DOES DEFENSE SPENDING REALLY PROMOTE AGGREGATE OUTPUT IN THE UNITED STATES?∗

∗ The views expressed in this paper are those of the authors and do not necessarily reflect the official policy or position of the US Air Force, the Department of Defense, or the US Government. The authors would like to express their thanks for the comments from Craig Stone and the other participants at the 2006 Canadian Economics Association conference, as well the comments received from three anonymous referees and the editor.

Abstract

Many studies have examined the relationship between defense spending and growth in real aggregate output with mixed results. Most recently, Atesoglu (2002) finds a positive relationship between defense spending and output. Capturing the error correction term as the long run adjustment parameter and including the long run adjustment in the relationship, we do not find evidence that defense spending promotes growth in real output. Instead, defense spending responds to aggregate income shocks. We re‐estimated the relationship and dummied all US military conflicts with similar results concerning military spending's effect on output. Interestingly, we find trade‐offs between defense and non‐defense government spending during war time.

∗ The views expressed in this paper are those of the authors and do not necessarily reflect the official policy or position of the US Air Force, the Department of Defense, or the US Government. The authors would like to express their thanks for the comments from Craig Stone and the other participants at the 2006 Canadian Economics Association conference, as well the comments received from three anonymous referees and the editor.

Keywords:

• Aggregate output
• Defense spending
• Cointegration
• Vector error correction model
• Military conflicts

JEL Codes:

• C32
• E12
• H56

Notes

∗ The views expressed in this paper are those of the authors and do not necessarily reflect the official policy or position of the US Air Force, the Department of Defense, or the US Government. The authors would like to express their thanks for the comments from Craig Stone and the other participants at the 2006 Canadian Economics Association conference, as well the comments received from three anonymous referees and the editor.

¹ Mehanna (2004) also does not find cointegration in a relationship that models the growth in real GDP per capita as a function of growth in military spending.

² Lau and Sin (1997) use two lags with annual data; Mehanna (2004) uses two lag lengths with quarterly data; Payne et al. (1994) use five lags with quarterly data; Karagol and Palaz (2004), conducting a similar exercise using Turkish data, use one lag with quarterly data.

³ More precisely, the two government spending variables include both consumption and gross investment expenditures. Defense and non‐defense spending are provided in BEA Table 3.9.5 (lines 11 and 16, respectively). Non‐defense spending contains Federal spending only; state and local government spending are not included. The real Aaa rate is calculated as the difference between the nominal rate and the annualized percentage change in the chained GDP deflator, where the percentage change is measured from the previous quarter to the current quarter. It has been suggested that a more appropriate inflation index may be to compare the GDP deflator for the current quarter versus the same quarter in the previous fiscal year. While year over year comparisons are of interest, we believe inflation expectations are set not as compared to the same time last year, but from a more short‐term viewpoint of what was inflation last quarter (or even month). Inflation could also be computed as a moving average of the last four quarters, for instance, but we find no consensus in the literature for a preferred method. All are seasonally adjusted.

⁴ Moody's Seasoned Aaa Corporate Bond Rate.

⁵ The Schwarz information criterion suggests using two lags and the Akaike information criteria suggests four lags. Estimating the model using 16 lags, as in Atesoglu, yields results that are similar when examining the cointegrating vector. However, there is weak evidence of cointegration (fail to reject the null of no cointegration with a p‐value of 0.11). Results, using 16 lags and Atesoglu's data range, also suggest that there is non‐adjustment in real GDP to changes in military spending in the long run.

⁶ Atesoglu (2002) also fails to find error‐correction behavior in non‐defense spending and the real Aaa rate (See Table II in Atesoglu, 2002).

⁷ The specific time horizons used are two quarters (i = 1 to 2), four quarters (i = 1 to 4), and eight quarters (i = 1 to 8) ahead.

⁸ The war dates were derived as follows: Kim (2005) reports that US troops were deployed 27 June 1950, while the armistice was signed 27 July 1953, thus the dates for the Korean war covers the period 1950:Q3 through 1953:Q2. The second period, 1965:Q2 through 1972:Q4 denotes the Vietnam War, as chronicled by Karnow (1983). The first combat troops arrived in March 1965, while the ceasefire was signed January 1973. The first Gulf war spans 1990:Q3 to 1991:Q1, as reflected in Yetiv (1997). The Global War on Terrorism essentially began after the September 11 attack.

⁹ The war dummy is interacted with the error‐correction term only in this VECM. It is not included in the cointegrating vector. The error‐correction in this VECM is derived from the results in Table I.