Military Spending: Is the Peace Dividend Real or Illusory
by
Dale Bremmer
Rose-Hulman Institute of Technology
and
Randall Kesselring
Military Spending: Is the Peace Dividend Real or Illusory
Abstract
This paper investigates the relationship between defense spending and real per capita gross domestic product for the G-7 nations during the years 1963-1990. Using a two-way, fixed-effects regression model, time series and cross sections are pooled. Per capita real GDP is assumed to be a function of per capita military spending, per capita government spending for nonmilitary purposes, per capita capital stock, per capita investment, mean years of secondary education, percentage of the population between the ages of 15 and 65, and population growth rate. Results from the fixed effects model indicate a statistically significant, inverse relationship between defense spending and per capita GDP. Results also tend to verify the importance of investment, human capital, and size of the labor force relative to total population.
I. Introduction
With the demise of the
In investigating the relationship between defense spending and
GDP, the post World War II period offers a rich data sample that has seen
defense spending in industrialized nations with market economies come almost
full circle. At the end of World War II,
the emergence of the Cold War saw the beginnings of the arms race and the
growth in defense spending among those nations belonging to NATO and the Warsaw
Pact, in general, and in the
A brief literature review of the relationship between defense expenditures and economic performance is provided in the next section. Part three provides an explanation of the statistical model used in the estimation procedures along with a description of the data. Results are analyzed in the fourth section, which is followed by some tentative conclusions and recommendations for additional work.
II. Literature Review
The impact of defense spending on investment
Past papers on the impact of an industrialized nation’s defense spending on its economy have analyzed how military spending affects investment, employment and economic growth.[1] Studies have concluded that, in the case of developed nations, investment and military expenditures are substitutes (Kennedy 1983, p. 198). Smith (1977, 1978, 1980, and 1983) argues that workers resist cuts in private consumption and public welfare. Given an exchange rate and a level of capacity utilization, the remainder of the nation’s output is divided between defense spending and investment. Smith argues that higher levels of military spending would necessarily imply reduced investment. In a study of 15 countries between 1960 and 1970, Smith found that investment and defense spending were negatively correlated (-0.73). Lindgren (1984) points out that most studies are in agreement on the negative correlation between defense spending and investment spending for industrialized, market economies.
The impact of defense spending on employment
Regarding the impact of military expenditures on employment, the Marxist critique of the 1960s was that defense spending was a necessary, albeit wasteful, policy to stabilize and expand capitalism (Baran and Sweezy 1968). They argued that raising defense spending could solve the problems of underconsumption and the unemployment associated with it. Furthermore, capitalism would resort to such spending in an effort to reduce class conflict. The hypothesis that increased military spending indirectly creates employment in the armaments industry and directly creates more jobs in the armed forces does not necessarily need to rely on the tenants of Marxist economics for support. Indeed, one of the primary concerns regarding disarmament at the end of the Cold War was its hypothesized relationship to rising unemployment.
In spite of this intuitive relationship,
The Impact of Defense Spending On GDP
Given the long-accepted, theoretical direct relationship between investment and economic growth, if defense spending has a negative impact on investment, then it would seem reasonable that defense spending would have an adverse impact on economic growth. This was exactly the findings of two studies published in the seventies, Szymanski (1973) and Lee (1973). Some studies attribute the negative effect of defense spending on economic growth to reduced investment.[2] Another study argues that defense spending restricts export growth and economic growth because military expenditures compete for the same resources used in the production of exports.[3]
However, other studies were unable to find any stable
relationship between military spending and economic growth.[4]
III. The Econometric Model and Data
The two-way fixed
effects model
To analyze the relationship between defense spending and economic growth in each of the countries belonging to the G-7, annual data between 1964 and 1990 was used in a “two way” fixed and random effects model that pools cross sections and time series. The specific form of the fixed effects model estimated was
Yit is per capita real GDP in country i observed in time period t. Xit represents a vector of independent variables hypothetically related to GDP. The symbol, ai (i = 1, ¼7), represents a set of binary (dummy) variables used to capture country-specific effects while the symbol, lt (t = 1964, ¼1990), represents another set of binary variables used to capture time-specific effects.[5]
Of course, xit is the random error associated with each observation for country i in time t.
The two-way random effects model
The two-way, random-effects model assumes that the group effects (countries) and time effects (years) impact the random error of the regression. The model becomes:
Here ui represents the error term associated with the random group effects, and wt represents the error term associated with the random time effects. Once again Cit represents an array of variables used to explain GDP.
The Data
Annual raw data for all the variables in both models were obtained from various issues of the government publication World Military Expenditures and Arms Transfers. This report is published every two or three years by the U.S. Arms Control and Disarmament Agency. The 1964-1995 data series is the longest period of data that this agency reports. To augment these variables a data set created by Nehru and Dhareshwa (1993) was downloaded from the World Bank’s web site. This data set provides several variables that are not usually available (human capital variables for example) from ordinary sources. Unfortunately, the time periods of available data change from country to country and from variable to variable. Thus, the latest usable annual data available from this source was 1990.
Table 1 provides a list of the variables used in the estimation routines. The variables fall into three categories: physical capital, labor force, and government spending indicators. The only valid, available physical capital variable was investment (INVEST). Consequently, its per capita value was used both in its current form and in its lagged form. Three labor force variables were selected. The average number of years of secondary education (EDUCATE) attained by the population, the ratio of the population between the ages of 15 and 65 to the total population (LFORCE), and the difference between total population and its lagged value in log form (POPGROWTH). Two variable were used to represent the government spending category, per capita military spending and per capita central government spending less military spending.
IV. Estimation Results
The estimation results for the two-way, fixed-effects model are reported in Table 2 and Table 3.[6] Table 2 lists the estimated coefficients and t-scores for the variables included in the model. Most of the results were as expected. The estimated coefficient for INVEST is positive and significant at the 5% level. The estimated coefficient for the lagged investment term (INVEST[-1]) is also positive but not quite significant for a two-tailed test. The human capital variable--EDUCATE--also has a positive and significant coefficient. Also, as expected, LFORCE--the percentage of the population between ages 15 and 64--had a positive and
significant impact on GDP. The coefficient for the remaining population variable--POPGROWTH--had the expected negative sign but was insignificant at any conventionally accepted level of significance. The estimated coefficient for government spending less military spending--GOVERNMENT--was positive but insignificant. None of these results were unexpected or at odds with most other published studies. Of course, the most interesting result
produced by this estimation procedure was the negative and significant coefficient for the military spending variable--DEFENSE. In fact, according to the estimates a one percent increase in military spending reduces per capita GDP by $3,695--not an insignificant amount.
Specification tests
The
appropriateness of the two-way, fixed-effects model was tested in several
ways. First, likelihood ratio tests were
used to investigate the significance of applying various restrictions (group
effects, time effects, and classical regression fit) to the model. These tests result from assuming five
alternative possible models. These five
alternatives are:
1. Per capita GDP = f(constant term only)
2. Per capita GDP = f(group effects only)
3. Per capita GDP = f(constant, regressors)
4. Per capita GDP = f(constant, regressors,
group effects)
5. Per capita GDP = f(constant, regressors,
group effects, time effects)
The restrictions and relevant Chi-squared statistics are
provided in Table 3. All the
restrictions except the time effects restriction are statistically significant
at the one percent level. So, there is
no statistical difference between the two-way, fixed effects model with both
time and group effects and the one-way fixed effects model with only group
(country) effects.
So, the
results of a one-way, fixed-effects model are presented in Table 4. The results are very similar to those
produced by the two-way procedure. One
difference is that EDUCATE retains a positive coefficient but that coefficient
is no longer significantly different from zero.
Also, the coefficient for government spending takes a negative sign and
becomes significant. There is virtually
no change in the estimated coefficient for military spending. It is still negative and significant.
One final specification test was performed. A Hausman test was used to provide some evidence on the choice between a fixed-effects and random-effects estimator. The calculated Chi-squared value was 43.25 (See Table 2). With seven degrees of freedom, this score easily meets the requirements of a one-percent test of significance. Consequently, for this set of data a fixed-effects model is the best choice.
V. Conclusions
This paper indicates that an inverse relationship between per capita GDP and military expenditures existed in the G-7 countries between 1964 and 1990. The implication is that there is a real economic cost to maintaining a significant military presence and that cost is in reduced levels of per capita GDP. Society faces a distinct tradeoff between perceived national security needs, international prestige, and domestic economic performance.
Several enhancements could possibly add some weight to the conclusions of this research. First, the data set could be expanded to include all the OECD nations. A broader sample might give a different picture of the relationship between defense spending and economic growth among both the industrialized and the developing nations. If the time portion of the data set could be extended it might be possible to test for coefficient stability over time. Finally, the most difficult question to explore is that of the theoretical cause of the inverse relationship between defense spending and economic performance? Is this inverse relationship due to diminished investment, reduced export growth, reduced efficiency in the labor market, or inappropriate physical and human capital formation? Without doubt many of these questions will be further illuminated by future researches.
References
Ahmed, S. “Temporary and Permanent Government Spending in an Open Economy,” Journal of Monetary Economics, 17, 2, 1986, pp. 405-419.
Baran, Paul and Paul Sweezy. Monopoly Capital. Harmondsworth: Penquin, 1968.
Blackaby, Frank and Thomas Ohlson. “Military Expenditure and Arms Trade: Problems of Data,” Bulletin of Peace Proposals, 13, 4, 1982, pp. 291-308.
Bremmer, Dale and Randy Kesselring, “The Opportunity Cost of Super Power Status: The Tradeoff Between Defense Spending and Economic Prosperity,” unpublished manuscript, March 1998.
Cappelen, Adne, Nils Petter Gleditsch and Olav Bjerkholt. “Military Spending and Economic Growth in the OECD Countries,” Journal of Peace Research, 21, 4, 1984, pp. 375 - 387.
Chowdhury, Abdur. “A Causal Analysis of Defense Spending and Economic Growth,” Journal of Conflict Resolution, 35, 1, 1991, pp. 80-97.
deGrasse Jr., Robert W. Military Expansion Economic
Decline: The Impact of Military Spending
on U.S. Economic Performance.
Dunne, P. and R. Smith. “Military Expenditure and Unemployment in the OECD,” Defense Spending, 1, 1990, pp. 57-73.
Faini, Riccardo, Patricia Annez, and Lance Taylor. “Defense Spending , Economic Structure and Growth: Evidence Among Countries and Over Time,” Economic Development and Cultural Change, 32, 3, 1984, pp. 487-498.
Hausman, J. "Specification Tests in Econometrics," Econometrica, 46, 1978, pp. 1251-1271.
Judge,
George G., R. Carter Hill, William E. Griffiths,
Helmut Lütkepohl, and Tsoung-Chao
Lee. Introduction to the
Theory and Practice of Econometrics.
Kennedy,
Gavin. Defense Economics.
Lee,
Jong Ryol. “Changing
National Priorities of the
Society,
Bruce M. Russett and Alfred Stepan,
eds.
Lindgren, Göran. “Armaments and Economic Performance in Industrialized Market Economies,” Journal of Peace Research, 21, 4, 1984, pp. 375-387.
Nardinelli, Clark and Gary B. Ackerman. “Defense Expenditures and the Survival of American Capitalism,” Armed Forces and Society, 3, 1, 1976, pp. 13-16.
Nehru, Vikram, and Ashok Dhareshwar. "A New Database on Physical Capital Stock: Sources, Methodology and Results," Revista de Analisis Economico, 8, 1, 1993, pp. 37-59.
Paul, Satya. “Defense Spending and Unemployment Rates: An Empirical Analysis for the OECD,” Journal of Economic Studies, 23, 2, 1996, pp. 44-54.
Rotschild, Kurt W. “Military Expenditure, Exports, and Growth,” Kyklos, 26, 4, 1973, pp. 804-813.
Smith,
Ron P. “Military Expenditure and Capitalism,”
Smith,
Ron P. “Military Expenditure and Capitalism: A Reply,”
Smith, Ron P. “Military Expenditure and Investment in OECD Countries, 1954-1973,” Journal of Comparative Economics, 4, 1980, pp. 19-32.
Smith Ron P. and George Georgiou. “Assessing the Effect of Military Expenditure on OECD Economies: A Survey,” Arms Control, 14, 1, 1983, pp. 3-15.
Szymanski, Albert. “Military Spending and Economic Stagflation,” American Journal of Sociology, 79, 1, 1973, pp. 1-14.
Weede, Erich 1983. “Military Participation rations, Human Capital Formation, and Economic Growth: A Cross-national Analysis,” Journal of Political and Military Sociology, 11(Spring), 11-19.
Table 1
List of Variables
Variable |
Definition |
PCGDP |
Gross Domestic Product Per capita |
INVEST |
Log of the dollar value of investment per capita |
DEFENSE |
Log of defense expenditures per capita |
GOVERNMENT |
Log of the difference between per capita government spending and per capita defense spending |
EDUCATE |
Log of the average number of years of secondary education |
LFORCE |
Log of the ratio of the working age (15-64) population to the total population |
POPGROWTH |
Difference between the log of total population and its lagged value |
Table 2
Estimation Results for the Two-Way, Fixed-Effects Model
(dependent variable: per-capita GDP)
([-1] indicates a one period lag)
Variable |
Coefficient |
t-ratio |
Constant |
-19030.25 |
-2.17** |
INVEST |
3892.55 |
2.46** |
INVEST[-1] |
2490.51 |
1.37 |
DEFENSE[-1] |
-3695.19 |
-2.58** |
EDUCATE |
2386.97 |
1.92*** |
GOVERNMENT[-1] |
313.27 |
0.26 |
POPGROWTH |
-51089.22 |
-1.09 |
LFORCE |
10494.06 |
3.11* |
R2 |
0.98 |
|
Hausman |
43.25* |
|
n |
168 |
|
*Significant at the 1% level for a two-tailed test.
** Significant at the 5% level for a two-tailed test.
***Significant at the 10% level for a two-tailed test
Table 3
Likelihood Ratio Tests on Alternative Specifications of the Fixed-Effects Model
Description |
H0 |
HA |
Chi-squared statistic |
degrees of freedom |
P-value |
No group effects |
Model 1 |
Model 2 |
522.66 |
6 |
0.0000 |
No regression fit of Y on X’s |
Model 1 |
Model 3 |
452.87 |
7 |
0.0000 |
No group effects or regression fit |
Model 1 |
Model 4 |
628.74 |
13 |
0.0000 |
Group effects but no regression fit |
Model 2 |
Model 4 |
106.08 |
7 |
0.0000 |
Regression fit but no group effects |
Model 3 |
Model 4 |
175.86 |
6 |
0.0000 |
Regression fit and group effects but no time effects |
Model 4 |
Model 5 |
12.86 |
23 |
0.9850 |
Regression fit but no time or group effects |
Model 3 |
Model 5 |
188.73 |
30 |
0.0000 |
Table 4
Estimation Results for the One-Way, Fixed-Effects Model
(group effects included, time effects omitted)
(dependent variable: per-capita GDP)
([-1] indicates a one period lag)
Variable |
Coefficient |
t-ratio |
INVEST |
3937.16 |
4.04* |
INVEST[-1] |
2298.09 |
1.96* |
DEFENSE[-1] |
-2426.43 |
-3.51* |
EDUCATE |
690.05 |
0.96 |
GOVERNMENT[-1] |
-1339.40 |
-1.83*** |
POPGROWTH |
-38133.22 |
-0.92 |
LFORCE |
11057.09 |
4.42* |
R2 |
0.98 |
|
n |
168 |
|
*Significant at the 1% level for a two-tailed test.
** Significant at the 5% level for a two-tailed test.
***Significant at the 10% level for a two-tailed test
[1]For a recent study of the relationship between military expenditures and economic growth in developing nations, see Chowdhury (1991).
[2]See Smith (1977 and 1978), Smith and Georgiou (1983) and Cappelen, et al (1984).
[3]See Rotschild (1973).
[4]See Nardinelli and Ackerman (1976), Faini et al. (1984), and de Grasse (1983).
[5]Because the estimation procedure includes an overall constant, as well as a group effect for every group and a time effect for every time period, it is necessary to impose the restriction E"i = E8i = 0 to eliminate the problem of perfect multicollinearity.
[6]The estimated coefficients for the large number of dummy variables in the model are not included in this paper. Those results will be provided to the interested reader.