321+ 74% + 758m1+ 7612a

+77Dr + "rsDt + 79071 + ’Ui + Sit (i=mif) (219)

where P" is the continuous latent variable associated with the outcome that house-
hold member 2' supplies a positive amount of labor to secondary job or incidental
work. The random fixed effect estimation can obtained after substituting (2.18)

into (2.19).

 

12Information about two different forms of multiple jobs is available in RLMS. One is about
formal secondary job, which is based on the question, “Tell me please, do you have some other
kind of work ”. followed by the questions about tenure, occupation and ownership of the employer.
We define this type of secondary job as “secondary formal job ”. The other is informal economic
activity: “Tell me please, in the last thirty days did you engage in some additional kind of work
for which you got paid? Maybe you sewed someone a dress, gave someone a ride in a car, assisted
someone with apartment or car repairs, purchased and delivered food, looked after a sick person,
or did something else that you were paid for ”. We define this type of secondary job as “secondary
informal job.”

13The more general form of this model is later called “Chamberlain’s method ”.

83

We can also use specification (2.18) and (2.19) to estimate the impact of wage
arrears in primary jobs on job turnovers since the change of primary job(s) is
another possible labor response to wage arrears in primary jobs and it can be

modelled similarly as secondary job holding as outlined in the theory section.

Analyzing the Effects of Wage Arrears in Primary Job on Household
Consumption

To assess if secondary job holdings serve an insurance function, we also analyze the
effect of secondary job holdings on the correlation between consumption and wage
arrears shocks. Following Kochar (1999), we first estimate the overall effect of wage
arrears shocks on household consumption by running a reduced-form regression of

consumption on a set of covariates that include wage arrears shocks:

log Cht = (10 + atht + O2Smt + a3Sft + a4Dr + 0151):, + 060” +

+Uh + Eht (2.20)

where H ht stands for household h characteristics, including age, sex and education
of household head, the number of household members in different age/sex cate-
gories, non-labor assets of the households. To estimate the wage arrears impact
(12 and a3, we use household fixed effect estimation and it allows the unobserved
time-invariant household characteristics to be correlated with wage arrears. If the
secondary job(s) helps the household to smooth consumption in the face of wage

arrears in the primary job, we expect the coefficient of 012 and 013 to be insignificant

84

or small in magnitude.

The role of secondary job holding is examined through a structural regression that
conditions on secondary job holdings of the male and female head of the household

’1, Jhmt and tht I

log Cht = 00’ + ai’Hht + az’Shmt + 013/51. ft + a4’Dr + as’Dt + ae’D-rt +

+a7/Jhmt + agthft + “h + Eht (2.21)

Secondary job holdings are clearly endogenous variables; and hence, the structural
regression (2.21) is estimated using instrumental variable procedures. We use the
presence of employment service center in the community as well as the interaction
term of this variable with the education level of female and male household member
as the instrument for secondary job holdings. The identifying assumption is that
the household preference is not affected by the presence of employment service
and the interaction terms once the household characteristics and the regional level

aggregated shocks are controlled for.

In the empirical estimation, instead of using the instrumental variables directly in
the second stage regression, we first use the instrument variables as well as the
exogenous explanatory variables to predict the secondary job holdings in a probit
model and then use the predicted values of secondary job holdings in the second
stage regression. Such procedure can take into account the nature of bivariate

variables and efficiently estimate the coefficient of endogenous variables without

85

the need of correcting for standard errors (Wooldridge 2001).

2.4 Data and Summary Statistics

2.4. 1 Sample Creation

The data comes from the Russian Longitudinal Monitory Survey (RLMS), which
was conducted by the Population Center at the University of North Carolina. The
RLMS is a household-based survey designed to measure the effects of Russian
reform on the economic well-being of households and individuals starting from
1992. Beginning in 1994, RLMS was designed to provide a longitudinal study of
populations of dwelling units. The data used for analysis are for the years 1994-

1996, 1998, 2000 and 2001 (Round IV-X).

The sample of couples used in the estimation has been created in three steps.
First, data from the RLMS household file is merged to the individual dataset.
The information from the household file include the identification numbers of the
household head and the spouse of the household head. Second, every individual
who is a household head or a spouse of the head is extracted from the merged data
in step one. Men between ages of 21 to 60 and women between ages of 21 to 55
are kept in the sample. Third, couples are created by merging together individuals
with the same family identification number in each year. This results in a sample

of 8008 couple-year observations.

86

The final data set is created by deleting observations with missing data or logically
wrong information. With the definition of “shock ”to be wage arrears for the first
time 14, observations in 1994 are not used in the regression analysis because it
is treated as the initial year. Couples with at least two observations are used in
the final data set. The number of observations deleted and the reason for deleting

them are as follows:

1. 170 observations are deleted because not both of husband and wife are

older than 21 years old.
2. 387 observations are deleted due to missing data in assets variables.

3. 27 observations are deleted because the reported total working hours per

month are greater than 450 hours.

4. 35 observations are deleted because the reported main occupation is

farmers.

5. 25 observations are deleted because the reported total years of schooling are

greater than 50.

 

l“Wage arrears for the first time is defined in the following steps. First, the individual was paid
at their work in the previous survey. If the answer to the question in the previous period, “Did
you work at your primary place of employment in the last 30 days?” is “Yes ”, then this person
must also have answered “Yes ”to the question “Tell me, please, at your primary place of work
in the last 30 days did you receive some amount of money in the form of wages, bonuses, grants,
benefits,revenues, profits? ”. Second, the individual had no wage nonpayment at the time of last
survey. That means their answer to the question, “At the present time, does your place of work
owe you any money, which for various reasons was not paid on time? ”is “No ”. Third, at the
first time of survey, this individual is working but not paid wage at work.

87

 

6. 597 observations are deleted because working hours is missing for the

respondents who reported to work.

7. 513 observations are deleted because wage arrears variable is missing for

the respondents who reported working at a primary job.

8. 658 observations are deleted because wage arrears variable is missing for
respondents who reported being involved in other economic activities than

formal jobs.

9. 189 observations are deleted due to missing data in community information

on closure of government factory, presence of employment services and

banks.

10. 1250 observations in 1994 are deleted.15

11. 513 observations are deleted from couples that don’t have at least two years

of observations.
12. 74 observations are deleted because wage arrears occurred after not working

in the previous year.

The resulting sample contains 3570 observations on 1605 couples with 584 of

them having two years of observations; 399 of them having three years of

 

’5 Since “the first time wage arrears ”is defined by getting wage arrears for the first time during
the survey period, we treat the year of 1994 as initial year. As a result the 1994 data is not
included in our estimation.

88

observations; 267 of them having four years of observations and 266 of them

having five years of observations.

2.4.2 Summary Statistics

Table 2.3 reports the distribution of employment states of the married couples in
our analysis sample. During 1995-2001, on average 83% of husbands and 75%
wives were employed in the formal sector. Conditional on being employed in
formal sectors, 87.7% of men and 92.5% of women have had only one job; 4.5% of
men and 3.8% of women had secondary formal job. 7.7% of men and 3.1% of
women had incidental work besides their first job in the formal sectors. During
the financial crisis in 1998, more women held secondary informal jobs and the
rate of secondary informal job holding increased to 6.6%. Over the years, less
than 1% of the people in formal sector had three jobs simultaneously.16 Exclusive
employment in the informal sector on average accounted for 5.2% for men and
3.3% for women. The rates were rising over the years: it rose from 4.5% in 1995
to about 6% in 2000’s for men and from 3% to almost 5% for women in the same

period.

The last row of Table 2.3 presents the rate of job change during the survey

period. Since job change is defined as any change in job compared to the last

 

16Paxson and Sicherman(1996) reported that using PSID, dual job holding rates in US. are
21% for men and 12% for women. While by CPS, the rates are 7% and 6% respectively.

89

survey 17, the job change rate is much higher in 1998 and 2000 when there were
two year gaps from the last survey. Especially during 1998-2000 period, when the
economy experienced financial crisis first and then recovery, as high as 66.4% of
men and 61.9% of women changed their original jobs. Overall, women are less
likely to change job than men. From 1995 to 1996 18, 15.3% of men and 10.7% of

women reported having changed their jobs.

In Table 2.6, we present the comparison of individual characteristics and
household characteristics by whether the husband or wife or both is still owed
wage by their primary working place. Average age for husband who never had
wage arrears is 37.24, compared to 38.2 for husband who had wage arrears. As
expected, the education is higher for husband without any wage nonpayment. It
is also not surprising to find that the employment rate for husbands with the
experience of wage arrears (92.5%) is much higher than that for husbands
without (72.9%), since people are subject to wage arrears only when they have
jobs. Moreover, husbands who had wage arrears are more likely to take
secondary formal or informal job than husbands who had no wage nonpayment,

and the difference in holding secondary informal jobs is statistically significant.

Wives were younger and more educated if the husband had never had wage

nonpayment. Not surprisingly, the incidence of wage arrears of wives and

 

17The job change is identified by the answer to the question, for example, in 1998: “Tell me,
please did you change your place of work or profession by comparison with December 1996 or
has everything remained the same? ”

18No question was asked in the survey of 1995 about the job change

90

husbands is highly correlated. 67.2% of the wives had wage nonpayment when
their husbands had wage nonpayment, whereas 42.2% of the wives had wage
arrears when their husband had never experienced wage arrears.19 Consistent
with the younger age of the wife, the number of the younger children is higher
and the number of children older than 7 is lower in the no wage arrears group.
Although family nonlabor income is not significantly different among the
households with or without wage nonpayment,20 household wealth, measured by
assets value in the base year (1994) is significantly higher for household which

had never had wage nonpayment.

When comparing between households whose female head had wage arrears to
households whose female head had no wage arrears, we find a similar pattern to
the comparison we made to the male household head. One striking difference is
that wife’s education is higher in the wage arrears group than in the no wage
arrears group. This finding, though, is consistent with Figure 2.4 where we saw
that women with primary and less education were far less likely than others to
have wage arrears in 1995. Four percent of the wives in the wage arrears group
hold secondary formal jobs and 3.8% hold secondary informal jobs. These rates
are 1.6% and 1.5% respectively in the no wage arrears group. The last two

columns reports the comparisons between the household without any wage

 

19Such high correlation might possibly be caused by marriage matching in the jobs. The
correlation of wife and husband’s occupations, according to International Labor Organization
Coding (four digits), is 0.22. But we can not identify the working place of the individuals.

20Family nonlabor income includes total household pension income, personal property sales,
rental income, capital investment income, insurance payment, alimony received, help from orga-
nizations and help from relatives.

91

arrears and the households with wage arrears. The findings are consistent with
the previous results. Husbands are more likely to take secondary informal jobs
and wives are more likely to take secondary formal jobs in the wage arrears
group. Such differences might be caused by increased labor supply from the wives

to their wage nonpayment.

2.5 Empirical Results

This paper presents three sets of results. We first consider the effects of first time
wage arrears, first own effects, and, second, cross effects and, separately, the
effects of general wage arrears on secondary job holdings,which include both
formal secondary job holdings and the informal secondary job holdings. We then
look at the response of changing jobs to wage arrears. The last set of results

involves the consumption regressions.

2.5.1 Labor Supply Response to Wage
Arrears—Secondary Job Holdings and Job Changes

Tables 2.5 to 2.8 report the estimated effect of wage arrears on the likelihood of

holding a secondary formal job or informal job using Chamberlain’s method of

random effects probit model.

92

First, wage nonpayment for the first time had no effect on the tendency of
husbands or wives to hold more than one formal job (Table 2.5). It is not
surprising since it might be difficult to locate another formal job within one
month after the wage arrears occurred. But we do find that wife’s tendency to
take secondary job takes a quadratic form: it increases with husband’s age and
reaches the maximum when the age of husband is about 52. Also wife’s tendency
to take secondary job when husband has wage arrears for the first time increases
with her education. So here we find a cross-effect of husband’s wage nonpayment
on the wife’s tendency to have secondary job. Wife’s age impact on this
cross-effect might be explained by experience increasing the opportunity for
women to find a secondary job and so does education. The findings for the
impact of wage arrears in general on holding secondary job are similar (Table
2.8). Husbands are not more likely to hold secondary job if they have wage
arrears or if their wives have wage arrears. One different result reported in Table
2.8 is that wives are found to be significantly more likely to take secondary job
when they themselves have wage arrears and the likelihood of having secondary

job also increases with education when their husband has wage arrears.

In terms of the tendency to hold secondary informal job (Table 2.7 and 2.8), we
find that it increases when husband was hit by wage nonpayment for the first
time. Although wives are not found to be more likely to take the informal jobs

when they got wage nonpayment for the first time, we do find that they are more

93

likely to take informal jobs when we use any owed wage arrears as the

measurement of wage arrears shocks.

Table 2.9 and 2.10 report the results on change of primary jobs after wage arrears
shock. The ideal data for estimating the effect of wage arrears on changing jobs
is monthly or quarterly data, where we could have complete information on
individual’s jobs and thus be able to more precisely detect the effect of wage
arrears. Unfortunately, we only have one observation of job information for each
year of the survey. So we won’t be able to tell how many times individual change
their jobs between two survey years, although we know if he or she changes his or
her job from the pervious survey. Moreover, some households were missing in
some of the surveys, which left us with an unbalanced panel. Given the data
structure we have, we can only estimate the job change one year or two years

after wage arrears.

If husbands have their first wage arrears in previous survey, they are not
necessarily more likely to change their job. The response of changing job varies
by education of husbands. Husbands with higher education are more likely to
report changing their jobs in the survey right after they have wage arrears for the
first time in the previous survey. 21 This might be explain by the fact that
individuals with more education have more outside opportunities. When we use

wage arrears in general, instead of using first time wage arrears, we find that

 

21The coefficient of the interaction term of first time wage arrears and husband’s education
(column 2 of Table 2.10) is 0.0852 and significant at 10% confidence level.

94

 

both husband and wives are more likely to change their previous jobs where they

have wage nonpayment.

We can not interpret the difference in the impacts on job changes of wage arrears
in general and first time wage arrears as an adjusting process to wage
nonpayment since, as we emphasized earlier, that there is one year or two years
gap between two surveys and even for the first time wage arrears, we are looking
at the job change response to wage arrears that happened one or two years ago.
The difference might just be caused by the fact that there are more observations

having wage arrears in general than having first time wage arrears.

2.5.2 Effects of Wage Arrears on Household Consumption

The overall impact of wage arrears is first estimated in the reduced form
regression as shown in the first column of Table 2.11 and 2.12. The results show
that the time profile of household consumption is relatively smooth; much of the
sample variation in consumption is explained by the region-year dummy variables
and a set of demographic preference shifters. In particular, male household
head’s wage arrears shock, measured either by first-time wage arrears or wage
arrears in general has no significant effect on consumption at all. The only
exception is that female household head’s wage arrears in general is estimated to

reduce the household consumption by about 0.05% and the coeflicient is

95

marginally significant at 10% level. This suggests that household consumption is

protected from wage arrears shocks.

Secondary job holdings, which is included in the structural equation, includes
both formal secondary jobs and incidental jobs. In predicting the secondary job
holdings with instrument variables and other exogenous regressors, the F test of
the significance of the instrumental variables of community employment service
center and interaction term of employment center and education variable is 4.89
and 5.20 for male and female secondary job holdings respectively , which are

significant at 7% and 8% level (Appendix Table 2.13).

Similar to the results from the reduced-form regression on first-time wage arrears
on consumption, the results from structural equation (column 2 in Table 2.11 )
shows that household consumption is very well protected from wage arrears
shocks. Although men taking secondary jobs (as shown in Table 2.11) helps to
increase the household consumption by over 60%, the result still implies that
conditioning on such big effect of secondary job holdings, first-time wage arrears
don’t significantly reduce the household consumption. As we mentioned earlier,
labor supply is one of the many mechanisms that households can fall upon in the
time of income shocks. Although it is puzzling that the big effect of secondary
job holdings by male household head doesn’t seem to explain the smoothness of
consumption at first time wage arrears, we can’t rule out that households use

dissaving, borrowing, and selling assets to smooth consumptions when they first

96

experience wage arrears shocks. In contrast to the results from the reduced-form
regression on general wage arrears on consumption, the results from structural
equation (column 2 in Table 2.12) shows that household consumption falls with
the experience of wage arrears shocks. The results from the structural regression
imply that wage arrears experienced by the female household head reduce
household consumption by about 10%. This decline in consumption in the
absence of adjustments secondary job holdings is consistent with the positive and
significant effects of male and female secondary job holding on consumption.
This result suggests that overall insignificant effect of wage arrears on

consumption reflects adjustment of secondary job holdings.

2.6 Conclusion

This paper documents the features of wage nonpayment in Russia during
economic transition since 1994. In particularly, it examines the labor supply
response, including multiple job holdings as well as job changes, of married
couples to their wage arrears for the first time and the wage arrears in general.
Wage arrears apparently affect the economy, among other things, by inducing
more labor supply response from working men and women in terms of secondary
job holdings and job changes. Such labor responses, especially secondary job
holdings, are found to help to smooth household consumption in the presence of

general wage arrears in primary jobs; but they don’t seem to explain the

97

 

smoothness of household consumption at first wage arrears. The results of this
paper suggests that different households consumption smoothing mechanism
might function simultaneously in Russia, but the development of the labor
market may well help the households to smooth income directly. The latter
finding is particularly important given the frequent income shocks that
households in the transition economy are facing. Public policies which directly or
indirectly improve the labor market conditions, such as lessening or eliminate the
control of population migration and increase in education investment, programs
that can enhance the ability of individuals to engage in labor markets, may help

households to better cope with uncertainties.

98

Somdoomdmmdwamfimmmfi.33 535m 33:52 Rampsfimaoq gamma—m H830m
.mamoficeama E was macho “5.898% duos—.398: ewe? :o 3:056:00 mm Ema go: 93:05 We 52:5: omega mEBA

 

 

6R8 :88 :28 83.8 82.8 38.8
$3 82. as... we; ”was ”was Sees-

88.8 $58 33.8 628 3:8 88.8
8.0..” was R3 «was 33. as: 52-
293—835 Edam «oz mow—£5

C88 388 88.8 88.8 $8.8 $88
em; 83 83 8.2 2.2 83 eeaeai

$8.8 88.8 88.8 :58 38.8 88.8
$3 $3 82 $3 and $8 5:.
Juflcz uflmhh—JO aw uflmaxfiwam un—Ogumg mac—MOB

$8.8 38.8 :38 :88 88.8 85.8
:2 32. Sea 33 and $2 $8851

38.8 :88 :88 :88 :58 83.8
memo and need awed was need 8:.
mum—$5 @950

58 88 we: 82 32 $2

 

$8.94. mw§> m0 325 was moseanH ”mm meB

99

—c»— age 18-29 —e— age 30-39
-- —o- - , age 40-49 — - — - age 50-60

 

 

 

19’94 1ng 1996 19'98 2600 2061
year

Figure 2.1: Currently Owed Wage by Age (Men)

—o—— age 18—29 _._._.,__ age 30-39
.. - age 4049 — — — — age 50-55

 

 

 

19’94 19’95 19’96 19'98 2600 2061
year

Figure 2.2: Currently Owed Wage by Age (Women)

100

—e— university and above—s— sp_ecial secondary
.. _e— — . general secondary - — — — primary and less

 

 

 

19'94 19‘95 19% 1698 2o'oo 2061
year

Figure 2.3: Currently Owed Wage by Education (Men)

—e— university and abov<+e— special secondary
.. —e —- general secondary - — — — pnmary and less

 

 

 

\§?\\: 7 ‘ ‘D

\ RX

2 1 \
19'94 19’95 19'96 19198 2050 2003

year

Figure 2.4: Currently Owed Wage by Education (Women)

101

—e— tenure: <1 year —.rr— tenure: 1-10 years
~ -0 — tenure: 10-20 years— — — — tenure: >20 years

 

 

 

1094 1905 1906 1908 2000 2001
year

Figure 2.5: Currently Owed Wage by Tenure (Men)

-—e—— tenure: <1 year —e— tenure: 1-10 years
-- —0 — tenure: 10-20 years— — — - tenure: >20 years

 

 

 

1904 1095 1906 1908 2000 2001
year

Figure 2.6: Currently Owed Wage by Tenure (Women)

102

—+>— wage: 1st quartile —.~e—— wage: 2nd quartile
.. —e — wage: 3rd quartile — — - — wage: 4th quartile

.8“

 

 

 

1994 1995 1996 1998 2000 2001
year

Figure 2.7: Currently Owed Wage by Wage (Men)

—e~— wage: 1st quartile _.._. wage: 2nd quartile
.5 —D- — wage: 3rd quartile — — — — wage: 4th quartile

 

.,\‘
\ U
\ .—

 

 

19'94 1995 1996 1998 2000 2001
year

Figure 2.8: Currently Owed Wage by Tenure (Women)

103

 

104

 

Anaev Ame:v Toemv Amemv
83 8:8 8:8 :88 :8 85888.: 8888. 8:0
Away A88 Amemv A88
888 8 888 82 :8 2829.85 2888. 8288
:88 :8 A88 88 :8 88892: $888.
83 8:8 88 88 :28: use 88:8 as
:88 :88 :88 :88 :8 88832:
83 88 N88 888 3882 858 8:858: :08 8293a
:88 $8 :88 :88 8982 .883 8883.88: 88:88
83 8:8 83. 8.88 :8 88888: $50
Av:mv Toma Tampa TowHV
82 88 88 :88 :8 seesseea 8298:.
AmeHV Acav AHOMV “88:8
83 88 8:8 88 :8 289885 88: as 85:8 a:
88 88 A88 :88 :8 aeeeaeea
88 :88 88 8.8.8 8:98 8:858: 8988882 888:
Amo:8 A888 AeCHV A:mmv
N88 23 888 83 :8 5882 8580
A88 :88 a: :8 :38 5882 828.50
:88 88 88 888 .8880 888 888:8:
A88 A888 AQHV Aem:v
:8 8:8 88 8.8 :8 see: 888
3:8 :osmasooo 0.: £388
5:83 :22 5835 :22 308880 3:288:80
88 as? 88 seem:

 

 

 

Awmmfi aefiieaomonv :ofiaasooo 3 Emmet/x ewm>> ”Wm 23:9

an:
Ed
38
88
at
88
as
88

$803

$3
88
:3
38
2:
$8
$5
38
5s
38
so:
88
83
:8

88:
88
am;
88

8a
88

a:
88

Go:
88
88
$5
88

E;
28
$9
88
A8:
88
av

3a
88

:3
88
£8
88
8:
88

A3
88

5803

A8:
88
SE
88

2:
88
88V
88
38V
88
a5
88
SE
88

65
88
38V
8.8
6.3
88
83
88

:32

Ammmfiv

:8
38V
88
0.3
88
:8
88
83
88
av
88
33
:8

38 mumEEmmmao. was muouwhmao 2:382
C8 @5885 BEEmanOOBJSmEH
83330 paflm$uwaosfim
Ackvnzflsxmfimumfiob woo? $58805 woo:
8810.3 8989 Umpflom 98 £80 550
at 285$ 885
95 £388: .aoEBBL

88 858:8 0d 86-3

fowvudo 3:283:80

Amt 8&8?» @2382 98 :3on
St 88103 $ng

wEEBm Una aoBQmem

A848 £883 827,;

28 8330:? 33m

33 macaaSmaoEmQ

was mcoflmamofimimvoz
232383

835m @388on «Ed 3:885

Amvv £85 838m 58890
Ad; 8120 8&0

.5803 8 289

105

.flmofififida E $539,530 mo :55: Z

.3833”?32.3263 355 Essa: asasxwafi Ewan 528m

 

68
m2
Gm:
awed
3va
33
:5
:3

83v
£2.

at
Swd
A33
ammo
3%
$2

33
Ed
$5
.83
8mg
23
2m:
and

:wmv
82
a:
$3
933
93.0

prmfiv

wood

33 fioamafib ,wstsuofisawz
50535300 .wEEE E $2093

$8805an 5935 EB Ehsfisothx
szwfigflo osmmfiovfiovsg 88$
mqoswazooo mausofiflm mmomimm Una 83m
Amwzmafio Baggage, 883

88230 EaE-=noE 98 Egta

.929; mm 2an

106

4832:5me 8985:2888 8.8: .1: 8:3 8: «EH
888558888 5 28 £85 @3qu5

 

888.8 88.8 :88 88.8 88.8 :88 888.8 88.8 888.8 8888 88:88 85828
828 8888 8888 888 8:8 888 888 888 f .I 8:8 888 8:5 388 88:20
:88 888.8 88.8 888.8 28.8 888.8 :88 888.8 88.8 888.8 888.8 888.8

888 888 8888 828 888 828 :88 828 8:8 888 888 888 888858 82
:88 888.8 88.8 888.8 :88 888.8 :88 888.8 88.8 888.8 888.8 888.8

8888 8888 888 888 888 :88 88.8 888 8888 8888 8.8.8 8888 813888888
8888 888.8 888.8 888.8 288.8 888.8 988.8 888.8 888.8 888.8 888.8 8888 A8 .85 88.
888 888 888 8888 888 8888 8888 8888 888 888 888 8888 88:85 8 888888
8888 888.8 888.8 888.8 888.8 888.8 888.8 888.8 888.8 888.8 888.8 888.8 88. 82:88
888 8888 888 :88 8888 888 888 8888 888 888 888 888 8888888 8 885:8!

888.8 888.8 888.8 888.8 888.8 288.8 888.8 888.8 288.8 288.8 288.8 288.8 888 88.88 888888 8 88
8888 888 888 888 8888 8888 8888 8888 8888 8888 8888 8888 88.58 8888888 8 888:8 I

8888 888.8 888.8 888.8 888.8 888.8 888.8 888.8 888.8 888.8 888.8 888.8 88. 88.58
888 :88 888 888 8888 888 888 888 8888 888 888 888 888888 8 888:8 I
8888 888.8 888.8 888.8 888.8 888.8 888.8 888.8 888.8 888.8 8.888 888.8

888 888 £88 888 8888 8888 8888 8888 888 888 8888 :88 88 88. 885:8 l

:88 88.8 :88 88.8 88.8 :88 :88 88.8 :88 888.8 888.8 888.8
888 8888 88.8 888 58 8888 888 888 8888 8888 88.8 8888 E 888 88:88 8 B88888.
magnum enmahoagm
88> 88888 88> 85.8.88 88> 88888: 88> 888888 88> 888888 88> 8882.8
888 8888 8888 8888 8888 =<

85> 8:8 853832 mo magnum uamfizofiam mo nossntuma «Mm 858%

107

 

 

$888.8 $888.8 #8888 I888 :88 8888
8888 888 888 8888 888 8888 888 88888888 8.885
88.8 I I I 1888 £838 £85.
888 I I I 888 8888 m88> 888 88>
82.8 888.8 $888.8 8,888.8 88.8 888.8
88.2 888.8 88.2 38.2 88.2 888.2 888888 8.885
#8888 1.88.8 #8888 1888.8 1888.8 888.8
88.88 82.8 8.888 88.88 88.88 8.888 888. 8.88>
8,888.8 8,888.8 :8888 $888.8 $888.8 $8888 88. 83:88
888 8888 8888 8888 E88 888 8888888 888 888.88
888.8 888.8 888.8 888.8 888.8 888.8 88. 8888
888 888 888 :88 888 8888 8888888 888 858888
£88.: $888.8 888.8 88.8 $888.8 $888.8 88:
8882 888.88 83: 8283 8882 88882 88883 38. 8.88288
$888.8 $888 #8888 $888.8 $8888 $888
8888 8888 888 888 8888 888 838 8888888 8.88.38:
88.8 I $88.8 .1888 I | 8882
:88 I 8888 8888 I I 8885 888 888888
88.8 88.8 888.8 888.8 188.8 8.8888
88.88 888.8 888.9 88.2 88888 888.2 888888 8.88288
$888 I888 888 8.8.8 $888.8 1,888.8
8888 888.8 88888 88.8 838 888.8 88.8 8.8888:
mudmuu< mempu/x mud®uu<
033 BE om8>> 8m mummt< 888.18 8.822% ow8>>
88> 5 88> 8 mum88> 8888> 888 88> 88: 88 852
88388 85888 88; 88> $52 88> 8888:: 893888

 

 

 

 

£88.58 mw8>> an 80583308830 203$:on 93 126365 N: @388

108

 

 

 

 

 

.732 endomemm “RS 98 88m 88 5:28 ”.5353 59c 85:56 3.283338“ u. 98 3 .EBfiqflwa E fiche «8.8855 5qu
$38888t8888888 1888888 :88888 88.888 88.888
8888 88:8: 8888888 88888 88.8888 8.8888 882 8888 88:8
:33 88888 8888 88.8: 88888 888.28
88888 8.23 888:: 88.23 88888: 88.82 8:88 88882 88888
888.8 88.8 88.8 88.8 888.8 88.8
828 828 828 888 828 828 8885 888 8 88882
888.8 :88 888.8 888.8 888.8 888.8
888 8888 888 888 :88 888 58 888 8 88882
88.8 88.8 88.8 88.8 :88 88.8 5885 8.8
8: 883 82 883 82 883 88883 8 88882
88.8 88.8 88.8 88.8 88.8 88.8 888 83
83 8:8 8:8 882 882 8:8 8:883 8 8852
£888 i838 $888.8 1.88.8 .1888 1888.8 88 88 8 888 88388
888 888 8888 8888 888 88.8 88880 8 88882
1.888 188.8 $88.8 $888 :888 1.88.8
8888 88.8 8888 8888 888 58 8.88880 8 8852
8.88.8 888.8 #8888 1,888.8 :8888 5.8888 88. 88:88
888 888 8888 888 8888 888 8888888 88 88>
L888 1888.8 $888.8 #8888 888.8 888.8 88. 88:88
888 :88 888 888 8888 8888 8888888 88: 83>
$888.8 188.8 188.: 8.88.8 888.8 82.8 888
88: 8888 88.5 88.88 88.888 88.88 8:883 888. p88>

8.888 8.8 888.

109

 

82. 58288 888: 85>

non iwwsooom mezzo: $85533

 

288.8 688.8 288.8 $8.8 $8.8 58.8
858.? 88.8 888. 888. 38.9- 880.9- @885 882-82 Amos
5388 588.8 2.288 689.8 858.8 658.8
888 588 888 28.8 :88- :88- M85> 8 8?
328.8 838.8 888.8 388.8 888.8 888.8
583- 3:8- 85.8 853 $28 88.8 85> 8 83
$558.8 £3588 £9588 888.8 288.8 858.8
38.8 $8.? 88.? 28.8 888. 28.8- 28825 8 83
32:58 £858 $3858 5588 588.8 358.8
882 SS; 88: 583 SS: 283 888:: 8 a?
65.8 52.8 $82.8 $82.8 28:8 25:8
33 $88 388 $88 883 £88 m85> 8 82.83%
$2.8 @828 A828 82.8 33.8 83.8
88.0- £88- 33¢. ”:38 8:3 888 2.388 8 82885
328.8 588.8 $288 $88 3088 888.8 ma5>l
8:3 :83 882. 8.8.8 853- 88.8. .885. 85> 88>
3858 A858 8588 58.8 85.8 $288 2888:!
583. 82$ 8:8 583- «god- 88.8 285. 85> E;
m m H m m H

 

 

28288 «8wa SEE >35 8 Son. Eavaooom wEEomv omcoammm .5qu mo coSwEEmm— QESQ 809:3 ”Wm £an

110

83:55: :omww:\8w.m 3:“ Ewhbmzuowaué: mm? :5 @5353 “o

33 “Ed :03835 3333 3333:“ m23£3> 555558 @833 35* 355558 m3: 5 :53 555588 23 3 38.0 >588

u:m::::w>ow§£:5:58 m3» an 835% u:m:§o~9:w€_o3om:o3 033 :3 :2:\:o:83 £32m .30 53:5: $3... 6303503 $33 :3

5503\55 mww wEvtoB 30 83:5: m3» 6.030953 v3... :3 SA. .30 owm am 55:30 .30 53:5: 2: "2039503 23 :3 ed 30 www

an 55:30 .30 83:5: 333:3 3:252:55 3535-55 mo 0?:wa 23 25 aomwmouwm: m3... 5 @2535 85338, 5380 .m

332 $3 am 5.85%6 .396— oAh am 550$:wmwi £83535: 5 Ba :85 “555% awn—30¢:

 

Elam onmm

Emm Emm

onmm cpmm 30 .30 53:52

 

mmvd movd
3.2 3.2

35.3
ms;
:85on
«32.

Game
885
85.8
Ewe-

mwmd mmvd

wwfifi 2.2

awzé
Std-
802.3
33o

vmvd mwvd 53>:
833555 3:55:55

3.2 5.3 3:2:8/5838 we vwfio><
853$:me 55H

om>>lm5mt< owa>>x

~858st @833

wqw3msmlm$mb< mw§>x

:os‘mozvm 983$

oh>>lu£dmh< www>>x

:os‘mgvm m.v:m3m=m

v:m3m:mlm8mt< mwm>>x

5585a M58353

Ammmfi .8 £§>Ium5wt<
mwm H .o- “~wa x gamma:
So: .8 @53m:mlm5ob<
unwed www>> x macaw/V.

5.88 :omaomumu:m

.5203 3 2a:

111

 

30w ESEOBW 3:203 @333

33. 5358mm wEEom £5535:

 

:88 :88 58.8 €88 $8.8 $8.8
Soo- oooo- ooooo- oooo- oooo- ooooo- @883 Sosaoz $08
38.8 38.8 :ooo8 $8.8 $8.8 $8.8
oooo oooo voooo Hooo- Hooo- ooooo- ers o0 no?
:38 638 @968 :38 88.8 83:8
vowo- oooo- goo- oooo Hooo ozoo 35> oo o?
$8.8 $8.8 $8.8 $8.8 $8.8 $8.8
fioo- :oo- ozoo- Hooo- Boo- fiooo- 3825 o0 No?
1.33.8 1688 £558 :38 38.8 $398
on: 83 oooo: no; vodo 52o oooomom oo o?
32.8 82.8 85.8 :38 6:8 5:8
oooo oooo ooooo oooo aoo ooooo as: oo 838:8
$2.8 3:8 A338 :28 2:8 33:8
oooo- oooo- goo- :oo oooo- voooo- 288:; o0 838%:
£828 £328 Looodoo $2.8 $2.8 2:28 mos/I
opoo «So 38o oooo- ovoo- goo- $22 owes
:88 82.8 $8.8 $2.8 328 8:28 Esmoml
vooo- o2.o voto moo vooo ooooo $22 owes
o o H o m H

 

 

m5mb< momma 3 A33. >53:oomm 9:22.: om:o:mom 5353 Moo :oSmEEmm £38m EovSWm ”9m 2358

112

83:55: :o_wo:\8mm 358 88583838.: or? 3:8 358353 mo

wwm :8 :2858 3:3 3838835 8388:, 33:55:80 883: :8 5:55:80 83» 8 :83 «33:55:80 83» 8 8.88 8088

585853955558 83» 8 83:8 58539586338503 83: :3 55:58:83 3820 80 83:5: 83: €_o3mm503 m3: 5

8503\85 ow: m:283 mo 83:5: 83: “2038803 83... 5 SA. mo .88 8 :8239 Ho 83:5: 83» 63038503 83... :3 3.0 we mm:

.8 85:8 .30 83:5: 833.28.» 38398985 3853.855 mo 8888 833 8.8 806888 83» :_ 8:305 8582.? 8330 .m

.88— voS .8 3:83:86 «883 Km 8 5805881 28588: :3 88 30:8 583588 853363

 

 

Emm 3mm onmm oumm 05mm ommm 30 mo 83:52
mmvd movd amvd mmvd vmvd cmvd 8387:
833853 558855

8.2 $2 moo: m2: No.2 83: 85:18:80 woos;
"8:805:9m 55H

$3.8 om>>lm88u< owa>>x
wwod- :ofiwosvm 98:3
* 1.38.8 $3.8 v83m5mlw882< owa>>x
mid mad :omowoscm 98:?
38.8 mm>>lm88u< .835 x

mvod- :oSmofium $58385

38835318883. momma x

838:3 £88353;

38.8 28.8 mm>>lum8mh<

owed mood- 882/ x mummmaq.

88.8 93.8 v:m3m:mlm8m:<

mnod mood- wwaXmommm<

EhmeL ”OmHUGhmuflH

8:88 oo 228

113

 

fig 3532: $36: $35

383 158385 53233 £53352

$83 6333 $8.3 $8.3 $8.3 $8.3 .
283 383 2:5 883. $8.3 $8.? 2:85 522-82 303
$8.3 $8.3 $8.3 $8.3 $8.3 238.3
553 383 $83 883 883 883 3:5 30 my?
2233 8333 $3.3 $3.3 $3.3 $3.3
$3.3 33.3 23.3 3.33 333 335 3:5 .6 a?
.2383 238.3 .2383 $8.3 $8.3 $83
383. 38.0- 383. :83- £83- :83- 28%: mo 33.
$3.3 $3.3 $3.3 $3.3 $3.3 233.3
3.3.: $2: 3%.: some- 52- 33.3 288: mo .3
$2.3 $5.3 $5.3 $8.3 $8.3 $83
332:. 333 :33 383 383 3:3 35> 30 838:5
$2.3 $2.3 $23 8333 $333 $333
283 283 28.0 wowed $83 $23 289: Mo 858%:
$3.3 $3.3 $3.3 $3.3 $3.3 $33 .327
:33 333 33... 28.3 883. 383 $85. .833 ii
$3.3 $3.3 $33 .2333 .2533 .2333 955:!
32.0- 33.3 333. 333 833 33.0 322 was: Ea
m m H m m H

 

 

3885. 8.335 8:38 8:5 3 $83 338385 wEEomv 8503833 .8353 .30 :oSaESmm— £38m 503533 ”Nam 233B

114

wflEfiSv somwmuidw» ES Ewhbmzuowmmt v33 «:8 @539: mo

33 98 .8382? at? 88335 moEfig 5:55:80 39: 98 5:58:80 23 3 x53 chains—Eco 23 am @320 @308

pamEEm>ow55588 23 as 83.8w EmEhOEEoqunmvE—o: m5 5 5E\=oEoB bumEm m0 835:: 2: ”29395: 23 E

5803\58 ow“ music? m0 538:: 23 ”20:34:03 2: E 3.» mo wwm am 5:55 Mo .8855 on» ”203820: 2: E oé mo mwa

aw 56:5 “0 82,5: 333:; anmvsomvwfi undigéfis .«0 $9295 23 m3 3063“me 23 E 39:05 81153 850 .m

.35: $2 pa Edowfiwmm 392 gm an endowmzwwmi £85553 5 was E85 Eavsfim awn—pom;

 

oumm 05mm

opmm 3mm

2mm 05mm 30 Mo 53832

 

coed mocd
8.: No.3

€33
33¢-
$3.3
38d-

$3.3
Seed
game
286

need Sad

No.2 9&2

$8.8
an:
8398
v83

83 no; 2:33
853:3 aqvvqwamvfl
EaES/éEwH mo mwfim><
"8:505:me ”Ewen.
maglmammhxx www>>x
858va Emma
vqgmzmlfiaohax wwm>>x
:oEwosvm mbfia
oralmfimmbaa www>> x
338:3“ $95333
vawnmsmlfi‘mobx‘ mwaa x
cosmozvm M65525

«5.3 no.2

60...... H .8 mm>>|m5mb<
owmod owa>> x meomm<
Ammofi .8 finanmsmlmuaot<
unwed mw§> x mummm<

"mEuoH. :omuowuoacm

.5283 S 03%

115

 

foe 3:833 wagon $83

x88 EEmEoE wEEom 98:83::

68.8 38.8 88.8 88.8 85.8 823.8
83 83 $83 :3- :3- :83- ESE 80.2-82 808
$8.8 388 A888 $8.8 $8.8 $88
83 83 R83 883 883 883 8:5 8 83
$38 888 A838 $8.8 $88 6838
83- 383- 883- 83 83 883 85> 8 a?
...,:88 $988 ...388 88.8 :88 88.8
23- 383- 883- 83- 83- 283- 888m 30 mm?
68.8 388 $38 88.8 88.8 58.8
82 83 as: 83- 83- 3:3. 288m 8 88.
8:8 3:8 A828 £88 888 A888
83 E3 883 83 23 883 85> 8 8:853
8:8 328 3:38 £88 :88 838
R3- 83- 883- 83 383 :83 828m mo 888%
32.2.8 3,338 ......§8.8 82.8 :28 33:8 oral
83 83 883 83 83 383 £85. 083
$28 82.8 8838 £28 £838 638 288ml
33 :3- 383 83 83 823 $22 $83
3 m H m m H

 

 

$58.18 «3wa on. 333 1388205 wEEomv mmcoammm 5qu mo :oBaESmm 85on Sow—Em uwN @388

116

885:6 somwouiavh 98 Emhbmzaommmfiu wt? «:8 finanmsn mo

wma 98 :ompwosvw :33 330835 8535? 3:55:50 v85 98 3358800 2: pa x55 £35888 2: am @820 iouofl

azoghgomwbazfifioo 23 an 835m 35880353303850: 2: E :oE\:mEoB 352m we 53:5: 0:... ”20596: «.5 E

5503\55 emu wanton, mo 85:5: 2: ”20:03.03 25 5 5A. mo awn am 55:40 .«0 .8285: 2: ”22385: 2: 5 ed mo mww

pm Etc—Eu mo .8355: 833:3 acvvcmamvfi unamgéfifi «o mwfiozm 2: 2w maommmemmz 2: E 89:05 83£8> 550 .m

.732 0&3 ad unwoficwfi «:32 “Nb 3 undocfimii £85588 5 was $0th 6:6:ng 885,3:

 

 

onmm onmm onmm Emm Emm onmm Bo mo 83:52
nwmd ommd 38¢ mgd mmvd :vd @3878
853:3 uaqumaovfi

mmg 8.: 3.2 3.2 wag 922 ESE>¢§B mo mwSo><
”853$:me ”Ewen.

88.8 83>lm8mb< www>>x
owed- ~838sz Mora
23.8 @canmsmlflawt< owa>>x
Sod 833sz work,
$8.8 ark/[85:48 «3wa x

vood ”~330sz Mvamnmsm

Ammodv vamnmsmlmuwouhm mwa>> x

v8.0- 358va @9533:

8.88 E88 088L522

wmod- mwod oma>>x$omm<

$8.8 $8.8 cgnmsmlfldmbxx

Sod- wmod- wwwzwxfimwm<

umEHmH—r QOmHONHQHQH

.3258 3 28.8

117

Table 2.9: Random Probit Estimation of Labor Response (Change of Job) to First

Wage Arrears

 

 

 

 

Husband’s Wife’s
Change of Job Change of Job
1 2 1 2
First Time Wage Arrears 0.029 0.029 0.045 0.050
in Previous Period—Husband (0.166) (0.167) (0.187) (0.188)
First Time Wage Arrears -0.021 -0.021 0.241 0.244
in Previous Period—Wife (0.163) (0.173) (0.174) (0.173)
Education of Husband 0.025 0.027 0.003 0.005
(0.055) (0.057) (0.059) (0.080)
Education of Wife 0.023 0.021 -0.023 -0.014
(0.057) (0.054) (0.063) (0.066)
Age of Husband -0.313 -0.341 0.271 -0.271
(0.344) (0.344) (0.373) (0.374)
Age2 of Husband -0.001 -0.001 0.002 0.002
(0.004) (0.004) (0.004) (0.004)
Age of Wife 0.326 -0.316 0.374 0.371
(0.366) (0.363) (0.405) (0.405)
Age2 of Wife —0.001 -0.002 -0.005 -0.005
(0.004) (0.007) (0.004) (0.005)
(log) Non-labor Income —0.016 -0.023 -0.032 -0.031
(0.014) (0.012)* (0.015)M (0.014)"

118

Table 2.9 (cont’d).

Interaction Terms:

Husband’s Educationx

First Time Wage Arrears—Husband
Husband’s Educationx

First Time Wage Arrears—Wife

0.023
(0.012)*
-o.oo3
(0.026)

 

Wife’s Educationx 0.024
First Time Wage Arrears~—Husband (0.033)
Wife’s Educationx -0.039
First Time Wage Arrears—Wife (0.034)
Joint Significance:

Average of Time-Variant 120.82 120.71 69.54 69.33
Independent Variables

p—value 0.000 0.000 0.000 0.000
Number of obs. 2617 2617 2617 2617

 

 

 

1.Robust standard errors are in parenthesis.

level.

"significant at 5% level;* significant at 10%

2.0ther variables included in the regressions are the average of time-variant independent

variables number of children at age of 0-6 in the household; the number of children at age

of 7-17 in the household; the number of working age men/ women in the household;

the number of elderly women/ men in the household; employment service at the community;

government factory closed at the community, bank at the community, and these community

variables interacted with education and age of husband and wife respectively; year

and year/ region dummies.

119

Table 2.10: Random Probit Estimation of Labor Response (Change of Job) to
Wage Arrears

 

 

 

Husband’s Wife’s
Change of Job Change of Job
1 2 1 2
Wage Arrears in Previous Period 0.270 0.271 -0.063 -0.070
—Husband (0.096)” (0.091)" (0.109) (0.109)
Wage Arrears in Previous Period -0.082 -0.082 -0.096 -0.087
——Wife (0.098) (0.098) (0.111) (0.111)
Education of Husband 0.022 0.024 0.003 0.005
(0.055) (0.057) (0.060) (0.059)
Education of Wife 0.027 0.027 -0.024 -0.015
(0.057) (0.058) (0.063) (0.066)
Age of Husband -0.346 -0.346 -0.291 -0.291
(0.345) (0.345) (0.371) (0.371)
Age2 of Husband -0.0004 -0.0004 0.002 0.002
(0.004) (0.004) (0.004) (0.004)
Age of Wife 0.334 0.334 0.411 0.408
(0.368) (0.368) (0.403) (0.404)
Age2 of Wife -0.0002 -0.0002 -0.005 -0.005
(0.004) (0.004) (0.004) (0.004)
(log) Non-labor Income -0.017 -0.017 -0.032 -0.039
(0.014) (0.014) (0.015)" (0.015)”

120

Table 2.10 (cont’d).

Interaction Terms:

 

Husband’s Educationx 0.024

Wage Arrears—Husband (0.057)

Husband’s Educationx 0.027

Wage Arrears—Wife (0.058)

Wife’s Educationx 0.022
Wage Arrears—Husband (0.033)
Wife’s Educationx -0.039
Wage Arrears—Wife (0.034)
Joint Significance:

Average of Time-Variant 120.63 120.54 69.01 68.66
Independent Variables

p-value 0.000 0.000 0.000 0.000
Number of obs 2617 2617 2617 2617

 

1.Robust standard errors are in parenthesis. “significant at 5% level;* significant at 10%
level.

2.0ther variables included in the regressions are the average of time-variant independent
variables number of children at age of 0-6 in the household; the number of children at age
of 7-17 in the household; the number of working age men /women in the household;

the number of elderly women / men in the household; employment service at the community;
government factory closed at the community, bank at the community, and these community
variables interacted with education and age of husband and wife respectively; year

and year/ region dummies.

121

Table 2.11: Consumption Regression (First Time Arrears)

 

Reduced Structure

 

 

 

Form Equation
Fixed Ef— Fixed Ef-
fect fect
Estimation IV Esti-
mation
First Time Wage Arrears -0.051 -0.026
—Husband (0.039) (0.042)
First Time Wage Arrears -0.035 -0.048
—Wife (0.041) (0.045)
Secondary Job Holding 0.699
—Husband (0.251)“
Secondary Job Holding 0.336
———Wife (0.215)
Age of Household Head —0.048 -0.050
(0.032) (0.034)
Number of Male Children -0.200 -0.190
(<18 years old) (0.027)“ (0.029)“
Number of Female Children -0.124 -0.126
(<18 years old) (0.022)“ (0.024)“
Number of Working Age Male -0.107 -0.097
(age 1860) (0.030)“ (0.032)“
Number of Working Age Female -0.101 -0.118
(age 1855) (0.031)“ (0.033)“
Number of Senior Male 0.033 0.070
(260 years old) (0.089) (0.095)
Number of Senior Female -0.216 -0.214
(255 years old) (0.051)“ (0.057)“
No-Labor Income 0.009 0.009
(0.003)“r (0.003)“
Number of obs 3570 3570
F(78,2510) 7.16
Wald chi2(80) 936246.45

 

1.Standard errors are in parenthesis. “significant at 5% level;* significant at 10% level.
2. Other variables included in the regressions are the year and year/ region dummies

122

Table 2.12: Consumption Regression (Wage Arrears)

 

 

Reduced Form Structure Equation

 

 

 

Fixed Effect Fixed Effect
Estimation IV Estimation

Wage Arrears 0.025 -0.028
-—Husband (0.029) (0.034)
Wage Arrears -0.047 -0.088
—Wife (0.029)* (0.032)“
Secondary Job Holding 0.726
—Husband (0.383)*
Secondary Job Holding 0.725
—Wife (0.323)”r
Age of Household Head -0.107 -0.119

(0.051)“ (0.054)“
Number of Male Children -0.257 -0.264
(<18 years old) (0.038)“ (0.041)”
Number of Female Children -0.135 -0.152
(<18 years old) (0.029)“ (0.032)“
Number of Working Age Male -0.126 -0.108
(age 18—60) (0.037)“ (0.041)“
Number of Working Age Female -0.109 -0.163
(age 18-55) (0.038)“ (0.044)“
Number of Senior Male 0.025 -0.008
(260 years old) (0.114) (0.126)
Number of Senior Female -0.170 -0.158
(255 years old) (0.062)“ (0.076)“
No—Labor Income 0.011 0.009

(0.004)“ (0.004)“
Number of obs 3570 3570
F(78,2510) 4.87
Wald chi2(80) 675238.84

 

1.Standard errors are in parenthesis. "significant at 5% level;* significant at 10% level.
2. Other variables included in the regressions are the year and year/ region dummies

123

BIBLIOGRAPHY

Alfandar, G. and M. E. Schaffer (1996), “‘Arrears ’in the Russian Enterprise
Sector, ”Enterprise Restructuring and Economic Policy in Russia ed.
Commander, S., Q. Fan, and ME. Schaffer, Washington DC. World Bank.

Bellman, R. (1957), Dynamic Programming Princeton, N .J .: Princeton
University Press.

Blundell, R. and T. MaCurdy (1999), “Labor Supply: A Review of Alternative
Approaches, ” Handbook of Labor Economics Volume 3, ed. O. Ashenflter and
D.Card.

Blundell, Richard, T. Magnac and C. Meghir (1997), “Savings and Labor-Market
'Iiansitions, ”Journal of Business 85 Economic Statistics Volume 15, No.2:
153-164.

Clarke, S. (1998), Structural Adjustment Without Mass Unemployment?:
Lessons from Russia Northampton, Mass: E.Elgar.

Commander, S., S. Dhar, and Y. Ruslan (1996), “How Russian Firms Make
Their Wage and Employment Decision, ” Enterprise Restructuring and Economic
Policy in Russia ed. S.Commander, Q. Fan, and M. E. Schaffer, Washington
DC. World Bank.

Desai, P. and T. Idson (2000), Work Without Wages— Russia’s Nonpayment
Crisis The MIT press.

Desai, P. (2000), “Why did the Ruble Collapses in August 1998?,” The American
Economic Review 90.2: 48-52.

Earle, J. S., and K. Z. Sabirianova (2002), “How Late to Pay? Understanding
Wage Arrears in Russia, ”Journal of Labor Economics v20, n3: 661-707.

Foley, M. C.(1997), “Labor Market Dynamics In Russia,” Economic Growth
Center, Yale UniversityCenter Discussion Paper No.780.

124

’ Gimpelson, V., and D. Lippoldt (2001), The Russian Labor Market: Between
Transition and Turmoil Lanham, Md. and Oxford: Rowman and Littlefield.

Grosfel, I. S., T. Verdier, S. Kolenikov, and E, Paltseva (2001), “Workers’
Heterogeneity and Risk Aversion: A Segmentation Model of the Russian Labor
Market, ”Journal of Comparative Economics 29: 230-256.

Heckman, J. J., and T. E. MaCurdy (1980), “A Life Cycle Model of Female
Labour Supply, ”Review of Economic Studies 49: 659-660.

Honoré, B. E. (1992), “Trimmed Lad and Least Squares Estimation of Truncated
and Censored Regression in Models with Fixed Effects, ” Econometrica 13:
533-565.

Jensen, R., and K. Richter (2002), “Social Security, Income Volatility and
Health: Evidence from The Russian Pension Crisis ”Working Paper.

Kochar, A. (1999), “Smoothing consumption by smoothing income:
hours-or-work responses to idiosyncratic agricultural shocks in rural India,
”Review of Economics and Statistics Vol. 81(1): 50-61.

Koumakhov, R., and N. Boris (2001), “Labor Hoarding in Russia: Where Does It
Come From?,” William Davidson Institute at the University of Michigan Business
School Working paper 394.

Lehmann, H., J. Wadsworth, and A. Acquisti (1999), “Crime and Punishment:
Job Insecurity and Wage Arrears in the Russian Federation, ”Journal of
Comparative Economics v22, n4: 595-617.

Lugovoy, O. (2002), “Arrears In A Titansition Economy: The Mechanism of
Contagion, ”RECEP working paper.

Lundberg, S. (1985), “The Added Worker Effect, ”Journal of Labor Economics
Vol. 3, Issue 1: 11-37.

MaCurdy, T. E.(1985), “Interpreting Empirical Models of Labor Supply in an
Intertemporal Framework with Uncertainty, ”Longitudinal Analysis of Labor
Market Data ed. J. Heckman and B. Singer, Cambridge: Cambridge University
Press.

125

Mackenzie, D. J. (2003a), “Aggregate Shocks and Labor Market Responses:
Evidence from Argentinas Financial Crisis, ”CREDPR Working Paper No. 176,
forthcoming, Economic Development and Cultural Change.

Mackenzie, D. J. (2003b), “How do Households Cope with Aggregate Shocks?
Evidence from the Mexican Peso Crisis, ” World Development 31(7): 1179-99.

Mroz, T., L. Henderson, and B. Popkin (2001), “Monitoring Economic Conditions
in Russian Federation: The Russia Longitudinal Monitoring Survey 1992-2000,
”Report submitted to the US. Agency for International Development. Carolina
Population Center, University of North Carolina at Chapel Hill, North Carolina.

Mroz, T., B. Osmolovsii, and B. Popkin (2002), “Monitoring Economic
Conditions in Russian Federation: The Russia Longitudinal Monitoring Survey
1992-2001, ”Report submitted to the US. Agency for International Development.
Carolina P0pulation Center, University of North Carolina at Chapel Hill, North
Carolina.

Rose, E. (2001), “Ex ante and ex post labor supply response to risk in a
low—income area, ”Journal of Development Economics Vol. 64: 371-388.

Russian Economic Trends Working Center for Economic Reform, Government of
the Russian Federation. London,U.K.: Whurr Publishers. Various issues
1999-2002.

Smith, J. P., D. Thomas, E. Frankenberg, K. Beegle, and G. Teruel (2002),
“Wages, employment and economic shocks: Evidence from Indonesia, ”Journal of
Population Economics 15: 161-93.

Spletzer, J. R. (1997), “Reexamining the Added Worker Effect,” Economic
Inquiry Vol.XXXV: 417-427.

Stephens, M. Jr.(2002), “Worker Displacement and the Added Worker
Effect,” Journal of Labor Economics v20, n3: 504-537.

Woodbury, S. A. (1983), “Substitution Between Wage and Nonwage Benefits,
”The American Economic Review 73:166-182.

Wooldridge, J. M. (2002) Econometric Analysis of Cross Section and Panel Data
MIT press: Cambridge, MA.

126

APPENDIX

Table 2.13: First Stage Regression of Multiple Job Holdings

 

 

Husband Taking
Secondary Job

Wife Taking
Secondary Job

 

 

Presence of Community -0.071 -0.116
Employment Service Center (0.210) (0.252)
Employment Service Center 0.029
Interacted with Husband’s Education (0.014)"
Employment Service Center 0.036
Interacted with Wife’s Education (0.018)“
Age of Household Head -0.019 -0.011
(0.003)" (0.004)"
Number of Male Children -0.043 -0.259
(118 years old) (0.128) (0.159)
Number of Female Children -0.158 -0.042
(118 years old) (0.074)” (0.100)
Number of Working Age Male 0.034 -0.088
(age 18—60) (0.045) (0.052)*
Number of Working Age Female 0.053 0.064
(age 18-55) (0.029)* (0.033)*
Number of Senior Male -0.040 0.008
(g,=60 years old) (0.059) (0.067)
Number of Senior Female 0.046 0.034
(1,=55 years old) (0.058) (0.067)
Non-labor Income 0.014 0.005
(0.007)“ (0.007)
Statistics for Joint Test
of Instrumental VariablesChi2(2) 4.89 5.2
Prob>chi2 0.0867 0.0744

 

1.Standard errors are in parenthesis. “significant at 5% level;* significant at 10% level.
2. Other variables included in the regressions are the year and year/ region dummies

127

Chapter 3

Estimation With or Without
Strict Exogeneity Assumption in
Unequally Spaced Panel Data

3.1 Introduction

In unequally spaced panel data, observations are missing entirely in certain
periods and they are only available on non-consecutive basis.1 Unequally spaced
panel data are normal rather than exceptions in many surveys carried out in
developing countries where the survey was interrupted because of funding issues
or other reasons. Mckenzie (2001) documented some examples of unequally
spaced surveys in developing countries. There are two other well-known surveys

which we would add to this list: Indonesia Family Life Survey (IF LS), whose

 

1 Different from unequally spaced panel data, unbalanced panel data, according to Arellano
and Bond (1991), is a sample in which consecutive observations on individual units are available,
but the number of time periods available may vary from unit to unit as well as the historical
points to which the observations correspond.

128

survey periods are 1993, 1997 and 2000; and Russian Longitudinal Monitory
Survey (RLMS), whose survey periods are 1992, 1993, 1994, 1995,1996,1998,2000,

2001, 2002.

For unbalanced panel data, Arellano and Bond (1991) noted that nothing
fundamental changes in the econometric methods, provided a minimal number of
continuous time periods are available for each unit. Wooldridge (2002, Chapter
17) also remarks that, provided the decision to select units out of a panel is made
randomly, any differencing method on any subset of the observed panel is
consistent and the usual test statistics are valid. For the unevenly spaced panel,
all observations are non-consecutive because in certain periods, the entire
cross-section may not be observed. Such a pattern of data missing would cause
complications when current outcome of certain variable depends on the lagged
values of the variable itself or the lagged values of other variables. Baltagi and
Wu (1999) considered the case in panel data where the disturbances following a
stationary AR(1). They develop a feasible generalized least squares (GLS)
procedure that takes into account that under the normal procedure, the
transformed disturbances are still heteroskedastic in the unequally spaced data.
Mckenzie (2001) studies AR(1) models with an unequally spaced pseudo-panel.
In this case, the differencing approach can no longer be directly applied and,
furthermore, unequal spacing in a dynamic model imposes nonlinear restrictions

on the parameters. In the pseudo-panel case, the population is divided into

129

certain cohorts and the mean of the cohorts is taken over the individual in each
cohorts. The resulting data is a pseudo—panel over the cohorts. And non-linear
least squares, minimum distance, and one-step estimators are used to estimate
the parameters. In genuine panel data, averaging can be taken over the entire
sample, but the estimates of the parameters can only be based on small

observations in the averaged model.

This paper focuses on a dynamic model, where the outcome of the dependent
variable depends on both the current value and the lagged value of the
explanatory variables. Section 2 considers a model with a strict exogeneity
assumption. A classic minimum distance estimation method and an one-step
GMM method are proposed for the consistent estimates of the parameters. The
two-step GMM method is also discussed. In section 3, the same model is studied
without strict exogeneity assumption. Arellano and Bond (1991) method is used
to account for the violation of strict exogeneity. The classic minimum distance
estimation then follows to recover the structural parameters. It also shows that

the same one—step GMM method also applies. Section 4 concludes the paper.

130

3.2 Estimation Under Strict Exogeneity
Condition

3.2.1 Static Model

Consider the following panel data regression model:
y” :xiifi'l'ci'l‘l‘id (i=1,...,N;t=1,2,...,T) (3.1)

, where mm is a 1 x k-vector of explanatory variables, ,8 is a k x 1-vector of
parameters to be estimated; c,-, the time-invariant unobservables of unit i are
allowed to be arbitrarily correlated with l‘z‘t; flit is an idiosyncratic disturbance.

The strict exogeneity assumption is :

E(#z‘,t|$z', 0:) = 0 (3-2)

So far, the model setup is a standard basic fixed effect model. When all the
observations are missing for certain periods, as in the unequally spaced panel,
this model becomes a special case of unbalanced panel covered by Wooldridge
(2002). Following his notation, the selection indicators (Sit) in the unevenly

spaced panel certainly depend only on exogenous rules, so the assumption

“Malawi, 32') = 0 (3-3)

131

 

is satisfied. In this case, the selection indicators are the same for all observations
and they are all zero in certain periods, thus we can omit the subscript t in the
indicator, that is: st = 0 for some t’s. Assume that all the observations are not
missing in the first period (t=1) and the last possible period (t=T). Then the
number of time periods observed for all observations should satisfy: T,- Z 2,

where T,- = Esit.

Fixed effect estimation can be extended easily to the unevenly spaced panel and
the FE estimator is consistent under strict exogeneity assumption in equation
(3). Furthermore under homoskedasticity and serial independence assumption for
the error term, the normal inference also holds, as shown in Wooldridge (2002,

Section 17.2).

3.2.2 Dynamic Model with Lagged Explanatory Variables

Now assume a dynamic model with lagged explanatory variables:

yiat = $i¢fi + 11545-10 + Ci + ”Lt (2:1, ..., N,t=1,...,T) (3.4)

where 50,3, is a 1 x k vector of explanatory variables that are dated
contemporaneously with the dependent variable gm; w,,t_1 is a scalar, which is

dated one-period lag of the dependent variable.

132

The strict exogeneity assumption in this model is:
E(Hitl$iawia Ci) = 0 (3-5)

This assumption is the key assumption for the estimation method we propose in
this section. So far the model is standard. In the unevenly spaced panel, we need
some notation to specify the pattern of missing data. Let us denote the observed
time periods by t3, where s = 1,2, ...,r and r g T. That is, y,,ts,x,,ts,wi,ts are
observed. Then the model specified in equation(3.4), in the unevenly spaced

panel case, should be:

this = witsfl + wits—10‘ + C: + Mtg (i=1,-~.N,8 = .-- T) (3-6)

The 7' — 1 equations in this model can be written as:

312' = $15 + ”wt—101 + CijT + M (i = 1, N) (3-7)
,where jT is a (r — 1) x 1 vector of ones.
In this dynamic model, w“ s_1 may not be observed. The closest observed period

to the (ts — 1)“ period is period (ts_1). We can use the value in the closest

period to predict w” 8_1.
The assumption we make about the data generating process of wig is that the
linea projector, L(w,,t3|wi,ts_m), is stationary for m 2 1. That is, we assume:

wl,ts : me’i,t3 -m + Vits (3‘8)

133

for m = 1, 2, This assumption does not put restrictions on the coefficient of
the lagged dependent variable if the lag length is different, and it only assumes
the coefficient to be the same for the same lag length. By definition of a linear

projection, the error term 11,18 has the following properties:
Ems) = 0 (3-9)

COU(1Ui,t3—ma VitS) = 0 (3.10)

At the same time, since 3321 does not help predict w” for all t, we also have the
following property of Vits

Cov(u,'ts,x,;) = 0 (3.11)
Based on the model specified in equation (3.8), we can predict 101-$3-1 by the
following equation:
w,,t8_1= ”(ts—1)‘(ts_1)wts—1+ ”its—1 (i=1, 2, ..., N; S =1,2,..., T) (3.12)
Substitute equation (3.12) into equation (3.6), we can write:

yi,t3 = $i,t3fi + wt3_17r(t3—1)-(ts_1)a + Ci + ”its + Quilts—1) (3-13)

If t, — 1 = t3_1, then (3.13) reduced to a standard model in an equally spaced
data. In vector form, let h be the number of distinct lag length between period

(t3_1) and period (ts — 1), where (h > 1). And also denote the magnitude of the

134

gap by mj with m = (ts — 1) — (ts_1) ,j =1,...,h. Then T — 1 equation in

(3.13) can be written as:

where w,,_1 = (wiy2_1,w,,t3_1, ...,w,,tT_1)' is a (r — 1) x 1 vector of the lagged
explanatory variables that we want to predict; II = f (7r) = (7r,n1,7rm2, ..., nmh)’ is
a h x 1 vector of parameters in predicting w,,_1. And wz_1 is a (r — 1) x h
matrix, where the non-zero element w,- t1,w,- t2: ..., w,- t is located in column

s 9 a 7"].
mj; mj is the gap in time period between the element in wit-1 and the
corresponding element in w,’_1. Lastly, e, = u,- + oil/,- is a (r — 1) x 1 vector of

is

error terms. Now the error term will consist of an extra term if w,,ts_1

estimated from wi,t3— 1.

For example, if t1 = 1; t2 = 3; t3 = 6; t4 = 7; t5 = 9, the estimation equation in

(3.14) will be:

yi,3 132's 0 ’wz',1 0 1 Ms + 0521.2
yrs 332'6 fl 0 0 w2:3 . M6+a€z°5
9 = a + a “-1 a + . + a ’

yi,7 1171.7 was 0 0 7,2 6’37 um
31239 331,9 0 101.7 0 #239 + 05i,8

The first method we consider to estimate ,8 and a is the classic minimum
distance (CMD) estimator. Letting z, = (33,-, 711:4) be a (r — 1) x (k + h) matrix
of exogenous variables, and 6 = (fi’, 7')’ be (k + h) x 1 vector of parameters to be

estimated, where 7 = Ha. The model in (3.14) can then be expressed as:

y,- =zi6+cij7+ei (i=1,...,N) (3.15)

135

Under strict exogeneity assumptions in (3.5) and the conditional mean
assumptions in (3.10) and (3.11), together with usual rank conditions, the fixed
effect estimator for (3.15) is consistent. We call this fixed effect estimator éFE-
And 6‘“; = (2,2,82;,,82,,,S)-1(2,2,szggtsym) , where the double dots stands for

demeaned form of variables.

Without further assumptions on the serial correlation in the error term, the
demeaned form of error term «'5', might be serially correlated and heteroskedastic.
Thus the robust variance matrix estimator should be applied for any statistical
inferences. The estimated robust variance matrix of the fixed effect estimator for
this model is:

5‘2 = AWE”) = (Z Elia—WE 2:204 (3.16)
where, V = Z,- jig-g2,- and 2; is the error term from fixed effect estimation, that
is,

e,- = y,- — 2,8 F E. This variance matrix is valid in the presence of any

heteroskedasticity or serial correlation, provided that r is small relative to N

(Wooldridge 2002).

The fixed effect estimator S F E is a (k + h) x 1 vector of estimated coefficients in
the “reduced ”form. We denote the “structural ”form of coefficients to be a
(k + 1) x 1 vector: 60 = (B’, a). To denote that the “reduced ”form of coefficients
0 = (B’,rrm1a,...,7rmha)' is a non-linear function of 60, we write 0 = f(60),

f : Rk+1 —+ Rk+h is a continuously differentiable function. Since éFE is a

136

consistent estimator of 6, we can recover the estimates of the 60 from éFE- The
classical minimum distance estimator of 60, denoted by 60 M D is the solution to

the problem:
MinaoléFE - f(9o)l’f?_1léFE — f (90)] (3-17)

,where {2‘1 is given in (3.16).

The solution do M D solves the first order condition:

[Vsof(éCArD)l'9'lléFE - “germ” = 0 (3-18)

,where Vgof(SCMD) is a (k + h) x (k + 1) Jacobian of f(éCMD)- Hansen (1982)
establishes that the resulting estimator 90 M D is consistent and asymptotically
normal. The estimated variance of SC M D takes the form:

A

AvafléCMD) = N —1{lV00f (genroll'fl-l [Vaof (écarolll (3-19)

The second method we consider is joint estimation of 7r, B, a through generalized
method of moment. This is an application of GMM framework for sequential
estimation studied by Newey and McFadden (1994). In this application, the
feasible estimation of B and (1 depends on the estimation of 7r in the first place.
By applying the general GMM formula to simultaneously estimate the two sets of

parameters, we can get consistent standard error estimators for the estimation of

B and a.

137

The first set of moment conditions does not include parameter B and a. It only
involves using observed period in the estimation of 7r. For example, the following

regression contains 7r1 and M in previous example:

wi,3 0 10231 W1 52,3
10137 = 101-"6 0 [ “2 J + 81,7
10239 0 101,7 51,9
In general, we write
w,- = 105:1“ + E, (3.20)

The dimension of w, is l x 1 where l 2 h; and luff—1 is a l X h matrix of observed
values of w as well. In other words, the dimension of w,- and wail depend on the
lag length in observed data as well as the missing period. In order to be able to
use the estimators from the observed period, the lag length between observed
periods must include the lag length needed to estimate the missing period. The
number of moment conditions depends on the lag length of observed periods as

well as the lag length needed to estimate the missing period.

We use f1(w, 7r) to denote the first set of moment conditions which involves

parameter 7r only:

f1(w. H) = Elw“2,-1(w. — wailnn = 0 (3.21)

The second set of moment condition, denoted by f2(w, x, B, a, II), is specified in
the differenced form of (3.14), where the unobserved time-invariant heterogeneity

is differenced out. This set of moment conditions involve the estimation of B, a as

138

well as IT:

f2(w,a:, H,fi,a) = E [ wjf ] [Ayi — (Asa-B — Aw,”_ll‘la)] 1' 0 (3.22)
1 .

Define the sample version of f1 and f2 by f1 and f2, where

—Zw**’ ,- _1(w --— wfiln) = 0 (3.23)

N I
f2(w3x1H,/Baa)=I—1V-Z [ mil/:1. 1 :l [AZ/i — (Altifi "" sz_1Hoz)] = 0 (3.24)

The moment conditions for simultaneous estimation of H, B, a are

f(w. x. H, a a) = {fun}. H)’. f2(w. x. H. a a)’]’ (3.25)

The GMM estimate of the parameters thus is the solution to the following
minimum criteria:
min(f)'W(f)’ (3.26)

and the optimal weighting matrix W = @[flm 1:, II, B, (1)].

Further, let 6 = (B'a)' and call the GMM estimates of 9 and II, 6 and f1. Define

A

the sample Jacobian terms by F11 = an1(w,fl);1"21= an2(w, FLO) and

[‘22 = ng2(w,$, 110) and ‘II = —f’1_11f1. In our application,

1
I‘11 = _TV_ Z(w{*_1'wf*_1) (3.27)

139

’A
r2,:_ 1 “EN: (11:? [1051) (3.28)
0221:1111 lAu’i,—l)

and

1‘22 2 _i ZiNzl($i,A$l) Zi="(1($ilAw:-1H)A (3 29)
25:1(u’Z—1'A13z‘) Zi=1< u’i,—1’Awi,—1H)

The variance of 6 is given by Newey and McFadden (1994) as the following:

A

Var(6)=(-1F22

ZIH

N
{2] (f2+1“21\11> (f2+F21‘1’)’(l} Fee (3.30)

It is worth mentioning the alternative of estimating the parameters through a
two-step GMM estimation. The first step is to estimate 11 by the moment
conditions given in (3.21). The second step is to estimate B and a by the
moment conditions given in (3.22), taking H from the first step as given. Newy
and McFadden (1994) derives the condition, under which the standard error of
the estimates of B and a is not affected by the estimation of II from (3.21) only.
The condition is F21 = 0. In our application, it means that if and only if a = 0,
does the first step have no effect on the second-step asymptotic variance. In
empirical application, when the null hypothesis that a = 0 can not be rejected

from the two step estimation, the standard error from the second step need not

140

be corrected. In other cases when a 76 0, the one-step GMM estimation outlined
above is no less efficient than the two-step GMM and the standard error is
correct. One concern in terms of the computation of the one-step GMM is that
the iteration may not converge since the moment conditions are not linear in
parameters. This may be alleviated by using the two-step estimates as the initial

value in the iteration.

3.3 Estimation Without Strict Exogeneity
Condition

3.3.1 Static Model

We are still look at the following model in the unevenly spaced panel:

yi,t3 = $i,t,sfi + a; + mi, (331)

but the assumption we make on the error term “Lts is the following:

E(Hi,t3l$i,ts—1’xi,t3—2a -°-»$z',1aci) = 0 (3.32)

By this assumption, “its is uncorrelated with error terms pm]. for t; < tj,
conditional on time invariant heterogeneity. In other words, “is is
predetermined. So, the correlation of the error terms with contemporaneous and
future values of the explanatory variables is allowed. One example would be the
regression of working hours on wage arrears shock, when working hours affect the

possibility of wage arrears shocks in the subsequent period.

141

The general method used for the model with violation of strict exogeneity
assumption also applies in the unevenly spaced panel. First difference the data in
the unevenly spaced panel gives us:

31mg — yi,t3_1 = (30mg — $i,t3_1)fl + (Hats - Mi,ts_1) (3-33)

and the observed level or change in the lagged value of arms can be used as
instrument variables. That is, $i.ts_2v$i,ts_3v--:Or (332'.ts_2 -— $i,ts,3),

(“$3-3 — x,,ts_4)...can be used as instrument variables for (sums — mists-1) in a
GLS estimation. Or in the Arellano and Bond (1992) framework, more of the

instruments can be explored in the GMM estimation.

3.3.2 Dynamic Model

Consider the following model:
yi,ts = $i,ts(3+wz',ts—1Q+Cz' +#z',ts (i=1,---,N,8 =1,---,T) (334)

We assume that conditional on individual time invariant heterogeneity,

explanatory variables 3:, and w,- are predetermined with respect to the error term:

E(#it|$i,ts—1,$z',ts-2, 502,1; wi,t3—1awi,ts—2a ..., 101,1; Ci) = 0 (3-35)

The within estimator for (3.34) is inconsistent because the within transformed
error is a function of predetermined information and thus is not orthogonal to the

within transformed explanatory variables. A common econometric approach for

142

handling violation of strict exogeneity is to first differencing this equation and
apply instrument variables on the differenced equation. The differenced equation

can be written as:
Ail/ms = Al‘msb’ + Aims—101 + Aims (335)

,where A denotes the first difference operation. That is, Ayms = yi,ts — yi,ts_1§
Axus = 13mg - 33i,t3_1; Awi,t3—1 = was-1 — “Jets—2 and

Alias = flats — Hi.ts_1-

As we mentioned before, mm 3-1 may not be observed in the unevenly spaced
panel and the most recent recent period that is observed is period t3_1. Based on
the assumptions of (3.8) and (3.10) for 10, we can predict wi,t s_1 by using
observed period. But we need to be careful in the prediction when ”wit is not
strict exogenous with respect to the error term. For example, when t3_1 = t3 — 1,
that is the previous one period is the most recent period observed for some 3, the
fact that E(w1,ts_1#i,ts_1) # 0 will cause the predicted term to be correlated
with the error term. In this case, we need to use one lag period before the most
recent period, that is "Luz-$84, to predict wi,t—1- But if all the gap between two
periods are greater than one, then we can still use the most recent period to
predict the missing period. To simply the notation, we assume that the gap
between two observed period is greater than one, thus we can use the most recent

period as predictor. The first differenced form of the model in terms of observed

143

period can be written as:

yi.t3 - yi.t3_1 = (36mg - $i.ts_1)fi + iwz’,t8_17r(tS—1)—(t3_1) (3-37)
—wzf,t3_277(ts_2)_(ts_2)la + (Hus — Hi,t3_1)

+(Vi,t3—1 — Vi.ts—2)0

In matrix form, (3.37) can be written as:
Ayz- = Ari-[1’ + Awilfla + A6.) (3.38)

where, A6,- = Au,- + aAui.

There are two issues that need to be addressed. One is that the correlation
between Arms and Anus is nonzero, because

E(AIi,t3A/1ni.ts) = E($i,t3 — Ii,t3_1)(#i,ts — #i,ts_1) =

E($i,tsllz',t3) — E($1,t5#~i,t3_1) -+- E($i,ts_1#i,ts_1)- All these three terms are not
equal to zero, according to the assumption that x,- are predetermined. Another
issue is that the parameter Ha is nonlinear and we want to “recover ”the

“structural ”coefficient oz.

The first issue can be addressed easily by instrument variable estimation. The
possible instruments for Amati; are [Hiya/“Lt? Marl-$34]. For example, starting

from period t3,in the Arellano and Bond (1991) framework, the matrix of

144

instruments, denoted by Z,- is :

i$i,t1a(wi,t2 -— wi.t1)l 0
Z1 = '

0 [13211 a 113m? ---33i,t1._2, (wi,t.,_1 — 10:39-2”

The estimated weighting matrix is:
<i> = 2,2,f(&,)(&,)'z, (3.39)

A5,: is residuals from the preliminary consistent estimator, such as 2SLS using a

couple of lags as instrument variables. The consistent estimator, denoted by

OGMM = ((BIGAIM’ HOGAIAI)’, is:

éGMM = [(131723 Aw:1),Zi(i>,er(A$iaAW:1)l-1l(A$iaAw:1)’Zi(i)’ZiA3/il (3.40)
A consistent estimate of the asymptotic variance of 66 M M is given by:

O = Avar(6GMM) = [(Ax,, Awil)'Z,-<I>’Z£(A:ri, Aw:1)]-1 (3.41)

To recover the parameter 60 = (13’, a) from 63 M M, we can apply the classic

minimum distance estimator in (3.17) and (3.18)to obtain consistent estimator of

60.

The one-step GMM method proposed in section 2 can also be applied in this
case. The only difference is that the set of moment conditions in (3.22) will be in

the following form:

I

132/T21. ] [Ag/2' - (A3316 — AwZ_1Ha)] = 0 (3.42)

w 2,—1

f2,(w,2:,l'l,fi,a) = E [

145

Starting from t3,

i$i,t1ia 0

l$i,t ,332',t a]
JUL-2: 1. 2

0 [x.,.,,x.,.,,...x.,.._21

Thus 6, a as well as II can be estimated through:

600,23, [1.5.60 = [fi(w,H)’, fi'(w,x,H,fi,a)'l' (3-43)

The one-step GMM method is no more complicated in the case without strict
exogeneity than the case with strict exogeneity. The only difference is there will
be fewer possible moment conditions in the second set of moments containing the
parameters 6 and a. The variance of the estimates of 6 and a given in (3.30)

also applies.
3.4 Conclusion

Estimation in unevenly spaced panel in a static model does not involve more
complication. In a dynamic model, the estimation involves predicting the missing
lagged value. In this paper, we consider a model with one lagged explanatory
variable. The classic minimum distance estimation can be applied to recover the
structural parameters from the consistent estimator of the reduced form

parameters. The consistent estimator for the reduced form parameters can be

146

obtained from either fixed effect estimation or GMM estimation, depending on
the assumption about the exogeneity of the explanatory variables with respect to

the error term. Also, the one-step GMM method can be applied in both cases.

147

BIBLIOGRAPHY

Angrist, J. D., and W. K. Newey (1991),“Over-Identification Test in Earnings
Functions With Fixed Effects, ”Journal of Business 85 Economic Statistics Vol.
9, No.3: 317-323.

Arellano, M. and S. R. Bond (1991), “Some Specification Tests for Panel Data:
Monte Carlo Evidence and an Application to Employment Equations, ” Review of
Economic Studies 58: 277-298.

Baltagi, B. H., and Y. Change (1994),“Incomplete Panels—A comparative study
of alternative estimators for the unbalanced one—way error component regression
model, ”Journal of Econometrics 62: 67-89.

Baltagi, B. H., and P. X. Wu (1999),“Unequally Spaced Panel Data Regressions
with AR(l) Disturbances, ”Econometric Theory, 15: 814-823.

Baltagi, B. H. (2001), Econometric Analysis of Panel Data, John Wiley & Sons,
LTD.

Mckenzie, D. J .(2001), “Estimation of AR(l) Models with Unequally Spaced
Pseudo-panels, ”Econometrics Journal Vol 4: 89—108.

Newey, W., and D. McFadden (1994), “Large Sample Estimation and Hypothesis
Testing, ”in Handbook of Econometrics Vol. IV, ed. R. Engle and D. McFadden:
2113-2241.

Wooldridge, J. M. (2002), Econometric Analysis of Cross Section and Panel
Data, MIT press: Cambridge, MA.

148

 

 

 

 

 

 

 

 

m 9 , .
mi18 H . , . .
. will)” , .

nifllel.4 . . .
U‘ilo . .

. E .
T . .

. ul.3 ,
5'19 . .
“H2 , 1 .

. Gi 111 .
mil. . , .. -
ml3 . . .