U.S. real GDP is forecast to grow approximately 3% per year throughout 1999 and 2000.  The forecast is based on a Keynesian aggregate expenditures model which assumed fixed interest rates, consumption spending, government spending, and net exports.  The MPC and MPI are 56% and 23% estimated from 1992.1-1998.3 quarterly data.  This implies an historically low Keynesian autonomous expenditures multiplier of 1.52.  The interest elasticity of investment is (positive)16%, also estimated from 1992.1-1998.3 data. Investment spending is projected into the future with a linear time trend as all the other components of aggregate expenditures are held constant.  GDP growth will moderate slightly over the next two years, alleviating the current labor shortage without creating significant new unemployment.  The government and private firms should avoid measures to substitute capital and other factors for labor.  The Federal Reserve should attempt to keep prices and interest rates as stable as possible. (JEL: E22, E66)
 

This paper forecasts U.S. gross domestic product (GDP) for the years 1999 and 2000. The explanatory variable is real private nonresidential fixed investment. The approach is based on the Keynesian aggregate expenditure model using new estimates of the marginal propensity to consume (MPC) and import (MPI) based on data for 1992-1998.  A Keynesian investment function is also estimated, providing an estimate of the interest elasticity of investment.  The Keynesian approach benefits from simplicity of implementation, and exploits the role of private investment expenditures in driving aggregate economic activity.

Real GDP is the most comprehensive measure of national income and output.  Continued growth of real GDP is necessary to sustain low unemployment and a rising standard of living.  A forecast downturn of GDP would predict a recession.  The forecast horizon of two years was chosen to minimize the possibility of external factors impacting the economy in an unforeseen way.  The forecast horizon is short enough to avoid seriously overstating or understating future GDP.

The rest of this paper is organized as follows: part 2. presents the data used to forecast GDP, estimate MPC and MPI, and estimate the investment function; part 3. presents the theoretical basis for the approach adopted in forecasting GDP; part 4. presents forecasts of GDP for 1999 and 2000; part 5. evaluates the importance of the forecast for the economy; and part 6. discusses conclusions for economic policy.
 

All variables are taken from the Federal Reserve Bank of St. Louis Federal Reserve Economic Data (FRED). The measures of GDP, Private Consumption Expenditures, Imports of Goods and Services, and Gross Private Domestic Investment are FRED variables GDPC92, PCEC92, IMPGSC92, and GPDIC92, are all real variables, and are given in billions of chained 1992 dollars at seasonally adjusted annual rates (SAAR). The primary source is the U.S. Department of Commerce Bureau of Economic Analysis.

The interest rate data are the secondary market 3-month T-bill rates given in FRED variable TB3MS, which is given as a percent discount. The primary source is the Board of Governors, U.S. Federal Reserve System. These are quarterly variables except for consumption (PCEC92) and the T-bill rate (TB3MS), which are monthly. The value given for the first month of each quarter (January, April, July, and October) was taken as the value for that quarter.

The sample period for the data used is from the first quarter of 1992 to the third quarter of 1998. This sample period runs from the end of the 1990-91 recession to the end of the available data. The forecast horizon is two years into the future. The data was first-differenced to estimate the MPC and MPI. The regression of interest rates on investment was done in logarithms because the relationship is probably nonlinear, and to estimate the interest elasticity of investment.

Real GDP was used as a measure of national output and income because it is the most encompassing and most commonly used measure. Alternative measures were considered, including real GNP, real net private income, and real disposable income, and any could have been used in this case, though they may not have provided equally good results. Alternative measures of consumption, investment, imports, and interest rates were also available. Private Consumption Expenditures and Private Domestic Investment, which exclude government investment and purchases of consumption goods, were used because the Keynesian model depends strongly on the consumption behavior of households and the investment behavior of private firms.

Although government spending also plays a major role in the Keynesian model, it is considered independent of GDP. (Gross) Imports of Goods and Services was used instead of Net Exports, because Gross Imports should respond more clearly to changes in U.S. GDP. The 3-month T-bill rate was used because it is the one interest rate the Federal Reserve System exerts very much control over, since the Fed is the primary buyer and holder of U.S. Treasury securities, and which also strongly influences market interest rates. For example, the Fed also controls the discount rate, but that has less direct influence on market interest rates.
 

This forecast assumes the interest rate will remain constant for the next two years and will have no impact on investment spending or GDP. The annual growth rates for investment spending will be calculated over the sample period of the data. The average annual growth rate will be used to project investment spending into the future for the eight quarters of 1999 and 2000 to provide a short-term projection of investment. The values for projected future investment will be used to calculate GDP forecasts for each future quarter using the Keynesian autonomous expenditures multiplier:

This is a simple model appropriate for forecasting short-term GDP growth.  The forecast should be interpreted as trend GDP in the absence of a recession.  This model is probably not adequate to forecast a recession, especially one caused by unanticipated real aggregate supply shocks.

The marginal propensity to save (MPS) is 0.43, from a Keynesian consumption function estimated with 1992.1-1998.3 quarterly data.  The Keynesian consumption function is written as:

where C is real aggregate consumption expenditures and Y is real GDP.  The consumption function was estimated in first-differences to address the effects of serial correlation of the data:

This equation was estimated with a constant or intercept to avoid the unrealistic restriction that the regression line passed through the origin.  The MPS = 1 - MPC.

The marginal propensity to import (MPI) is 0.23, from a Keynesian net export function estimated with 1992.1-1998.3 quarterly data.  The Keynesian net export function is written as:

where X is real net exports, EX is real gross exports, IM is real gross imports and is assumed to be a function of real GDP, and Y is real GDP.  The gross import function is written as:

First-differenced, the gross import function becomes:

The MPI in the gross import function is the same as in the net export function.  Like the consumption function, the first-differenced gross import function was estimated with an intercept to avoid the unrealistic restriction that the regression line passes through the origin.

The investment function is assumed to be nonlinear in a time index (t), which starts at zero for 1992.1 quarter and counts up by one for each subsequent quarter, and the secondary market three-month treasury bill interest rate (i):

which is written as it will be estimated in logarithms as:

The coefficient c is the interest elasticity of investment and shows the percentage change in total real investment spending for every one percent change in the interest rate.

One of the shortcomings of this paper's approach is that investment spending is known to be highly volatile and responds to changes in interest rates and other factors which are particularly difficult to predict. Thus, the GDP forecasts here are best interpreted as long-term trend real GDP, that is, what GDP should be in the absence of a recession. Actual GDP may deviate from its long-term trend. Since investment may not grow at a constant rate, and since the MPS and MPI may not be constant over long periods of time, a relatively short forecast horizon, here two years, is the most appropriate use of this model.

Dornbusch and Fischer (1978, p. 143) estimated an MPS of 12% = 0.12. They cite estimates ranging from 65% = 0.65 in the short run to 28% = 0.28 in the long run (Dornbusch and Fischer, 1978, p. 152), based on data for the 1960s and 1970s.  This should make it clear that it is very difficult to estimate the MPS, MPC, and MPI.
 

The consumption function, equation 2, was estimated with 1992.1-1998.3 quarterly data.  The regression estimate is (t-statistics in parentheses):

The estimated MPC is the coefficient on Y, indicating a value of 43% = 0.43 for the MPS.  The adjusted R-square of the estimate is 0.312, indicating approximately 30% of the variation of C is explained by variation in Y.  Low R-squares are not surprising for regressions with first-differenced data, which tend to amplify the impact of noise in the data.  The t-statistic of the MPC is greater than three, indicating strong rejection of the null hypothesis that the MPC = 0.  The t-statistic of the intercept is less than one, indicating failure to reject the null hypothesis that the intercept equals zero.

The gross import function, equation 3, was also estimated with 1992.1-1998.3 quarterly data.  The regression estimate is (t-statistics in parentheses):

The estimated MPI of 23% = 0.23 is the coefficient on Y.  The adjusted R-square of the estimate is 0.207, indicating approximately 20% of the variation of gross imports is explained by variation in real GDP.  As with the consumption function, low R-squares are not surprising for regressions with first-differenced data.  The t-statistic of the MPI is nearly three, indicating strong rejection of the null hypothesis that the MPI = 0.  The t-statistic of the intercept is nearly two, indicating rejection of the null hypothesis that the intercept equals zero.  One interpretation of this non-zero intercept is that the first-differenced gross import function is misspecified, which would mean our MPI estimate is biased upward.  Note this was not the case for the consumption function.

Based on the estimates of equations 2 and 3, MPS = 0.43 and MPI = 0.23.  The Keynesian autonomous spending multiplier 1/[MPS + MPC] = 1.52.

The investment function, equation 4, was also estimated with 1992.1-1998.3 quarterly data.  The regression estimate is (t-statistics in parentheses):

The estimated interest elasticity of investment is 16% = 0.16 indicates raising the interest rate by one percentage point causes investment spending to rise by 16%.  Theoretically, the interest elasticity should be negative, so this estimate will not be used for forecasting investment spending conditional on different choices of the rate of interest.  White (1956) cites survey data suggesting very low interest elasticity of investment, which may explain a positive estimate.
 
The investment function will only be used to project investment spending into the future assuming no change in interest rates.   The t-statistic for the constant and the interest elasticity of investment indicate strong rejection of the null hypotheses that the coefficients equal zero.  The t-statistic for the time index coefficient rejects the null hypothesis that the coefficient equals zero only at the 15% significance level or higher.

The adjusted R-square is 0.8399, indicating 84% of the variation in investment spending is explained by variation in interest rates and a linear time trend.  The F-statistic is 69.2, indicating strong rejection of the null hypothesis of zero slopes.

Equation 4 is used to project investment spending into the future for 1998.4-2000.4.  Real investment spending is projected to increase each quarter by an amount equal to the coefficient on the time index.  Changes in projected investment spending are multiplied by the Keynesian autonomous expenditures multiplier , as in equation 1, to get changes in real GDP:

which are added cumulatively to real GDP for 1998.3 quarter to get projected real GDP.  Results of this calculation are given in table 1.  All numbers are given in billions of chained 1992 dollars.
 

Table 1  Forecast Real GDP and Investment, 1998.4-2000.4
Quarter	Projected Real Investment (I)	Projected DI	Projected Change in Real GDP (Y)	Projected Real GDP
1998.4	1054	-277.6	-421.9	7145
1999.1	1058	4.08	6.20	7151
1999.2	1062	3.95	6.01	7157
1999.3	1065	3.83	5.83	7163
1999.4	1069	3.73	5.67	7168
2000.1	1073	3.62	5.51	7174
2000.2	1076	3.53	5.37	7179
2000.3	1080	3.44	5.23	7184
2000.4	1083	3.35	5.10	7189

 
This forecast projects investment spending to rise 14% throughout 1999 and 13% throughout  2000.  Real GDP is projected to increase 3.2% in 1999 and 2.9% in 2000.  Although a mild slowdown is predicted these growth rates are not unfavorable and are consistent with the consensus value of 3% for long-term U.S. real GDP growth.  This forecast suggests the U.S. economy will continue to expand at a moderate and sustainable rate, assuming no change in interest rates.
 

GDP forecasts are interesting to a broad range of people because GDP is the most comprehensive and encompassing measure of economic activity. Because the forecast presented here can be interpreted as trend GDP, it could be compared to the U.S. Department of Commerce's projected GDP data to provide early warning of a recession.  The forecast predicts a mild slowdown in the growth of real GDP and investment, but nothing too drastic.  The forecast suggests little possibility that the continued low unemployment of the current expansion can rise to the level where output prices are driven up.  Slightly higher unemployment levels seem to be indicated for the next two years, making it easier for employers to find employees, probably without creating much involuntary unemployment.
 

Real GDP is forecast to rise at approximately 3% per year, with real investment spending rising approximately 13-14%, for the next two years.

Assuming the forecast turns out to be correct, the many firms seeking to fill vacant job positions face some relief.  The situation is not going to get worse, at least not over the next two years.  It probably will become at least a little easier to fill job vacancies in the near future.  No very dramatic change is forecast.  Because of the shortage of workers, some firms are desperate to hire, and some have abandoned job searches.  Firms are well advised to avoid or delay extreme measures to substitute other resources for labor, such as capital.  The current labor shortage may be a persistent, long-term problem, but it is not going to get worse in the next two years.

Workers may not be able to continue to command such high wage increases, starting salaries, and signing bonuses, but the outlook for labor is similarly rosy.  Unemployment should not increase above, and perhaps not even up to, the natural rate of approximately 6%.  Although upward wage pressures should remain moderate compared to the past three years, the economic expansion and favorable job market appear far from over.

The forecast assumes no significant change in government spending over the next two years, which seems likely given projected budget surpluses.  The government is not likely to increase spending dramatically, which would cause the expansion to overheat, or decrease it, which could cause a recession.

The forecast assumes no change in interest rates over the next two years.  The Federal Reserve should be extremely careful not to change the interest rate unless it is clearly justified, and any change should be extremely moderate.  The interest rate is currently very low, and lowering farther would promote more investment, lower unemployment, and threaten inflationary pressures.  Raising the interest rate would also have negative effects of choking off the investment-fueled expansion, and raise unemployment, perhaps above the natural rate.