Full text of Working Papers (Federal Reserve Bank of Atlanta) : Evaluating Wall Street Journal Survey Forecasters : A Multivariate Approach, Working Paper 2002-8a

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Evaluating Wall Street Journal Survey Forecasters:
A Multivariate Approach
Robert Eisenbeis, Daniel Waggoner, and Tao Zha
Working Paper 2002-8a
July 2002

Working Paper Series

Federal Reserve Bank of Atlanta
Working Paper 2002-8a
July 2002

Evaluating Wall Street Journal Survey Forecasters:
A Multivariate Approach
Robert Eisenbeis, Daniel Waggoner, and Tao Zha
Federal Reserve Bank of Atlanta

Abstract: This paper proposes a methodology for assessing the joint performance of multivariate forecasts of
economic variables. The methodology is illustrated by comparing the rankings of forecasters by the Wall Street
Journal with the authors’ alternative rankings. The results show that the methodology can provide useful insights
as to the certainty of forecasts as well as the extent to which various forecasts are similar or different.
JEL classification: C53
Key words: Wall Street Journal, joint forecast, probability, ranking, correlation, variance, multivariate
assessment

Bevin Janci provided invaluable research assistance in gathering and organizing the data used in this article. The views
expressed here are the authors’ and not necessarily those of the Federal Reserve Bank of Atlanta or the Federal Reserve
System. Any remaining errors are the authors’ responsibility.
Please address questions regarding content to Robert Eisenbeis, senior vice president and director of research, Research
Department, Federal Reserve Bank of Atlanta, 1000 Peachtree Street, N.E., Atlanta, Georgia 30309-4470, 404-498-8824, 404498-8956 (fax), robert.a.eisenbeis@atl.frb.org; Daniel Waggoner, research economist, Research Department, Federal
Reserve Bank of Atlanta, 1000 Peachtree Street, N.E., Atlanta, Georgia 30309-4470, 404-498-8278, 404-498-8810 (fax),
daniel.f.waggoner@atl.frb.org; or Tao Zha, assistant vice president, Research Department, Federal Reserve Bank of Atlanta,
1000 Peachtree Street, N.E., Atlanta, Georgia 30309-4470, 404-498-8353, 404-498-8956 (fax), tao.zha@atl.frb.org.
The full text of Federal Reserve Bank of Atlanta working papers, including revised versions, is available on the Atlanta
Fed’s Web site at http://www.frbatlanta.org. Click on the “Publications” tab and then “Working Papers” in the navigation
bar. To receive notification about new papers, please use the on-line publications order form, or contact the Public Affairs
Department, Federal Reserve Bank of Atlanta, 1000 Peachtree Street, N.E., Atlanta, Georgia 30309-4470, 404-498-8020.

Evaluating Wall Street Journal Survey Forecasters:
A Multivariate Approach
I.

Introduction
Economic forecasting usually involves making simultaneous predications of key

financial and real macro economic variables at intervals of months, quarters or even years
into the future. Yet even when models are estimated simultaneously, forecasters typically
focus on the out-of-sample accuracy of individual variables or dimensions of economic
performance and not on the overall accuracy of their description of the economy.
Bluechip Economic Indicators collects forecasts from a panel of experts monthly, and the
forecasted values of many series are presented, but no summary measures of joint
accuracy are provided. In contrast, twice a year at the beginning of January and July, the
Wall Street Journal (WSJ) surveys a group of forecasters for their forecasts of several
key macroeconomic variables designed to characterize what the performance of the
economy will be. The Journal publishes the individual forecasts and does provide a
ranking of a few of the top forecasters, based on how close the forecasts of the variables
are to their realized values. The actual methodology used to provide these rankings has
changed over time, but at present it simply ranks the forecasters on the sum of the
weighted absolute percentage deviation from the actual realized value of each series,
where the weight for each series is simply the inverse of the actual realized value of the
series. This performance assessment method may become distorted, and even undefined,
when the realized value is close to or equal to zero. Moreover, it does not consider the
correlations in the data among the variables being forecast. This latter consideration is

important because accuracy should reflect internally consistency in predicting the
performance of the economy and not merely good luck on one particular dimension.
In this paper we propose a methodology, which not only yields a measure of joint
forecast performance, but also provides a single measure of how similar a joint forecast is
to those of other forecasters. The method also allows us to assess the collective forecast
accuracy of all the forecasters and the accuracy of individual forecasters over time. The
procedure is not even dependent upon having all forecasters represented in each forecast
period. Finally, it provides some indication of how tightly the forecasts are clustered
around the realized values, and can be used to compare judgmental forecasts as well as
those of formal econometric models. The next section describes the proposed
methodology and subsequent sections illustrate its use with data from the Wall Street
Journal.

II. Methodology
Two considerations are important in evaluating the accuracy of a joint forecast of
several economic variables. First, some variables are inherently less stable than others
and thus are harder to forecast than others. For instance, the unemployment rate is both
persistent and does not vary significantly from quarter to quarter. Hence, it is easier to
predict on average than a highly volatile variable like GDP growth. Whatever measure
used to compare forecasts should take into account this difference in variability by
penalizing forecast errors in easy-to-forecast variables more than similar size errors in
hard-to-forecast variables.

Second, because many important economic variables are correlated; certain
combinations of these variables are more or less likely to occur together than others.
For instance, because the CPI and short-term interest rates tend to be positively
correlated, any model that reflects this underlying structure in the data should generate
forecast errors in these two variables that would also likely be positively correlated. A
forecast that over-estimated CPI inflation while under estimating interest rates should be
penalized more than a forecast that over estimated both. That is, going out on a limb and
missing on a key dimension that did not reflect the underlying data structure should be
penalized more because such errors are less likely, on average.
The most common measure of variability is variance and the corresponding measure
of correlation between two variables is covariance. In a multivariate setting the
variance-covariance matrix can be formed with the variance of each variable along the
diagonal and the covariances of the variables in the off diagonal entries. This variancecovariance matrix Ω can be used to form a multivariate distance chi-squared statistic
commonly used in statistical inference, if we are willing to assume that the forecast errors
are multivariate normally distributed with mean zero. The statistic is of the form:
χˆ 2 = ( yˆ t − y t )′ Ωt−1 ( yˆ t − y t ) : chi2 ( n ) ,

where yˆ t is the time t forecast of a vector of economic variables, y t is the realized value
of the forecast vector, and Ω −t 1 is the inverse of the variance-covariance matrix, and n is
the number of variables in y t . is distributed as chi2 (n).
The remaining problem is to devise an estimate of the variance matrix, which we
approach by decomposing it into more tractable components. The exact details are
described in Appendix 1.

Given a vector of forecast errors associated with a particular forecast, we can
compute the p-value for its associated chi2 and call it an “accuracy score.” The summary
measure provided by the computed accuracy score has several useful properties. First, it
is a probability that is invariant to the underlying scale of the errors. Second, it can be
interpreted as a measure of how similar, or close, the joint forecast is to the realized
values in the economy. Third, we can go even further and interpret the p-value,
expressed as a percentage, associated with a particular forecast as indicating that it is
closer to the true value than p % of all possible forecasts. Fourth, it can be used to
compare and rank forecasts. And finally, the distribution of forecasts across forecasters
can be compared both within a forecast period and across periods. The next section
illustrates how the methodology can be used in a simple two-variable case, and then it is
extended in Section IV to the entire set of Wall Street Journal forecast variables.

III. Empirical Illustration – Two variable case
To illustrate what the methodology is doing, we first present a two dimensional
forecast example in Chart 1. The Wall Street Journal publishes semiannually the
forecasts of between 30 and 50 economic forecasters, who submit their projections for
many key economic variables. Chart 1 plots just the forecasts of two variables from the
July 1999 Journal survey: the 3 month T-bill for December 31, 1999 and the dollar/yen
exchange rate for December 31, 1999.
Several features of this chart are noteworthy. First, the ellipse, centered around
the true, realized values being forecast, represents the two-third probability surface
showing how similar forecasts lie. Any forecast lying on this ellipse can be said to be

closer to the true value (the gray square) than two thirds of all possible forecasts.
Forecasts on an inner concentric ellipse (not drawn) outperform those lying on the twothird ellipse, and forecasts outside the ellipse under-perform those on the ellipse.
It will also be noticed that the probability surface is not a circle, indicating that
dispersions (variances) are not equal. Furthermore, because the ellipse is tilted upward,
there is a positive correlation between the two variables, which is 0.48 according to the
bottom panel of Table 2. The methodology considers these correlations in its calculation
of the measure of joint forecast accuracy. Finally, it is clear that while the forecasts are
generally fairly tightly grouped, most are outside the two-third ellipse and only two are
reasonably close to the joint realization. Several forecasts are reasonably close on the 3month T-bill rate, but most of the forecasts show significant errors on the dollar/yen
exchange rate. Because the T-bill rate had a smaller variance than the exchange rate,
errors on that dimension will be less severe than errors on the exchange rate. But the best
forecasters did a substantially better job on both dimensions and stand out above the rest
of the pack. The distributions of individual forecasters’ joint forecast accuracy scores for
this two variable example are shown in Chart 2. It can be clearly seen that three of the
forecasters had a much higher score than did the other forecasters. Moreover the
distribution of the forecasts is quite spread out. As will be shown in the next section, the
pattern of these probabilities varies significantly from forecast period to forecast period.

III. Empirical Results – Multivariate
As indicated earlier, forecasters usually predict many economic variables, and the
proposed methodology for assessing accuracy is robust enough to handle a large number

of variables. We explore the properties of the Wall Street Journal forecasts using the
proposed methods. The variables included in the forecast survey collected by the Wall
Street Journal have changed over time, and again, the proposed methodology can take
this into account by simply dropping the appropriate row and column from the variancecovariance matrix for each variable not in the sample. Because of the sheer volume of
data, only the results for a few of top forecasters in one Wall Street Journal survey will
be discussed (Chart 3), but we will provide some summary information of the key
features of the forecasts over time (Charts 4 and 5 and Table 3).
Chart 3 presents the joint forecast accuracy scores and each forecaster’s rank for a
selected few of the top forecasters according to our method (labeled EWZ rank) together
with the rankings according to the WSJ’s selection criteria for July 1999. Except for the
two best forecasters, Fosler and Sinai, the EWZ ranking is different from the Wall Street
Journal ranking. The Journal’s placement of Ramirez third compares with our
placement of only 19th , and Orr ranks sixth compared with our placement of 44th . This
raises some interesting questions concerning the differences in the forecasts on the
different variables and how these differences are weighted.
Table 1 displays relevant forecasts and their realized values at the time the rankings
were done. 1 For this survey, the Journal evaluation first computes the absolute value of
the difference between forecast and actual value for each variable and then weights the
error by the inverse of the actual value. The smaller the weighted sum is, the higher a
forecaster is ranked. The weights vary over time as the values of the variable change.
However, if the actual value of a variable is close to zero (which is not uncommon for a

1

In some instances, data are revised. The tables here include the real-time data available to the Journal at
the time the evaluations were made.

variable like GDP growth rate), the weight assigned to the GDP error becomes arbitrarily
large, and can be misleading.
In the absence of correlations between forecast variables, our methodology is similar
to the Journal method except that the weights applied to the errors are equal to the
inverse of the forecast variance for each variable rather than the value of the variable
itself. According to the covariance matrix reported in top panel of Table 2, the weights
assigned to the squared differences between forecast and actual values would be, ignoring
the correlations for the time being, 5.39 for the 3-month Treasure bill rate, 8.90 GDP
growth, 21.00 CPI inflation, 45.12 the unemployment rate, 19.46 the 30-year Treasure
bond yield, and 0.13 the exchange rate. The weight assigned to forecast errors of the
unemployment rate is large because this series does not vary as much, and the forecast
variability is relatively low. The weight given to errors in the exchange rate is small
because this series is more volatile and hard to forecast.
The following examples illustrate the importance of using the variance as a weight as
well as taking account of the correlations in the variables being forecast. Consider first
that Ramirez (Table 1) is ranked 19th by our method because the errors in her forecasts of
both 30-year Treasury bond yield (7%) and the exchange rate for yen (125 ) compared
with the realized values (6.48% and 102) are large relative to the variances of these
variables. Similarly, Orr is placed at 44th by our ranking largely because of the large
error in his exchange rate forecast of (135) relative to the actual value (102). Reynolds is
ranked Number 5 by our method as compared to 20th by the Journal method, mainly
because of the negative correlation, -0.34, between GDP and unemployment (see the
bottom panel of Table 2). He under-forecast the average annual rate of GDP growth in

the 3rd quarter of 1999 and over-forecast the unemployment rate in November 1999. The
negative correlation implies that this kind of forecast error is expected, and thus should
not punished as much. The case of Thayer, ranked sixth by our method and 37th by the
Journal method is more complicated. He over-forecast the exchange rate and underforecast the 30-year Treasury bond rate, which is contrary to the positive correlation of
forecast errors of these two variables. But his under-forecast of the 3-month Treasury bill
rate, GDP growth, the 30-year Treasure bond rate, combined with his over-forecast of the
unemployment rate, are consistent with the pair wise negative correlations reported in the
top panel of Table 2. Furthermore, his under-forecast of both interest rates is consistent
with the positive correlation of forecast errors in both interest rates.
The methods we are proposing can also be used to explore the forecast performance
of individual forecasters over time. Charts 4 and 5 compare the forecasts and model
rankings for two particular forecasters: Ramirez and Yardeni. Ramirez (Chart 4) had a
mean EWZ rank of 27.3 and a mean accuracy score of 58.8%, meaning that on average
she would be expected to out-perform about 59% of the forecasters.
Yardeni (Chart 5) had a mean EWZ rank of 27.6 and a mean accuracy score of
57.4%. Both these forecasters had similar performance, and like Rameriz, Yardeni was
also recognized for his forecasting performance, but on two rather than only one
occasion. For the July 1998 survey he was ranked 6th by the Wall Street Journal,
whereas our method would have ranked him 23rd, and for the July 1999 survey he was
ranked 1st by the Journal whereas we would have ranked him 8th .
Table 3 presents the mean EWZ rank and average accuracy scores, together with their
respective standard deviations, for those forecasters that appeared in the Wall Street

Journal forecast for at least 4 periods between July of 1986 and January of 2002. The
scores and ranks for January 2002 are also provided. The data are sorted, so that those
active forecasters who provided a survey for January of 2002 appear first. Among the
active forecasters, several performed quite well. Soss and Kudlow both have low mean
ranks, but these were largely accumulated in the late 1980’s and early 1990’s, and may
not reflect their current expected performance. In fact, for the January 2002 survey,
Kudlow dropped to 53 rd; again this illustrates the difficulty of maintaining performance
over time. Hoffman not only had the fifth lowest mean rank over a very long period of
time, but also had a high mean accuracy score with relatively lower standard deviations
on both, and especially on his mean rank. Considering the entire table, the people with
the superior performance record tend to be those whose forecasts covered a short period
of time in the early to mid-1980s. Interestingly, this was a relatively more volatile period
than the 1990’s, but also it is worth noting that the variables forecast were different and
the number of variables was smaller.
The charts on individual performance can also be used to highlight those instances
when forecasters take extreme positions. Yardeni made a big point about his concern for
Y2K and the consequences if the US and the rest of the world didn’t make the necessary
preparations. His concerns were reflected in his forecasts in Chart 5 for the July 1999
and January 2000 Journal surveys: the accuracy scores are extremely low when
compared with both those of other forecasters and how the economy actually performed.
But not all forecasting accuracy problems are due to taking extreme positions. This is
illustrated by the lower performance in terms of accuracy scores for all forecasters in
January 1995 and in July 1990 . This highlights the difficulty in predicting turning

points. All forecasters had trouble with turning points, which is shown in Chart 6.
containing the mean accuracy scores for all the forecasters, as well as the top and bottomranked 5 forecasters. All forecasters made large errors in their January 1995and July
1990 forecasts. More recently, forecasters made big errors in their forecasts for January
2001, and clearly also had some difficulty with their January 2002 forecasts, as the
strength of the economy was under-estimated. In addition, while there was a lot of
agreement among the forecasters January 2001 as the distribution of the forecasts was
reasonably tight, all were also systematically off the mark.
This chart also illustrates that at times there is more unanimity among forecasters than
at others. For example the bottom 5 and top 5 forecasters were closer to each other in
some periods than in others, suggesting that the variation in the forecasts may serve as an
indication of how much uncertainty there may be about where the economy is going. The
dispersions, for example, widened considerably during the Asian crisis in the summer of
1997.

IV. Conclusion
In this paper we have offered a systematic approach to evaluating a forecaster’s
performance relative to others, and illustrated the methodology in the context of specific
examples. Our approach formalizes a way of assessing forecast accuracy, but could be
applied to a number of different multivariate performance assessment problems. One
may differ on how to estimate the variance-covariance matrix but once it is reasonably
approximated, our approach provides not only the ranking results but also the probability

of how close to the actual data a particular forecast is in comparison with all other
potential forecasts.

Appendix 1

When we make a forecast today of the values of a set of economic variables at some
points in the future, we would not expect our forecasts to be perfect even if we had
perfect knowledge of the inner workings of the economy. There are always events, such
as political or natural disasters, that are impossible to predict and effect the economy.
More formally, we could not give perfect forecasts even if we knew the “correct” model
of the economy. We will use the notation Ω Et to denote the variance-covariance matrix of
the forecast errors inherent in the economy and Ω Ft to denote the variance-covariance
matrix of the forecast errors made by individual forecasters. If y t is the n-vector of
variables to be forecast and yˆ t is the forecast of y t , we assume that both y t and yˆ t have
the same mean yt , the variance-covariance matrix of y t is Ω Et , and the variancecovariance matrix of yˆ t is Ω Ft .
If we make a rather mild assumption that forecast errors inherent in the economy
( y t − y t ) are uncorrelated with those made by forecasters ( yˆ t − y t ), then the total variance
matrix of the forecast errors ( yˆ t − y t = ( yˆ t − y t ) − ( y t − y t ) ) will be Ω t = Ω

E
t

+ Ω tF .

The advantage of having forecasts from many different forecasters is that the crosssectional variance-covariance matrix gives us an estimate of Ω Ft . Note that this does not
depend on having a time series of the individual forecasters. Forming an estimate of Ω Et
is more delicate. For the exercises in this paper, we use the variance-covariance matrix
estimated from the reduced-form Bayesian dynamic, multiple-equation model described
in Robertson and Tallman (1999). The covariance matrix Ω Et is simulated with ten

thousand simulations whose computing time takes about 95 minutes for each survey on a
Pentium III 800 PC. Intuitively, this model incorporates the features of random walk,
unit root, and cointegration inherent in the data, and thus it offers a good benchmark.
The sample used by the model begins at January 1959. From January 1961 to February
1977, the 30-yr Treasury bond yield is replaced by the 20-yr bond yield; from January
1959 to December 1960, it is replaced by the 10-yr bond yield. As for the exchange rate
between Euro and US$ from January 1959 to September 1979, it is extrapolated from the
January 1980-September 2001 regression of the synthetic Euro exchange rate with US$
on the Mark exchange rate with US$. We take account of both parameter (model)
uncertainty and randomness in future shocks in simulating the variance-covariance matrix
Ω Et at each forecast date (Waggoner and Zha, 1999). Take the July 1999 Journal survey

as an example. When the survey was published, forecasters had only the data released in
June 1999. That means that they had some data (such as financial data) up to June 1999
and some data (such as CPI) up to May 1999 while GDP was available only up to the first
quarter of 1999. To be comparable with the information set used by all forecasters, the
model uses the data set as though it was available at the end of June 1999.

Appendix 2
We have spent a lot of time describing the estimation of the variance-covariance
matrix of forecast errors Ω Et using simulation methods, which ideally should be reestimated each time a forecast is made for a new period. It turns out, however, that
experiments with re-estimation of the matrix is not really necessary. The rank
correlations for the forecast rankings with and without re-estimation are so high that the
rankings are reasonably robust to changes in this matrix over time. For example,

comparing rankings one year apart the Spearman rank correlation is .99, and even 5 years
apart is .98. Hence we have supplied the current estimate that could be used by any
interested party for some time into the future, when combined with the variancecovariance matrix Ω Ft estimated from the Wall Street Journal published forecast data.
These two matrices can be combined as shown in Appendix 1, so that any interested party
could replicate our rankings for the forthcoming survey. Below is the estimated Ω Et
matrix
Variance-Covariance Matrix
T-Bill
DGP
CPI
CUR
T-Bond
Yen/US$
US$/Euro

T-Bill
3.2324
0.79906
-0.3147
-0.32571
1.1581
9.6655
-0.0525

DGP
0.79906
8.3368
0.068871
-0.61587
0.31782
-2.0202
-0.004986

CPI
-0.3147
0.068871
0.7801
0.02141
-0.11968
-3.0315
0.021998

CUR
-0.32571
-0.61587
0.02141
0.29975
-0.12797
-0.067084
0.0075195

T-Bond
1.1581
0.31782
-0.11968
-0.12797
0.96538
5.0814
-0.024588

Yen/US$
9.6655
-2.0202
-3.0315
-0.067084
5.0814
94.593
-0.29445

US$/Euro
-0.0525
-0.004986
0.021998
0.0075195
-0.024588
-0.29445
0.0038811

The variables in the above table are defined as follows:

T-Bill 3 = 6-month-ahead forecast of 3-Month Treasury Bills, Secondary Market (% p.a.)
DGP = One-quarter ahead forecast of quarterly Real GDP growth
CPI = 5-month-ahead forecast of annual CPI inflation rate (prior to 12 months ago)
CUR = 5-month-ahead forecast of Civilian Unemployment Rate (SA, %)
T-Bond = 6-month-ahead forecast of 10-Year Treasury Bond Yield at Constant Maturity
(% p.a.)
Yen/US$ = 6-month-ahead forecast of Yen/US$ exchange rate
US$/Euro = 6-month-ahead forecast of US$/Euro exchange rate

References

Eisenbeis, Robert A., and Robert B. Avery, 1973. “Two Aspects of Investigating Group
Differences in Linear Discriminant Analysis.” Decision Sciences, Vol 4, 487-493.
Robertson, John C., and Ellis W. Tallman, 1999. “Vector Autoregressions: Forecasting
and Reality.” Federal Reserve Bank of Atlanta Economic Review 84 (First Quarter): 418.
Waggoner, Daniel F., and Tao Zha, 1999. “Conditional Forecasts in Dynamic
Multivariate Models.” Review of Economics and Statistics 81(4) (November), 639-651.

Forecasters

Fosler

Daane

7

Levy

Lonski

Gross/McCulley

Synnott III

6

Berson

Ratajczak

Hymans

Hummer

5

Blitzer

Sohn

Thayer

4

Wyss

Laufenberg

Mueller

Zandi

3

Coons

Yen/Dollar Exchange Rate, Dec 1999 .
100

Englund

Bussman

Shepherdson

2

Walter

Harris

Smith

Yamarone

Camilli

Score (%)

Chart 1 - Individual Forecasts for July 1999 WSJ Survey

160

150

140

130

120

110

90
Actual Value

80

70

60
8
9

3-Month T-Bill Rate, Dec 1999

Chart 2 - Accuracy Scores July 1999 WSJ Survey

100

90

80

70

60

50

40

30

20

10

0

Chart 3 - Ranking and Scores for July 1999 WSJ Survey
100

50

44

90
80

40

60

30

50

19

40

Rank

Score (%)

70

20

30
7
4

10

4

7

6

5

3

1 1

10

6

5

3

2 2

0

0
Cahn

Dudley

Fosler

Littmann

EWZ Rank

Ramirez Reynolds

WSJ Rank

Sinai

Thayer

Accuracy Score

Chart 4 - Rameriz's Forecast
Performance Over Time

100
90
80
70
60
50
40
30
20
10
0

60
50
40
30
20
10

Date
EWZ Rank

WSJ Rank

Accuracy Score

Mean Rank

Jul-00

Jan-00

Jul-99

Jan-99

Jul-98

Jan-98

Jul-97

Jan-97

Jul-96

Jan-96

Jul-95

Jan-95

Jul-94

Jul-93

Jan-93

0
Jul-92

Score (%)

Mean Accuracy Score=58.8
Mean EWZ Rank = 27.3

Orr

Rank

20

Jul-86
Jan-87
Jul-87
Jan-88
Jan-89
Jul-89
Jan-90
Jul-90
Jan-91
Jul-91
Jan-92
Jul-92
Jan-93
Jul-93
Jan-94
Jul-94
Jan-95
Jul-95
Jan-96
Jul-96
Jan-97
Jul-97
Jan-98
Jul-98
Jan-99
Jul-99
Jan-00
Jul-00

Score (%)
100
90
80
70
60
50
40
30
20
10
0
30

20

10

0

Date

EWZ Rank
WSJ Rank
Accuracy Score
Mean Rank

Rank

Mean Acuracy Score = 57.4
Mean EWZ Rank = 27.6

Chart 5 - Yardeni's Forecast
Performance Over Time
60

50

40

Table 1 - Forecast Performance for Top Forecasters July 1999 WSJ Survey
3-Month GDP
Civilian
30-Year Yen/Dollar
T-Bill Growth
CPI
Unemployment T-Bond Exchange
Accuracy
EWZ Rank Score WSJ Rank
Rate
Rate Inflation
Rate
Rate
Rate
Cahn
4.90
3.63
2.4
4.0
6.50
120
7
73.84
4
Dudley
5.00
3.27
2.5
4.0
5.80
115
4
86.85
6
Fosler
5.25
3.73
2.3
4.0
6.40
110
1
99.64
1
Littmann
4.75
3.27
2.7
4.2
6.15
116
3
87.43
7
Orr
5.45
3.80
2.5
4.2
6.35
135
44
12.73
5
Ramirez
5.40
3.67
2.5
4.0
7.00
125
19
52.31
3
Reynolds
5.20
2.57
2.8
4.7
6.40
117
5
80.29
20
Sinai
5.23
3.13
2.5
4.1
6.55
110
2
99.40
2
Thayer
4.40
2.80
2.2
4.4
5.20
112
6
79.08
37
Actual data 5.32
3.70
2.6
4.1
6.48
102

Table 2 - July 1999 WSJ Survey

3-Month T-Bill
GDP Growth
CPI
Unemployment
30-Year T Bond
Yen/Dollar

3-Month
T-Bill Rate
3.27
0.35
-0.18
-0.36
0.87
10.26

Variance-Covariance Matrix
Civilian
GDP
CPI
Unemployment
Rate
Growth Rate
Inflation
0.35
-0.18
-0.36
1.98
0.03
-0.29
0.03
0.84
0.02
-0.29
0.02
0.39
0.12
0.13
-0.11
-0.04
-2.44
0.06

30-Year
T-Bond Rate
0.87
0.12
0.13
-0.11
0.91
3.09

Yen/Dollar
Exchange
Rate
10.26
-0.04
-2.44
0.06
3.09
140.16

30-Year
T-Bond Rate
0.51
0.09
0.14
-0.18
1.00
0.27

Yen/Dollar
Exchange
Rate
0.48
0.00
-0.22
0.01
0.27
1.00

Correlation

3-Month T-Bill
GDP Growth
CPI
Unemployment
30-Year T Bond
Yen/Dollar

3-Month
T-Bill Rate
1.00
0.14
-0.11
-0.32
0.51
0.48

GDP
Growth Rate
0.14
1.00
0.02
-0.34
0.09
0.00

CPI
Inflation
-0.11
0.02
1.00
0.03
0.14
-0.22

Civilian
Unemployment
Rate
-0.32
-0.34
0.03
1.00
-0.18
0.01

Table 3

Overall Forecast Performance For Those Forecasters With Four or More Available
Forecasts

Periods
Average
Forecasters
Covered
Score
July 88 - Jan 92, Jan 94, Jan
01 - Jan 02
Kudlow**
69.6
Resler
Jan 86 - Jan 02
68.3
Jan 88 - Jan 94, July 01 - Jan
02
Soss*
68.2
July 00 - Jan 02
DiClemente
67.4
Hoffman
Jan 88 - Jan 02
66.5
Jan 86- Jan 02
Levy
63.7
Jan 89 - July 99, July 01 - Jan
02
Wyss
63.6
Harris
July 86 - Jan 02
62.8
Hymans
July 86 - Jan 02
62.3
July 98 - Jan 02
Swonk
61.7
Littmann
Jan 93- Jan 02
61.5
July 94 - Jan 02
Perna
61.4
Jan 86 - Jan 02
Sinai
61.0
Berson
Jan 90-Jan 02
60.7
Jan 90 - Jan 02
Rippe
60.7
July 88 - Jan 02
Daane
60.5
Karl
Jan 94 - Jan 02
60.5
Jan 86 - Jan 02
Hyman/Lazar
59.7
Jan 94 - Jan 02
Cosgrov e
59.7
Ramirez
July 92 - Jan 02
59.4
July 94 - Jan 02
Sy nnott
59.4
Jan 94-Jan 02
Berner/Greenlaw
59.2
Wilson
Jan 86, Jan 01 - Jan 02
58.4
July 94, July 99 - Jan 02
Sterne
57.4
Muell er
July 91- Jan 02
56.3
McCulley
July 94- Jan 02
55.0
Jan 94 - Jan 02
Coons
54.9
Hummer
Jan 93 - Jan 02
54.7
Thay er
July 99 - Jan 02
54.3
Jan 94- Jan 02
Lonski
53.9
Zandi
July 95 - Jan 02
52.9
Dudley
Jan 96 - Jan 02
51.7
Jan 98 - Jan 02
Sohn
51.4
Smith
Jan 87 - Jan 02
49.0
Herrick
July 94 - Jan 02
48.4
July 99 - Jan 02
Shepherdson
47.4
Laufenberg
July 95 - Jan 02
47.0
Fosler
Jan 91 - Jan 02
45.6
July 97 - Jan 02
Steinberg
45.6
Ev ans
Jan 94 - July 96
45.5
Jan 99 - Jan 02
Gallagher
45.1
July 93-Jan 02
Ally n
44.3
Orr
July 99 - Jan 02
40.0
July 99 - Jan 02
Camilli
39.9
July 99 - Jan 02
Yamarone
39.8
Shilling
Jan 86 - Jan 02
39.7
July 98 - Jan 02
Wesbury
35.9

Standard
Deviation
of Scores

Average
Rank

Standard
Deviation
of Ranks

Number of
Forecast
Surveys

Jan 2002
Survey
Score

Jan 2002
Survey
Rank

38.78
28.28

19.75
18.15

16.42
12.17

12
33

11.4
86.9

53
3

26.74
44.60
26.72
29.38

23.21
14.25
19.89
22.18

15.20
19.81
9.44
10.79

14
4
28
33

66.1
76.3
33.5
33.5

15
9
38
37

27.86
31.35
32.81
29.30
32.34
31.33
32.15
28.88
31.22
30.32
31.63
28.46
32.99
28.85
31.34
31.59
30.67
29.93
31.30
30.93
31.12
28.61
21.83
31.81
26.06
37.99
32.25
35.65
30.74
28.87
35.45
34.17
29.66
32.88
28.92
31.77
36.90
37.80
33.62
33.71
33.35

25.48
21.68
21.34
19.57
22.72
21.25
25.29
22.52
24.28
24.33
24.25
25.68
25.65
26.58
22.13
28.09
16.25
22.00
26.68
25.87
26.65
28.37
21.17
27.82
30.10
32.38
24.89
32.72
32.94
25.50
29.57
32.57
29.40
33.71
33.43
34.24
31.00
29.17
29.33
32.90
33.02

12.95
12.59
14.20
16.97
14.83
14.48
15.52
13.24
11.23
14.33
13.74
12.13
16.48
16.48
16.80
18.23
14.29
13.35
14.94
13.23
15.45
12.12
15.05
15.40
11.83
19.16
17.25
17.38
15.19
19.17
20.68
19.64
14.99
17.50
12.50
12.53
15.85
16.94
18.82
16.94
20.18

29
31
29
7
18
16
31
25
25
27
16
31
17
19
16
17
4
9
19
15
17
19
6
17
13
13
9
29
16
6
14
23
10
7
7
17
6
6
6
31
7

60.3
42.7
80.8
71.9
82.8
37.4
1.9
26.1
76.8
20.6
46.6
72.4
62.9
68.8
69.8
65.6
29.6
89.0
49.1
49.4
40.5
65.5
21.5
56.2
52.7
65.5
12.6
26.0
39.3
25.5
92.3
32.7
45.7
18.3
15.1
36.7
40.8
44.5
21.5
3.7
30.2

20
30
6
12
5
34
55
43
8
49
26
11
19
14
13
16
41
2
25
24
32
17
47
22
23
18
52
44
33
45
1
39
27
50
51
35
31
29
48
54
40

** Kudlow's record was accumulated ov er largely the late 1980s and early 1990s. His ranks in Jan. and July of 2001 were 41st and 5th, respectiv ely .
* Soss's record was accumulated over largely t he late1980s and early 1990's. His rank in July 2001 was 51st.

Table 3 (cont.)

Overall Forecast Performance For Those Forecasters With Four or More
Available Forecasts

Forecasters
McDevitt
Angell
Englund
Cahn
Blitzer
Bussman
Moskow itz
Brow n
Yardeni
Walter
Platt
Reynolds
Braverman
Worseck
Williams
Karczmar
Boltz
Leisenring
Bostian
Palash
Straszheim
Kellner
Dederick
Laughlin
Evans
Reaser
Robertson
Wahed
Kahan
Keran
Ciminero
Gramley
Vignola
Barbera
Hoey
Eickhoff
Lerner
Jones
Melton
Michaelis
Jordan
Schott
Cooper
Pate
Nathan
Hunt
Maude
How ard

Standard
Standard Number of
Periods
Average Deviation Average Deviation Forecast
Covered
Score
of Scores Rank
of Ranks Surveys
July 96- July 01
59.0
29.93
26.89
13.99
9
July 94-July 01
58.8
34.99
23.67
16.25
15
Jan 94 - July 01
46.2
26.24
33.27
13.87
15
July 95 - Jan 01
59.2
29.12
27.40
14.01
10
July 94-Jan 01
55.8
30.05
27.64
17.82
14
July 97 - Jan 01
44.7
29.83
32.83
8.47
6
Jan 86- July 99, July 00
65.2
29.45
19.79
10.53
28
Jan 92 - July 00
59.6
33.12
27.39
17.80
18
July 87 - July 00
57.0
30.35
27.25
15.05
28
Jan 97 - July 00
54.9
35.58
27.75
22.77
8
July 88 - Jan 00
68.7
28.00
21.26
13.20
23
July 86 - Jan 00
65.0
29.56
24.59
14.42
27
Jan 86 -Jan 99
71.6
27.31
20.00
13.88
28
Jan 89 - Jan 99
62.6
36.24
24.11
18.64
19
Jan 94 - Jan 99
56.4
31.40
28.09
12.77
10
July 93 - July 98
47.7
32.38
36.00
14.82
10
Jan 86 -Jan 98
62.0
30.27
25.24
15.59
25
July 87- July 97
73.7
22.95
19.52
11.69
21
Jan 94 -July 97
64.4
29.24
23.63
17.11
8
July 95 - July 97
60.5
20.52
34.80
19.06
5
July 86 - Jan 97
70.7
26.55
16.05
10.30
21
Jan 86 - Jan 97
62.7
33.24
24.14
13.55
22
July 86 - July 96
72.1
29.66
18.62
12.13
21
July 94- July 96
58.6
28.05
24.20
15.94
5
Jan 94 - July 96
50.0
33.43
31.00
17.48
6
July 92 - Jan 96
72.8
25.62
15.50
8.42
8
Jan 86 - Jan 96
69.4
30.73
17.45
10.36
20
July 89 - Jan 96
65.4
34.91
22.42
12.92
12
Jan 87 - Jan 96
62.0
35.20
22.24
15.11
17
Jan 94- Jan 96
51.3
21.19
27.40
17.99
5
Jan 94, Jan 95 - Jan 96
48.2
37.25
25.50
14.71
4
Jan 87-Jan 95
72.5
31.33
16.00
11.29
17
July 92 - July 94
66.9
32.02
24.20
15.61
5
Jan 90-Jan 94
56.2
32.56
29.11
8.65
9
Jan 86 - Jan 94
54.0
33.66
25.00
11.58
16
July 91 - July 93
93.2
10.73
10.20
11.65
5
July 86- July 93
62.5
26.52
23.73
13.12
15
July 86 - Jan 93
71.1
28.72
17.47
9.01
15
Jan 86- Jan 93
64.7
33.49
20.87
10.64
15
July 87- Jan 92
78.4
28.54
11.50
7.44
10
Jan 89 - Jan 92
56.6
39.35
21.00
12.37
7
Jan 86 - Jan 91
76.6
26.49
10.00
5.55
11
Jan 86 - July 90
69.2
32.01
12.10
9.30
10
Jan 87 - July 89
82.3
25.67
11.33
9.42
6
July 96- Jan 89
64.6
30.55
19.50
6.72
6
Jan 86 - Jan 89
41.5
31.12
26.43
6.70
7
July 86- July 88
65.5
26.48
22.00
8.34
5
Jan 86 - July 87
40.7
35.96
21.25
8.46
4