kg It .. V’LfauFHYﬁilid. . .
.. a: a v. 972...»...2

v.49...

 

 

 

 

..
L}. A
L n 7:: 3K

 

1:1: .1.
3.3.7.. .

 

.mliiungnnuni 1.4.). 91.x
5 2.. 2.2594 citieg
.3! 6?" If..."
. .elrilvoniiilzf
St. I

 

 

 

    

   

 

. . . . , . . . ... .41... . ,
‘ . . AV: 0 D. . ‘. r . ‘ tr . a.
. it ...; - , . . . . : . .. , t . .... A . .u. finch. .. .. . .. . . z. : .r .A.
rtﬁf O. . . .. . .. . . .4 .u”. .ﬁ}...l 5‘5; ‘~ ms. .. I!
.n “Lb: ... . . . . . t. i 2.4... s .7. . 1: f . p: it: 32...): a J... A?! .I .1.

 

 

 

J. IllllllllllllllllWillllllllllllllllllllllllllllll

3 1293 02061 1764

This is to certify that the

thesis entitled

Forecasting Accuracy, Rational Expectations,and
Market Efficiency in the US Beef Cattle Industry

presented by

Matthew P. Schaefer

has been accepted towards fulﬁllment
of the requirements for

Master of ScienceAiegree inAgricultural Economics

mm

j r profess

 

Date May 19, 1999

 

0-7639 MS U is an Afﬁrmative Action/Equal Opportunity Institution

 

 

 

LIBRARY

Michigan State

University

 

 

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

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

11/00 animus-M4

FORECASTING ACCURACY, RATIONAL EXPECTATIONS, AND MARKET
EFFICIENCY IN THE US BEEF CATTLE INDUSTRY

By

Matthew P. Schaefer

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE

Department of Agricultural Economics

1999

ABSTRACT

FORECASTING ACCURACY, RATIONAL EXPECTATIONS, AND MARKET
EFFICIENCY IN THE US BEEF CATTLE INDUSTRY

By

Matthew P. Schaefer

Recent studies have tested whether futures prices respond to US. Department of
Agriculture inventory reports in accordance with the efﬁcient market hypothesis. These
studies use survey forecasts to identify the anticipated and unanticipated information
contained in a report. However, this approach implicitly assumes the survey forecast to be
an unbiased and efﬁcient predictor of the data in the USDA report. Furthermore, previous
studies have not tested the bias and efﬁciency properties of USDA preliminary estimates
as predictors of ﬁnal revised USDA ﬁgures. This study introduces a framework for
conducting tests of the efﬁcient markets hypothesis in the presence of biased and
inefﬁcient survey forecasts, and preliminary USDA estimates that are biased and ineﬂicient
predictors of ﬁnal revised ﬁgures. The approach is applied to the US beef cattle industry

and results are quite different from those obtained using the conventional analysis.

TABLE OF CONTENTS

LIST OF TABLES .................................................... v
DEFINITIONS ...................................................... vii
CHAPTER ONE ...................................................... 1
1.1 Introduction ................................................ 1
CHAPTER TWO ..................................................... 4
2.1 Introduction ................................................ 4

2.2 Theoretical Considerations ..................................... 5

2.3 Evidence of An Announcement Effect ............................. 8

2.3.1 Hypothesis 1 ......................................... 8

2.3.2 Hypotheses 2 and 3 ................................... 10

2.4 Conclusions ................................................ 14
CHAPTER 3 ........................................................ 15
3.1 Introduction ............................................... 15

3.2 The Literature .............................................. 17

3.3 Theoretical Framework ....................................... 19

3.3.1 Hypothesis 1: Unbiasedness .......................... 20

3.3.2 Hypothesis 2: Efﬁciency ............................... 21

3.3.3 Constructing an optimal forecast of ﬁnal revised cattle on-feed . 22

3.4 Data ..................................................... 22

3.4.1 Preliminary and ﬁnal revised estimates .................... 22

3.4.2 Public data ......................................... 23

3.4.3 Private data ........................................ 23

3.5 Results ................................................. 24

3.5.1 Test of unbiasedness of USDA preliminary estimates ......... 24

3.5.2 Test of efﬁciency .................................... 27

3.5.3 Forecast evaluation ................................... 30

3.5.4 Summary of section 3.5 ............................... 33

3.6 Implications ............................................... 34

3.6.1 Test of the rational expectations hypothesis ................ 34

3.6.2 Tests of announcement effects .......................... 35

3.6.3 Tests of market efﬁciency .............................. 36

3.7 Conclusion ................................................ 37
CHAPTER 4 ........................................................ 39
4.1 Introduction ............................................... 39

4.2 The Literature .............................................. 42

4.3 Theoretical Framework ....................................... 45

iii

4.3.1 Hypothesis 1: Unbiasedness .......................... 46
4.3.2 Hypothesis 2: Efﬁciency ............................... 46
4.3.3 Constructing an optimal forecast of ﬁnal revised cattle on-feed . . 47
4.3.4 Constructing an optimal forecast of unanticipated information

................................................. 47
4.3.4.1 Scenarios 1,2, and 3: .......................... 48
4.4 Data ..................................................... 49
4.4.1 Survey forecasts ..................................... 49
4.4.2 Preliminary and ﬁnal revised estimates .................... 50
4.4.3 Other public data .................................... 50
4.4.4 Private data ........................................ 50
4.5 Results ................................................... 51
4.5.1 Analysis of industry survey forecasts ...................... 51

4.5.1.1 Test of unbiasedness of Knight-Ridder industry survey
forecasts .................................... 5 1

4.5.1.2 Test of efficiency: Knight-Ridder industry survey

forecasts ................................... 52
4.5.1.3 Forecast evaluation ........................... 57
4.5.1.4 Summary of section 4.5.1 ....................... 59
4.5.2 Analysis of the Knight-Ridder individuals’ forecasts .......... 59
4.6 Implications For Other Research ................................ 62
4.6.1 Tests of announcement effects .......................... 62
4.6.2 Tests of market efﬁciency .............................. 63
4.7 Conclusion ................................................ 64
CHAPTER 5 ....................................................... 66
5.1 Introduction ............................................... 66
5.2 Market Efﬁciency ........................................... 68
5.2.1 Weak form efficiency: the random walk model .............. 68
5.2.2 Semi-strong form test of market efﬁciency ................. 69
5.2.3 Strong form test of market efficiency ..................... 70
5.3 Conventional Approach ...................................... 70
5.4 Limitations of the Existing Methodology .......................... 78
5.5 An Alternative Approach ...................................... 79
5.6 Conclusions ................................................ 88
CHAPTER 6 ........................................................ 88
6.1 Conclusion ................................................ 88
BIBLIOGRAPHY .................................................... 92

iv

Table 1.

Table 2.

Table 3.

Table 4.

Table 5.

Table 6.

Table 7.

Table 8.

Table 9.

Table 10

Table 11

Table 12.

Table 13.

Table 14.

Table 15.

Table 16.

Table 17.

LIST OF TABLES

Descriptive statistics for live cattle nominal daily release day futures

price changes in levels, by month; 1980:01-1989: 12 .................... 9
Descriptive statistics for live cattle nominal daily release day futures

price changes in levels, by month;1990:01-1990:05 ..................... 9
Means, variances and F- statistics for hypothesis 1 .................... 10

Descriptive statistics for live cattle absolute relative non-release day

futures price changes in levels, by month; 1990:01-1990:05 .............. 12
Descriptive statistics for live cattle absolute relative release day ﬁxtures

price changes in levels, by month; 1990:01-1990:05 ................... 12
Means, variances, and statistical tests for hypotheses 2 and 3 ............. 13
Test of unbiasedness: USDA preliminary estimates .................... 25
Deﬁnitions of variables ......................................... 26

Release Dates and Contents of USDA Statistical Bulletin, Cattle: Final

Estimates ................................................... 3 1

. Forecast evaluation: RMSE comparison ........................... 33

. Test of unbiasedness of Knight-Ridder industry survey forecasts ......... 52
Deﬁnitions of variables ........................................ 53
Forecast evaluation: RMSE Comparison ........................... 58
Properties of the Knight-Ridder individuals’ forecasts ................. 61
Deﬁnitions of variables ........................................ 74
Results from test of semi-strong efficiency using the Pearce and Roley

methodology ............................................... 75

Hypotheses test results for the Pearce and Roley semi-strong form efﬁciency
test ....................................................... 76

Table 18. Results from test of announcement effects using Pearce and Roley’s

methodology ............................................... 77
Table 19. Hypotheses test results for announcement effects from Pearce and

Roley model ................................................ 77
Table 20. Parameter estimates and statistics from econometric forecasting models . . . 81
Table 21. Results from test of strong efficiency ............................. 85
Table 22. Hypotheses test results for strong form efﬁciency test ................. 86

Table 23. Results from test of announcement effects using Pearce and Roley’s
methodology ............................................... 87

Table 24. Hypotheses test results for announcement effects .................... 87

vi

DEFINITIONS

intention - an intention is deﬁned as one agent’s ex-ante anticipated action.

Examples include; the number of acres an agent intends to plant; or

the number of sows an agent intends to farrow.

survey forecast - a survey forecast is deﬁned as the average of many

individuals’ ex-ante expectations of the actual value of an

economic variable.

USDA preliminary estimate -

USDA ﬁnal revised estimate -

a USDA preliminary estimate is deﬁned as the
USDA’s ﬁrst ex-post estimate of the actual value of
an economic variable, and is subject to revision by

the USDA.

a USDA ﬁnal estimate is deﬁned as the USDA’s
ﬁnal ex-post estimate of the actual value of an
economic variable. The ﬁnal estimate is published
after a certain period, which has typically been 5
years in the case of the USDA Cattle on Feed
report. The ﬁnal estimate is considered to be the

best approximation of the variable’s actual value.

vii

INTRODUCTION

CHAPTER ONE

1.1 Introduction

This study is designed to help policy makers, econometricians, and market analysts
evaluate the usefulness of information released in the US Department of Agriculture
Cattle on Feed (COF) report, as well as private information, and other publically available
information available from the USDA on cattle market conditions. Policy makers must
make decisions about what information should be collected and reported to ensure fed
cash prices and live cattle ﬁrtures prices reﬂect available information about the supply and
demand of live cattle. Econometricians must determine which information should be used
to conduct tests of rationality, announcement effects, and the efﬁcient market hypothesis.
Producers and market analysts must make decisions regarding their positions in cash and
ﬁrtures markets with respect to survey forecasts and USDA data, relative to other

available public and private information.

Chapter 2 of this study, Does the USDA Cattle on Feed Report Still Provide New
Information to Market Participants?, investigates whether the COF report is fulﬁlling one
of its public policy objectives, which is to provide new information for the formation of
live cattle futures prices. First and second moments of release day live cattle futures price
changes are compared to those of non-release days. The objective is to determine whether

release day live cattle futures price changes have greater variance than non-release days,

indicating the Cattle on Feed report does provide new information for the formation of

live cattle futures prices.

Chapter 3, Are Preliminary Estimates in the USDA Cattle on Feed Report
Rational Forecasts of USDA Final Re vised Estimates?, tests whether USDA preliminary
estimates of cattle on-feed, placements and marketings are rational forecasts of USDA
ﬁnal revised estimates by testing them for unbiasedness and efficiency. The objective of
conducting tests of unbiasedness and efﬁciency is to determine whether USDA preliminary
estimates optimally reﬂect available public and private information useful for the
prediction of ﬁnal revised estimates of cattle on-feed, placements and marketings. The
results from this chapter have important implications for what information should be used
to conduct tests of rationality, tests of announcement effects, and tests of the efﬁcient

market hypothesis.

Chapter 4, Are Knight-Ridder Pre-Release Survey Forecasts of the USDA Cattle
on Feed Report Rational Forecasts of USDA Final Revised Estimates?, tests whether
Knight-Ridder Financial News industry survey forecasts are rational forecasts of USDA
ﬁnal revised estimates using the same approach as used in chapter 3. Again, the objective
is to determine whether the survey forecasts optimally reﬂect available public and private
information useful for the prediction of cattle on—feed, placements, and marketings
numbers. Furthermore, we determine whether a private ﬁrm’s forecasts are strong form

rational relative to their own private information. The results from tests of rationality

conducted for the survey forecasts also have important implications on how tests of

announcement effects and tests of the efﬁcient market hypothesis should be conducted.

Finally, in chapter 5, Market Efﬁciency in the US Beef Cattle Industry, we
introduce a framework for conducting tests of the efﬁcient market hypothesis in the
presence of biased and inefﬁcient survey forecasts and preliminary USDA estimates. The
methodology is employed to determine whether live cattle futures prices respond to the
USDA Cattle on Feed report in accordance with the strong form of the efficient market

hypothesis, and to test whether the report contains new information.

DOES THE USDA CATTLE 0N FEED REPORT STILL
PROVIDE NEW INFORMATION TO MARKET PARTICIPANTS?

CHAPTER TWO

2.1 Introduction

The objective of this chapter is to determine whether the USDA Cattle on Feed
(COF) report still provides new information for the formation of live cattle futures prices.
The ﬁrst and second moments of futures price changes surrounding the release of the
report are analyzed. For the period January 1980 through December 1989, Grunewald,
McNulty, and Biere (1993) reported that live cattle futures prices one day following the
COF report recorded a lock-limit price move over 15% of the time in the near-term, ﬁrst
deferred, and second-deferred contract horizons. In contrast, between January 1990 and
May 1998 lock-limit moves occurred less than 1.3% of the time; 4 in the near-term, and
zero in the ﬁrst and second deferred contracts. Based on this evidence, market agents

appear to be responding less dramatically to the information in the COF report since 1990.

If the information provided by the COF report does not alter the supply and
demand situation of a suﬁicient number of traders, then the information may be considered
irrelevant for price formation. Moreover, if the information contained in the COF report
is not relevant for price formation, then testing whether prices efficiently respond to
information released in the COF report would be irrelevant. Therefore, this analysis will

help determine whether the COF report is fulﬁlling its public policy objective, which is to

provide new market information for the formation of live cattle futures prices, as well as

determine whether a test of market efficiency is still relevant.

Our results indicate price changes one day following the COF report do exhibit
greater volatility in the 1980's than in the 1990's. However, release day price changes
during the 1990's do still exhibit greater absolute mean value and volatility than non-
release day price changes. This evidence suggests the COF report still provides new
information for the formation of near-term, ﬁrst- and second deferred contracts in the

1990's.

2.2 Theoretical Considerations
The task in this section is to test three null hypotheses which will provide empirical
evidence on whether the COF report still provides important news for the formation of

live cattle futures prices.

Hypothesis 1: Release day price changes over the period January 1980 through December
1989 have equal variance as release day price changes over the period January 1990
through May 1998. This is tested using an F-test of equal variances across two

independent samples.

Hypothesis 2: Over the period January 1990 through May 1998, the distribution of live

cattle ﬁrtures price changes on release days have equal absolute mean value as non-release

days. This is tested using t-statistics, and the Kruskal-Wallis one-way analysis of variance

by ranks test.

Hypothesis 3: Over the period January 1990 through May 1998, the distribution of live
cattle ﬁxtures price changes on release days have equal variance as non-release days. This
is tested using an F-test of equal variances across two independent samples, and the

Savage test for positive random variables.‘

The empirical approach used here has been presented in previous studies by
Sumner and Mueller (1989) and Fortenbery and Sumner (1993). The approach is based
on the assumption that market participants’ reservation prices are a function of their
supply and demand information. If the release of a USDA production report alters the
supply/demand perception of a sufficient number of traders, then the newly perceived

supply/demand situation should be reﬂected in a market price change (F ortenbery and

 

1The distribution of price changes are truncated on both the left and right hand sides of the distribution.
The absolute value of each price change limits the left-hand side of the distribution to zero. Also, the
Chicago Mercantile Exchange limits daily price changes at $1.50/cwt. The daily price limit truncates the
right hand side of the distribution at 1.50. The truncated distribution is therefore non-normal and all test
statistics based on the normal distribution will yield erroneous results. For this reason t-and F—statistics
will be compared to the Kruskal-Wallis and Savage non-parametric tests. The Kruskal-Wallis one-way
analysis of variance by ranks test is equivalent to the Mann-Whitney U test when k=2, where k is the
number of independent samples. If the Kruskal-Wallis test is signiﬁcant, it indicate: there is a signiﬁcant
difference between at least two of the sample medians in the set of k medians. As a result, the researcher
can conclude there is a high likelihood that at least two of the samples represent populations with different
median values (Sheskin, pg.397). The Savage test is designed speciﬁcally for positive variables. Under

the null hypothesis, two distributions, f (x) and It - f (M) are identical; 1 = 1 . Under the
alternative hypothesis, the ﬁrst distribution has a greater variability and greater mean than the second;
A. > 1 . The modiﬁed savage statistic follows the standard normal distribution.

Sumner, 1993). The average market price change the day following the report will likely
be zero. However, the average of the absolute value of price changes will be greater than
zero. In addition, the magnitude of the change in traders’ perceptions will be reﬂected in

the magnitude of the price change (Fortenbery and Sumner, 1993).

The empirical analysis uses the movements of closing prices for near-term, ﬁrst-
deferred, and second-deferred live cattle ﬁrtures contract horizons traded at the Chicago
Mercantile Exchange. Daily data for trading days ranging from January 1980 through
May 1998 were used. Release days are deﬁned as the day following the COF report
release since the report is released at the end of the trading day. Therefore, the relevant
measured price change for an announcement on January 20 is the difference between the
price at which the market closed on the 21" and the 20‘“. In addition, the sample includes
price changes for non-release days, deﬁned here as four days prior and four days after the
price change on the announcement day. Since hypothesis 1 tests the equality of variances

across two independent samples a price change in levels variable is used;

(1) AP = (RH. - P...)

r+l.t

where P represents futures price and i denotes the day of month t. Since hypotheses 2
and 3 test the equality of variance and mean, an absolute value of the price change is used

and it is expressed in relative terms (proportional change). This variable is deﬁned as:

APHI!
(2) —' = ABS[(P,+L, — P,

L!

 

where ABS refers to the absolute value.

2.3 Evidence of An Announcement Effect

This section provides separate descriptive statistics for release and non-release day
price changes by contract used to test hypotheses 1, 2, and 3, as well as, results from t-
tests, F-tests, and non-parametric tests discussed in section 2.2. The results suggest I)
nominal price changes in levels over the period January 1980 through December 1989
have greater variance than the same price changes over the period January 1990 through
May 1998, and 2) over the period January 1990 through May 1998, the absolute value of
release day relative price changes have greater mean value and variance than non-release
days. This provides evidence the COF report does cause above normal price ﬂuctuations
in the live cattle futures market, despite the diminishing number of limit moves in the

1990's.

2. 3. I Hypothesis I

Separate descriptive statistics for nominal live cattle futures price changes in levels
one day following the COF report are given for the period January 1980 through
December 1989 in table (1), and the period January 1990 through may 1998 in table (2).
Results ﬁ'om the test of hypothesis 1 appear in table (3). Results indicate live cattle
futures price changes displayed more volatility in response to the COF report during the

1980's than in the 1990's.

 

Table 1. Descriptive statistics for live cattle nominal daily release day futures
price changes in levels, by month; 1980:01-1989:12

 

 

Statistics for A I; 1.: Near-Term First-Deferred Second-Deferred
Nobs_ 120 120 120
Mean -0.048000 -0.010417 0.034333
SE, of Mean 0.778 0.0782 0.07242
Median -0.010000 0.0600000 0.100000
Maximum 1.50 1.50 1.50
Minimum -1.50 -1.50 -1.50
Std, Dev, 0.852326 0.857011 0.793467
Skewness 0.074382 -0.03 983 -0. 135402
Kurtosis 2.295284 2.359175 2.623996
Jarque-Bera 2.593775 2.085015 1.073570
Probability 0.273381 0.352569 0.584625

 

 

Table 2. Descriptive statistics for live cattle nominal daily release day futures
price changes in levels, by month;l990:01-1990:05

 

 

Statistics for A B+ 1.: Near-Term First-Deferred Second-Deferred
Nobs. 101 101 101
Mean «0.074851 -0.031 188 -0.049505
SE, of Mean 0.06276 0.05341 0.04159
Median -0.080000 0.030000 -0.200000
Maximum 1.37 1.20 0.98
Minimum -1.50 -1.40 -1.15
Std. Dev, 0.630886 0.536849 0.418018
Skewness -0. 160547 -0.327507 -0.210317
Kurtosis 2.837339 2.956755 2.802243
Jarque-Bera 0.545234 1.813433 0.909171
0.761384 0.403848 0.634711

Probability

 

 

Table 3. Means, variances and F- statistics for hypothesis 1

 

Stats for % Change Near-Term First-Deferred Second-Deferred

 

1980:01 - 1989212

 

Variance 0.720406 0.728347 0.624343
1990:01 - 1998:05

Variance 0.394076 0.285353 0.173009
F-stat 1.83” 2.55" 3.61”

DF (119,100) (119,100) (119,100)

 

Notes: Double asterisk denotes signiﬁcance at the 5 percent level. The F -test of equal variances is given
as, F = max[s,’ ,s§]/min[s,2 ,s,’] , where. respectively, 3,2 and 322 are the variances of the two samples.

The F-test tests the null hypothesis of equal variances. vaariances of the two groups are not statistically
different, then a t-test using a pooled variance will be used instead of the t-test given above. Degrees of

frwdorn for the F-test are ﬁgured as ml for both numerator and denominator.

2. 3.2 Hypotheses 2 and 3

Descriptive statistics for the absolute value of live cattle relative release day and
non-release day price changes for the period January 1990 though May 1998 are given in
tables (4) and (5) respectively. Results from the test of hypotheses 2 and 3 appear in

table (6).

Both the absolute value of release and non-release day relative price changes for
each contract are non-nonnal, as was expected due to truncation on both sides of each
distribution. Simple inspection of the descriptive statistics indicates release day price
changes for near-term, and ﬁrst- and second-deferred contracts each have higher mean and
variance values than non-release days. The t-tests (F-tests) testing the null hypothesis of
equal means (variances) across release and non-release days reject the null in every case.

Results from these hypotheses tests indicate the absolute value of release day relative price

10

changes have greater mean and variability than non-release days for each contract horizon.
Results from the Kruskal-Wallis test are consistent with results from the t-tests of equal
means. It is possible to reject the null hypothesis of equal median values for release and
non-release day absolute relative price changes at the p-value of 0.002 level or lower for
each contract horizon. As well, results from the Savage test are consistent with F-tests of
equal variances; it is possible to reject the null hypothesis of equal variance and mean
value across the distributions at the p-value of 0.001 level. Based on this evidence, we
can conclude release day and non-release day price changes in the 1990's still belong to

different distributions.

1]

 

Table 4. Descriptive statistics for the absolute value of live cattle relative non-
release day futures price changes in levels, by month;l990:01-l990:05

 

 

 

 

. Near-Term First-Deferred Second-Deferred
Statistics for ”"’

i.t
Nobs, 707 707 707
Mean 0.005357 0.004461 0.003527
SE, of Mean 1.79E-4 1.52E-4 1.23E-4
Median 0.003938 0.003346 0.002825
Maximum 0.026643 0.002535 0.002439
Minimum 0 0 0
Std, Dev, 0.004759 0.004035 0.003278
Skewness 1.38346 1.65288 2.05264
Kurtosis 5.07824 6.572695 10.2567
Jarque-Bera 352.763 694.1049 2047.76
Probability 0.000001 0.000001 0.000001

 

 

Table 5. Descriptive statistics for the absolute value of live cattle absolute relative
release day futures price changes in levels, by month; l990:01-l990:05

 

_ Near-Term First-Deferred Second-Deferred
Statistics for ”l"
Nobs. 101 101 101
Mean 0.007193 0.006152 0.004723
S,E_ ofMean 5.62E-4 4.6lE-4 3.76E-4
Median 0.005380 0.005215 0.003958
Maximum 0.024867 0.002021 0.001655
Minimum 0 0 0
Std, Dev, 0.005658 0.004635 0.003786
Skewness 1.07805 1.13548 0.898610
Kurtosis 3.58183 3.87806 3.22595
Jarque-Bera 20.9883 24.9484 13.8077
Probability 0.000001 0.000001 0.000001

 

 

 

 

12

 

Table 6. Means, variances, and statistical tests for hypotheses 2 and 3

 

 

 

Stats for % Change Near-Term First-Deferred Second-Deferred
Non-Report Days

Mean 0.005357 0.004461 0.003527

Variance 2.3E-05 1.6E-05 1.1E-05
Report Days

Mean 0.007193 0.006152 0.004723

Variance 3.2E-05 2. lE-OS 1.4E-05
t—Stat 3.12" 3.50" 3.03Ml

DF 122 122 122
F-Stat 1.40" 1307* 1.322“

DF (100,706) (100,706) (100,706)
Kruskal-Wallis Test

lz-Stat 11.37 15.822 9.435

prob, 0.0007 0.0001 0.0021
Savage Test

t-stat 3.47 3.71 3.89

prob. 0.001 0.001 0.001

 

Notes: Double asterisk denotes signiﬁcance at the 5 percent level, b Asterisk denotes signiﬁcance at the 10

percent level. The t-test of equal means, degrees of freedom for the t-test, and F-test of equal variances

(Sf/"1+ 522/”2)2
(Si/"IY/("I " 0+ (“"22 /"2)2/(”2 " l) ,

F = maxls,’,s,’]/min[s,’,s§] , respectively, where, respectively, )7, and f, are the means of release and
non-release day samples; nI and n2 are the number of observations from each sample;

 

maven... (a- We". + 3.2/t, m:

and $12 and s: are the variances of the two samples. The F -test tests the null hypothesis of equal

variances. vaariances of the two groups are not statistically different, then a t-test using a pooled
variance will be used instead of the t-test given above. Degrees of freedom for the F-test are ﬁgured as n-l
for both numerator and denominator.

13

2.4 Conclusions

The evidence suggests that release day nominal near-term, and ﬁrst- and second
deferred live cattle ﬁrtures contract price changes occurring one day following the COF
report, do exhibit greater volatility in the 1980's than in the 1990's. However, the absolute
value of release day relative live ﬁitures price changes during the 1990's do still exhibit
greater mean value and variance than the absolute value of non-release day price relative
price changes in each contract horizon. We conclude from these results the USDA Cattle
on Feed report is fulﬁlling its public policy objective which is to provide new information

to market participants.

14

ARE PRELIMINARY ESTIMATES IN THE USDA CATTLE 0N FEED REPORT
RATIONAL FORECASTS OF USDA FINAL REVISED ESTINIATES?

CHAPTER 3

3.1 Introduction

Economic policy makers, market participants, and econometricians must often
make decisions or inferences based on preliminary or incomplete data. Each agent faces a
problem of judging the informational content of preliminary announcements based on all

available information (Mankiw, Runkle, Shapiro, 1984).

In this paper, preliminary estimates of the USDA Cattle on Feed (COF) report are
tested for rationality as predictors of ﬁnal revised estimates relative to public and private
information. While there has been much discussion on the impact of private information
on the formation of expectations, little empirical evidence has been presented which
indicates whether USDA preliminary estimates optimally reﬂect private information. The
rationality test considered here is particularly interesting because we use USDA

preliminary estimates as expectations of USDA ﬁnal revised ﬁgures.

Given the widespread use of USDA data in tests of announcement effects and tests
of market efficiency, an important empirical issue is the extent to which USDA estimates
of economic variables are consistent with the rational expectations hypothesis. Tests of

the rational expectations hypothesis are usually classiﬁed as weak form or strong form

15

tests (Lovell, 1986)? Weak form tests examine whether expectations of an economic
variable are unbiased predictors of the subsequent realized values of the variable. Weak
form tests also examine whether expectations of economic variables are eﬂicient, in the
sense past values of the economic variable cannot be used to explain the error between
expected and the realized value (i.e. whether the forecast errors are auto-correlated).
Strong form tests of rational expectations examine efﬁciency by testing whether the
forecast errors can be explained by other theoretically relevant economic variables. If the
expectations are strong form rational, all relevant variables are orthogonal to the forecast

error (Colling, Irwin, and Zulauf, 1992).

Therefore, if expectations of an economic variable are rational they must satisfy
both unbiasedness and efficiency conditions. Rational expectations therefore are optimal,
in the sense the expectations’ conditional mean is equal to the true mean of the economic
variable, and that they have minimum variance in the class of all such linear unbiased

expectations.

Our results show that preliminary estimates of cattle on-feed, placements, and
marketings are not strong form rational relative to public and private information. This
study is presented in the following way. Section 3 .2 discusses the literature. Section 3.3

presents the theoretical framework. Section 3.4 discusses the implementation of the

 

2Weak and strong form tests of rationality should not be equated with Fama’s deﬁnition of weak,
semi-strong, and strong form market efficiency. While they are related, they are distinct topics.

16

framework for data on cattle on-feed. Section 3.5 provides empirical results. Section 3.6
discusses the implications of our ﬁndings for other research. Section 3.7 provides

concluding remarks.

3 .2 The Literature

In the current literature there has developed a number of terms each referring to
expectations made by private individuals, private ﬁrms, or the US Department of
Agriculture. While each can denote an expectation of the same commodity, each are
distinct in that they represent sets of different information, usually at different points in
time. It is critical that a precise deﬁnition of each is clearly understood now to insure as
little confusion as possible. A clear deﬁnition of terms as they have been presented in the

literature, and in this study, is presented at the start of this thesis.

Preceding, Mankiw et. al develop a theoretical framework for analyzing
preliminary estimates of economic data and then apply this framework to the money
stock. The authors ﬁnd Federal Reserve preliminary estimates of the money stock are
best characterized as measured by classic errors-in-variables, implying the revision is not
correlated with the true value (ﬁnal revised estimate) of the series. However, the authors
conclude the revision is forecastable using available public information. This implies the
preliminary estimates are not strong form rational estimates of ﬁnal revised estimates,

even though they are unbiased.

l7

Runkle (1991) tests whether USDA Hogs and Pigs report preliminary estimates
of sows farrowed are rational forecasts of ﬁnal revised sows farrowed. Runkle concludes
that preliminary estimates are unbiased and efficient, and therefore are rational forecasts
of ﬁnal revised ﬁgures. It is not speciﬁed whether he performs a weak form or strong
form test of rationality. In one other study, Baur and Orazem (1994) test whether USDA
reported producers’ intentions of orange production are strong form rational forecasts of
ﬁnal revised orange production’. They conclude USDA reported producer intentions of
orange production, for each month they are collected during the growing season, are
unbiased and efﬁcient relative to speciﬁc public information. Thus, producer intentions of

orange production are strong form rational forecasts of USDA preliminary estimates.

This study will extend the current literature in two respects. First, we test whether
USDA Cattle on Feed report preliminary estimates of cattle on-feed, placements and
marketings are unbiased and efficient forecasts of USDA ﬁnal revised ﬁgures. Secondly,
we conduct direct tests of efficiency relative to public and private information. Our results
indicate USDA preliminary estimates of cattle on-feed, placements and marketings are not
strong form rational. Finally, we discuss the implications of using these preliminary
estimates to conduct tests of the efficient market hypothesis and tests of announcement

effects.

 

3In this case, the USDA ‘ﬁnal’ estimate of orange production can be considered both a preliminary
estimate and ﬁnal estimate. The USDA releases only one estimate of orange production at the end of each
production season and this ﬁgure is not revised. In this sense, it is both preliminary estimate and ﬁnal
estimate.

18

3.3 Theoretical Framework

USDA preliminary estimates of cattle on-feed, placements and marketings reported
in the COF report are subject to revision at both regular and irregular intervals. In
January, all monthly and quarterly estimates are reviewed and subject to revisions. The
reviews are primarily based on slaughter data, state check-off or brand data, and any other
data that may have been received after the original estimate was made. A random sample
of US producers is surveyed to provide data for these estimates. The COF report
estimates are prepared by the Agricultural Statistics Board after reviewing
recommendations and analysis submitted by each state office. National and state survey
data are reviewed for reasonableness with each other and with estimates from the previous
month and/or quarter when setting the current estimates. After a period of what has
historically been 5 years, the USDA releases ‘ﬁnal’ estimates for a number of past years

which are not revised again.

The preliminary estimates reported by the USDA are estimations, not simple
weighted averages of raw data provided by private participants. Therefore, tests of
rationality are appropriate. Were the preliminary estimates weighted averages, the
appropriate hypothesis test would be a classic errors-in-variables test as demonstrated by

Mankiw, Runkle, and Shapiro (1984).

19

 

0

Let F' P

i+n,t ’ i+n.! ’

and 114;“ denote ﬁnal revised ﬁgures of USDA reported cattle

on-feed, placements, and marketings for month t, which are released on day i +n. Let

E1, 13‘: , and M“, denote the preliminary announced estimates for the same month,

which are released on day i. Note that ﬁnal revised ﬁgures are available at some time
later than the preliminary estimates. The important question for policy makers, market
agents, and econometricians is whether preliminary estimates are rational, in the sense that
they are unbiased and efficient forecasts of ﬁnal revised estimates conditional on available
information. If the preliminary estimate is rational, it is equal to the mathematical

expectation of the ﬁnal revision conditioned on the information which was known at the

 

time it was made: that is, in the case of on-feed, E“: = E (F ‘25”) where t5” denotes

i+n.!

all available information on day i concerning month t, and E is the mathematical

expectations operator.

3. 3. 1 Hypothesis 1 .' Unbiasedness
The ﬁrst implication of rationality is that forecasts must be unbiased. To determine
if the preliminary estimates of cattle on-feed are unbiased the following model is

estimated. The same method is used for evaluating placements and marketings.

(3) F‘ = a. + all??? + u,

i+n.!

20

 

The unbiasedness property requires the coefficients in equation (3) may be

restricted to do = 0, and a1 = l, and that u, satisfy the following condition,
E(u,|¢,,,) = o.

3. 3.2 Hypothesis 2: Eﬂiciency

Efficiency imposes the additional requirement that any variable known at time (i, t)

or before be orthogonal to u, , that is ,6 = 0 for any vector Z” 6 1.1' The preliminary

estimates are rational forecasts of the ﬁnal revised ﬁgures only if both unbiasedness and
efﬁciency conditions are met. The efficiency hypothesis is tested by applying least squares

to the following equation:

(4) F — F“ = a0 +(al—1)Ej+ ﬂ'Zu + u,

for months t = 1,2,.... T, which is constructed by simply subtracting E‘: from both sides of
equation (4) and including variables in the set Z” 6 2.1: Efficiency implies

that (a, - 1) = ﬂ = 0, and that u, satisfy the following condition, E(u,|¢,’,) = 0. This

implies the revision, F t -

m I E‘j, is uncorrelated with available information, Z” 6 H

21

3. 3. 3 Constructing an optimal forecast of ﬁnal re vised cattle on-feed

Deﬁne the optimal linear (minimum variance) forecast of F. . conditional on

r+n,t

information {17,3 , Z” } to be:

(5) 17,5 =13, + ,5; + [392.

The values of 50 ,5, , and ,6 are found by applying least squares to equation (5),

I l . I
over a penod of Wthh If," , rs known.

3.4 Data
3.4.1 Preliminary and ﬁnal revised estimates

The preliminary estimates used in this study include estimates of cattle on-feed,
placements, and marketings reported in the monthly USDA COF report from January
1986 through December 1994. The seven state C 0F report is released each month by the
Agricultural Statistics Board of the US. Department of Agriculture (USDA). Each report
contains preliminary estimates of the number of cattle on feed the ﬁrst day of each month,
the number placed on feed during the previous month, and the number marketed during
the previous month. The ﬁnal revised estimates are published in the ERS Statistical

Bulletin, Cattle: Final Estimates, approximately every 5 years.

22

3. 4.2 Public data

Pre-release survey forecasts of cattle on-feed, placements, and marketings
generated by Knight-Ridder Financial News are used to approximate what the entire
market expects the COF report to announce. Knight-Ridder obtains forecasts of each
component by surveying approximately 20 industry experts for their expectations before
the release of each report, then computes a simple average of the responses. This is what
Knight-Ridder denotes as the industry survey forecast for each component. The
individuals’ forecasts and the industry survey forecasts are made available by electronic

mail approximately two days prior to the release of the C OF report.

Other public information used in this study includes net returns to feedlots which is
a proﬁtability measure, and lagged preliminary estimates of cattle on-feed, placements, and
marketings. Net returns to feedlots is calculated as the difference between the selling price
of fed cattle and returns to variable costs of high plains feedlots. Net returns data was
collected from the USDA high plains cattle feeding simulator reported in the Livestock,

Dairy, and Poultry report.

3. 4.3 Private data

One of the innovations in this paper is the inclusion of private information in tests
of efficiency. The private data used in this study include raw monthly private feedlot
survey data for cattle on-feed, placements, and marketings, as well as feedlot capacity.

The data are provided by Professional Cattle Consultants (PCC). PCC collects the data

23

from individual feedlots approximately one week prior to the release of the USDA Cattle
on Feed report. PCC compiles the responses and returns the aggregated results back to
each feedlot in a newsletter format. Over the period of this study, the PCC survey
constitutes between 1.3 and 2.2 million head of cattle. This represents approximately 20%
of all cattle on—feed, placed, and marketed from the seven largest cattle feeding states in
the United States each month. In this study the raw PCC data for each component is
divided by feedlot capacity to eliminate variation due to feedlots entering and exiting the

survey over the sample period.

3.5 Results
3. 5. 1 Test of unbiasedness of USDA preliminary estimates

Each variable was ﬁrst tested for a unit root, and no evidence of a unit root was
found in any of the data series so they are assumed stationary. Results of the unbiasedness
tests are shown in table (7). With respect to preliminary estimates of cattle on feed, F-
tests reject the null hypothesis of unbiasedness at essentially any signiﬁcance level. With
respect to preliminary estimates of placements and marketings, F-tests fail to reject the
null hypothesis of unbiasedness at the 5% level, but not at the 10% level, in each case.
The conclusion is that USDA preliminary estimates of cattle on-feed, placements, and
marketings are not unbiased estimates of the ﬁnal revised ﬁgures reported at some later

date.

24

 

Table 7. Test of unbiasedness: USDA preliminary estimates

 

 

a0 a1 [72 Q(20)‘ H,:a, = 0.0:, = 1

On-feed 164.9 0.984 0.98 206.59’” F(2,106) Prob > F

(85.64)b (0.01) 13.39 0.00001

Placements 11.61 0.996 0.99 36.44” F(2,106) Prob > F
(12.23) (0.007) 2.55 0.08

Marketings 79.02 0.95 0.94 39.66'Ml F(2,106) PfOb > F
(37.75) (0.02) 2.92 0.06

 

‘Q(20) denotes the Ljung-Box Q-statistic computed at lag twenty which tests the null hypothesis that auto-
correlation is zero with degrees of freedom equal to the number of lags, b standard errors in parentheses,“
denotes signiﬁcance at the 0.01 level, * denotes signiﬁcance at the 0.05 level.

25

 

Table 8. Definitions of variables

 

FILM USDA ﬁnal revised cattle on-feed
13:“ USDA ﬁnal revised placements
M11“ USDA ﬁnal revised marketings
1‘7: USDA preliminary estimates of cattle on-feed
g: USDA preliminary estimates of placements
1w]: USDA preliminary estimates of marketings
1415’“ private PCC cattle on-feed
19.37,: private PCC placements
M1]: 1.: private PCC marketings
15;sz Knight-Ridder survey on-feed forecast
13:” Knight-Ridder survey placements forecast
44:2,: Knight-Ridder survey marketings forecast
NR” net returns to feedlots (variable cost minus the selling price of fed cattle)
E5 optimal linear forecast of actual revised cattle on-feed at time it
135 optimal linear forecast of actual revised placements at time it
M1 optimal linear forecast of actual revised marketings at time it

 

26

3. 5.2 Test of efﬁciency
To formally test whether preliminary estimates of cattle on-feed are efﬁcient

predictors of ﬁnal revised cattle on-feed ﬁgures the following model is estimated:

t

_F"

(6) F 1,1 = b0 + (b1 - 1)}7: + IB'Zu + “1

1+ r1,t

F" F" Fe P“ P" 13:2,,M.“ M,‘j,_,,,M.‘

1.1-1’ r,r—12’ r—2.r’ 1.1—1’ 1.1-12’ 1.1-h r-2,t’
whereZ” = and

P P P P P P
171-74, Fir-12 ’ P14.” P1—7.r-12 2 M14,“ Mr—7,r-12

all variables are deﬁned in table (8). The parameter estimates are as follows:

0

Em, — E‘: = — 162.13- 0.1 1 IFS + 0.055 F34 + 0.005 E142 + 0.06 13:2,
’ ’ (327.12) (0.064) ’ (0.123) ' (0.055) ' (0.123) ’
'1' 90820591114 + 993917.142 — 8833-1: + 8:995) Mir—1
a _ e P _ P
+9.32; Mm raw», + as (3.23%
(7) P P p _ p
‘1' 7359519? 8-7,: + (39633 8-11-12 '1’ %Z££? Mr-m 4&69035? M-zr—rz
+ 0.84 u,_l

(0.101)

D.W.= 2.49, S. E. R.= 69.88, R2 = 0.70, F = 13.60

An F-test of the joint hypothesis that each slope coefficient and lagged residual

coefficient equals zero has a p-value of 0.00001. Therefore, we reject the null hypothesis

27

that USDA preliminary cattle on-feed estimates optimally use all available public and

private information at essentially any signiﬁcance level.

To formally test whether the preliminary placements estimates are efficient

predictors of ﬁnal revised ﬁgures the following model is estimated:

0

(8) P P“ = b. +(br -1)P.‘I + W... + 6,

i+n.! 1.1

where Z” = {P0 P“ 13:2,;sR—p7,r’1)rf7,t_12,NR,-J_n}. Parameter estimates are

i,t-I ’ i,t-12 ’

as follows:
PM - Pu = - 1.7.7,- taaéa + 9.9811” + 9.9.2.912- + W-..
P _ p
(9) + zi167‘3688)5 R77" 3(9?4§)O I)I'--7,t—12 + (104612) NIQU_12

D.W.= 1.91, S.E.R.= 27.25, 752 = 0.11, F = 2.76

An F-test of the joint hypothesis that each slope coefficient and lagged residual
coefﬁcient equals zero has a p-value of 0.012. Therefore, we reject the null hypothesis
that USDA preliminary placements estimates optimally use all available public and private

information at the 5% level, but not at the 1% level.

28

Likewise, the following model is estimated for marketings:

(10) M' - M? = b0 +(b.-1)M,»‘f. + P7... + 6.

r+n.t

_ a a e p p
where 2,, - M,,,_,,M,.,,_,,,M,_2,,,M,._ 7,,M,_ 7,_ ”,2: aUP ,,_ _,. . Parameter

estimates of the sum of lagged preliminary placements estimates are estimated as a second
order polynomial distributed lag constrained to zero at the far end. Placements will be
marketings in four to eight months and are assumed to follow a quadratic function.

Parameter estimates are as follows:

My,“ — M, = 36.-06 0234M“ + 0.072 M;;_, —-0019M;;',_ 2+ 011 M,_,,
. "(73.93)(0.)11”' W007
+ 656. 66 Mi - 7214114,P,,_12 -0.0007PDL1,,_,
(365 3.0 7" 2(59 75) ( 0.005) ’
(11) +00..03PDL2,4+024{,_,
(.)00023 (.)0126

D.W.= 2.01, S.E.R.= 26.78, R2 = 0.24, F = 3.88

where PDLI and PDL2 denote the polynomial distributed lag terms. An F-test of the

joint hypothesis that each slope coefficient and lagged residual coefficient equals zero has

a p-value of 0.0005. Therefore, we reject the null hypothesis that USDA preliminary

29

markings estimates optimally use all available public and private information at essentially

any signiﬁcance level.

From these tests we conclude USDA preliminary estimates of cattle on—feed,
placements and marketings are not strong form rational forecasts of ﬁnal revised
estimates. Strikingly, the USDA, on average, does not even account for serial correlation
from past errors when making preliminary estimates. Since the errors are made in a
predictable manner, the preliminary estimates are not optimal forecasts. In section 3.3.3,
we provide the models whereby optimal forecasts of cattle on-feed, placements and

marketings can be computed.

3. 5. 3 Forecast evaluation

In this section we generate optimal forecasts of ﬁnal revised estimates conditional
on available public and private information available at the time the COF report is
released, and at the time the Knight-Ridder survey forecasts are released. Finally, we
evaluate the RMSE of the optimal forecasts relative to preliminary estimates and survey
forecasts. Due to the unavailability of ﬁnal revised estimates of cattle on-feed, parameter
estimates of model (5) in section 3.3.3 can only be re-estimated approximately every 3-5

years; this also applies to placements and marketings over the past ﬁve years.

30

 

Table 9. Release Dates and Contents of USDA Statistical Bulletin, Cattle: Final

 

Estimates
Available On-feed, Placements, and Release Date
Marketings Final Revisions
1980:01-1983:12‘ 1985201
1984201 - 1988212” 1990201
1989101 - 1993212c 1995201
1994:01 - 1998:12" 1998:01

 

'ERS Statistical Bulletin # 720, bERS Statistical Bulletin # 798, cERS Statistical Bulletin # 905, ‘ ERS
Statistical Bulletin # 953.

Approximately every ﬁve years the Economic Research Service (ERS) issues a statistical
bulletin entitled, Cattle: Final Estimates, which provides ﬁnal revised estimates for
monthly 7-state cattle on-feed, placements, and marketings. Table (9) shows when ﬁnal

revised data have been made available since 1985.

We estimate parameter estimates for model (5) using data over the sample period
1986201 through 1988:12. Since the ﬁnal revised data were made available on January
1990, and not again until January 1995, we then use the estimated model (5) to make
cattle on-feed forecasts for the period 1990:02 through 1994:12. The same method is
applied to placements and marketings. This insures that forecasts only use information
available at the time the forecasts are being made. The forecasting models for cattle on-

feed, placements, and marketings were speciﬁed as follows:

31

F1:n,t=b0+bFa+bFi7l+bF712+bF821+b17tl+b17-12
+b7sz 13+ng +b9Miat— 12+b10M:2.r-1+b11FP171

(12) 1.1-1
+b12Prf 71 + br3Mrp7r + u,

i+n.! _ +b113if1 + b 2P.t- 1+ b 387—12 + b41752J

(13)
+b5P1-73 + b6NRr'.t—12 + 51

M

i+n. t

=b0 +ble’,2+bMf,_l+b3,“M,_41,+bM;2,r
+b MMf,,+bPDL1,,_7,,4+bPDL2u—4’Pé

(14)
where all right hand side variables have been deﬁned previously in section 3.5.2, and

u, , 6', , and 5, have zero mean conditional on (15,, , . Out-of-sample forecasts of ﬁnal

revised cattle on-feed, placements and marketings were generated for the period 1990:02
through 1994: 12, and evaluated using root mean square error evaluation. The ability of
out-of-sample forecasts to predict ﬁnal revised numbers were compared relative to

preliminary estimates. Table (10) presents the results.

32

 

CC

f0

1111

CS

 

£17

 

pr

Uf

 

 

Table 10. Forecast evaluation: RMSE comparison

 

 

Preliminary Estimates Econometric Forecasts
On-feed 89.92 311.83
Placements 33 .80 37.15
Marketings 34.1 1 37.41

 

Results indicate none of the econometric forecasts beat preliminary estimates. The
relatively poor cattle on-feed forecasts can be explained by the fact that model (12)
exhibited signiﬁcant ﬁrst order auto-correlation in the errors. However, this cannot be
corrected out-of-sample because the econometrician does not have access to ﬁnal revised
data at the time the forecasts are made. This results in biased parameter estimates and
thus biased out-of-sample forecasts. Models (13) and (14) did not exhibit auto-
correlation. Thus, the difference in RMSE between preliminary estimates and econometric
forecasts of placements and marketing may be attributed to sampling error or the fact that

they are not optimal forecasts. This is further analyzed in chapter 5 using futures prices.

3.5.4 Summary of section 3.5

Based on tests of unbiasedness and efﬁciency we conclude the USDA preliminary
estimates of cattle on-feed, placements and marketings may not be strong form rational.
However, the RMSE error analysis contradicts this evidence. Here, we assume
preliminary estimates of cattle on-feed are strong-form rational, despite the fact they failed
unbiasedness and efﬁciency tests, because of the RMSE evaluation. It appears that the

bias and inefﬁciency of preliminary estimates of cattle on-feed estimates can be attributed

33

to auto-correlation in USDA ﬁnal revisions (i.e. ﬁnal revisions are adjusted up or down
together). However, since ﬁnal revisions are released at irregular intervals, the information
needed to account for the auto-correlation when forecasting is not generally available
when the forecasts have to be made. With respect to preliminary estimates of placements
and marketings, results from tests of unbiasedness, efficiency suggest these estimates are
not strong form rational, but RMSE evaluation does not. Based on these results, we are

indecisive about their rationality conditions.

3.6 Implications
In this section, we brieﬂy discuss alternative methods to conduct tests of
rationality, announcement effects and market efficiency when USDA preliminary estimates

are biased and inefficient estimates of ﬁnal revised numbers.

3. 6. 1 Test of the rational expectations hypothesis

An econometrician wishing to test the rationality of pre-release survey forecasts of
cattle on-feed, such as those provided by Knight-Ridder, must decide whether to use ﬁnal
revised or preliminary estimates data to evaluate the rationality of the survey forecasts.

Since we have shown that preliminary estimates of placements and marketings may not

t

 

 

i+n.!

satisfy the following conditions, R: at E (P t5”) and M f, i E (Miim, t5”), which

estimate should be used? The objective of tests of unbiasedness and efﬁciency is to

determine whether agents optimally use all available information to forecast the true value

34

of the variable. However, it is not clear in this case whether preliminary estimates are
themselves optimal. Final revised data are the best ex-post estimate of the actual numbers
which market participants are most interested in. Therefore, the ﬁnal revised data should
be used. The correct speciﬁcation of the model for testing whether Knight-Ridder pre-

release survey forecasts of placements are optimal forecasts is :

t

(15) P — 12:... = b. + (b. - 01352., + .42.... + a.

1+ "J

Using R: — Biz, as a dependent variable lead to inefficient parameter estimates,

causing t- and F-tests to be less reliable. This can result in falsely accepting or rejecting

the null hypothesis. The same logic holds true for testing marketings.

3. 6.2 Tests of announcement effects

To test whether the C OF report contains new information for the formation of live
cattle futures prices, a vector of price changes one day following the reports release can be
regressed on a vector of unanticipated information contained in the report. If the report
contains new information, the coefficients of unanticipated information should be
signiﬁcantly different from zero. Previous studies have deﬁned unanticipated information
has the difference between preliminary estimates and survey forecasts (see Colling and
Irwin (1990), Grunewald, McNulty and Biere (1993), and Colling, Irwin and Zulauf

(1996,1997) for examples). However, if preliminary estimates are non-optimal forecasts

35

of ﬁnal revised ﬁgures, this approach will lead to a non-optimal forecast of unanticipated
information. Since agents may be able to generate optimal linear forecasts of ﬁnal revised
estimates of placements ans marketings, but not cattle on-feed, the following model can be
estimated to test for announcement effects and compared to results from the existing

model presented by Pearce and Roley (1985):

p7... - p... = b. + b.(F.-,‘i - P2577) + 127(1’75 — P727.)

(16)
+b3(Mi.ft - Mic-2.1)+ (0,

where p, +1., - p” is the change in close-to-close live cattle futures prices one day

following the COF report. This model assumes agents are completely rational, in the
sense they understand preliminary estimates of cattle on-feed cannot be improved due to
auto-correlation in the USDA revisions, but that optimal linear forecasts of ﬁnal revised

placements and marketings can be generated.

3. 6. 3 Tests of market efficiency

Previous studies have turned a model such as (16) into a test of the efﬁcient
market hypothesis by including the survey forecasts as explanatory variables (see Pearce
and Roley (1985) and Colling and Irwin, (1990)). The efficient market hypothesis asserts
that information available before the release of a government report should already be

reﬂected by prices, and thus not useful for the prediction of the price change following the

36

report (Fama, 1970). Again, since agents may be able to generate the optimal linear
forecast of ﬁnal revised placements and marketings, the following model can be estimated
to test whether the market’s response to new information released in the COF report is

efficient:

Pr+r,r ‘ p7,: : b0 + b10713 " Friar) + b2(Pr.{ _ Bill)

(17)
+b3(M{, — M1:2.r)+ b45523 + 17511732.: + b6 Mic-2,1 + for

Market efficiency implies b4 = b5 = b6 = 0. This implies that price changes occurring

one day following the release of the USDA C 0F report cannot be signiﬁcantly explained
by information summarized by the Knight-Ridder survey forecasts. If biased and
inefﬁcient estimates of unanticipated information are used to estimate model (17),
parameter estimates will be inefficient, rendering t- and F-tests less reliable. This is

analyzed in chapter 5.

3.7 Conclusion

We ﬁnd that USDA preliminary estimates of cattle on feed, placements, and
marketings may not be strong form rational forecasts of USDA ﬁnal revisions, conditional
on available public and private information. Each fail the unbiasedness and efficiency
tests, however, out-of-sample forecasts of cattle on-feed, placements and marketings
cannot be generated which can out-forecast preliminary estimates. Here, we assume
preliminary estimates of cattle on-feed are strong form rational because it appears that

37

 

 

 

1h

11

 

 

their bias and inefficiency can be attributed to auto-correlation in USDA ﬁnal revisions
which is not generally available when the out-of-sample forecasts have to be made.
However, USDA placements and marketings revisions are not auto-correlated. Therefore,
we are cannot know whether the difference between the RMSE of preliminary estimates

and out-of-sample forecasts of ﬁnal revised numbers are statistically different.

These ﬁndings do have an inﬂuence on empirical tests of the rational expectations
hypothesis. The conclusion is ﬁnal revised estimates should be used to test expectations
for unbiasedness and efficiency when preliminary estimates do not optimally reﬂect
available information, because the objective of each test is to determine if agents optimally
forecast the true value of the variable, not the non-optimal preliminary estimate (i.e it is
presumable the actual (ﬁnal) number that rational agents are motivated to be interested in).
These ﬁnding also have an impact on event studies which test for announcement effects
and market efﬁciency. Optimal linear forecasts of ﬁnal revised estimates can be used to
approximate the true value of the economic variable in the presence of non-optimal
preliminary estimates. This will improve the efficiency of such models and make t- and F-

tests more reliable.

38

ARE KNIGHT-RIDDER PRE-RELEASE SURVEY FORECASTS OF THE USDA
CA TILE 0N FEED REPORT STRONG FORM RATIONAL FORECASTS OF
USDA FINAL REVISED ESTIMATES?

CHAPTER 4

4.1 Introduction

Market participants and econometricians must often make decisions or inferences
based on survey forecasts of USDA reports available prior to their release. The extent to
which survey forecasts optimally use available information should be a key criterion in this
decision process. In this paper, Knight-Ridder survey forecasts of cattle on-feed,
placements and marketings are tested for rationality relative to public and private
information. While there has been much discussion on the impact of private information
on the formation of expectations, little empirical evidence has been presented which

indicates whether survey forecasts optimally reﬂect private information.

Given the widespread use of the rational expectations hypothesis in tests of
announcement effects and tests of market efficiency, an important empirical issue has
become the extent to which expectations of economic variables are consistent with the
rational expectations hypothesis. Tests of the rational expectations hypothesis are usually
classiﬁed as weak form or strong form tests (Lovell, 1986)‘. Weak form tests examine

whether expectations of an economic variable are unbiased predictors of the subsequent

 

4Weak-and strong-form tests of rationality should not be equated with Fama’s deﬁnition of weak:
semi-strong, and strong form market efficiency. While they are related, they are distinct topics.

39

 

realized values of the variable. Weak form tests also examine whether expectations of an
economic variable are efficient, in the sense past values of the economic variable cannot be
used to explain the error between expected and the realized value (i.e. whether the
forecast errors are auto-correlated). Strong form tests of rational expectations examine
efficiency by testing whether forecast errors can be explained by other theoretically
relevant economic variables. If the expectations are strong form rational, all relevant

variables are orthogonal to the forecast error (Colling, Irwin, and Zulauf, 1992).

Therefore, if expectations of an economic variable are rational they must satisfy
both unbiasedness and efficiency conditions. Rational expectations therefore are optimal,
in the sense the expectations’ conditional mean is equal to the true mean of the economic
variable, and that they have minimum variance in the class of all such linear unbiased

expectations.

The private data used in this study is raw survey data collected directly from
feedlots. It was obtained from Professional Cattle Consultants (PCC), a private ﬁrm
which coincidently provides Knight-Ridder with monthly forecasts of cattle on-feed,
placements and marketings which are presumably made using the data from their survey as
well as other sources of public information. These forecasts are used, along with many
other ﬁrms’ and agents’ forecasts to compute the common Knight-Ridder average
“industry” forecast used in previous studies. As well as testing the industry survey

forecasts for rationality, we test whether PCC’s forecasts, are strong form rational relative

40

to their own private data. Surprisingly, the results suggest the ﬁrm may not use their own
data optimally. With regard to the survey forecasts themselves, results indicate Knight-
Ridder survey forecasts of cattle on-feed, placements and marketings are also not strong

form rational.

But do these results reﬁrte the rational expectations hypothesis or do they indicate
that market agents do not reveal their true expectations when surveyed? It is more likely
that market agents don’t accurately respond to surveys such as Knight-Ridder’s. It is very
difficult to understand why we can employ basic linear auto-regressive models and out-
forecast most of the individuals by such a signiﬁcant margin. Based on our results, we
conclude that Knight-Ridder industry survey forecasts of cattle on-feed, placements and
marketings are not very useful for predicting ﬁnal revised USDA values. We discuss the
implications of our results for tests of announcement effects, and tests of market

efficiency.

Section 4.2 discusses the literature. Section 4.3 presents the theoretical
framework. Section 4.4 discusses the data used. Section 4.5 provides empirical results.
Section 4.6 discusses the implications of our ﬁndings for other research. Section 4.7

provides concluding remarks.

41

4.2 The Literature

Two studies which test whether survey forecasts of agricultural supply variables
are weak form rational forecasts of USDA preliminary estimates include Colling, Irwin,
and Zulauf (1996, 1997). Colling et al. (1996) test whether Knight-Ridder survey
forecasts of corn, wheat and soybean exports are weak form rational forecasts of USDA
Export Inspections report preliminary estimates. Colling et al. (1996) conclude Knight
Ridder survey forecasts of soybeans exports are unbiased predictors of USDA preliminary
estimates, and that forecast errors are uncorrelated. The conclusion is the Knight-Ridder
survey forecasts of soybeans exports are weak form rational forecasts of USDA
preliminary estimates. However, Knight-Ridder survey forecasts for corn and wheat
exports failed the unbiasedness test, and therefore are not weak form rational forecasts.
Colling, Irwin, and Zulauf (1997) test whether Knight-Ridder survey forecasts of frozen
pork bellies are weak form rational forecasts of USDA Cold Storage report preliminary
estimates. Colling et al. (1997) conclude Knight Ridder survey forecasts of pork bellies
are unbiased predictors of USDA preliminary estimates, and that forecast errors are
uncorrelated. The conclusion is the Knight-Ridder survey forecasts of frozen pork bellies

are weak form rational forecasts of USDA preliminary estimates.

Several recent studies have tested whether survey forecasts of agricultural supply

variables are strong form rational expectations of USDA preliminary estimates, relative to

available public information. These studies include (Colling and Irwin (1990), Colling,

42

Irwin, and Zulauf (1992), and Grunewald, McNulty, and Biere (1993)). These studies

provide mixed evidence in support of the strong form rational expectations hypothesis.

Studies ﬁnding evidence in support of the strong form of the rational expectations
hypothesis include Colling and Irwin (1990), Grunewald et al, and Coiling et al. (1992).
Colling and Irwin (I990) conclude Futures World News survey forecasts of hog breedings
and marketings are unbiased and efficient forecasts of USDA Hogs and Pigs report
preliminary estimates of both. Efﬁciency was tested by regressing survey forecast errors
on the difference between survey forecasts and out-of-sample forecasts from auto-
regressive models speciﬁed using only public data. Test results indicate out-of-sample
forecasts contain no additional information useﬁrl for the prediction of USDA preliminary
estimates of breedings and marketings (i.e. the survey forecasts are efficient). The
conclusion is Futures World News survey forecasts of breedings and marketings are
strong form rational forecasts of USDA preliminary estimates. Following Colling and
Irwin’s (1990) methodology, Grunewald et al. conclude Knight-Ridder survey forecasts of
cattle on-feed are strong fonn rational forecasts of USDA Cattle on Feed report
preliminary estimates. And ﬁnally, Colling et al. (1992) conclude Futures World News
survey forecasts of marketings and breedings are unbiased and efﬁcient predictors of
USDA Hogs and Pigs report preliminary estimates of both. The authors conduct a direct
test of efficiency and conclude the survey forecast errors are not explained by theoretically

relevant public information. The conclusion is the Futures World News survey forecasts

43

of breedings and marketings are strong form rational forecasts of USDA preliminary

estimates of both.

Grunewald et al., using Colling and Irwin’s (1990) approach, provide evidence
reﬁ'uting the strong form of the rational expectations hypothesis. The authors conclude
Knight-Ridder survey forecasts of placements and marketings are unbiaseded, but
inefficient relative to available public data. The conclusion is Knight-Ridder survey

forecasts of placements and marketings are not strong-form rational forecasts of USDA

preliminary estimates of both.

This study will extend the work of past studies in three respects. First, none of the
studies mentioned have conducted direct tests of efficiency relative to public and private
data. We conduct tests of unbiasedness and efficiency to determine whether survey

forecasts are strong form rational, in the sense they optimally reﬂect available public and
private data. Secondly, none of the studies mentioned have tested whether a single private
ﬁrm uses their own data optimally. We test forecasts from a private ﬁrm represented in
the monthly Knight-Ridder survey for unbiasedness and efficiency relative to private data
which the company collects itself. Thirdly, tests of unbiasedness and efficiency are
conducted using ﬁnal revised USDA ﬁgures. Final revised ﬁgures are appropriate
because, as determined in chapter 3, USDA preliminary estimates of cattle on-feed,

placements and marketings are not rational forecasts of ﬁnal revised estimates.

44

Finally, we discuss the implications of using biased and inefﬁcient survey forecasts
and USDA preliminary estimates to conduct tests of the efficient market hypothesis and

tests of announcement effects.

4.3 Theoretical Framework

Let F P. and M i denote ﬁnal revised ﬁgures of USDA reported cattle

i+n.! ’ i+n.! ’ r+n,t

on-feed, placements, and marketings for month t, which are released on day i+ n. Let

a 0
F1112 [7,12

and 114,", denote the preliminary announced estimates for the same month t

which are released on day i. Let 11:52,, P732: ,and Min denote Knight-Ridder survey

forecasts of the same month t, which are released on day i-2. Note that ﬁnal revised
ﬁgures are available at some time later than the preliminary estimates. The same notation
is used to denote both individual and average forecasts. The important question for
market agents and econometricians is whether survey forecasts are rational, in the sense
that they optimally use all available information at the time they are made to predict the
ﬁnal revised USDA ﬁgures. If the survey forecast is rational, it is equal to the
mathematical expectation of the ﬁnal revised ﬁgure, conditioned on the information which
was known at the time it was made: that is, in the case of cattle on-feed,

i+n,t

7:... = 7(7

 

¢,_2 ,) where (IL2 I denotes available information, and E is the

mathematical expectations operator.

45

4. 3. I Hypothesis 1 : Unbiasedness
The ﬁrst implication of rationality is that forecasts be unbiased. To determine if
the survey forecasts of cattle on-feed are unbiased the following model is estimated. The

same method is used for evaluating placements and marketings.

(18) F’

1+ ",1

_ e
— a0 + a117,, + u,

The unbiasedness property requires that the coefficients in equation (18) may be restricted

to a0 = 0, and a, = 1, and that u, satisfy the following condition, E(u,|¢,_2',) = 0.

4. 3.2 Hypothesis 2: Efﬁciency

Efficiency imposes the additional requirement that any variable known at time (i-

2,t) or before be orthogonal to u, , that is ,6 = O for any vector 27-2.: 6 ¢,_2,, . The

survey forecasts are rational forecasts of ﬁnal revised ﬁgures only if both conditions are
met. The efficiency hypothesis is tested by applying least squares to the following

equation:

(19) F.

46

which is constructed by simply subtracting 13:1, from both sides of equation (19) and
including information in the set Z,_2,, e ¢,_2,,. Efficiency implies that (a, - 1) = ,3 = 0,

fir-2.1) = 0 . This implies the error,

 

and that u, satisfy the following condition, E (u,

0

F.

H- ",1

e . . . . .
112‘, , is uncorrelated wrth available1nformatron,Z,_2J E (11”,, .

4. 3. 3 Constructing an optimal forecast of ﬁnal revised cattle on-feed

Deﬁne the optimal linear (minimum variance) forecast of F ‘ conditional on

i+n,t

information {Z )to be:

i-2,t

f _ "t
(20) E-ZJ " 0 + :6 Zr-ZJ

where Z,_2, e ¢,_2, includes relevant survey forecasts. The values of b0 and [3 are

0

found by applying least squares to equation (20), over a period of which F is known.

i+n.t

4. 3. 4 Constructing an optimal forecast of unanticipated information
Based on the result from section 3.3.3 in chapter 3, and the previous section 4.3.3,

we now have the ability to create optimal linear (minimum variance) forecasts of

47

unanticipated cattle on-feed at the time the COF report is released under three different
scenarios. The result here is utilized in a later section to demonstrate how tests of
announcement effects and tests of market efficiency can be improved in instances where

USDA preliminary estimates and survey forecasts are non-optimal.

4. 3. 4.1 Scenarios 1,2, and 3:
Scenario 1: Both USDA preliminary estimates and Knight-Ridder survey

forecasts of cattle on feed are non-optimal forecasts of revised ﬁnal estimates conditional

i,t’ r.t

on information {Fa Z. } and {Zr-2,1}, respectively. Then the deﬁnition of

unanticipated information on cattle on-feed that is contained in the COF report is

(El: " Fig!)

Scenario 2: USDA preliminary estimates are non-optimal and Knight-Ridder
survey forecasts are optimal. Then the deﬁnition of unanticipated information on

unanticipated information on cattle on-feed contained in the COF report is

(E1 7 E32!)

48

Scenario 3: USDA preliminary estimates are optimal and Knight-Ridder survey

forecasts are non-optimal. Then the deﬁnition of unanticipated information on cattle on-

feed contained in the COF report is (F: — F172,, ).

4.4 Data
4.4.1 Survey forecasts

Each month Knight-Ridder Financial News conducts surveys of private livestock
market analysts and major retail commodity ﬁrms to forecast upcoming USDA cattle on-
feed, placement, and marketing numbers. Aﬁer dropping the highest and lowest forecasts
of each number, Knight-Ridder computes simple averages which it calls industry forecasts.
Monthly Knight-Ridder forecasts from January 1986 to December 1994 are used in this
paper, giving 108 observations. Knight-Ridder survey forecasts are widely disseminated

are considered to be public information.

Knight-Ridder also publishes the individual analysts’ forecasts used to calculate
the industry forecasts. Using these data, we analyze individuals’ forecasts for
unbiasedness, and root mean square error forecast performance. Because Knight-Ridder
does not include the same analysts in the survey every month, the number of observations
available for most analysts is less than 108. We evaluate the unbiasedness and forecast
performance of the 13 analysts which had contributed more than 60 observations over the

sample period.

49

4. 4. 2 Preliminary and ﬁnal revised estimates

The seven state COF report is released each month by the Agricultural Statistics
Board of the US. Department of Agriculture (USDA). Each report contains preliminary
estimates of the number of cattle on feed the ﬁrst day of each month, the number placed
on feed during the previous month, and the number marketed during the previous month.
Final revisions are published in the ERS statistical Bulletin Cattle: Final Estimates
approximately every ﬁve years. As determined in chapter 3, preliminary estimates over the
period January 1986 through December 1994 are not unbiased and efﬁcient predictors of
USDA ﬁnal estimates. This implies ﬁnal estimates should be used to conduct tests of the

rationality of survey forecasts.

4. 4. 3 Other public data

Other public information used in this study includes net returns to feedlots which is
a proﬁtability measure; and lagged values of USDA preliminary estimates of cattle on-
feed, placements, and marketings. Net returns to feedlots is calculated as the difference
between the selling price of fed cattle and returns to variable costs of high plains feedlots.
Net returns data was collected from the USDA high plains cattle feeding simulator

reported in the monthly Livestock, Dairy, and Poultry report.

4. 4.4 Private data
The private data used in this study include raw monthly private feedlot survey data

for cattle on-feed, placements, and marketings, as well as feedlot capacity. The data are

50

provided by Professional Cattle Consultants (PCC). PCC collects the data from individual
feedlots approximately one week prior to the USDA Cattle on Feed report, compiles the
responses, and returns the aggregated results back to each feedlot in a newsletter format.
Over our sample period, the PCC survey constitutes between 1.3 and 2.2 million head of
cattle. This represents approximately 20% of all cattle on-feed, placed, and marketed
from the seven largest cattle feeding states in the United States each month. In this study
the raw PCC data for each component is divided by feedlot capacity to eliminate variation

due to feedlots entering and exiting the survey over the sample period.

4.5 Results
This section presents empirical results for tests of unbiasedness, efﬁciency, and
forecast performance conducted for Knight-Ridder industry survey forecasts and the

individuals’ forecasts reported by Knight-Ridder.

4. 5. 1 Analysis of industry survey forecasts
4. 5. I . I Test of unbiasedness of Knight-Ridder industry survey forecasts

Each variable was ﬁrst tested for a unit root, and no evidence of a unit root was
found in any of the data series. Results from unbiasedness tests on the industry survey
forecasts are shown in table (11). F-tests reject the null hypothesis of unbiasedness for
cattle on—feed and marketings forecasts at the 5% level. We fail to reject unbiasedness for

placements based on the F-test. The conclusion is that Knight-Ridder industry cattle on-

51

 

Table 11. Test of unbiasedness of Knight-Ridder industry survey forecasts

 

a0 a, [—32 Q(20)‘ H,:a, = 0,rtl = 1
0.14mi 371.71 0.96 0.97 3490* F(2,106) Prob > F
(133.23)b (0.02) 10.76 0.0001
Placements 31.08 0.99 0.92 14.63 F(2,106) Prob > F
(48.29) (0.03) 0.53 0.59
Marketings 140.20 0.91 0.83 27.83 F(2,106) Prob > F
(70.95) (0.04) 3.78 0.03

 

‘Q(20) denotes the Ljung-Box Q-statistic computed at lag twenty to which tests the null hypothesis that
auto-correlation is zero with degrees of freedom equal to the number of lags, " standard errors in
parentheses, * denotes signiﬁcance at the 0.05 level.

feed and marketings survey forecasts are not unbiased estimates of USDA ﬁnal estimates,

while the placements are unbiased.

4.5.1.2 Test of efﬁciency: Knight-Ridder industry survey forecasts

Next, we test the industry survey forecasts for efficiency. Because a forecast must
be both unbiased and efficient to be rational, results from testing the Knight-Ridder cattle
on-feed, and marketings survey forecasts are presented simply to highlight information
they may not be using optimally; in particular the PCC data. Results from tests of
efﬁciency for the industry survey forecasts are presented below. The null hypothesis that

industry survey forecast errors are uncorrelated with available information,

is rejected for cattle on-feed, placements, and marketings. Information

Zi—2,r 6

1-2,r ’

ZEN ,¢,_2., is presented in table (12).

52

 

Table 12. Deﬁnitions of variables

 

P
1‘17.

l

P
P7,

USDA ﬁnal revised cattle on-feed

USDA ﬁnal revised placements

USDA ﬁnal revised marketings

USDA preliminary estimate of cattle on-feed (lagged values only)
USDA preliminary estimate of placements (lagged values only)

USDA preliminary estimate of marketings (lagged values only)

private PCC cattle on-feed

private PCC placements

private PCC marketings

Knight-Ridder survey on-feed forecast

Knight-Ridder survey placements forecast

Knight-Ridder survey marketings forecast

net retums to feedlots (variable cost minus the selling price of fed cattle)
optimal linear forecast of actual revised cattle on-feed at time it

optimal linear forecast of actual revised placements at time it

optimal linear forecast of actual revised marketings at time it

optimal linear forecast of actual revised cattle on-feed forecasts at time i-2,t
optimal linear forecast of actual revised placements forecasts at time i-2,t

optimal linear forecast of actual revised marketings forecasts at time i-2,t

 

53

To formally test whether Knight-Ridder industry survey forecasts of cattle on-feed
forecasts are efficient forecasts predictors of USDA ﬁnal revised cattle on-feed estimates,

the following model is estimated:

(21) F — F757,, = b. +(b7 —1)F:f7,. + P77-.. + 77,

a a p p a a P P
I37.1-1’}:r“,-l 12’F- 7,1 ’E-7J—lZ’RJ—l’RJ—12’R—7J’R-7J-12’

where Z,._2 , = a a . The
M Mi.t—12 ’ Milt, Milt-12

i,-t 1’

parameter estimates are as follows:

F;+n,-F;_2,=1589.03-0.57Ff_2 +0.467F"_1-.0026F;:_12+21328th
' ’ (.506 22) (0105) 2" (.0097) 0.029 (219.19)

—329.74Fi”7 _,2+0.247P‘:_1+0.18Ra_12+4218.56P.f7,

2(4.2 78) ’ (.00)45 (.)011 5(94. 52)
--1045.33Pf7 ,2 -0.39 Ila";l -.0574 M11 -1063.90 M67,

(22) (715.)95 ’ (0141) ”‘ (0141) (99939) ’ '

+1418.12M.’:7,_ 12

(.97587)

D.W.= 1.85, S.E.R.= 105.88, R2 = 0.63, F: 13.55

where standard errors are in parenthesis. An F-test of the joint hypothesis that each slope
coefﬁcient equals zero has a p-value of 0.00001. Therefore, we reject the null hypothesis
that Knight Ridder industry cattle on-feed survey forecasts optimally use all available

public and private information at essentially any signiﬁcance level.

54

To formally test whether Knight-Ridder industry placements survey forecasts are

efﬁcient predictors of USDA ﬁnal revised estimates the following model is estimated:
(23) P

_ R324 = b0 + (bl - 1)1)1'f2,t + ﬂ'Zi-ZJ + gt

_ a a p p -
where 214,: - {Bx-1’RJ-IZ’R-7J’R-7J-12’ Nag-12}- Parameter est1mates are as

follows:
Pi+n,1 — Pi-2,l : "1221332- 933913—21 + 90.1093? 8.1-1 + ((3:11125)P1.1-12
P _ P
(24) ”(5.9.1.1001 129.179.113-71-” + 942.3%”

D.W.= 1.76, S. E. R.= 86.67, I? = 0.30, F = 7.66

An F-test of the joint hypothesis that each slope coefﬁcient equals zero has a p-
value of 0.00001. Therefore, we reject the hypothesis that Knight-Ridder industry survey
forecasts of placements optimally use all available public and private information at

essentially any signiﬁcance level.

Likewise, the following model is estimated for marketings:

(25) M.‘

1+ n,(

' Mia-2.1 = b0 + ([71 — 1) Min + 16,214.: + 61

55

8
_ a p a p a
where Zi-2,l — MAM, MP7.” Mat-12: Mtg—”42,2 a. P . Parameter

1,1 1.1-j
J=4

estimates of the sum of lagged preliminary placement estimates are estimated as a second
order polynomial distributed lag constrained to zero at the far end. Placements will be
marketings in four to eight months and are assumed to follow a quadratic function.

Parameter estimates are as follows:

M , - 14:1,, = 182.11- 0.242 M,:,, + 061347761149

'+"' (100.4) (0.102) "M

- 0.136 44:-..

(0.102)

+ 1270.22 M157; + %5081 Mill-12 + 0(b00369?PDL1U’4

(451.73) 433.46)

(26) + 9.007? PDLZU.4

0.0032

D. W.= 1.92, S. E. R: 46.26, 1‘22 = 0.26, F = 5.79

where PDLl and PDLZ denote polynomial distributed lag terms. An F -test of the joint
hypothesis that each slope coefﬁcient equals zero has a p-value of 0.000015. Therefore,
we reject the null hypothesis that Knight-Ridder industry marketings survey forecasts
optimally use all available public and private information at essentially any signiﬁcance

level.

56

4. 5. 1. 3 Forecast evaluation

Based on our sample period, 1986:01 through 1994212, we estimate parameter
estimates for model (20) using data over the sample period 1986:01 through 1988: 12; the
same approach is applied to placements and marketings. Model (20) is estimated using
ﬁnal revised data from 1986201 through 1988: 12, and this estimated model is used to
make cattle on-feed predictions for the period 1990:02 through 1994: 12’. The forecasting

models for cattle on-feed, placements, and marketings were speciﬁed as follows:

i

Em; = be + b15524 + bZFi’j-l + b3F13-12 + b45157; + b533,:
(27) +b6RZ-l + b7Piz-12 + b88511 + bngiz‘, '1’ blO Mill-l
+bllMZI-12 + blei-fi + u,
(28) R+n,t : b0 + blR-e-ZJ + b2R3-l + b3R3—12 + b41357“
+Q~an+€
(29) Mm” '3 [)0 + blMie-ZJ + [)2 M:,_l + b3M1‘Z-12 + b4 Mill

+b5PDL1,.‘,_4 + bGPDLZW4 + 4:,

where all right hand side variables have been explained in section 4.5.2, and

u,, 8,, and 5, have zero mean conditional on ¢,_2., . Out-of-sample forecasts of ﬁnal

 

5 For an explanation of why this approach must be taken, see section 3.5.3, chapter 3.

57

revised cattle on-feed, placements and marketings were generated for the period 1990:02
through 1994:12, and evaluated using root mean square error evaluation. The ability of
out-of-sample predictions to predict ﬁnal revised ﬁgures were compared relative to

Knight-Ridder industry survey forecasts. Table (13) shows the results.

 

Table 13. Forecast evaluation: RMSE Comparison

 

KR Survey Forecasts Econometric Model
On-feed 189.04 338.11
Placements 103.30 96.67
Marketings 57.91 55.52

 

Results indicate the econometric placements and marketings forecasting models
out-predicted Knight-Ridder industry survey forecasts. The relatively poor cattle on-feed
forecasts can be explained by the fact that model (27) exhibited ﬁrst order auto-correlation
in the errors. However, this cannot be corrected out-of-sample because the
econometn'cian does not have access to ﬁnal revised data at the time the forecasts are
made. This results in biased parameter estimates and thus biased out-of-sample forecasts.
Models (28) and (29) did not exhibit auto-correlation. Thus, the difference in RMSE
between preliminary estimates and optimal forecasts of placements and marketing may be

attributed to sampling error. This is further analyzed in chapter 5 using futures prices.

58

4. 5. 1.4 Summary of section 4. 5. 1

Based on tests of unbiasedness and efﬁciency, and root mean square error forecast
evaluation, we conclude the Knight-Ridder survey forecasts of placements and marketings
are not strong form rational. This implies these Knight-Ridder forecasts do not optimally
reﬂect available public and private information. Therefore, it is possible incorrect
inference can result on the part of the econometrician when forecasting cattle supply,
testing the rational expectations hypothesis, and testing for announcement effects and
market efﬁciency, if Knight-Ridder forecasts of placement and marketings are used.
However, we conclude Knight-Ridder forecast of cattle on-feed are strong-form rational
despite the fact they failed unbiasedness and eﬁiciency tests. We reach this conclusion

based on RSME evaluation.

It is possible that incorrect inference can result on the part of the econometrician
when forecasting cattle supply, testing the rational expectations hypothesis, and testing for
announcement effects and market efﬁciency, if Knight Ridder pre-release industry survey

forecasts of placements and marketings are used. This is the topic of chapter 5.

4. 5. 2 Analysis of the Knight-Ridder individuals ’ forecasts

Results from unbiasedness and root mean square error analysis conducted for
Knight-Ridder survey forecasts appear in table (14). We report only the statistical
signiﬁcance of the F-test for unbiasedness and the forecasts’ root mean square error. We

reject unbiasedness in every case except for analyst 12's cattle on-feed forecasts. We did

59

test analyst 12's cattle on-feed forecasts for efﬁciency. We reject the null hypothesis at the
5% level. The conclusion is none of the individual Knight-Ridder forecasts are weak form
rational. What is particularly interesting about these results is that using PCC’s raw
feedlot data, we out-forecasted PCC’s forecasts of marketings and placements, and the
industry marketings forecasts over the sample period. Moreover, we tested the individual
PCC Knight-Ridder forecasts for efﬁciency: results suggest PCC is not optimally using

their own data to forecast cattle on-feed, placements, or marketings.

The fact we tested the individual forecasts for unbiasedness and efﬁciency using
ﬁnal revised estimates suggests the individual cattle on-feed forecasts failed this test for
the reason discussed earlier. Cattle on-feed estimates exhibit auto-correlated errors, which
is impossible to correct for since ﬁnal revisions are not available on a regular basis. It is
unclear whether all the individual forecasts are strong form rational. However, we can
conclude each set individual placements and marketings forecasts are not strong ﬁom
rational as ﬁnal revised estimates of placements and marketings do not exhibit serial

correlation and that we out-forecasted each of them wide a wide margin.

60

 

Table 14. Properties of the Knight-Ridder individuals’ forecasts

 

 

 

Unbiasedness Root Mean Square Errors
Analyst Obs. F P M F P M
1‘ 108 0.5E-2 0.6E-4 0.1E-6 152.08 214.26 97.37
2 92 0512-2 0.1E-1 0.1E-6 161.01 195.25 102.25
3 108 0712-3 0. 113-3 0.1E-6 163.89 214.80 99.97
4 72 0.1E-1 0.9E-2 0.6E-3 226.11 214.65 80.04
5 101 0613-3 0715-3 0. 113-6 155.73 209.31 96.83
6 90 0.4E-2 0.3E-2 0.1E-6 169.80 203.69 105.96
7 104 0.2E-1 0.3E-3 0.1E-6 167 195.88 91.25
8 74 0513-1 0.2E-2 0. 113-6 173.02 187.09 110.36
9 105 0213-4 0913-4 0. 113-6 172.88 199.40 99.94
10 89 0113-1 0.3E-1 0. 113-6 183 182.03 98.65
11 74 0113-1 0213-1 0.3E-5 195.74 185.29 101.11
12 64 0.2 0313-4 0113-4 183.36 213.02 90.33
13 78 0.1E-1 0.213-1 0.1E-6 184.83 177.18 106.58
Industry 108 189.04 103.30 57.91
A.R.' 24 338.1 1 96.67 55.52

 

‘ denotes auto-regressive forecasts.

 

61. PCC 2. Western Cattle Consultants 3. Cargill Investor Services 4. Farmer’s Grain and livestock Corp.
5. Refoo Research 6. Shearson Lehman Bros. 7. Merrill Lynch Futures 8. Linnco Futures 9. A.G.

Edwards and Sons Inc. 10. Dean Witter Rynolds 11. DEC Futures 12. Packers Trading Co. 13. AR Ring

andAssoc.

61

4.6 Implications For Other Research

In this section, we brieﬂy consider the implications of our results which indicate
USDA preliminary estimates and Knight-Ridder survey forecasts of placements and
marketings may be biased and inefﬁcient forecasts of USDA ﬁnal revised ﬁgures on tests

of announcement effects and tests of market efﬁciency.

4. 6. 1 Tests of announcement effects

To test whether the COF report contains new information for the formation of live
cattle futures pn’ce, a vector of price changes occurring one day following the report’s
release can be regressed on a vector of unanticipated information contained in the report.
If the report contains new information, the coefﬁcient of unanticipated information should
be signiﬁcantly different than zero. Previous studies have deﬁned unanticipated
information has the difference between the preliminary estimate and the survey forecast
(see Pearce and Roley (1985), Colling and Irwin (1990), Grunewald, McNulty and Biere
(1993), and Colling, Irwin and Zulauf(1996,1997) for examples). However, if USDA
preliminary estimates and survey forecasts are non-optimal forecasts of ﬁnal revised
ﬁgures, this approach will lead to a non-optimal forecasts of unanticipated information.
Since agents may be able to generate optimal linear forecasts of ﬁnal revised placements,
the following model can be estimated to test for announcement effects and compared to

results from the existing Pearce and Roley (1985) model:

62

p... - p... = bo + b.(F.: — ES...) + b202,? — PE...)

(30) ..
+b3(Mi,f1 _ M52111" ‘0:

where pH1 , - p” is the change in close-to-close live cattle ﬂitures prices one day

following the COF report. This model assumes agents are completely rational, in the
sense they understand preliminary estimates of cattle on-feed cannot be improved due to
auto-correlation in the USDA revisions, but that optimal linear forecasts of ﬁnal revised

placements and marketings can be generated.

4. 6.2 Tests of market efficiency

Previous studies have tumed a model such as (30) into a test of the efﬁcient
market hypothesis by including the survey forecasts as explanatory variables (see Pearce
and Roley (1985) and Colling and Irwin, (1990)). The efﬁcient market hypothesis asserts
that information available before the release of a government inventory report should
already be reﬂected by prices, and thus not useful for the prediction of the mice change
following the report (F ama, 1970). Again, since agents may be able to generate optimal
linear forecasts of ﬁnal revised placements, and marketings the following model can be
estimated to test whether the market’s response to new information released in the COF
report is efﬁcient and compared to results from the existing Pearce and Roley (1985)

model:

63

p... — p... = b. + b101,? — F152,.) + b.(P.5 — Hi.)

(31) ~ ..
+b3(Mi.f1 " M1{2.1)+ b4Fi:2.1 + 17513132.: + b6M1{2,: + $1

Market efﬁciency implies that b4 = bS = b6 = 0. This implies that price changes

occurring one day following the release of the USDA COF report cannot be signiﬁcantly
explained by information available summarized by the Knight-Ridder cattle on-feed survey
forecasts and the optimal placements and marketings forecasts. If biased and ineﬂicient
estimates of unanticipated information are used to estimate model (31), parameter
estimates will be relatively inefﬁcient, rendering t- and F-tests less reliable. This is

analyzed in chapter 5.

4.7 Conclusion

We ﬁnd that Knight-Ridder individual analysts’ forecasts of placements, and
marketings are not strong form rational forecasts of USDA ﬁnal revised ﬁgures,
conditional on available public and private information. Likewise, we ﬁnd that Knight-
Ridder industry survey forecasts of placements and marketings, calculated as the average
of individuals represented in each monthly Knight-Ridder survey, are not strong form
rational forecasts. The efﬁciency of Knight-Ridder survey forecasts of placements and
marketings can be improved by conditioning on a larger set of available information. In
the case of the placements industry survey forecasts, we ﬁnd that Knight-Ridder
placements, lagged preliminary placements estimates, and current and lagged private
placements are important for explaining the industry survey forecast error. In the case of

64

the industry marketings survey forecasts, we ﬁnd that Knight-Ridder marketings, current
private marketings, and lagged preliminary placements estimates are important for
explaining the industry forecast errors. Results also suggest Professional Cattle
Consultants don’t use their own private data optimally to form the forecasts of placements
and marketings which they provide Knight-Ridder. In the case of cattle on-feed, the
source of bias and inefﬁciency of Knight-Ridder forecasts as predictors of ﬁnal revised
estimates to due to auto-correlated errors, not available data. This implies the
econometrician is unable to improve Knight-Ridder cattle on-feed forecasts as the ﬁnal

revised data are not available at the time forecasts of ﬁnal revisions are made.

The conclusion is Knight-Ridder industry survey forecasts of placements and
marketings are not the optimal forecasts of the market’s subjective expectation of actual
placements, and marketings. This has implications on how tests of announcement effects
and tests of the efficient market hypothesis should be speciﬁed. An alternative approach
to conducting these tests is presented which uses optimal forecasts of ﬁnal revised USDA
ﬁgures to measure unanticipated information. This will improve the efﬁciency of these

tests and make t-tests and F-tests more reliable.

65

MARKET EFFICIENCY IN THE US BEEF CATTLE INDUSTRY

CHAPTER 5

5.1 Introduction

Several recent studies have used survey forecasts to distinguish between
anticipated and unanticipated information contained in USDA inventory reports to
investigate announcement effects and to test the efﬁcient markets hypothesis (Barnhart,
Colling and Irwin, Grunewald, McNulty, and Biere, and Colling, Irwin, and Zulauf).
However, the true anticipated and unanticipated components of a government report are
unobservable. The conventional approach introduced by Pearce and Roley, is to deﬁne
anticipated information as a survey forecast conducted prior to the report’s release, and
unanticipated information as the difference between the actual report numbers and the pre-
release survey forecast. This approach has led to very useful research analyzing whether
USDA inventory reports fulﬁll their primary public policy objective of providing new
information to market participants and testing whether the futures market’s respond
efﬁciently to information released in USDA reports. However, none of the existing
studies have tested whether USDA preliminary estimates (i.e. the information released in
the report) are rational forecasts of ﬁnal revised numbers provided later, after more
information has become available. If the USDA revises preliminary estimates in a
predictable way, and/or if survey forecasts are not unbiased and efﬁcient forecasts of the
ﬁnal revised USDA ﬁgures, then the convential tests of announcement effects and the

efﬁcient market hypothesis (EMH) may lead to incorrect inference. This is because, in

66

these circumstances, the conventional approach fails to properly decompose information
into components that can be predicted (anticipated) and another that cannot

(unanticipated).

In this chapter, two contributions are made. First, we develop a procedure for
testing announcement affects and market efﬁciency when survey forecasts are not
unbiased, efﬁcient, forecasts of ﬁnal revised USDA numbers, and/or when USDA
preliminary estimates are not unbiased, efﬁcient, forecasts of ﬁnal revised USDA
numbers. Second, we show how private information, not publically available, can be
included in the analysis and used to shed light on announcement effects and market
efﬁciency. The procedures are illustrated by applying them to test for forecasting
accuracy, announcement effects, and strong form market efﬁciency in the US beef cattle

industry.

Section 5.2 presents a brief description of market efﬁciency. Section 5.3 presents
the conventional approach for testing for announcements effects and market efﬁciency.
Section 5.4 discusses limitations of the conventional approach. Section 5.5 presents and
alternative approach for testing for announcement effects and market efﬁciency. And

section 5.6 presents conclusions.

67

5.2 Market Efﬁciency

In this section we formally introduce Fama’s models of market efﬁciency. The
deﬁnitional statement that in an efﬁcient market prices “fully reﬂect” available information
is so general that it has no empirically testable implications. To make the model testable,
the process of price formation must be speciﬁed in more detail. Below we brieﬂy present

Fama’s deﬁnition of weak, semi-strong, and strong form market efﬁciency.

5. 2. 1 Weak form efﬁciency: the random walk model

Weak form tests of market efﬁciency are concerned with whether successive one-
period returns are independent, and whether successive one-period returns are identically
distributed. The two hypotheses together constitute the random walk model. The random

walk model can be speciﬁed as follows:

(32) In p, — ln 117,,l = b0 + b,(ln pH — 1n p,_2)+ 5',

where 1n p, - In [71—1 is the rate of return in period i calculated as the difference between
the logan'thm of price in period i and period i-I; b0 and [71 are a constant and slope
parameter, respectively; 5‘, is a random error term with zero auto-correlation. Weak form
efﬁciency requires that the coefﬁcients in model (32) may be restricted to b0 = bl = 0 ,

and residuals display zero auto-correlation. The restriction [70 = 0 implies there is zero

68

opportunity cost of trading as margin requirements can be satisﬁed through investment in

government T-bills. The restriction bl = 0 is implies the current period’s returns cannot

be predicted on the basis of information reﬂected in the previous period’s return.

5. 2. 2 Semi-strong form test of market efﬁciency

Semi-strong form tests of the efﬁcient markets model are concerned with whether
current prices “ﬁilly reﬂect” all obviously publicly available information. This implies that
successive one-period returns cannot be predicted on the basis of publically available

information or last period’s return. The model is speciﬁed as follows:

(33) 10p, " 1n .0).) = be + b1(lnp1—1—lnp1-2)+ ﬂ, In VVi-l '1' 51'

where "/1—1 6 90-1 is a vector of available public data at time H and ¢1-1 denotes all
available information at time i-l; b0 and b] are a constant and slope parameter,

I

respectively; ,6 is a vector of slope parameters. Semi-strong form market efﬁciency

requires that coefﬁcients in model (33) may be restricted to b0 = [71 = ﬂ = 0 , and

residuals display zero auto-correlation.

69

5. 2. 3 Strong form test of market efficiency

Strong form tests of the efﬁcient markets model are concerned with whether all
available information is ﬁilly reﬂected in prices in the sense that no individual has higher
expected trading proﬁts than others because he has monopolistic access to some
information. This implies that successive one-period returns cannot be predicted on the
basis of private information, publically available information, or last period’s return. The

model is as follows:

(34) In p, - 1n .0.-1 = bo + b,(lnp,._l — 1n p,_2)+ ,6'Z,_l + g,

where Z,._l e ¢1—1 is a vector of available public and private data at time H and (25H
denotes all available information at time i-l; b0 and b1 are a constant and slope
parameter, respectively; ﬂ' is a vector of slope parameters. Strong form market

efﬁciency requires that coefﬁcients in model (34) may be restricted to b0 = bl = ﬂ = 0 ,

and residuals display zero auto-correlation.

5.3 Conventional Approach
Consider the conventional model for testing the EMH applied to investigate the

price response to the Cattle on Feed (COF) report . The model is as follows:

70

mp...,.-1np.,.=b.+b.(1nF.:-1n1~::..)+b.(1nP“-mei..)+b.(mM.:.—1nM.-:..)

1,1

(35)
+b.(1np.. — mp.-.,,)+ b. 11183,. + b. Inn-:2. + 41an.. + v.

where In pm is the natural log of the closing live cattle futures price on day i, the day

which the report for month t is released; In F a

1,! ’

In P: and In M“, are the natural log

of USDA preliminary estimates of cattle on-feed, placements and marketings, respectively,

for month t released on day i; In 17,.in , In If” and In Min are the natural log of

survey forecasts of cattle on-feed, placements and marketings, respectively, for month t

released on day i-Z (two days before the ofﬁcial report is released); and v, is a random

error term uncorrelated with information known at the close of the trading day i. Model
(3 5) is typically estimated in logarithms to minimize non-stationarity caused by increased
price dispersion at higher price levels (Grunewald et al.). Model (3 5) implicitly assumes

cattle on—feed survey forecasts are unbiased and efﬁcient implying,

E52,: = E(Fra

,l

 

fix-2,1); where (1514‘, denotes all available information at the time the

survey forecast is released two days before the COF report release. Likewise for

placements and marketings survey forecasts. So, under the EMH b4 ,b5 ,b6 and b7
should all be zero and v, should be serially uncorrelated, because all available (anticipated)

information, ¢,._2., , should already be reﬂected by day i prices. If the report contains new

71

information, then at least one of b1,b2 ,b3 should be different from zero. Model (3 5) is

estimated separately using near-term, ﬁrst-deferred, and second-deferred contract prices
over the sample period March 1990 through December 1994. Cattle are on feed
approximately four to ﬁve months so one would expect near-term contract price changes
to have a positive relationship with unanticipated shocks to marketings, and deferred
contract prices to have a negative relationship with unanticipated shocks to placements
(more placements means more supply and lower prices in four to ﬁve months). Each
variable was tested for stationarity, and no evidence of a unit root was detected.
Residuals from model (3 5) were tested for heteroscedasticity including ARCH effects, and
none was detected. Previous studies have estimated model (3 5) using the two-limit tobit
model to account for limit price moves. Over our sample period, the near-tenn contract
recorded just 4 limit moves, and the ﬁrst- and second-deferred contracts recorded zero
limit moves. The degree of inefﬁciency due to four limit price moves is expected to be

negligible, thus model (3 5) is estimated using OLS for each contract horizon.

Parameter estimates and hypotheses test results for market efﬁciency appear in
tables (16) and (17), respectively. In general, results indicate failure to reject the EMH:
survey forecasts are not useﬁil for predicting price changes occurring one, two and three
days following the release of the COF report for each contract horizon at conventional
signiﬁcance levels. Results indicate the only possible exception is a rejection for second-

deferred price changes one day following the C 0F report release at the 10% level, but not

72

at the 5% level. The conclusion is that prices do not respond to information already

known when the COF report is released, which is consistent with the EMH.

73

 

Table 15

. Definitions of variables

 

r—7,t

P
13-7.1

USDA ﬁnal revised cattle on-feed

USDA ﬁnal revised placements

USDA ﬁnal revised marketings

USDA preliminary estimate of cattle on-feed (lagged values only)
USDA preliminary estimate of placements (lagged values only)

USDA preliminary estimate of marketings (lagged values only)

private PCC cattle on-feed

private PCC placements

private PCC marketings

Knight-Ridder survey on-feed forecast

Knight-Ridder survey placements forecast

Knight-Ridder survey marketings forecast

optimal linear forecast of actual revised cattle on-feed at time i,t

optimal linear forecast of actual revised placements at time i,t

optimal linear forecast of actual revised marketings at time i,t

optimal linear forecast of actual revised cattle on-feed forecasts at time i-2,t
optimal linear forecast of actual revised placements forecasts at time i-2,t

optimal linear forecast of actual revised marketings forecasts at time i-2,t

 

74

 

.26. .5...me 05 .a A... .56. .88.. o. 3 00:85:»... 8.25.. .6. .358... .. 880558.. 5 Echo .65 ”.862

 

.. . .8 .....8 .88.... .....8 .28 .88 .88 .88 .o. .8

8.. 8... :28 2...... 8.... 8... 2.. 8.... 8... 8... 2.... 85.63.53,.
:2.-. $8.... .888 .88 .n . .8 .88 .88 .88 .68

8.. 2.... : . n... 8...... 8...... 8... n . ... 8...... 8... 2...... 2.. 85.6.9.8...
.. . .8 .28 38.8 .88 .2 .8 .88 .88 .88 . :8

8.. 8... :9... 2.... 8.... 2.... 8... 3.... 8... 8.... 2.... 5.2.52
33.8 no»... to 3.30 has .3. 9.62. 2.5 05 co 99:20 8....
.....8 .888 .....8 .28 .88 .....8 .88 .28

2... S... 8... 8...... 8...... 8.... 8.8 8... 8.9 o. .... 3:683:88
.....8 .48.... .n. .8 .28 .88 .88 .88 .~.8

z... 8.8 8... 8.... 8...... 8... 8.....- 8... 8...- 2.8 25.3-2:
.28 .88 .88.... .28 .88 .88 .88 .28

S. 8.? z... 8...... 8.....- 2...... 8.... n 8... 8...... A8.... 6.2.52
Oman—oh ENNK to 3:qu has he mesa: $6099» 95 co 093:0 90E
.....8 .258 2...... ... . .8 .88 .n ....8 3...... .2 .8

8.” z 8 n 8.... .8... 8.... .3... 8... :88 8... 8... 35.6.9888
6...... .88 .88.... 2...... .n . .8 .88 .88 .88

8.. .2. 8.... 8...... 2...... 8...... .8... :8... 8... n . .8 85.8.2...
8... .898 .88 .28 .88 .88 .88 .38

8.. .2. 8.... 8...... 2...... . 2. :8... 8.... Z... 2... 5.3.362
83.8 no»... to Show Ba. 5:. 9.65 an... 05 co owns... 82

Ba .m . .8 a s .8 s s s a .8

 

Bissau 20.0680

 

 

muse—2.59:. 3.3— a... 3......— o... «52. 35.95» 9:57.28 .8 .8. =8...— nzaom .3 e35.

 

75

 

Table 17. Hypotheses test results for the Pearce and Roley semi-strong form efﬁciency test

 

Near-term

F int-deferred

Second-deferred

Near-term

First-deﬁned

Second-deferred

Near-term '

First-deferred '

Seoond-deferred'

Tests for the first trading day after Came on Feed release
HOIb‘=b,=b6=b,=0
H02b‘=b,=b6=b,=0
H02b4=bs=b‘=b,=0

Tests forthcseoondtrading dayaﬂerCattIe on Feed release
Hoib‘=b,=b‘=b,=0
Ho:b4=b,=b‘=b,=0
H°:b‘=b5=b6=b,=0

Tests for the third trading day aﬂer Cattle on Feed release

Ho:b4=b,=b6=b,=b8=0
H02b4=bs=b6=b7=b,=0

Hozb4=bs=b6=b7=b,=0

F4131

1.27

1.04

2.31

0.33

1.17

1.22

1.69

1.12

1.33

P-value

0.30

0.40

0.07

0.86

0.33

0.31

0.16

0.36

0.27

 

'Model was corrected for third order auto-correlation. and this parameter was additionally tested equal to zero.

Next we imposed the EMH, ( b4 = b5 = b6 = b7 = 0 ), and tested for

announcement effects (bl = b2 = b3 = O ). Parameter estimates and hypotheses test

results for announcement effects appear in tables (18) and (19), respectively.

76

 

Table 18. Results from test of announcement effects using Pearce and Roley’s methodology

 

 

Cocﬂ'tcrent Estumtes

b. a b. b. [7’ W
Price change on the first trading day after Cattle on Feed release

Near-term -0.001 0.11 0024 0.078” 0.14 1.75
(0.001) (0.078) (0.023) (0.028)

First-deferred 0.0004 0.051 0038‘ 0.062” 0.15 1.70
(0.001) (0.069) (0.02 (0.025)

Second-deferred -0.0007 0.038 0029' 0.035‘ 0.11 2.01
(0.0008) (0.055) (0.016) (0.019)

 

Notes: Std. errors in parentheses, " Asterisk (') denotes signiﬁcance at 10 percent level, (”) at the Spercent level.

 

Table 19. Hypotheses test results for announcement effects from Pearce and Roley model

 

Tests for ﬁrst trading day after Cattle on Feed release Fostat P-value
Near-term H03bl = 1,2 -_- b3 = 0 3.01 0.04
First-deferred ”o3bi = b, = 53 = 0 3.23 ' 0.03
Second-deferred Haj] = b: = 1,3 = 0 2.25 0.09

 

Placements and marketings parameter estimates have expected signs and individual
explanatory power. Results suggest the COF report does provide new information to the
near-tenn and ﬁrst-deferred contracts. For these contract prices we reject the null
hypothesis below the 5% level, but not the 1% level. However, in the case of the second-
deferred contract we reject the null hypothesis below the 10% level, but not the 5% level.
The latter result suggests the C 0F report may provide less information relevant to longer

term prices.

77

5.4 Limitations of the Existing Methodology
Model (3 5) may generate inefﬁcient parameter estimates and incorrect inference if

.

 

i+n.!

¢,.,)and

the following two conditions are not satisﬁed: FI‘: = E (F

Fitz! = E(Pi:n,t

 

.
7-..), F...

is the ﬁnal revised USDA estimate of cattle on-

n,t

feed in month t, detemrined on day i+n.

We test USDA preliminary estimates and Knight-Ridder survey forecasts for
rationality by conducting tests of unbiasedness and direct tests of eﬂiciency relative to
public and private information. We conclude USDA preliminary estimates and KR survey
forecasts of cattle on-feed, placements and marketings may not be rational estimates of
ﬁnal revised numbers. One interesting aspect of our efficiency tests is the use of private
data provided by Professional Cattle Consultants (PCC). PCC conducts a feedlot survey
prior to the release of each monthly COF report, and returns compiled cattle on-feed,
placements and marketings responses to each feedlot in a newsletter format approximately
one week before the C 0F report release. Our sample of PCC data spans from January
1986 through December 1994. Over this sample period, the PCC survey represents
between 1.3 and 2.2 million head of cattle each month; approximately 20 % of all cattle
on-feed, placed and marketed from the seven largest cattle feeding states each month. The

PCC data is divided by feedlot capacity to eliminate variation due to feedlots entering and

78

exiting the survey over the sample period. Furthermore, no evidence of a unit root was

detected in any of the data series.

5.5 An Alternative Approach
Next we consider whether a model of market efﬁciency can be speciﬁed which
optimally reﬂects unanticipated and anticipated information. Following Mankiw et al.,

deﬁne the optimal linear (minimum variance) forecast of ﬁnal revised cattle on-feed,

F}; t , conditional on information available at time (i,t),

{Fa Fe Ff7J,Pf.,J,M’:7J},tobe:

1,1, i-2.t’ l t t

A A A A A

(36) 17.5 = b. + 12.13: + 12.12:... + an. + M25. + 5, M

where all variables have been deﬁned previously. Model (3 6) is estimated over the period

January 1986 through December 1988. To obtain optimal linear forecasts of ﬁnal revised

cattle on-feed ﬁgures conditional on information at time (i-2,t), 1172’, , model (36) was

simply estimated without sz . The same procedure is used for placements and

marketings. Deﬁne the optimal linear minimum variance forecasts of placements and

marketings conditional on information available at time (i,t) as follows:

(37) P.

79

A A A

(38) Mf = b0 + blM.“ + b2 Mil, + b3Mi’j7’,

1,! I
To obtain optimal linear forecasts of ﬁnal revised placements and marketings ﬁgures

conditional on information at time (i-2,t), 1172,, and MEN , models (3 7) and (38) were

simply estimated without 3‘: and M .a respectively. Parameter estimates from

1,1’

econometric models are presented in table (20). The parameter estimates from each
model are used to predict cattle on-feed over the period February 1990 through December
1994, and placements and marketings over the period January 1990 through December
1994, using updated data. This approach was necessary because the USDA released ﬁnal
revised estimates for the period January 1984 through December 1988 in January 1990,
and not again until January 1995. F-test results indicate the econometric models generate
more efﬁcient in-sample forecasts of cattle on-feed, placements and marketings than
USDA preliminary estimates and Knight-Ridder survey forecasts of each over the period
January 1986 through December 1988. Note that if auto-correlation is detected in any of
the econometric models, it cannot be used to improve forecasting because ﬁnal revised
data are not available at the time out-of-sample forecasts are made starting February 1990.
Based on the durbin watson statistics we can conclude parameter estimates from the cattle

on-feed

80

 

Table 20. Parameter estimates and statistics from econometric forecasting models

 

 

11mm We Models Conditional on Information (1.1) Econometric Models Conditional on Information (1.2.x)
“W“ 5:... MI... Er... ML...
Comm -24.11 13.83 862.89* 283.75"
s.e. (143.84) (66.13) (454.44) (132.52)
E: 1.12"
s.e. (0.063)
R3
8.8.
MI: 0.93M
s.e. (0.085)
1.7:“ -0.09 1.01“
s.e. (0.063) (0.04)
1332..
s.e.
M1123 0.005 0.69"I
s.e. (0.08) (0.11)
1357., 201.59 -641 .99
s. e. (130.54) (409.60)
[357" 390.04" 1127.05M
s.e. (173.31) (507.52)
ML” 6129 547.63“ -1145.53 1189.18"
s.e. (274.24) (230.31) (901.14) (480.75)

 

 

81

 

Table 20. Parameter estimates and statistics from econometric forecasting models

 

EZ 0.99 0.99 0.96 0.96 0.95 0.82
F- stat 1950.33 2389.58 269.66 207.61 304.06 73.84
p .. value 0.00001 0.00001 0.00001 0.00001 0.00001 0.00001
DW 0.74 1.90 1.87 1.52 1.86 2.01

 

 

Notes: Std. errors in parentheses, b Asterisk (‘) denotes signiﬁcance at 10 percent level, (”) at the Spercent level.

econometric models conditional on information at time (i,t) and (i-2,t) are unbiased but
inefﬁcient. Thus, an agent can determine ex-ante that out-of-sample forecasts from both
cattle on-feed models will not be minimum variance forecasts. However, econometric
models for placements and marketings conditional on information at time (i, t) and (i-Z, t)
do not exhibit signiﬁcant auto-correlation. T-tests from these models indicate PCC
marketings data is useful for predicting ﬁnal revised marketings at time (i,t) and (i-2,t),
and that PCC placements data is useful for predicting ﬁnal revised placements at time (i-
2,I). We can reasonably conclude that a rational agent would use out-of-sample forecasts
of ﬁnal revised placements and marketings from the econometric models conditional on
information at time (i, t) and (i-2,t). Based on these results it follows that unanticipated
and anticipated placements and marketings, as deﬁned in model (3 5) using USDA
preliminary estimates and KR survey forecasts, may be non-optimal estimates of
unanticipated and anticipated information. An optimal model of market efﬁciency is as

follows:

82

 

1m...- Inp... = b. + 13(11113-3- mat...) + b.(1n Rf, —1n12?;..)+ b.(1n Mt. — InME...)

(39> - -
+b.(1n p... -1np.-...)+ b. Inez... + 12.111122. + b. lnMi... + v.

where 1n Rf and In Mf, are the natural log of optimal forecasts of ﬁnal revised

placements and marketings, respectively, conditional only on available information at the

time the COF preliminary report numbers are released; In 1352,, and In M52“ are the

natural log of optimal forecasts of ﬁnal revised placements and marketings, respectively,
conditional only on available information at the time the KR survey forecasts are
released; and all other variables have been previously deﬁned. Proceeding, model (39) is

estimated to test for market efﬁciency in precisely the same manner as model (3 5). We

test the null hypothesis that coefﬁcients on available information, b4 ,b5 ,b6 and b , are

jointly equal to zero. Parameter estimates and hypotheses test results appear in tables (21)
and (22), respectively. First, we reject the EMT-I for near-term and ﬁrst-deferred contract
prices at the 10% level, but not the 5% level. Second, we reject the EMH for second-
deferred contract prices one day following the C 0F report at the 5% level, but not the 1%
level. The latter result is consistent, yet stronger, than our result from model (3 5); it
implies second-deferred price changes one day following the COF report are not

consistent with the strong form of the EMH. It is important to note that risk adjusted

83

 

proﬁts must be generated on the basis of the optimal forecasts of placements and

marketings before we can deﬁnitively conclude the live cattle futures market is inefﬁcient.

84

.36. 8.580% 05 .8 A38 495. 83.8; c. «a 3:85.33 8659 AL 8.8.53 8 88055.39 5 taco .Em 882
§8 3.88 2888 4.8.8 5.8 8.8 8.8 388 2:8

8.. 88 :88? 88 888 £88 :8. 88. .88 888 88. 25.88.8888
A... _ .8 8.8 388.8 a .88 3. .8 8.8 8.8 8.8 3.8

88.. 88 . :8 8888 .8888 88 8.8 .88. 88 88 88. 85.8.8.2:
2 _ .8 38.8 388.8 A: .8 3. .8 8.8 38.8 $8.8 88.8

88.. 88 :3 8- 88 888 88 88 888. 88 888 88.9 56:32

3.3.9.. Week to 2:60 baa .83 mice: 3.5 05 co 09:29 3:.—

3 .88 88.8 :88 a. .8 8.8 :88 288 E8

8.. 88 88 .888 8888. 88. 88. 88 88. :8. 35.68.888.88
38.8 $888 :88 8. .8 88 A88 888 .88

88.. 88- .88 .888 888. «88 8888. 88 88- :8. 8.3538;
8.8 88.8 c; .8 8. .8 8.8 8.8 2.8.8 888

8.. 88- 88 88.8 .8888. 888. .88. 88 8.8. I .8. 5.2.52

840.2 took to 8:39 be. .83 ”Eva: 9.83 2: co «mango 8E

 

 

 

a. a :8 :88. 888. 888 .88 88 .88. 88 88 8658688888
88 88.8 :88 a _ .8 8.8 8.8 8.8 5.8

88.8 88 :88. 888. 88888. 888. .888 .88 88 888 865.52:
:88 388.8 8.8 :88 8.8 8.8 $8.8 88.8

88 8. .8 :888. 888. 888 «.8 :88 88. .8 v8 85.562
33.8 monk no 3:60 6:- mnﬁ «535 as... 95 cc owed—.0 8E

25 8m . .8 .8 .8 .8 .8 .8 .8 ,8 .8
9:53 8.8886

 

.35.qu age: .8 .88 88.... 8.38— .: 938,—.

 

85

 

Table 22. Hypotheses test results for strong form efﬁciency test

 

Tests for the ﬁrst trading day after Cattle on Feed release F-stat P-value
Near-term 110;],4 = b, = [,6 = 1,7 = 0 2.35 0.07
First-deferred H03b‘ = b3 _—, be : [,7 = 0 2.29 0.07
Second-deferred Ho;b‘ _._ b3 = 56 = [,7 = 0 3.07 0.02

Tests for the second trading day after Cattle on Feed release

Nearm ”03% = 1,5 = [,6 = 57 = 0 0.53 0.71
First-deferred ”0:5‘ = b, = 1,6 = b. = 0 1.36 0.26
Second-deferred H031” = [,5 = [,6 = [,7 = 0 1.90 0.13

Tests for the third trading day after Cattle on Feed release

Nw°m . HOZb‘ = b5 = b6 = b7 = b8 : O 1.73 0.15
First-def“. H0:b‘ = b5 = b6 = b7 = b8 = 0 0.99 0.43
Second-deferred‘ ”01% = 55 = 56 = b, = b, = 0 1.53 0.20

 

‘ Model was corrected for third order auto-correlation, and this parameter was additionally tested equal to zero.

However, the ex-ante predictability of price movements found here is a necessary (but not

sufﬁcient) condition for the existence of arbitrage opportunities.

Next, we impose the EMH on model (39), [24 = b5 = b6 = [27 = O , and test the

null hypothesis that parameter estimates on unanticipated information, b1 ,b2 and b , are

jointly equal to zero using near-term and ﬁrst-deferred contract prices. In the case of
second-deferred contract prices we test the same null hypothesis, but without imposing the
EMT-I on model (3 9) since available information was found to be jointly signiﬁcant for
predicting these price changes. Failure to reject the null hypothesis indicates the COF

report does not contain new information for market participants. Parameter estimates and

86

hypotheses test results for announcement effects appear in tables (23) and (24),

respectively.

 

Table 23. Results from test of announcement effects using Pearce and Roley’s methodology

CocfﬁcigitﬁEstirmtes

bo I; bz b, E? DW
Price change on the ﬁrst trading day after Cattle on Feed release
Near-term -0.001 0.03 0.007 0.05 0.008 1.69
(0.001) (0.07) (0.02) (0.03)
F int-deferred 0.0005 -0.03 -0.01 0.04 0.002 1.66
(0.001) (0.06) (0.02) (0.03)

 

Notes: Std. errors in parentheses, ” Asterisk (‘) denotes signiﬁcance at 10 percent level, (") at the Spercent level.

 

Table 24. Hypotheses test results for announcement effects

Tests for the that trading day after Cattle on Feed release F-stat P-value
Near-term H03b1= b2 = b3 -.- 0 1.15 0.34
First-deferred ”0sz = b2 = [,3 .—_ 0 1.05 0.38
Sword-deferred ”0th _—. 52 = 53 -.- 0 2.35 0.08

 

Two results are strikingly different than those found using the conventional approach; we
fail to reject the null hypothesis that unanticipated information is not useﬁil for the
prediction of near-term and ﬁrst-deferred contract prices. Finally, we reject the null

hypothesis using second-deferred contract prices at the 10% level, but not the 5% level.

87

5.6 Conclusions

From these results we can reach two important conclusions. First, this evidence
suggests the COF report may not provide new information to agents with optimal linear
forecasts of ﬁnal revised estimates of placements and marketings. Second, this evidence
also suggests that agents with optimal forecasts of COF ﬁnal revised report numbers
(which use PCC data and other data) may be able to predict some price movements after

the preliminary COF report numbers are released.

88

CONCLUSION

CHAPTER 6

6.1 Conclusion

The purpose of this study was to help policy makers, econometricians and market
analysts evaluate the usefulness of information regarding the USDA Cattle on Feed report,
as well as private information and other public information reported by the USDA. We
analyzed several interrelated questions related to the rational expectations hypothesis to
determine whether the USDA and market agents optimally use available public and private
information to form expectations of the supply of cattle on-feed, placements and
marketings. We also analyzed several interrelated questions related to the efﬁcient
markets hypothesis to determine whether live cattle ﬁitures price changes occurring one,
two, and three days following the release of the C 0F report optimally reﬂect available

public and private information

In chapter 2 of this study, Does the Cattle on Feed Report Still Provide New
Information to Market Participants?, ﬁrst and second moments of release day live cattle
ﬁitures price changes were compared to those of non-release days. The objective was to
determine whether release day live cattle futures price changes have greater mean and
variance than non-release days. We concluded that release day price changes over the
period January 1990 through May 1998 have greater mean and variance than non-release

day price changes. This evidence suggests the Cattle on Feed report is fulﬁlling one of its

89

 

public policy objectives which is to provide to new information for the formation of live

cattle ﬁitures prices.

In chapter 3, Are Preliminary Estimates in the USDA Cattle on Feed Report
Rational Forecasts of USDA Final Revised Estimates?, we tested whether USDA
preliminary estimates of cattle on—feed, placements, and marketings are strong form
rational forecasts of USDA ﬁnal revised ﬁgures by conducting tests of unbiasedness and
efﬁciency relative to available public and private information. We concluded that USDA
preliminary estimates of each component may not be strong form rational relative to
available public and private information. However, a root mean square error forecast
competition indicated that preliminary estimates of each component could not be out-
forecasted by conditioning ﬁnal revised numbers on available public and private data
suggesting USDA preliminary estimates may be optimal estimates of ﬁnal revised numbers
Finally, we discussed the implications of using biased and inefficient USDA preliminary
estimates to conduct tests of the rational expectations hypothesis, tests of announcement

effects, and tests of the efficient market hypothesis.

In chapter 4, Are Knight-Ridder Pre-Release Survey Forecasts of the USDA
Cattle on Feed Report Rational Forecasts of USDA Final Revised Estimates?, we tested
whether Knight-Ridder Financial News industry survey forecasts are strong form rational
forecasts of USDA ﬁnal revised ﬁgures using the same approach used in chapter 3. After

conducting tests of unbiasedness and efficiency, we concluded that Knight-Ridder industry

90

 

survey forecasts of cattle on-feed, placements and marketings may not be strong form
rational forecasts, and thus may not be useﬁil for approximating an average expectation of
the market as a whole. Results from a root mean square forecasting competition suggest
Knight-Ridder cattle on-feed forecasts cannot be beaten by conditioning ﬁnal revised
numbers on available public and private information, but that Knight-Ridder placements
and marketings forecasts can be beaten. Next, we tested whether a private ﬁnn’s
forecasts of cattle on-feed, placements and marketings, as reported by Knight-Ridder, are
strong form rational relative to private data the ﬁrm collects itself. We concluded the
private ﬁrm may not optimally use their own data. Finally, we discussed the implications
of using biased and inefﬁcient survey forecasts to conduct tests of announcement effects,

and tests of the efficient market hypothesis.

Finally, in chapter 5, Market Efﬁciency in the US Beef Cattle Industry, we
introduced an alternative approach for conducting tests of announcement effects and tests
of the efficient market hypothesis in the presence of biased and ineﬂicient survey forecasts
and preliminary USDA estimates. The methodology was employed to determine whether
live cattle futures prices respond to the release of the USDA Cattle on Feed report in
accordance with the strong form of the efficient market hypothesis, and to test whether the
report contains new information. Then results were compared to those generated using
the conventional approach. Results from the conventional approach and the alternative
approach are quite different. Results from the conventional approach suggest prices are

efﬁcient relative to available public data, and the COF report provides new information to

91

 

agents with expectations similar to Knight-Ridder survey forecasts and USDA preliminary
estimates of cattle on-feed, placements and marketings. However, the alternative
approach indicates the COF report may not provide new information to agents with
optimal linear forecasts of ﬁnal revised estimates of placements and marketings. And
ﬁnally the evidence also suggests that agents with optimal forecasts of COF report
numbers (which use PCC data and other data) may be able to predict some price

movements after the C 0F report is released.

92

BIBLIOGRAPHY

Abel, AB. and F. Mishkin. “An Integrated View of Tests of Rationality, Market
Efﬁciency and the Short-Run Neutrality of Monetary Policy” Journal of Monetary
Policy. 11(1983): 3-24.

Anderson, R. W. “Some Determinants of the Volatility of Futures Prices.” Journal of
Futures Markets. 5(1985):332-348.

Baillie, R.T. “Econometric Tests of Rationality and Market Efficiency.” Econometric
Reviews. 8(2)(1989): 151-186.

Bamhart, S. W. “Effects of Macroeconomic Announcements on Commodity Prices.”
Amer. J. Agr. Econ. 71(1989):389-403.

Baur, RF. and PF. Orazem. “The Rationality and Price Effects of USDA Forecasts of
Oranges.” Journal of Finance. 49(1994): 681.

Bonham, CS. and DC. Dacy. “In Search of a ‘Strictly Rational’ Forecast.” Review of
Economics and Statistics. 73(May 1991): 245-53.

Brown, B. W. and S. Maital. “What Do Economists Know? An Empirical Study of
Experts Expectations.” Econometrica. 49(1981): 491-504.

Carlson, John A., “A Study of Price Forecasts.” Annals of Economic and Social
Measurement. 6(1977): 27-56.

Colling, P.L., and SH. Irwin. “The Reaction of Live Hog Futures Prices to USDA Hogs
and Pigs Reports.” American Journal of Agricultural Economics 72(1990):84-94.

Colling, P.L., Irwin, SH, and Carl R. Zulauf. “Weak and Strong-Form Rationality Tests

of Market Analysts’ Expectations of USDA Hogs and Pigs Reports.” Review of
Agricultural Economics l4(1992b):263 -70.

. “Informational Content of Government Hogs and Pigs Reports: Comment.”
Amer. J. Agr. Econ. 77(Aug. 1995): 698-702.

. “Reaction of Wheat, Corn and Soybean Futures Prices to USDA ‘Export
Inspections Reports’.” Review of Agricultural Economics . 18(1996): 127.

. “Futures Price Responses to USDA’s Cold Storage Report.” Agribusiness
13(1997d):393-400.

93

Cooley, T. F. and S. J. DeCanio. “Rational Expectations in American Agriculture, 1867-
1914.” Review of Economics and Statistics. (l976):9-17.

Ehrbeck, T. and R. Waldmann. “Why Professional Forecasters Biased? Agency Versus
Behavioral Explanations.” Quarterly Journal of Economics. 1(1996): 21-40.

Fama, E.F. “Efficient Capital Markets: A Review of Theory and Empirical Work.”
Journal of Finance. 25(1970):338-417.

Figlewski, S. and P. Wachtel. “The Formation of Inﬂationary Expectations.” Review of
Economics and Statistics. 63(1981): 1-10.

Fisher, B.S. “Rational Expectations in Agricultural Economics Research and Policy
Analysis.” American Journal of Agricultural Economics 64(1982):260-65.

Fortenbery, T. R. and D. A. Sumner. “The Effects of USDA Reports in the Futures and
Options Markets.” 13(1993): 157-174.

Garcia, P., Leuthold, R.M., Fortenbury, R., and Gboroton F. Sarassoro. “Pricing
Efficiency in the Live Cattle Futures Market: Further Interpretation and
Measurement.” American Journal of Agricultural Economics 70(1988): 162-169.

Gordon, J .D. The Distribution of Daily Changes in Commodity Futures Prices.
Washington DC: US. Department of Agriculture, ERS Technical Bulletin No.
1702, 1985.

Grossman, SJ. and Joseph E. Stiglitz. “Information and Competitive Price Systems.”
American Economic Review. 66(1976b):246-253.

. “On the Impossibility of Inforrnationaly Efficient Markets.” American
Economic Review. 70(1980):393-408.

 

Grunewald, 0., McNulty, MS, and Arlo W. Biere. “Live Cattle Futures Response to
Cattle on Feed Reports.” American Journal of Agricultural Economics.
75(1993):131-l37.

Hall, J. A., B. W. Brorsen, and SH. Irwin. “The Distribution of Futures Prices: A Test of
the Stable Paretian and Mixture of Normals Hypothesis.” J. Finance and

Quantitative Analysis. 24(1989): 105-1 6.

Hoffman, G., “The Effect of Quarterly Livestock Reports On Cattle and Hog Prices.” N.
Cent. J. of Agr. Econ. 2(1980):145-50.

94

 

Irwin, SH, and Cameron S. Thraen. “Rational Expectations in Agriculture? A Review of
the Issues and The Evidence.” Review of Agricultural Economics. 16(1994):133-
158.

Just, R. “The Impact of Less Data on the Agricultural Economy and Society.” Amer. J
Agr. Econ. 65(1993): 872-881.

Kenyan, D., K. Kling, J. Jordan, W. Scale, and Nancy McCabe. “Factors Affecting
Agricultural Futures Price Variance.” J. of Futures Markets. 7(1987):73-91.

Lovell, M.C. “Tests of the Rational Expectations Hypothesis.” American Economic
Review. 76(1986): 110-24.

Maddala, G.S. Limited-Dependent and Qualitative Variables in Econometrics. New York:
Cambridge University Press, 1986.

Malkiel, B.G. “Efﬁcient Market Hypothesis.” The New Palgrave; A Dictionary of
Economics. London: McMillan Press, 1982:2/ 120-23.

Mankiw, G.N., D.E. Runkle, and MD. Shapiro. “Are Preliminary Announcements of the
Money Stock Rational Forecasts?” Journal of Monetary Economics. 14(1984)
15-27. North Holland.

Mann, T. L. and R.J. Dowen. “Are Hog and Pig Reports Informative?” Journal Of
Futures Markets. 16(1996): 273-287.

. “The Impact of Proprietary-Public Information on Pork Futures.” Journal of
Futures Markets, l7(l997):417-432.

McDaniel, K. L. and SR. Koontz. “Forecasting Placements, On-Feed, and Marketings
Numbers in the USDA Seven-State Cattle On Feed Report.” Oklahoma State
University Department of Agricultural Economics Current Farm Economics.
May(1995).

Miller, KB. “The Relational Between Volatility And Maturity in Futures Contracts,”
Commodity Markets and Futures Prices. Ed. R.M. Leuthold. Chicago: Chicago
Mercantile Exchange, 1979.

Miller, S. “The Response of Futures Prices to New Market Information: The Case of Live
Hogs.” S.J. Agr. Econ. 2(1979): 67-70.

Mishkin, S. Fredrick. “Are Market Forecasts Rational?” American Economic Review.
71, June (1981): 295-306.

95

Muth, F. John. “Rational Expectations and the Theory of Price Movements.”
Econometrica. 29(1961): 315-35.

Pearce, D.K., and V.V. Roley. “Stock Prices and Economic News.” 1. Business.
58(1985):49—67.

Rosett, R.N. “Estimation of the Two-Limit Probit Regression Model.” Econometrica.
43(1975): 141-46.

Runkle, D.E. “Are F arrowing Intentions Rational Forecasts?” American Journal of
Agricultural Economics. Aug.( 1 99 l ): 594-600.

Samuelson, P.A. “Proof That Properly Anticipated Prices Fluctuate Randomly.”
Industrial Management Review. 6(1965):41-49.

Schoeder, T. J ., J. Blair, and J. Mintert. “Abnormal Returns in Livestock Futures Prices
Around USDA Inventory Report Releases.” N. Central J. Agr. Econ. 12(1990):
293-304.

Sumner, DA. and Mueller. “Are Harvest Forecasts News?” Amer. J. Agr. Econ.
71(Feb. 1989): 1-8.

Taylor, M. P. “What Do Investment Manager Know? An Empirical Study of
Practitioners’ Predictions.” Economica. 55(1988): 185-202.

96