Académique Documents
Professionnel Documents
Culture Documents
GDP data for the entire 1959-1991 period offer the basis to test the
ability of a simple constant growth model to describe the trend in GDP over
time. However, such a regression model cannot be used to forecast GDP over
any subpart of that period. To do so would be to overstate the forecast
capability of the regression model since, by definition, the regression line
minimizes the sum of squared deviations over the estimation period. To test
forecast reliability, it is necessary to test the predictive capability of a given
model over data that was not used to generate the regression model. These
data offer an interesting basis for evaluating the usefulness of the simple
growth model approach to economic forecasting. In the absence of GDP data
for the 1991-1995 period, the reliability of alternative forecast technique can
be illustrated by arbitrarily dividing historical GDP data into two subsamples:
a 1959-1983 test group, and a 1984-1991 forecast group. A regression model
estimated over the test group can be used to forecast actual GDP over the
1984-1991 period. Estimation results over the 1959-1983 subperiod provide
a forecast model that can be used to evaluate the predicitive reliability of a
constant growth model over the 1984-1991 forecast period.
Table 1. Gross Domestic Product in Current-Year (Nominal)
Dollars and in 1987 Dollars, 1959-91
B. Use the simple regression model approach to estimate, in turn, the linear
relationship between the natural logarithm of (1) nominal GDP and time,
and (2) real GDP (ln GDP) and time. Estimate each model over two time
intervals: the entire 1959-1991 period and the 1959-1983 test subperiod,
where
lnYt = b0+b1Tt+ut
and where lnYt is the natural logarithm of each respective measure of GDP
in year t, and T is a time trend variable (where T1 = 1959, T2 = 1960, T3 =
1961,…, and T25 = 1983); and u is a residual term that includes the effects
of all factors that have been omitted from the regression models and the
effects of random or stochastic elements. These are called constant growth
models because they are based on the assumption of a constant percentage
growth in economic activity per year. How well does each constant growth
model fit actual GDP data?
C. Create a spreadsheet that shows actual and forecast GDP values for the
1984-1991 period. Each GDP forecast is derived using the coefficients
estimates for each model in Part B along with values for each respective
time trend variable over the 1984-1991 period. Remember that T26 = 1984,
T27 = 1986, T28 = 1987,…, and T33 = 1991 and that each constant growth
model provides predicted, or forecast, values for the relevant lnYt variable.
To obtain values for Yt, simply take the antilog (exponent) of each
predicted lnYt variable. Place these forecast values in the spreadsheet
alongside actual figures. Then, subtract actual figures from forecast values
to obtain annual estimates of forecast error for each Yt variable, plus an
estimate of average forecast error for each Yt variable over the 1984-1991
period.
D. Compute the correlation coefficient between actual and forecast values for
each Yt variable over the 1984-1991 period. Also compute the sample
mean forecast error for each Yt variable. Based on these findings, how well
do constant growth models generated using data over the 1959-1983
period forecast actual GDP data over the 1984-1991 period?