Military expenditure and economic growth: peru 1970–1996

Abstract

The present study examines the effects of military expenditure on growth in Peru in the period from 1970 to 1996. By using a Deger‐type Simultaneous Equations Model it is possible to break up the net effect into supply‐ and demand‐side influences. The former consist of positive externalities of defence activities on the other sectors of the economy, while the latter can be described as crowding‐out of civilian investment. Estimations find the supply‐side effects to be insignificantly different from zero, while the crowding‐out effect of defence spending is significant and substantial. It is thereby established that defence expenditure has a negative overall effect on economic growth in Peru. Although several caveats – including specification problems of the Deger model, the quality of the data used, a relatively small sample and the presence of autocorrelation in the estimations – must be considered, these results turn out to be quite robust with respect to estimation methods (3SLS, 2SLS, OLS) and slight modifications to the model. They are also consistent with previous empirical findings from other countries and cross sectional studies.

Keywords:

• Military Expenditure
• Economic Growth
• Peru
• Deger‐type Simultaneous Equations Model

Notes

* E‐mail: [email protected]

For a discussion of the different channels of influence, see Sandler and Hartley (1995).

Good analytical summaries on this literature can be found in Sandler and Hartley (1995) and Ram (1995).

Dunne and Nikolaidou (2001); Dunne et al. (2000); Lebovic and Ishaq (1987); Scheetz (1991); Sezgin (2001).

Later versions of the model include additionally an equation on the trade balance, which will not be used here. As shall be commented on subsequently, the Deger‐model suffers from several theoretical inconsistencies; including the trade balance equation as proposed by Deger (1986) would further increase this problem, as it is not developed from a solid theoretical framework.

The scope of this paper does not permit a presentation and discussion of structuralist theory. A very condensed summary, further references and a justification of the inclusion of the agricultural growth variable can be found in Deger (1986, pp. 200–203). It should be mentioned, however, that the growth of any sector of the economy should be expected to be positively correlated with overall growth, as GDP growth can always be expressed as a weighted average of sector growth rates. Therefore, the significantly positive parameter estimates to the agriculture variable Deger and Smith (1983) and Deger (1986) find in their studies do nothing to support (or put into doubt) the basic concept of their model.

Of these three variables that represent the foreign sector, Deger and Smith (1983) and most other empirical studies based on this model use the trade balance. Moreover, as was already mentioned, the more recent versions of the model treat the trade balance ratio as an endogenous variable and include a fourth equation to the system (see, for example, Deger, 1986, or Dunne and Nikolaidou, 2001). Only Scheetz (1991), who also works with a four‐equation model, uses the current account variable. This study will follow his example, because b is the most adequate variable in the growth equation.

Deger and Smith (1983, p. 342).

Deger (1986) briefly mentions that inflation might cause ‘forced savings’; however, arguments that favour a negative effect of inflation on savings can easily be found; for example, inflation might induce agents to anticipate consumption expenditures to protect the purchasing power of their income.

Smith (1995, pp. 69–88).

The conflict between Peru and Ecuador was brought to an end in a peace accord in 1998.

Several other environmental variables not included in equation (3) might be experimented with. Scheetz (1990) argues that defence spending was especially high during the military rule under generals Velasco (1968–1975) and Morales Bermúdez (1975–1980). The same author includes a dummy for years of military dictatorship in Scheetz (1991). Alternatively, a dummy variable might be incorporated to capture eventual effects of the already mentioned armed border conflicts between Peru and Ecuador.

This paper is indebted to Macroconsult S.A., which kindly permitted the use of their transformed data, and to Apoyo Consultoría, which provided data on terrorist activity.

For an analysis of the economic crises of the 1980s, see Paredes and Sachs (1991) and Dornbusch and Edwards (1990). For an assessment of the stabilization programme from 1990 onwards, see Klein (1995).

See Brzoska (1995) and Ball (1984).

This view is shared by Scheetz (1990).

Unless stated otherwise, this paper will work with a significance level of 10% in double‐sided tests.

However, it can be argued that OLS estimates might nevertheless be considered: in small samples, alternative estimators will also be biased if normality cannot be taken for granted. The main difference between 2SLS and 3SLS consists of the fact that, in the former, (a single‐equations method), each structural equation is estimated separately, while in the latter (a systems method), they are estimated together as a set. While 3SLS is in general asymptotically more efficient than 2SLS, it is also more sensitive to misspecifications. If, for example, one structural equation of the system is misspecified, this will render all 3SLS‐estimators of the system inconsistent, while in 2SLS only the parameters of the flawed equation will be affected. In the present case, where the theoretical foundations of two equations of the system are being questioned, it is therefore convenient to report both results. For a condensed discussion of the methods used see Kennedy (1998, pp. 157–168).

Considering that in the case of these two variables strong a priori beliefs justify the assumption that their parameters should be non‐negative, one‐sided t‐tests could also be applied. Higher power then leads to significance of Chilean expenditure at the 10% level in the 3SLS‐estimation, and to significance at the 5% level of both variables in all other estimations. No other variables turn from insignificance to significance by this reasoning, with the exception of DNG in the 2SLS‐estimation.

Several other pure time‐series studies, like those of Dunne and Nikolaidou (2001) and Dunne et al. (2000), also find insignificance of the GDP variable in the military burden equation.

Note, however, that including the non‐military public spending variable leads to a problem of double counting in the savings equation.

Although methods have been devised for correcting for autocorrelation, this paper adopts the position of Mizon (1995), who shows that, because of the very strong restrictions such correction procedures impose on the time structure of a model, they generally lead to inconsistent estimates. Autocorrelation should therefore not be corrected for.