"EXPONENTIAL SMOOTHING: 11E4EFFECTOF INITIAL VALUES AND LOSS FUNEMZONSON POST—SAMPLE FORECASTING ACM:WADY'
by
Spyros MAKRIDAKIS"and
Michele HIBON"
N° 90/46/TM
Research Professor of Decision Sciences and Information Systems,INSEAD, Boulevard de Constance, Fontainebleau, 77300 Cedex, France
Research Associate at INSEAD, Boulevard de Constance, Fontainebleau,77300 Cedex, France
Printed at INSEAD,Fontainebleau, France
EXPONENTIAL SMOOTHING: THE EFFECT OF INITIAL VALUES AND LOSS
FUNCTIONS ON POST-SAMPLE FORECASTING ACCURACY
1
Spyros MAKRIDAKIS and Michele HIBON
INSEAD Fontainebleau France
EXPONENTIAL SMOOTHING: THE EFFECT OF INITIAL VALUES AND LOSS
FUNCTIONS ON POST-SAMPLE FORECASTING ACCURACY
ABSTRACT
This paper describes an empirical investigation aimed atmeasuring the effect of different initial values and loss functions(both symmetric and asymmetric) on the post-sample forecastingaccuracy. The 1001 series of the M-competition are used andthree exponential smoothing methods are employed. The resultsare compared over various types of data and forecasting horizonsand validated with additional data.
Exponential smoothing methods are widely used in many industrial applications
including production planning, production scheduling and inventory control (Brown,
1959; Brown, 1963; Brown, 1967; Gardner, 1985; Holt et al., 1960; Johnson and
Montgomery, 1974; Makridakis and Wheelwright, 1989). Although extremely simple
and easy to model, such methods have been found to be as accurate as more complex
and statistically sophisticated alternatives (Groff, 1973; Chatfield, 1978; Koehler and
Murphree, 1988; Makridakis and Hibon, 1979; Makridakis et al., 1982; Martin and Wit,
1989). Furthermore exponential smoothing methods are robust, easy to program and
require a minimum of historical data. In addition, forecasting systems based on these
methods can be easily maintained while the cost of running them on the computer is the
smallest of all available alternatives.
The purpose of this paper is to empirically investigate the effect of various
initial values and loss functions on the post-sample forecasting accuracy of three (Single,
Holt's and Dampened) exponential smoothing methods. The first part of the paper
2
3
reviews the literature and provides the reasoning for undertaking this study. The second
part describes the methodology used and formulates various hypotheses to be studied.
The third part presents and analyses the results. There is a concluding section which
discusses the implications of the findings, validates such findings with another set of data
provided by Fildes (1989) and presents possible avenues for further research.
LITERATURE REVIEW AND REASONS FOR UNDERTAKING THIS STUDY
Exponential smoothing methods were first proposed by Brown (1959) and Holt
(1960) and were extended by Brown (1963), Winters (1960) and others (Harrison 1967;
Gardner and McKenzie, 1985). Since their introduction the question of how to initialize
the first smoothed value(s) has always been posed (Cogger, 1973; McClain, 1981; Taylor,
1981; Wade, 1967). Several alternatives have been proposed in the literature. The most
common among them are the following (see Appendix for more details):
1. Least Square Estimates: The historical data available is used to estimate
ordinary least square estimates of the initial value(s) (Brown, 1959). In practice this is
the most widely used approach for computing them.
2. Backcasting: The data is inverted and forecasting starts using the most recent
data and going backwards forecasting the less recent ones. The forecast, or smoothed
values, at period 1 are then used as initial values to start the usual forecasting (Ledolter
and Abraham, 1984).
3. Training Set: The data is divided into two parts. The first part (usually the
smaller of the two) is used to estimate the initial values for the exponential smoothing
4
equation(s) used with the second part where the final forecasts (see Makridakis et al.,
1983) are being based.
4. Convenient Initial Values: Some convenient values can be used to initialize the
smoothing equation(s). For instance, the first data value can be used to initialize the
level, while the difference between the first and the second actual value (or the average
of the second minus the first and third minus the fourth) can be used to initialize the
trend. (Makridakis and Wheelwright, 1978).
5, 6 and 7 Zero Values: The initial values can be all set to zero, or alternatively
one can he set to zero and the other(s) can be initialized using one of the alternatives
described above. This set of values(s) can be used as benchmarks to judge the improved
accuracy of approaches 1 to 4 above.
Because of the widespread applications of exponential smoothing methods even
small reductions in their forecasting errors can bring big improvements in terms of lower
costs and/or better customer services (Gardner, 1990a). At present few guide-lines and
no empirical evidence exist to help users decide upon the best initialization procedure
(see Gardner, 1985). The present study aims to provide such empirical evidence and
propose guide-lines, if any exist, for selecting appropriate initialization approaches.
Forecasting and, in general, statistical models can be optimized using a number
of loss functions such as linear, quadratic, or higher order. As in the case of initial
values there is not much help or empirical evidence to guide the choice of the best loss
function to optimize a model's parameters (Granger, 1969; Montgomery and Johnson,
1976), although, in practice the great majority of computer programs employ a quadratic
loss that minimizes the sum of square errors when a model is fitted to historical data.
However, little is known about the influence of alternative loss functions on the post-
5
sample forecasting accuracy (Cogger, 1979). The aim of this paper is to study such an
influence as far as the three exponential smoothing methods utilized in the present study
are concerned. The obvious purpose of interest is whether or not post-sample
forecasting accuracy can be improved by the appropriate choice of a loss function.
Finally, the effects of non-symmetric loss-functions are investigated as the cost
of negative errors (i.e., underestimating demand) is usually considered more critical
than that of positive ones (i.e., overestimating demand). The purpose of such
investigation is to find out the influence of non-symmetric losses on the post-sample
forecasting errors and suggest guide-lines, if any exist, in using non-symmetric loss
functions to balance the cost of negative versus positive errors.
EXPERIMENTAL DESIGN AND METHODOLOGY
The three (Single, Holt's and Dampened) most commonly used exponential
smoothing method were selected for the study (see Appendix for a description of the
models involved). Seven types of initial values (see last section and Appendix) were
used for Holt's and Dampened smoothing and five for Single. In addition eight
optimization criteria (loss functions) were employed. They range from a linear to a
fourth power one (see Appendix). The optimization of the model parameter(s) was
done using a grid search algorithm which found the optimal smoothing constants through
finer and finer searches around a global optimum initially identified through the grid
search.
Table 1 shows the 56 possibilities tested for each of the three smoothing
methods (with the exception of Single where possibilities 6 and 7 were not tested). A
non-symmetric loss function was also applied by weighting positive errors less than
6
negative ones. Such weighting was clone at five levels (.35, .50, .65, .80, .95) while
computing the model fitted errors. Consequently the post sample forecasting accuracy
of each horizon and method was recorded and compared to that of symmetric
optimization.
INSERT TABLE 1 ABOUT HERE
The methodology employed consisted of using the 1001 series of the M-
Competition (see Makridakis et al., 1982) for each of the possibilities listed in Table 1.
For each series and experimental case the optimal model parameters were estimated
and subsequently used to forecast for periods 1, 2, ... , m (where m=6 for yearly data,
m=8 for quarterly data and m=18 for monthly data). These forecasts were then
compared to the actual values (known but obviously not used in developing the
forecasting model) so as to compute the post-sample forecasting errors for each of the m
forecasting horizons. Three post-sample accuracy measures were computed from such
errors: the Mean Absolute Deviations (MAD), the Mean Absolute Percentage Errors
(MAPE) and the Mean Square Errors (MSE). These accuracy measures were
calculated separately for yearly, quarterly and monthly data and were also summarized
for all data and forecasting horizons. Similarly, the same accuracy measures were also
computed when a non-symmetric loss function was used to optimize the parameter(s) in
the model fitting phase.
The approach used in this study is not different to that of real life applications
where m forecasts are made at period t (present) even though their accuracy can only be
found in the future when the actual data becomes available.
The hypotheses to be tested were the following:
A. If yo is the average post-sample accuracy when the initial values are found by
ordinary least squares (prevalent approach) and when the optimization is done
by a quadratic loss function (prevalent approach), then
H =Ho • o 1
HA /4A +ici
where yi is the average post-sample accuracy of an alternative initialization procedure.
B. If yo is the average post-sample accuracy when a quadratic loss function
(prevalent approach) is used to obtain optimal model parameters and when ordinary
least squares (prevalent approach) is employed to initialize the first value(s), then
H0 : = ,u•o o j
HA :#A+//j
where Aej is the average post-sample accuracy of an alternative loss function.
C. If yo is the average post-sample accuracy when the initial values are found by
ordinary least squares and the optimization is done through a quadratic loss
function, then
H = p••o • o
HA :14A+
7
8
where denotes post sample accuracies for specific forecasting horizons, averages of
such horizons (e.g., short, medium, long), specific types of data (e.g., monthly, quarterly,
yearly) or length of the time series.
PRESENTATION AND ANALYSIS OF THE RESULTS
Table 2 shows the MAPE of the best and worst initialization alternatives
together with that of least square estimates (the most widely used approach) for various
forecasting horizons. The optimization alternative used was that of minimizing a
symmetric quadratic (MSE) loss function. As it can be seen in Table 2 the differences in
forecasting accuracy between the best and worst alternatives are extremely small, and
statistically non-significant for Single and Dampened exponential smoothing as well as
for short (periods 1 to 6) forecasting horizons for Holt's. The same results can be
observed in Table 3 which shows the MAPE of the best and worst symmetric
optimization alternative together with that of the MSE (the most widely used approach)
for various forecasting horizons when the initialization approach employed was that of
ordinary least squares. None of the differences are statistically significant except for
those of Holt's for longer than six forecasting horizons.
INSERT TABLES 2 AND 3 ABOUT HERE
In conclusion, there is no evidence from the empirical results to reject the null
hypotheses stated in A and B except in the case of Holt's smoothing for periods longer
than six horizons.
9
The same conclusions can be drawn when the data is separated into yearly,
quarterly and monthly. All of the observed differences when various initial approaches
and loss functions are used are extremely small and none are statistically significant at
the 5% level or below except in the case of Holt's smoothing for periods longer than six
horizons. Table 4, for instance, shows the results of an analysis of variance for yearly
data (m=6 for such data) when Dampened smoothing was used. None of the
differences between initialization procedures (columns), optimization criteria (rows) or
their interaction is statistically significant (the smallest P-value is equal to 0.34). When
the same analysis of variance is conducted for Holt's smoothing the only statistically
significant difference comes from the horizon effect.
INSERT TABLE 4 ABOUT HERE
Table 5 summarizes the MAPE for the best and worst alternatives for the
various initialization values and loss functions. 'B' on the top right corner of each box
means 'Best' among the horizontal alternatives (i.e., optimization criteria) while 'W'
signifies 'Worst'. Similarly, 'B' and 'W' on the left, lower corner of each box means 'Best'
and 'Worst' alternative among the vertical ones (i.e., initialization values). Table 5(a)
presents the results of Single smoothing, 5(b) presents those of Holt's while 5(c) presents
those of Dampened.
The differences in Table 5(a) are extremely small (the largest is only 0.3%) and
none of them are statistically significant. Moreover, none of the alternatives perform
consistently "best" or "worst" in terms of the initialization and optimization alternatives
experimented with.
The differences in Table 5(b) are considerably bigger than those in Table 5(a).
Moreover, almost all differences between the best and worst alternatives are statistically
1 0
significant (at least at the 5% level). In addition the median is consistently the worst
optimization approach while power 1.5 or 2 (MSE) is consistently the best. Among the
different initialization alternatives the best results are found when both initial values are
set to zero, except in one case where the best result is when only one of the two is set to
zero.
Differences in Table 5(c) are less consistent than those in Table 5(b). The loss
function which does best most of the time is that of MAPE while the corresponding
'best' for initialization is that of least square estimates and convenient values.
Furthermore, the great majority of differences are not statistically significant.
There are no consistent results that hold across Tables 5(a), 5(b), and 5(c).
Thus, hypotheses C cannot be rejected from the experimental findings. In Single
smoothing, Table 5(a), the best results are found when the loss function is the median
and the worst the fourth power. There is no best initialization procedure although the
worst is when the first value is set to zero. However, it must be emphasized that all
differences are extremely small and statistically no significant. In Holt's smoothing,
Table 5(b), the worst loss function is the median (the opposite of Table 5(a)) and the
best is the 1.5 power in all but one case when the 2nd power (MSE) provides the best
results. In terms of initial values the best post-sample accuracies are found when the
first values are both set to zero, except in one case when only one of the two is set to
zero. The worst results are when convenient values are used to initialize.
For Dampened smoothing, Table 5(c), the results are closer to those of Single.
Thus the worst loss function is the fourth power (although not in all cases), the worst
initialization is with zero values (not in all cases) while the best is found with convenient
values (again not in all cases). The great majority of differences are extremely small and
11
statistically non-significant (in Single smoothing the differences are even smaller and
none are statistically significant).
Tables 6 shows results similar to those of Table 5 except the post-sample
accuracy is that of MAD instead of MAPE. Concerning Single smoothing the
differences in post-sample MADs are extremely small and statistically non-significant as
was the case in Table 5(a). However, there are no other consistent patterns between
Tables 5(a) and 6(a). For instance, in Table 6(a) there is a consistent improvement in
post-sample MAD when the model parameters are optimized through higher power
(2.5, 3, or 4) loss functions while in Table 5(a) this is not the case. Similarly, the initial
values that provide the most accurate results are not the same in Tables 5(a) and 6(a).
In Holt's smoothing the differences in MADs are bigger, however higher power
loss functions do not improve the results. Moreover, the median continues to be the
worst optimization alternative while the 1.5 or 2 power the best. There is also
consistency in initialization procedures where the results of Table 5(b) and 6(b) are
similar. Thus, setting both initial values at zero provides the best results most of the
time while the worst results are found when the initialization is done through a training
set.
Finally, there is little consistency between Tables 5(c) and 6(c) - Dampened
smoothing. In Table 6(c) the best optimization criterion is power 1.5 in all but one case,
while in Table 5(c) the best was MAPE (in all but two cases). Finally, there is no
initialization procedure which is consistently best in Table 6(c) while the best in 5(c) was
that of convenient values (in all but two cases).
Thus, it can be concluded that few consistent results can be reported between
Tables 5 (a) and 5 (c) and 6 (a) and 6 (c). That is whatever, if anything, influences post-
12
sample MAPEs does not consistently influence MADs. This is not, however, the case
with Tables 5 (b) and 6 (b) - referring to Holt's smoothing - where the results are fairly
consistent.
INSERT TABLE 6 ABOUT HERE
Post-sample forecasting accuracies were also computed with MSE. The values
found were, however, large and extremely unstable. The averages were often reduced
by a factor of 10,000 by excluding as few as six series. Given the large number of series
and forecasting horizons involved (almost 14,000 in total) such large fluctuations make
the use of MSE inappropriate as a comparative measure (see Chatfield, 1988).
Furthermore, no consistent or insightful results could be deduced by examining the
various tables of post-sample MSE values even when large errors were excluded. This is
why such tables are not reported in this paper.
The non-symmetric optimization was done using ordinary least square estimates
for initial values and a quadratic (MSE) loss function. Five levels of non-symmetric
losses were used by adding to the sum 35%, 50%, 65%, 80% or 95% of the square error
at period t, when such error was positive while adding the entire square error when it
was negative (see Appendix for more details). As usual the parameter(s) that
minimized the sum of square errors were chosen and were used to make m forecasts and
subsequently compute the post-sample accuracies.
The differences in post-sample accuracies when a non-symmetric loss function
was used were extremely small for all three exponential smoothing methods. In Single
smoothing the great majority of such differences were in the second decimal. In Holt's
smoothing there were some small improvements in post-sample accuracies for longer
than twelve forecasting horizons when the non-symmetric loss function, at the 35% level,
13
was used. However, such differences were not statistically significant while the best
overall results were still obtained when a symmetric loss function was employed. In
Dampened smoothing the best overall results were found with a non-symmetric loss
function at the 50% level. Furthermore, for twelve or longer forecasting horizons the
improvements were considerably larger than those of Single or Holt's smoothing and
consistent; however, they were not statistically significant.
DISCUSSION
If the median and the fourth power are excluded as loss functions to base the
optimization of model parameters few consistent or statistically significant differences
can be found in post-sample forecasting accuracies whether such accuracies are
measured in terms of MAPE, MAD or MSE. Moreover, no consistent patterns could be
found when MAPEs, MADs or MSEs criteria were used to optimize the model's
parameters and MAPEs, MADs or MSEs measures were employed to compute post-
sample accuracies. Thus, there was no correspondence between the type of loss function
used during the model fitting and the accuracy measure employed to compute the post-
sample errors.
These results are surprising. In the forecasting literature the initialization
procedure and the optimization criteria have been considered to influence post-sample
forecasting accuracies. It has been also advocated that there must be a correspondence
between the loss function used in the model fitting and the corresponding post-sample
accuracy used to measure forecasting errors (Zellner, 1986).
From a practical point of view the prevalent approaches of using MSE as a loss
function and ordinary least square estimates to initialize the starting values seems
adequate as the differences between such approaches and the best of the alternatives
14
are small and statistically non-significant, in the great majority of cases. Furthermore, as
these approaches (MSE as the loss function and least square estimates for initialization)
are easy to program and require little computer time to apply there is no motivation to
change them. On the other hand, it makes no sense to consider more elaborate
alternatives such as backcasting for initial values or medians for optimizing the model's
parameter(s) since such alternatives are more difficult to program and require more
computer time when used to obtain forecasts.
In addition to the various results reported in the last section several other
hypotheses were tested during our study. For instance we found that sample size did not
exhibit any consistent influence on the magnitude of post-sample forecasting errors or the
choice of the best initialization or optimization alternatives. In addition, if frequency
distributions of the differences in post-sample errors between the various approaches
were made it was found that the great majority of them were less than 1% - this was in
particular true with Single and Dampened smoothing. Furthermore, no obvious
patterns of such differences could be deduced and no important factors could be found
that could explain the larger than 1% errors.
Another hypothesis tested was whether a specific set of initial values or loss
functions was best for yearly, quarterly or monthly data. But again no consistent
conclusion that hold among the three methods could be reached. Similarly, no
forecasting horizons could be better predicted than others by the appropriate choice of
specific initial values or loss functions.
The practical implications of our study suggest that there are few benefits, if
any, in attempting to find optimal ways to initialize the values of exponential smoothing
15
methods (at least the three we studied). Moreover, the choice of a best loss function is
of no consequence as long as the median and the fourth power are excluded. As
Gardner (1990 b) explained "the reason that starting values and loss functions don't
make any difference is that the optimal smoothing parameter(s) found compensate for
various starting values and different loss functions".
These findings mean that the prevalent approach of initializing by ordinary least
squares and optimizing by a quadratic loss (MSE) function provide satisfactory results
which cannot be improved in any consistent way that holds constant across methods,
data types, forecasting horizons or sample sizes. These conclusions are both good and
bad news. The good news is that exponential smoothing methods (and in particular
simple and Dampened) are easy, accurate and robust forecasting techniques that can be
used across a wide range of actual forecasting applications. The bad news is that
theoretical expectations do not seem to hold empirically for reasons that are not always
clear. Thus, research efforts must concentrate on better understanding such reasons and
in developing alternative methods and approaches that can more accurately predict real
life time series. Somehow it must be possible to beat Single smoothing for longer
forecasting horizons and Dampened for shorter and medium ones. Moreover, research
efforts must be directed in better understanding the effects of one-period-ahead versus
two, three, ... , m-period optimization and their consequence on post-sample forecasting
accuracy (Makridakis, 1990). Finally, more research needs to be done to better
understand the lack of consistency between various loss functions used in model
optimization and the resulting post-sample accuracies. For instance, one would have
expected a correspondence between the type of loss function used in model fitting and
the best results found when post-sample accuracies were measured in the same fashion;
however, none were found in our study.
16
An interesting question in all types of empirical work is whether or not the
results found can be generalized and can also hold with other types of data. In order to
validate the generality of the findings it was therefore decided, after the present results
were found, to test all the 56 possibilities listed in Table 1 with the data of Fildes (1989)
Such data are not at all similar to those of the M-Competition. They consist of 261
monthly series all coming from a Single source (AT&T). Moreover, all series exhibit a
strong negative trend and include little or no seasonality.
The similarities in the results between the two data sets is striking as far as
Single and Holt's smoothing are concerned (see Table 7). That is the magnitude of the
difference between the various experimental cases is very similar while the best and the
worst alternatives are practically the same. With Dampened smoothing the best
initialization procedure, most of the time, was that of the least squares (this was not so
with the M-Competition data) while there was no loss function which provided in a
consistent way the best or the worst results as it was also the case with the M-
Competition data (see Table 7)
Conclusion
This study has shown few differences in post sample forecasting accuracies
when different initialization values and optimization (loss) functions have been used. In
addition non-symmetric loss functions did not change in any statistically significant
fashion the post-sample results. Apart from the conclusion that the median and the
fourth power produced inferior results, no other pervasive finding holds across the
experimental possibilities tested. Finally, concerning the differences observed the
biggest ones were for longer than six forecasting horizons and were mostly concentrated
to Holt's exponential smoothing. All differences between the various experimental cases
17
in Single smoothing were small and statistically insignificant while the magnitude of
those very few of Dampened which were statistically significant was a small fraction of
those of Holt's.
The practical implications of this study suggest dropping existing concerns about
initial values and loss functions and instead concentrating on more important issues
affecting post-sample forecasting accuracy such as optimizing for more than one-step-
ahead forecasting horizons and using actual post-sample measures to base the model
selection process.
To allow replication and/or extensions of the present study both the M-
Competition and the Fildes data can be obtained at no cost by writing to Spyros
Makridakis at INSEAD.
References
Brown, R. G., Statistical Forecasting for Inventory Control, New York: McGraw-Hill,1959.
Brown, R. G., Smoothing, Forecasting and Prediction of Discrete Time Series, EnglewoodCliffs, NJ: Prentice-Hall, 1963.
Brown, R. G., Decision Rules for Inventory Management, New York: Holt, Rinehart andWinston, 1967, Part 2.
Chatfield, C., 'The Holt-Winters forecasting procedure', Applied Statistics, 27(1978), 264279.
Chatfield, C., 'Apples, Oranges and Mean Square Error', International Journal ofForecasting, 4(1988), 515-518.
Cogger, K. 0., 'Extensions of the fundamental theorem of exponential smoothing',Management Science, 19(1973), 547-554.
Cogger, K. 0., 'Time Series Analysis and Forecasting with an Absolute Error Criterion',TIMS Studies in Management Science, 12(1979), 189-201.
Fildes, R., 'Evaluation of Aggregate and Individual Forecast Method Selection SelectionRules' Management Science, 35(1989), 1056-1965.
Gardner, E. S., 'Exponential Smoothing: The State of the Art', Journal of Forecasting,4(1985), 1-28.
Gardner, E. S., 'Evaluating Forecasting Performance in Inventory Control Systems',Management Science, 36(1990 a), 490-499.
Gardner, E. S., Private Correspondence, (1990 b)
Gardner, E. S. and Dannenbring, D. G., 'Forecasting with exponential smoothing: someguide-lines for model selection', Decision Sciences, 11(1980), 370-383.
Gardner, E. S. and McKenzie, E., 'Forecasting Trends in Time Series', ManagementScience, 31(1985), 1237-1246.
Gilchrist, W. G., 'Methods of estimation using discounting', Journal of the RoyalStatistical Society (B), 29(1967), 355-369.
Gilchrist, W. G., Statistical Forecasting, London: Wiley, 1976.
Granger, C. W. J., 'Prediction with a generalized cost of error function', OperationalResearch Quarterly, 20(1969), 150-161.
Groff, G . K., 'Empirical comparison of models for short-range forecasting', ManagementScience, 20(1973), 22-31.
Harrison, P. J., 'Exponential smoothing and short-term sales forecasting', ManagementScience, 13(1967), 821-842.
18
Holt, C. C., et al., Planning Production, Inventories, and Work Force, Englewood Cliffs,NJ: Prentice-Hall, 1960, Chapter 14.
Johnson, L. A. and Montgomery, D. C., Operations Research In Production Planning,scheduling and inventory Control, New York: Wiley, 1974.
Koehler, A.B. and Murphree, E.S., 'A Comparison of Results from State SpaceForecasting with Forecasts from the Makridakis Competition', International Journal ofForecasting, 4(1988), 45-58.
Ledolter, J. and Abraham, B., 'Some comments on the initialization of exponentialsmoothing', Journal of Forecasting, 3(1984), 79-84.
Makridakis, S., 'A New Approach to Times Series Forecasting', Management Science,36(1990), 505-512
Makridakis, S. and Wheelwright, S. C., Forecasting Methods for Management (5th ed),New York, Wiley, 1989.
Makridakis, S. and Wheelwright, S. C., Interactive Forecasting: Univanate andMultivariate Methods (2nd ed), San Francisco: Holden-Day, 1978.
Makridakis, S. and Hibon, M., 'Accuracy of forecasting : an empirical investigation (withdiscussion)', Journal of the Royal Statistical Society (A), 142(1979), 97-145.
Makridakis, S., Anderson, A., Carbone, R., Fildes R., Hibon, M., Lewandowski R.,Newton, J., Parzen, R., and Winkler, R., 'The accuracy of extrapolation (time series)methods: results of a forecasting competition', Journal of Forecasting, 1(1982), 111-153.
Makridakis, S., Wheelwright, S. C. and McGee, V., Forecasting: Methods andApplications (2nd edn), New York: Wiley, 1983.
Martin, C.A., and Wit, S.F., 'Forecasting Tourist Demand: A Comparison of theAccuracy of Several Quantitative Methods", International Journal of Forecasting,5(1989), 7-20.
McClain, J. 0., 'Restarting a forecasting system when demand suddenly changes', Journalof Operations Management, 2(1981), 53-61.
Montgomery, D. C. and Johnson, L. A., Forecasting and Time series Analysis, New York:McGraw-Hill, 1976.
Taylor, S. G., 'Initialization of exponential smoothing forecasts', AIIE Transactions,13(1981), 199-205.
Wade, R. C., 'A technique for initializing exponential smoothing forecasts', ManagementScience, 13(1967), 601-602.
Winters, P.R., 'Forecasting sales by exponentially weighted moving averages',Management Science, 6(1960), 324-342.
Zellner, A., 'A Tale of Forecasting 1001 Series: The Bayesian Knight Strikes Again',International Journal of Forecasting, 2(1986), 491-494.
19
- 1 -
APPENDIX
A.Exponentjal Methods Used
Single
et = X t - Ft
where Xt is the actual data at period t andFt is the one-step-ahead forecast at period t-1 for period t.
St = St-1 + aet
where a is the smoothing constant whose value is 0 a 1
and F t+i = S t = 1, 2, 3, ... , m
where m = 6 for yearly data, m = 8 for quarterly and m = 18 for monthly
fiott's Smoothing
et = X t - Ft
St S t_i + Tt_i + aet
Tt = Tt-1 Oet
where p is a smoothing constant whose value is 0 13 1and Ft+i = S t + iTt
Damoened Trend
et = Xt - Ft
S t = + OT t . 1 + aet
Tt = OTt-1 Ret
F i+m = S t + y, OLrt1=1
-2-
B.jnitial values used
1. least Square
For Single exponential smoothing the initial value Si was found as :
Xt
Si — t=in
where n is the number of historical data available.
For Holt's and dampened exponential smoothing S 1 and T 1 are found as :
n nnE tXt - E tY, Xt
t=1 t=1 t=1n
112
nE t2t=1 ,=t
n n
EX t E tS 1 = 'in
This initialization approach is referred to as the ''prevalent" one as it is the mostwidely used in forecasting applications.
2. j3 ackcastinp
The data is inverted and the most recent data value becomes period 1 while theleast recent (i.e., period 1) becomes the last one (i.e., period n). Consequently thevalues of S 1 , or S 1 and T 1 are found as above and the appropriate equation(s) is(are)used to forecast. The last values of Sn, or Sn and Tn are used for initial estimates in theregular forecastings except that the sign of the value of Tn is reversed. Thus, in Singlesmoothing
S 1 = Sn
while in Holt's and Dampened
Si = SnT1 =0 - Tn
T1 –
and
-3-
3 . Training set
The data is separated into two sets (the first set makes up one third of thehistorical data while the second makes up the remaining two thirds). The initial valuesfor S1 or S 1 and T 1 for the training set are found as in 1 above.
If f is the last period of the training (first) set then the values of the S t or S 1 andTi for the remaining data are found as:
= Sfor
SI SfTI Tf
4. Convenient values
The value of S 1 or S i and T 1 are simply set as follows :
S i =XIor
S =XIT1 = (X2 - X1 + X4 - X3) / 2
5. Zero values
The initial values are set as follows:
S = 0or
S1 = 0T1 = 0
6 . Zero Value (for Holt's and Dampened smoothing only)
S 1 = 0T 1 = Least square estimate (see 1 above).
7 . Zero value (for Holt's and Dampened smoothing only)
S 1 = Least square estimate (see 1 above)T 1 =0
-4-
C.Symmetric Loss Functions
The one-period-ahead forecasting errors e t were computed as :
et = Xt - Ft
Consequently the smoothing parameters a, a and 0, or a, j3 and cl) were chosenin such a way as to minimize the corresponding model fitting loss function outlinedbelow :
I Mean Absolute Deviation (MAD)
N- I ett=1
II Mean Absolute Percentage Error (MAPE)
vn et I X
t=1t
n
III Median Absolute Percentage Error (Median)
The smoothing parameter(s) corresponding to the middle value (median) whenall absolute percentage errors were arranged from the smallest to the largest waschosen.
IV I.5th Power
I et 11.5
t=1
V Mean Square Error (MSE)
Et=i
-5-
VI 2.5th Power
VII 3th Power
Eiett.i
VIII 4th Power
4t=i
D.Non-Svmmetric Loss Functions
The prevalent initialization procedure (least square estimates see 1 above) and theprevalent optimization function (MSE, I above) were used with the following non-symmetric loss functions.
n
where 141= 1 when e t > 0
and tv=c when et
TABLE 1: INITIAL VALUES AND LOSS FUNCTIONS
Symmetric loss functions
InitialValues I II HI IV V VI VII VIII
MAD MAPE MEDIAN 1.5th MSE 2nd 2.5th Cubic 4thpower power power power power
1 Least square Single Single Single Single Single Single Single Singleestimates Holt's Holt's Holt's Holt's Holt's Holt's Holt's Holt's
Dampened Dampened Dampened Dampened Dampened Dampened Dampened Dampened
2 Backcasting Single Single Single Single Single Single Single SingleHolt's Holt's Holt's Holt's Holt's Holt's Holt's HolesDampened Dampened Dampened Dampened Dampened Dampened Dampened Dampened
3 Training set Single Single Single Single Single Single Single SingleHolt's Holt's Holt's Holt's Holt's Holt's Holt's Holt'sDampened Dampened Dampened Dampened Dampened Dampened Dampened Dampened
4 Convenient Single Single Single Single Single Single Single Singlevalues Holt's Holt's Holt's Holt's Holt's Holt's Holt's Holt's
Dampened Dampened Dampened Dampened Dampened Dampened Dampened Dampened
5 s =0, or Single Single Single Single Single Single Single Singles=0 and t=0 Holt's Holt's Holt's Holt's Holt's Holt's Holt's Holt's
Dampened Dampened Dampened Dampened Dampened Dampened Dampened Dampened
6 s =0 Holt's Holt's Holt's Holt's Holt's Holt's Holt's Holt'st =least square Dampened Dampened Dampened Dampened Dampened Dampened Dampened Dampened
7 s = least square Holt's Holt's Holt's Holes Holt's Holt's Holt's Holt'sT=0 Dampened Dampened Dampened Dampened Dampened Dampened Dampened Dampened
TABLE 2
COMPARISON OF INITIALIZATION
OPTIMIZATION BY MSE
1 3
FORECASTING HORIZONS
6 8 12 18 Average(All horizons)
All SINGLE Least Squares (1) 8.7 13.3 19.7 18.0 16.9 26.1 17.0Data Best: Convenient (4) 8.5 13.1 19.4 17.9 16.9 26.1 16.9
Worst: zero (5) 8.8 13.2 19.6 18.4 16.9 25.8 17.0
All HOLT Least Squares (1) 8.7 12.9 21.3 22.7 21.3 33.6 19.8Data Best: Both zero (5) 8.6 12.5 19.4 20.6 19.4 34.8 18.7
Worst: Convenient (4) 8.8 13.4 22.1 25.4 26.0 42.5 22.8
All DAMPEN Least Squares (1) 8.5 12.4 18.5 18.1 17.2 27.4 17.0Data Best: Convenient (4) 8.4 12.5 18.7 18.2 17.2 27.5 17.0
Worst: Both zero (5) 8.7 12.9 19.2 18.8 17.2 27.1 17.3
TABLE 3
COMPARISON OF OPTIMIZATION
INITIALIZATION BY PREVALENT VALUES
1 3
FORECASTING HORIZONS
6 8 12 18 Average(All horizons)
All SINGLE MSE 8.7 13.3 19.7 18.0 16.9 26.1 17.0Data Best: MEDIAN
Worst: etMEDIA 8.6
8.713.013.4
19.419.8
17.918.0
16.816.9
25.926.1
16.917.0
All HOLT MSE 8.7 12.9 21.3 22.7 21.3 33.6 19.6Data Best: et 1.5 8.6 12.6 20.5 22.1 20.9 34.0 19.4
Worst: MEDIAN 9.5 15.5 24.6 28.3 33.0 50.0 27.6
All DAMPEN MSE 8.5 12.4 18.5 18.1 17.2 27.4 17.0Data Best: MAPE 8.7 12.2 18.6 17.8 17.5 24.8 16.8
Worst: MEDIAN 8.5 12.4 19.1 19.3 17.9 26.7 17.4
TABLE 4: TWO-WAY ANALYSIS OF VARIANCE
Source Sum of d.f. Mean Computed
Squares Square F-value p-value
Columns 2266. 5. 453. .87 .498
Rows 2940. 5. 587. 1.13 .340
Row X Col 1199. 25. 47. .09 1.000
Error 3062580. 5904. 519.
TOTALS 3068990. 5939.
165 series of yearly data with DAMPEN-Trend Method
Average Error on 6 Forecasting Horizons
Horizontal values: Errors for each OPTIMIZATION criteria
Vertical values: Errors for each STARTING VALUE
THE DIFFERENCES ARE NOT STATISTICALLY SIGNIFICANT
TABLE 5(a): SINGLE SMOOTHING, AVERAGE MAPE FOR ALL FORECASTING HORIZONS AND TIME SERIES
Symmetric loss functions
InitialValues I II III IV V VI VII VIII
MAD MAPE MEDIAN 1.5thpower
MSE 2nd 2.5thpower power
Cubicpower
4thpower
Least square 16.9Bestimates
16.9B w16.9B 16.9B w17.0W w17.0W 17.0W B17.0W
2 Backcasting 16.9 816.8 B16.7B B16.8 B16.9 w17.0W 17.0W B17.0W
3 Training set 16.9 B16.8B 16.8B 16.9 w17.0 w17.0 w17.1 w17.1W
4 Convenient B16.8Bvalues
B16.8B 16.8B 16.88 B16.9 B16.9 B16.9 B17.0W
5 s=0, or w17.0s=0 and t=0
w17.1W w16.9B w17.0 w17.0 w17.0 17.0 B17.0
6 s = 0t =least square
7 s = least squareT=0
TABLE 5(b): HOLTS SMOOTHING, AVERAGE MAPE FOR ALL FORECASTING HORIZONS AND TIME SERIES
InitialValues
Symmetric loss functions
I H III IV V VI VII VIII
MAD MAPE MEDIAN 1.5thpower
MSE 2nd 2.5thpower power
Cubicpower
4thpower
1 Least squareestimates
19.8 19.9 27.6W 19.4B 19.8 20.1 20.3 22.7
2 Backcasting 20.7 21.5 27.8W 20.2B 20.3 21.8 21.6 26.7
3 Training set 21.4 22.8 30.5W 20.5 B 20.8 21.3 22.5 24.0
4 Convenientvalues
w23.5 w23.6 w29.6W w23.0 w22.8B w24.4 w24.8 w25.7
5 s =0, ors=0 and t=0
B19.0 B18.9 27.0W B 18.4B B18.7 B19.2 B19.7 B21.0
6 s =0t = least square
19.1 19.1 27.1W 18.8B 18.9 19.4 B19.7 21.2
7 s =least squareT=0
19.2 19.3 B26.8W 18.6B 21.8 21.1 23.5 25.0
TABLE 5(c): DAMPENED SMOOTHING, AVERAGE MAPE FOR ALL FORECASTING HORIZONS AND TIME SERIES
Symmetric loss functions
InitialValues I II III IV V VI VII VIII
MAD MAPE MEDIAN 1.5thPower
MSE 2nd 2.5thPower Power
CubicPower
4thPower
1 Least square 16.9estimates
16.8B 17.4W B 16.9 B17.0 B17.0 B172 B17.4W
2 Backcasting 17.0 B16.7B 17.4 B16.9 17.1 17.1 17.3 17.5W
3 Training set 17.0B 17.0B 17.4 17.1 17.3 17.3 17.4 18.4W
4 Convenient B 16.8values
B16.7B 17.4 B16.9 B17.0 17.1 B172 w19.0W
5 s =0, or w17.3s=0 and t=0
w17.2B w17.7" w17.2B w17.3 17.3 17.4 B17.4
6 s =0 17.1Bt = least square
w17.2 17.6W 17.1 B 17.1B 17.1B B 17.2 17.5
7 s =least square 16.9BT=0
17.1 B173 17.0 17.2 w17.4 w17.5 17.9W
"B" at the upper, right hand side of each box signifies Best while "W" signifies Worst accuracy."B" at the lower, left hand side of each box signifies Best, while "W" signifies Worst accuracy.
TABLE 6(a): SINGLE SMOOTHING, AVERAGE MAD FOR ALL FORECASTING HORIZONS AND TIME SERIES(Values have been divided by 1000)
Loss functions
InitialValues I II III IV V VI VII VIII
MAD MAPE MEDIAN 1.5thpower
MSE 2nd 2.5thpower power
Cubicpower
4thpower
1 Least square 15.4estimates
15.5W 15.3 15.4 W 15 .4 15.2B w 15 .4 15 2BW •
2 Backcasting 15.4W 15.4W B15.1 15.4W 15.3 B15.1 B15.08 15.1
3 Training set B 15.3W 13 15.2 15.3W B 15.3W B15.2 B15.1 B15.0 14.9B
4 Convenient 15.5Wvalues
15.4 15.5W 15.4 15.3 B15.1 B15.0 B14.88
5 s =0, or w15.6Ws=0 and t=0
w15.6W w15.6W w15.6W w15.4 w15.3 15.1 14.9B
6 s =0t =least square
7 s =least squareT=0
TABLE 6(b): HOLTs SMOOTHING, AVERAGE MAD FOR ALL FORECASTING HORIZONS AND TIME SERIES(Values have been divided by 1000)
InitialValues
Loss functions
I II III IV V VI VII VIII
MAD MAPE MEDIAN 1.5thpower
MSE 2nd 2.5thpower power
Cubicpower
4thpower
1 Least squareestimates
12.4 13.2 B173W 12.4 12.4 12.3B 15.1 15.2
2 Backcasting 11.9 B12.7 21.9W 11.6B 11.8 11.8 w15.8 15.5
3 Training set w15.5 w18.7 w28.1W w13.0B w14.3 w14.6 14.9 15.1
4 Convenientvalues
13.6 15.6 23.5W 12.8 11.7B 11.8 14.4 w15.7
5 s = 0, ors = 0 and t =0
B11.3B 12.9 20.4w B 11.3B B11.5 /311.6 B13.9 14.9
6 s =0t = least square
12.2B 13.6 22.8W 12.2B 12.2B 12.2B 14.7 15.1
7 s = least squareT =0
13.0B 13.9 18.6W w13.0B 13.6 13.1 B13.9 B14.1
TABLE 6(c): DAMPENED SMOOTHING, AVERAGE MAD FOR ALL FORECASTING HORIZONS AND TIME SERIES(Values have been divided by 1000)
Loss functions
InitialValues I II III IV V VI VII VIII
MAD MAPE MEDIAN 1.5thpower
MSE 2nd 2.5thpower power
Cubicpower
4thpower
1 Least square 13.4estimates
14.9W 14.2 B 10.8B B12.2 B12.7 12.9 w13.5
2 Backcasting 13.3 14.3 14AW 12.5 B 12.8 13.2 w13.2 13.1
3 Training set 13.5 813.2 15.8W 12.7B 13.1 13.1 13.1 812.7
4 Convenient B 13.2values
14.3 w18.7W 11.8 B 12.9 13.1 w132 13.1
5 s=0, or 13 .6s=0 and t=0
W I -5 .6W 15.1 11.9B 12.4 13.0 12.9 13.1
6 s =0 13.9t = least square
14.8 B15.0W 11.7B 12.6 12.9 B124 12.9
7 s = least squarew15.0T=0
15.3W B15.0 w15.0 w14.1 w14.2 13.0B 13.0B
TABLE 7(a): SINGLE SMOOTHING, AVERAGE MAPE FOR ALL FORECASTING HORIZONS AND TIME SERIES
InitialValues
FILDES DATA
I II III IV V VI VII VIII
MAD MAPE MEDIAN 1.5thpower
MSE 2ndpower
2.5thpower
Cubicpower
4thpower
Least squareestimates
18.1 18.1 18.1 18.1 18.2 18.2 18.2 18.2
2 Backcasting 18.1 18.1 18.2 18.1 18.1 18.2 18.2. 18.3
3 Training set 18.1 17.9 18.1 18.1 18.1 18.1 18.1 18.2
4 Convenientvalues
18 . 1 17.9 18.2 18.1 18.1 18.2 18.2 18.2
5 s =0, ors=0 and t=0
18.1 18.1. 18.1 18.1 18.1 18.1 18.1 18.1
6 s =0t =least square
7 s = least squareT=0
TABLE 7(b): HOLVS SMOOTHING, AVERAGE MAPE FOR ALL FORECASTING HORIZONS AND TIME SERIES
FILDES DATA
InitialValues I H III IV V VI VII VIII
MAD MAPE MEDIAN 1.5thpower
MSE 2nd 2.5thpower power
Cubicpower
4thpower
1 Least square 9.1estimates
12.4W B11.5 8.9B 9.7 10.2 10.4 10.5
2 Backcasting 8.8B 12.5 17.1W 8.9 9.2 9.9 10.3 10.8
3 Training set 8.9B 10.2 18.0W 9.6 10.1 10.2 10.7 11.1
4 Convenient 9.6Bvalues
w12.7 w20.0W 10.0 10.3 10.0 10.5 10.9
5 s =0, or w18.1s=0 and t=0
11.6B 16.9 w18.9 w18.9 w19.0 w19. w19.1W
6 s = 0 B7.3Bt = least square
9.3 15.4W B7.8 B8.2 B8.6 B8.8 B8.9
7 s= least square 10.4T=0
B9.0 17.7W 8.013 8.9 9.8 9.6 9.5
TABLE 7(c): DAMPENED SMOOTHING, AVERAGE MAPE FOR ALL FORECASTING HORIZONS AND TIME SERIES
FILDES DATA
InitialValues I II III IV V VI VII VIII
MAD MAPE MEDIAN 1.5thPower
MSE 2nd 2.5thPower Power
CubicPower
4thPower
1 Least square 12.6estimates
13.4 13.6W B 10A B10.4 B10.5 B9.4B B10.1
2 Backcasting B 12.5 813.2. 13.8W 12.6 12.6 13.2 13.5 12.1B
3 Training set 14..3W 13.9 W 14 3 W13.6 13.3 13.4 10.8B 11.2
4 Convenient 13.4W.values
8 13.2 B 13.1 12.3 12.3 13.0 13.4W 11,6
5 s=0, or w18.0s=0 and t=0
w17.4 13.9E w18.1 W w18.1W w18.1 W 18.0 18.0
6 s =0 w18.0t = least square
w17.4 13.9E W 18.1. w18.1 w18.1 w18.2 w18.3W
7 s = least square 15.0WT=0
14.3 B13.1 12.6 12.6 11.3 11.5 11.5
"B" at the upper, right hand side of each box signifies Best while "W" signifies Worst accuracy."B" at the lower, left hand side of each box signifies Best, while "W" signifies Worst accuracy.
INSEAD VOREING PAPERS SERIES
1986
86/01 Arnoud DE MEYER
86/02 Philippe A. NAERTMarcel VEVERBERGHand Guido VERSVIJVEL
86/03 Michael BRIMS
86/04 Spyros MAKRIDAKISand Michele HIBON
86/05 Charles A. VYPLOSZ
86/06 Francesco CIAVAllI,Jeff R. SHEEN andCharles A. VYPLOSZ
"The R 6 D/Production interface".
"Subjective estimation in integratingcommunication budget and allocationdecisions: a case study", January 1986.
"Sponsorship and the diffusion oforganizational innovation: a preliminary viev".
"Confidence intervals: an empiricalinvestigation for the series in the M-Competition" .
"A note on the reduction of the vorkveek",July 1985.
"The real exchange rate and the fiscalaspects of a natural resource discovery",Revised version: February 1986.
86/16 B. Espen ECKBO andHerwig M. LANGOHR
86/17 David B. JEMISON
86/18 James TEBOULand V. MALLERET
86/19 Rob R. VEITZ
86/20 Albert CORHAY,Gabriel HAVAVINIand Pierre A. MICHEL
86/21 Albert CORHAY,Gabriel A. HAVAVINIand Pierre A. MICHEL
86/22 Albert CORHAY,Gabriel A. HAVAVINIand Pierre A. MICHEL
"Les primes des offres publiques, la noted'information et le marche des transferts decontrele des societes".
"Strategic capability transfer in acquisitionintegration", May 1986.
"Towards an operational definition ofservices", 1986.
"Nostradamus: a knowledge-based forecastingadvisor".
"The pricing of equity on the London stockexchange: seasonality and size premium",June 1986.
"Risk-premia seasonality in U.S. and Europeanequity markets", February 1986.
"Seasonality in the risk-return relationshipssome international evidence", July 1986.
86/07 Douglas L. MacLACHLANand Spyros MAKRIDAKIS
86/08 Jose de la TORRE andDavid H. NECKAR
86/09 Philippe C. HASPESLAGH
86/10 R. MOENART,Arnoud DE MEYER,J. BARGE andD. DESCHOOLMEESTER.
86/11 Philippe A. NAERTand Alain BULTEZ
86/12 Roger BETANCOURTand David GAUTSCHI
86/13 S.P. ANDERSONand Damien J. NEVEN
"Judgmental biases in sales forecasting",February 1986.
"Forecasting political risks forinternational operations", Second Draft:March 3, 1986.
"Conceptualizing the strategic process indiversified firms: the role and nature of thecorporate influence process", February 1986.
"Analysing the issues concerningtechnological de-maturity".
"From "Lydiametry" to "Pinkhamization":■isspecifying advertising dynamics rarelyaffects profitability".
"The economics of retail firms", RevisedApril 1986.
"Spatial competition a la Cournot".
86/23 Arnoud DE MEYER
86/24 David CAUTSCHIand Vithala R. RAO
86/25 H. Peter GRAYand Ingo WALTER
86/26 Barry EICHENGREENand Charles VYPLOSZ
86/27 Karel COOLand Ingemar DIERICKX
86/28 Manfred KETS DEVRIES and Danny MILLER
86/29 Manfred KEYS DE VRIES
86/30 Manfred KETS DE VRIES
86/31 Arnoud DE MEYER
"An exploratory study on the integration ofinformation systems in manufacturing",July 1986.
"A methodology for specification andaggregation in product concept testing",July 1986.
"Protection", August 1986.
"The economic consequences of the FrancPoincare", September 1986.
"Negative risk-return relationships inbusiness strategy: paradox or truism?",October 1986.
"Interpreting organizational texts.
"Vhy follow the leader?".
"The succession game: the real story.
"Flexibility: the next competitive battle",October 1986.
"Comparaison Internationale des marges brutesdu commerce", June 1985.
"How the managerial attitudes of firms withFMS differ from other manufacturing firms:survey results", June 1986.
86/31 Arnoud DE MEYER,Jinichiro NAKANE,Jeffrey G. MILLERand Kasra FERDOVS
86/32 Karel COOLand Dan SCHENDEL
"Flexibility: the next competitive battle",Revised Version: March 1987
Performance differences among strategic groupmembers", October 1986.
86/14 Charles VALDMAN
86/15 Mihkel TOMBAK andArnoud DE MEYER
86/33 Ernst BALTENSPERGERand Jean DERMINE
86/34 Philippe HASPESLAGHand David JEMISON
86/35 Jean DERMINE
86/36 Albert CORHAY andGabriel HAVAVINI
86/37 David GAUTSCHI andRoger BETANCOURT
86/38 Gabriel HAVAVINI
86/39 Gabriel HAVAVINIPierre MICHELand Albert CORHAY
86/40 Charles VYPLOSZ
86/41 Kasra FERDOVSand Wickham SKINNER
86/42 Kasra FERDOWSand Per LINDBERG
86/43 Damien NEVEN
86/44 Ingemar DIERICKXCarmen MATUTESand Damien NEVEN
1987
87/01 Manfred KETS DE VRIES
87/02 Claude VIALLET
87/03 David GAUTSCHIand Vithala RAO
87/04 Sumantra GHOSHAL andChristopher BARTLETT
87/05 Arnoud DE MEYERand Kasra FERDOWS
"The role of public policy in insuringfinancial stability: a cross-country,comparative perspective", August 1986, RevisedNovember 1986.
"Acquisitions: myths and reality",July 1986.
"Measuring the market value of a bank, aprimer", November 1986.
"Seasonality in the risk-return relationship:some international evidence", July 1986.
"The evolution of retailing: a suggestedeconomic interpretation".
"Financial innovation and recent developmentsin the French capital markets", Updated:September 1986.
"The pricing of common stocks on the Brusselsstock exchange: a re-examination of theevidence", November 1986.
"Capital flows liberalization and the EMS, aFrench perspective", December 1986.
"Manufacturing in a new perspective",July 1986.
"FMS as indicator of manufacturing strategy",December 1986.
"On the existence of equilibrium in hotelling'smodel", November 1986.
"Value added tax and competition",December 1986.
"Prisoners of leadership".
"An empirical investigation of Internationalasset pricing", November 1986.
"A methodology for specification andaggregation in product concept testing",Revised Version: January 1987.
"Organizing for innovations: case of themultinational corporation", February 1987.
"Managerial focal points in manufacturingstrategy", February 1987.
"Customer loyalty as a construct in themarketing of banking services", July 1986,
"Equity pricing and stock market anomalies",February 1987.
"Leaders who can't manage", February 1987.
"Entrepreneurial activities of European MBAs",March 1987.
"A cultural view of organizational change",March 1987
"Forecasting and loss functions", March 1987.
"The Janus Bead: learning fro, the superiorand subordinate faces of the manager's job",April 1987.
"Multinational corporations as differentiatednetworks", April 1987.
"Product Standards and Competitive Strategy: AnAnalysis of the Principles", May 1987.
"METAFORECASTING: Rays of improvingForecasting. Accuracy and Usefulness",May 1987.
"Takeover attempts: what does the language tellus?, June 1987.
"Managers' cognitive maps for upward anddovnvard relationships", June 1987.
"Patents and the European biotechnology lag: astudy of large European pharmaceutical firms",June 1987.
"Why the EMS? Dynamic games and the equilibriumpolicy regime, May 1987.
"A new approach to statistical forecasting",June 1987.
"Strategy formulation: the impact of nationalculture", Revised: July 1987.
"Conflicting ideologies: structural andmotivational consequences", August 1987.
"The demand for retail products and thehousehold production model: new views oncomplementarity and substitutability".
87/06 Arun K. JAIN,Christian PINSON andNaresh K. MALHOTRA
87/07 Rolf BANZ andGabriel HAVAVINI
87/08 Manfred KETS DE VRIES
87/09 Lister VICKERY,Mark PILKINCTONand Paul READ
87/10 Andre LAURENT
87/11 Robert FILDES andSpyros MAKRIDAKIS
87/12 Fernando BARTOLOMEand Andre LAURENT
87/13 Sumantra GHOSHALand Nitin NOHRIA
87/14 Landis GABEL
87/15 Spyros MAKRIDAKIS
87/16 Susan SCHNEIDERand Roger DUNBAR
87/17 Andre LAURENT andFernando BARTOLOME
87/18 Reinhard ANGELMAR andChristoph LIEBSCHER
87/19 David BEGG andCharles VYPLOSZ
87/20 Spyros MAKRIDAKIS
87/21 Susan SCHNEIDER
87/22 Susan SCHNEIDER
87/23 Roger BETANCOURTDavid GAUTSCHI
87/32 Arnoud DE MEYER
87/33 Yves DOZ andAmy SHUEN
87/34 Kasra FERDOWS andArnoud DE MEYER
87/35 P. J. LEDERER andJ. F. THISSE
"German, French and British manufacturingstrategies less different than one thinks",September 1987.
"A process framework for analyzing cooperationbetween firms", September 1987.
"European manufacturers: the dangers ofcomplacency. Insights from the 1987 Europeanmanufacturing futures survey, October 1987.
"Competitive location on netvorks underdiscriminatory pricing", September 1987.
87/36 Manfred KETS DE VRIES 'Prisoners of leadership", Revised versionOctober 1987.
87/24 C.B. DERR andAndre LAURENT
"The internal and external careers: atheoretical and cross-cultural perspective",Spring 1987.
87/41 Gavriel HAVAVINI and "Seasonality, size premium and the relationshipClaude VIALLET betveen the risk and the return of French
common stocks", November 1987
87/29 Susan SCHNEIDER andPaul SHRIVASTAVA
"The robustness of !IDS configurations in theface of incomplete data", March 1987, Revised:July 1987.
"Demand complementarities, household productionand retail assortments", July 1987.
"Is there a capital shortage in Europe?",August 1987.
"Controlling the interest-rate risk of bonds:an introduction to duration analysis andimmunization strategies", September 1987.
"Interpreting strategic behavior: basicassumptions themes in organizations", September1987
87/42 Damien NEVEN andJacques-F. THISSE
87/43 Jean GABSZEWICZ andJacques-F. THISSE
87/44 Jonathan HAMILTON,Jacques-F. THISSEand Anita WESKAMP
87/45 Karel COOL,David JEMISON andIngemar DIERICKX
87/46 Ingemar DIERICKXand Karel COOL
"Combining horizontal and verticaldifferentiation: the principle of max-mindifferentiation", December 1987
"Location", December 1987
"Spatial discrimination: Bertrand vs. Cournotin a model of location choice", December 1987
"Business strategy, market structure and risk-return relationships: a causal interpretation",December 1987.
'Asset stock accumulation and sustainabilityof competitive advantage", December 1987.
87/25 A. K. JAIN,N. K. MALNOTRA andChristian PINSON
87/26 Roger BETANCOURTand David GAUTSCHI
87/27 Michael BURDA
87/28 Gabriel HAVAVINI
87/30 Jonathan HAMILTON
"Spatial competition and the Core", August
W. Bentley MACLEOD
1987. 1988and J. F. THISSE
87/31 Martine OUINZII andJ. F. THISSE
87/37 Landis GABEL
87/38 Susan SCHNEIDER
87/40 Carmen MATUTES andPierre REGIBEAU
"On the optimality of central places*,September 1987.
"Privatization: its motives and likelyconsequences", October 1987.
"Strategy formulation: the impact of nationalculture", October 1987.
"Product compatibility and the scope of entry",November 1987
88/01 Michael LAWRENCE andSpyros MAKRIDARIS
88/02 Spyros MARRIDAKIS
88/03 James TEBOUL
88/04 Susan SCHNEIDER
88/05 Charles VYPLOSZ
88/06 Reinhard ANGELMAR
88/07 Ingemar DIERICKXand Karel COOL
88/08 Reinhard ANGELMARand Susan SCHNEIDER
88/09 Bernard SINCLAIR-DESGAGN6
88/10 Bernard SINCLAIR-DESGAGN6
88/11 Bernard SINCLAIR-DESGAGN6
"Factors affecting judgemental forecasts andconfidence intervals", January 1988.
"Predicting recessions and other turningpoints", January 1988.
"De-industrialize service for quality", January1988.
"National vs. corporate culture: implicationsfor human resource management", January 1988.
"The swinging dollar: is Europe out of step?",January 1988.
"Les conflits dans les canaux de distribution",January 1988.
"Competitive advantage: a resource basedperspective", January 1988.
"Issues in the study of organizationalcognition", February 1988.
"Price formation and product design throughbidding", February 1988.
"The robustness of some standard auction gameforms", February 1988.
"Vhen stationary strategies are equilibriumbidding strategy: The single-crossingproperty", February 1988.
87/39 Manfred KETS DE VRIES "The dark side of CEO succession", November1987
88/12 Spyros MAKRIDAKIS
"Business firms and managers in the 21stcentury", February 1988
88/13 Manfred KETS DE VRIES "Alexithymia in organizational life: theorganization man revisited", February 1988.
88/14 Alain NOEL "The interpretation of strategies: a study ofthe impact of CEOs on the corporation",March 1988.
88/15 Anil DEOLALIKAR andLars-Hendrik ROLLER
88/16 Gabriel HAVAVINI
88/17 Michael BURDA
88/18 Michael BURDA
88/19 M.J. LAWRENCE andSpyros MAKRIDAKIS
88/20 Jean DERMINE,Damien NEVEN andJ.F. THISSE
88/21 James TEBOUL
88/22 Lars-Hendrik ROLLER
88/23 Sjur Didrik FLAMand Georges ZACCOUR
88/24 B. Espen ECKBO andHerwig LANGOHR
88/25 Everette S. GARDNERand Spyros MAKRIDAKIS
"The production of and returns from industrialinnovation: an econometric analysis for adeveloping country", December 1987.
"Market efficiency and equity pricing:international evidence and implications forglobal investing", March 1988.
"Monopolistic competition, costs of adjustmentand the behavior of European employment",September 1987.
"Reflections on "Wait Unemployment" inEurope", November 1987, revised February 1988.
"Individual bias in judgements of confidence",March 1988.
"Portfolio selection by mutual funds, anequilibrium model", March 1988.
"De-industrialize service for quality",March 1988 (88/03 Revised).
"Proper Quadratic Functions with an Applicationto AT&T", May 1987 (Revised March 1988).
"P.quilibres de Nash-Cournot dans le marcheeuropken du gaz: un cas on les solutions enboucle ouverte et en feedback coincident",Mars 1988
"Information disclosure, means of payment, andtakeover premia. Public and Private tenderoffers in Prance". July 1985, Sixth revision,April 1988.
"The future of forecasting", April 1988.
88/26 Sjur Didrik FLANand Georges ZACCOUR
88/27 Murugappa KRISHNANLars-Hendrik ROLLER
88/28 Sumantra GHOSHAL andC. A. BARTLETT
"Semi-competitive Cournot equilibrium inmultistage oligopolies", April 1988.
"Entry game with resalable capacity",April 1988.
"The multinational corporation as a network:perspectives from interorganizational theory",May 1988.
88/29 Naresh K. MALHOTRA,Christian PINSON andArun K. JAIN
88/30 Catherine C. ECKELand Theo VERMAELEN
88/31 Sumantra GHOSHAL andChristopher BARTLETT
88/32 Kasra FERDOWS andDavid SACKRIDER
88/33 Mihkel M. TOMBAK
88/34 Mihkel M. TOMBAK
88/35 Mihkel TOMBAK
88/36 Vikas TIBREWALA andBruce BUCHANAN
88/37 Murugappa KRISHNANLars-Hendrik ROLLER
88/38 Manfred KETS DE VRIES
88/39 Manfred KETS DE VRIES
88/40 Josef LAKONISHOK andTheo VERMAELEN
88/41 Charles WYPLOSZ
88/42 Paul EVANS
88/43 B. SINCLAIR-DESGAGNE
88/44 Essam MAHMOUD andSpyros MAKRIDAKIS
88/45 Robert KORAJCZYKand Claude VIALLET
88/46 Yves DOZ andAmy SHUEN
"Consumer cognitive complexity and thedimensionality of multidimensional scalingconfigurations", May 1988.
"The financial fallout frog Chernobyl: riskperceptions and regulatory response", May 1988.
"Creation, adoption, and diffusion of
innovations by subsidiaries of multinationalcorporations", June 1988.
"International manufacturing: positioningplants for success", June'1988.
"The importance of flexibility inmanufacturing", June 1988.
"Flexibility: an important dimension inmanufacturing", June 1988.
"A strategic analysis of investment in flexiblemanufacturing systems", July 1988.
"A Predictive Test of the HBO Model thatControls for Non-stationarity", June 1988.
"Regulating Price-Liability Competition ToImprove Welfare", July 1988.
"The Motivating Role of Envy : A ForgottenFactor in Management, April 88.
"The Leader as Mirror : Clinical Reflections",July 1988.
"Anomalous price behavior around repurchasetender offers", August 1988.
"Assymetry in the EMS: intentional orsystemic?", August 1988.
"Organizational development in thetransnational enterprise", June 1988.
"Group decision support systems implementBayesian rationality", September 1988.
"The state of the art and future directionsin combining forecasts", September 1988.
"An empirical investigation of internationalasset pricing", November 1986, revised August1988.
"Prom intent to outcome: a process frameworkfor partnerships", August 1988.
88/47 Alain BULTEZ,Els GIJSBRECHTS,Philippe NAERT andPiet VANDEN ABEELE
88/48 Michael BURDA
88/49 Nathalie DIERKENS
88/50 Rob WEITZ andArnoud DE MEYER
88/51 Rob VEITZ
88/52 Susan SCHNEIDER andReinhard ANGELMAR
88/53 Manfred KETS DE VRIES
88/54 Lars-Hendrik ROLLERand Mihkel M. TOMBAK
88/55 Peter BOSSAERTSand Pierre HILLION
88/56 Pierre HILLION
88/57 Vilfried VANHONACKERand Lydia PRICE
88/58 B. SINCLAIR-DESGAGNEand Mihkel M. TOMBAK
8R/59 Martin KILDUFF
88/60 Michael BURDA
88/61 Lars-Hendrik ROLLER
88/62 Cynthia VAN HULLE,Theo VERMAELEN andPaul DE WOUTERS
"Asymmetric cannibalism between substituteitems listed by retailers", September 1988.
"Reflections on 'Wait unemployment' inEurope, II", April 1988 revised September 1988.
"Information asymmetry and equity issues",September 1988.
"Managing expert systems: from inceptionthrough updating", October 1987.
"Technology, work, and the organization: theimpact of expert systems", July 1988.
"Cognition and organizational analysis: who'sminding the store?", September 1988.
"Whatever happened to the philosopher king: theleader's addiction to power, September 1988.
"Strategic choice of flexible productiontechnologies and welfare implications",October 1988
"Method of moments tests of contingent claimsasset pricing models", October 1988.
"Size-sorted portfolios and the violation ofthe random walk hypothesis: Additionalempirical evidence and implication for testsof asset pricing models", June 1988.
"Data transferability: estimating the responseeffect of future events based on historicalanalogy", October 1988.
"Assessing economic inequality", November 1988.
"The interpersonal structure of decisionmaking: a social comparison approach toorganizational choice", November 1988.
"Is mismatch really the problem? Some estimatesof the Chelvood Cate II model with US data",September 1988.
"Modelling cost structure: the Bell Systemrevisited", November 1988.
"Regulation, taxes and the market for corporatecontrol in Belgium", September 1988.
88/63 Fernando NASCIMENTOand Vilfried R.VANHONACKER
88/64 Kasra FERDOWS
88/65 Arnoud DE MEYERand Kasra FERDOVS
88/66 Nathalie DIERKENS
88/67 Paul S. ADLER andKasra FERDOWS
1989
89/01 Joyce K. BYRER andTavfik JELASSI
89/02 Louis A. LE BLANCand Tavfik JELASSI
89/03 Beth H. JONES andTavfik JELASSI
89/04 Kasra FERDOVS andArnoud DE MEYER
89/05 Martin KILDUFF andReinhard ANGELMAR
89/06 Mihkel M. TOMBAK andB. SINCLAIR-DESGAGNE
89/07 Damien J. NEVEN
89/08 Arnoud DE MEYER andHellmut SCHUTTE
89/09 Damien NEVEN,Carmen MATUTES andMarcel CORSTJENS
89/10 Nathalie DIERKENS,Bruno GERARD andPierre BILLION
"Strategic pricing of differentiated consumerdurables in a dynamic duopoly: a numericalanalysis", October 1988.
"Charting strategic roles for internationalfactories", December 1988.
"Duality up, technology dovn", October 1988.
"A discussion of exact measures of informationassymetry: the example of Myers and Majlufmodel or the importance of the asset structureof the firm", December 1988.
"The chief technology officer", December 1988.
"The impact of language theories on DSSdialog", January 1989.
"DSS software selection: a multiple criteriadecision methodology", January 1989.
"Negotiation support: the effects of computerintervention and conflict level on bargainingoutcome", January 1989."Lasting improvement in manufacturingperformance: In search of a new theory",January 1989.
"Shared history or shared culture? The effectsof time, culture, and performance oninstitutionalization in simulatedorganizations", January 1989.
"Coordinating manufacturing and businessstrategies: I", February 1989.
"Structural adjustment in European retailbanking. Some view from industrialorganisation", January 1989.
"Trends in the development of technology andtheir effects on the production structure inthe European Community", January 1989.
"Brand proliferation and entry deterrence",February 1989.
"A market based approach to the valuation ofthe assets in place and the growthopportunities of the firm", December 1988.
89/11 Manfred KETS DE VRIESand Alain NOEL
89/12 Vilfried VANHONACKER
89/13 Manfred KETS DE VRIES
89/14 Reinhard ANGELMAR
89/15 Reinhard ANGELMAR
89/16 Untried VANHONACKER,Donald LEHMANN andFareena SULTAN
89/17 Gilles AMADO,Claude FAUCHEUX andAndre LAURENT
89/18 Srinivasan BALAK-RISHNAN andMitchell KOZA
89/19 Wilfried VANHONACKER,Donald LEHMANN andFareena SULTAN
89/20 Wilfried VANHONACKERand Russell VINER
89/21 Arnoud de MEYER andKasra FERDOVS
89/22 Manfred KETS DE VRIESand Sydney PERZOV
89/23 Robert KORAJCZYK andClaude VIALLET
89/24 Martin KILDUFF andMitchel ABOLAFIA
89/25 Roger BETANCOURT andDavid GAUTSCHI
89/26 Charles BEAN,Edmond MALINVAUD,Peter BERNHOLZ,Francesco GIAVAll1and Charles WYPLOSZ
"Understanding the leader-strategy interface:application of the strategic relationshipinterview method", February 1989.
"Estimating dynamic response models when thedata are subject to different temporalaggregation", January 1989.
"The impostor syndrome: a disquietingphenomenon in organizational life", February1989.
"Product innovation: a tool for competitiveadvantage", March 1989.
"Evaluating a firm's product innovationperformance", March 1989.
"Combining related and sparse data in linearregression models", February 1989.
"Changement organisationnel et realitèsculturelles: contrastes franco-americains".March 1989.
"Information asymmetry, market failure andjoint-ventures: theory and evidence",March 1989
"Combining related and sparse data in linearregression models",Revised March 1989
"A rational random behavior model of choice",Revised March 1989
"Influence of manufacturing improvementprogrammes on performance", April 1989
"Vhat is the role of character inpsychoanalysis? April 1989
"Equity risk premia and the pricing of foreignexchange risk" April 1989
"The social destruction of reality:Organisational conflict as social drama"April 1989
"Two essential characteristics of retailmarkets and their economic consequences"March 1989
"Macroeconomic policies for 1992: thetransition and atter", April 1989
89/27 David KRACKHARDT andMartin KILDUFF
89/28 Martin KILDUFF
89/29 Robert GOGEL andJean-Claude LARRECHE
89/30 Lars-Hendrik ROLLERand Mihkel M. TOMBAK
89/31 Michael C. BURDA andStefan GERLACH
89/32 Peter HAUG andTavfik JELASSI
89/33 Bernard SINCLAIR-DESGAGNE
89/34 Sumantra GHOSHAL andNittin NOHRIA
89/35 Jean DERMINE andPierre HILLION
89/36 Martin KILDUFF
89/37 Manfred KETS DE VRIES
89/38 Manfrd KETS DE VRIES
89/39 Robert KORAJCZYK andClaude VIALLET
89/40 Balaji CHAKRAVARTHY
89/41 B. SINCLAIR-DESGAGNEand Nathalie DIERKENS
89/42 Robert ANSON andTavfik JELASSI
89/43 Michael BURDA
89/44 Balaji CHAKRAVARTHYand Peter LORANGF
89/45 Rob VEIT7. andArnoud DE MEYER
"Friendship patterns and cultural attributions:the control of organizational diversity",April 1989
"The interpersonal structure of decisionmaking: a social comparison approach toorganizational choice", Revised April 1989
"The battlefield for 1992: product strengthand geographic coverage", May 1989
"Competition and Investment in FlexibleTechnologies", May 1989
"Intertemporal prices and the US trade balancein durable goods", July 1989
"Application and evaluation of a multi-criteriadecision support system for the dynamicselection of U.S. manufacturing locations",May 1989
"Design flexibility in monopsonisticindustries", May 1989
"Requisite variety versus shared values:managing corporate-division relationships inthe M-Form organisation", May 1989
"Deposit rate ceilings and the market value ofbanks: The case of Prance 1971-1981", May 1989
"A dispositional approach to social networks:the case of organizational choice", May 1989
"The organisational fool: balancing a leader'shubris", May 1989
"The CEO blues", June 1989
"An empirical investigation of internationalasset pricing", (Revised June 1989)
"Management systems for innovation andproductivity", June 1989
"The strategic supply of precisions", June 1989
"A development framework for computer supportedconflict resolution", July 1989
"A note on firing costs and severance benefitsin equilibrium unemployment", June 1989
"Strategic adaptation in multi-business firms",
June 1989
"Managing expert systems: a framework and casestudy", June 1989
89/46
89/47
Marcel CORSTJENS,Carmen MATUTES andDamien NEVEN
Manfred KETS DE VRIESand Christine MEAD
"Entry Encouragement", July 1989
"The global dimension in leadership andorganization: issues and controversies",
89/64(TM)
89/65(TN,AC, PIN)
Enver YUCESAN andLee SCHRUBEN
Soumitra DUTTA andPiero BONISSONE
"Complexity of simulation models: A graphtheoretic approach", November 1989
"MARS: A mergers and acquisitions reasoningsystem", November 1989
April 1989
Damien NEVEN andLars-Hendrik ROLLER
"European integration and trade flows",August 1989
89/66 B. SINCLAIR-DESGAGNE(TH,EP)
"On the regulation of procurement bids",November 1989
Jean DERMINE "Home country control and mutual recognition",July 1989
89/67 Peter BOSSAERTS and(PIN) Pierre BILLION
"Market microstructure effects of governmentintervention in the foreign exchange market",December 1989
Jean DERMINE "The specialization of financial institutions,the EEC model", August 1989
89/48
89/49
89/50
89/51 Spyros MAKRIDAKIS
89/52 Arnoud DE MEYER
89/53 Spyros MAKRIDAKIS
89/54 S. BALAKRISHNANand Mitchell KOZA
89/55 H. SCHUTTE
89/56 Vilfried VANHONACKERand Lydia PRICE
89/57 Taekwon KIM,Lars-Hendrik ROLLERand Mihkel TOMBAK
89/58 Lars-Hendrik ROLLER(EP,TM) and Mihkel TOMBAK
89/59 Manfred KETS DE VRIES,(08) Daphna ZEVADI,
Alain NOEL andMihkel TOMBAK
89/60 Enver YUCESAN and(TM) Lee SCHRUBEN
89/61 Susan SCHNEIDER and(All) Arnoud DE MEYER
89/62 Arnoud DE MEYER(TM)
89/63 Enver YUCESAN and(TM) Lee SCHRUBEN
"Sliding simulation: a new approach to timeseries forecasting", July 1989
"Shortening development cycle times: amanufacturer's perspective", August 1989
"Vhy combining works?", July 1989
"Organisation costs and a theory of jointventures", September 1989
"Euro-Japanese cooperation in informationtechnology", September 1989
"On the practical usefulness of meta-analysisresults", September 1989
"Market growth and the diffusion ofmultiproduct technologies", September 1989
"Strategic aspects of flexible productiontechnologies", October 1989
"Locus of control and entrepreneurship: athree-country comparative study", October 1989
"Simulation graphs for design and analysis ofdiscrete event simulation models", October 1989
"Interpreting and responding to strategicissues: The impact of national culture",October 1989
"Technology strategy and international R 6 Doperations", October 1989
"Equivalence of simulations: A graph theoreticapproach", November 1989
90/16 Richard LEVICH and "Tax-Driven Regulatory Drags European1990 FIN Ingo WALTER Financial Centers in the 1990's", January 1990
90/01TM/EP/AC
B. SINCLAIR-DESGAGNE "Unavoidable Mechanisms", January 1990 90/17FIN
Nathalie DIERKENS "Information Asymmetry and Equity Issues",Revised January 1990
90/02EP
Michael BURDA "Monopolistic Competition, Costs ofAdjustment, and the Behaviour of EuropeanManufacturing Employment", January 1990
90/18MKT
Wilfried VANHONACKER "Managerial Decision Rules and the Estimationof Dynamic Sales Response Models", RevisedJanuary 1990
90/03TM
Arnoud DE MEYER "Management of Communication in InternationalResearch and Development", January 1990
90/19 Beth JONES and "The Effect of Computer Intervention and TaskTM Tavfik JELASSI Structure on Bargaining Outcome", February
90/04 Gabriel HAVAVINI and "The Transformation of the European Financial 1990
PIN/EP Eric RAJENDRA Services Industry: Prom Fragmentation toIntegration", January 1990 90/20
TMTavfik JELASSI,
Gregory KERSTEN and"An Introduction to Group Decision andNegotiation Support", February 1990
90/05 Gabriel HAVAVINI and "European Equity Markets: Tovard 1992 and Stanley ZIONTSFIN/EP Bertrand JACOUILLAT Beyond", January 1990
90/06PIN/EP
Gabriel HAVAVINI and "Integration of European Equity Markets:Eric RAJENDRA Implications of Structural Change (or Key
90/21FIN
Roy SMITH and
Ingo WALTER"Reconfiguration of the Global SecuritiesIndustry in the 1990's" • February 1990
90/07
Market Participants to and Beyond 1992",January 1990
Gabriel HAVAVINI "Stock Market Anomalies and the Pricing of
90/22
FINIngo WALTER "European Financial Integration and Its
Implications for the United States", February1990
FIN/EP Equity on the Tokyo Stock Exchange', January1990
90/23 Damien NEVEN "EEC Integration towards 1992: SomeEP/Sli Distributional Aspects", Revised December 1989
90/08TM/EP
Tavfik JELASSI and "Modelling vith MCDSS: What about Ethics'",B. SINCLAIR-DESGAGNE January 1990 90/24 Lars Tyge NIELSEN "Positive Prices in CAPM", January 1990
FIN/EP
90/09 Alberto CIOVANNINI "Capital Controls and International Trade
EP/PIN and Jae VON PARK Finance", January 1990 90/25 Lars Tyge NIELSEN "Existence of Equilibrium in CAPM", JanuaryFIN/EP 1990
90/10 Joyce BRYER and "The Impact of Language Theories on DSSTM Tavfik JELASSI Dialog", January 1990 90/26 Charles KADUSHIN and "Why netvorking Pails: Double Binds and the
90/11TM
Enver YUCESAN "An Overviev of Frequency Domain Methodologyfor Simulation Sensitivity Analysis",
January 1990
OB/BP
90/27TM
Michael BRIM
Abbas FOROUGHI andTavfik JELASSI
Limitations of Shadow Networks", February 1990
"NSS Solutions to Major Negotiation StumblingBlocks", February 1990
90/12 Michael BURDA "Structural Change, Unemployment Benefits and
EP nigh Unemployment: A U.S.- European 90/28 Arnoud DE MEYER "The Manufacturing Contribution toComparison", January 1990 TM Innovation", February 1990
90/13 Soumitra DUTTA and "Approximate Reasoning about Temporal 90/29 Nathalie DIERKENS "A Discussion of Correct Measures ofTM Shashi SHEKHAR Constraints in Real Time Planning and Search',
January 1990
FIN/AC Information Asymmetry", January 1990
90/30 Lars Tyge NIELSEN "The Expected Utility of Portfolios of90/14 Albert ANCEHRN and "Visual Interactive Modelling and Intelligent FIN/EP Assets", March 1990TM Hans-Jakob LOTHI DSS: Putting Theory Into Practice',
January 1990
90/15 Arnoud DE MEYER, "The Internal Technological Reneval of a90/31MKT/EP
David GAUTSCHI andRoger BETANCOURT
"What Determines U.S. Retail Margins?",February 1990
TM Dirk DESCHOOLMEESTER, Business Unit with a Mature Technology",Rudy MOENAERT and January 1990Jan BARBE
90/32SM
Srinivasan BALAK-RISHNAN and
"Information Asymmetry, Adverse Selection andJoint-Ventures: Theory and Evidence",
Mitchell KOZA Revised, January 1990
90/33
OBCaren SIEHL,David BOWEN and
"The Role of Rites of Integration in ServiceDelivery", March 1990
Christine PEARSON
90/34 Jean DERMINE
"The Gains from European Banking Integration,
FIN/EP
a Call for a Pro-Active Competition Policy",
April 1990
90/35 Jae Won PARK
"Changing Uncertainty and the Time-VaryingEP
Risk Premia in the Term Structure of Nominal
Interest Rates", December 1988, Revised
March 1990
90/36 Arnoud DE MEYER
"An Empirical Investigation of ManufacturingTM
Strategies in European Industry", April 1990
90/37 William CATS-BARIL
"Executive Information Systems: Developing anTM/OB/SM
Approach to Open the Possibles", April 1990
90/38 Vilfried VANHONACKER "Managerial Decision Behaviour and theHKT
Estimation of Dynamic Sales Response Models",
(Revised February 1990)
90/39 Louis LE BLANC and
"An Evaluation and Selection Methodology forTM Taufik JELASSI
Expert System Shells", May 1990
90/40 Manfred KETS DE VRIES "Leaders on the Couch: The case of RobertoOB
Calvi", April 1990
90/41 Gabriel HAWAWINI, "Capital Market Reaction to the AnnouncementFIN/EP Itzhak SWARY and
of Interstate Banking Legislation", March 1990
Ik WAN JOG
90/42 Joel STECKEL and
"Cross-Validating Regression Models inMKT Wilfried VANHONACKER Marketing Research", (Revised April 1990)
90/43 Robert KORAJCZYK and "Equity Risk Premia and the Pricing of ForeignFIN Claude VIALLET
Exchange Risk", May 1990
90/44 Gilles AMADO, "Organisational Change and Cultural Realities:OH Claude FAUCHEUX and
Franco-American Contrasts", April 1990
Andre LAURENT
90/45 Soumitra DUTTA and
"Integrating Case Based and Rule BasedTM Piero BONISSONE
Reasoning: The Possibilistic Connection",
May 1990
Page 1Page 2Page 3Page 4Page 5Page 6Page 7Page 8Page 9Page 10Page 11Page 12Page 13Page 14Page 15Page 16Page 17Page 18Page 19Page 20Page 21Page 22Page 23Page 24Page 25Page 26Page 27Page 28Page 29Page 30Page 31Page 32Page 33Page 34Page 35Page 36Page 37Page 38Page 39Page 40Page 41Page 42Page 43Page 44Page 45Page 46Page 47
Top Related