Does defence spending matter to employment in Taiwan?

Abstract

This paper investigates an important but neglected issue regarding the economic role of defence spending on employment in Taiwan. The study herein adopts official time series data of yearly defence spending, employment in the private sector, GDP, average monthly salary from 1966 to 2002, and the Autoregressive Distributed Lag (ARDL) approach to the cointegration proposed by Pesaran and Shin (1998) and Pesaran et al. (2001). The main finding of this study is that defence spending is able to benefit the employment situation in the long run, but damages employment in the short run, which is reasonable but different from the finding in Turkey provided by Yildirim and Sezgin (2003). In addition, the change in real GDP has a positive and significant influence on employment in both the short run and long run

Keywords:

• Defence spending
• Employment
• Real monthly salary
• Taiwan
• JEL code: H56, J21, O53

Notes

‡ E‐mail: [email protected]

† E‐mail: [email protected].

† E‐mail: [email protected].

* The authors acknowledge the financial support provided for this research by the National Science Council, Taiwan (NSC‐91‐2415‐H‐004‐008). The authors would like to thank anonymous reviewers of Defence and Peace Economics for their constructive comments and suggestions. The authors also wish to express their appreciation for the comments and suggestions made by all participants in the seminar held in the Department of Public Finance at National Chengchi University, Taipei, Taiwan, on 15 March 2004, and Mr. Jin‐Xyu Lin's assistance during the production of this paper

The reason for this consequence is probably that Taiwan is shirking some of the burden of defending the island, trusting that a US security guarantee is sufficient to deter any attack by China, and also more than 90% of Taiwanese think the chances of a Chinese attack are low, making it difficult for legislators to allot more money for defence.

This suggestion is based on the assumption that, in the intense desire for profit, a mature capitalist economy has a tendency towards overproduction or, conversely, underconsumption, due to depressed growth of wages.

There are three explanations for the negative relationship between military expenditure and economic growth: crowding‐out effects, the constraints of industrial capacity, and the objectives of social consumption.

The other three channels are the possible reduction of investment, the application of military technology less concerned with the economic rules of profitability, and the growth of effective demand.

Paul (1996) further indicated that defence spending has a favorable impact on the unemployment rate in Germany and Australia, whereas in Denmark it damages the employment situation. In the Netherlands, Japan, Italy, Spain, Austria, New Zealand, Sweden, Canada, and the US, there is no significant causal relationship between the unemployment rate and defence spending.

According to official data, Taiwan's economic growth rate was − 2.18% in 2001. However, in the same year, Singapore, another Asian dragon, had an even lower rate, an economic decline of 2.37%

All real variables utilized in this paper are deflated by CPI at the base year of 2001.

Taiwan is more democratic than it was, and as more and more legislators express their opinion on defence spending, defence spending faces a more competitive situation as it competes with other items of government spending. Many political and social organizations request Taiwan's government to lower its defence spending budget for the consideration of mitigating any possible conflict between Taiwan and China and in order to establish a more secure society.

According to data provided by the Ministry of National Defence, Taiwan, this share was 48.24%, 50.02%, 54.03%, 51.62%, 54.04% and 54.54% for 1998–2003, respectively.

Taiwan's government adopted a disarmament policy, called ‘Armed Forces Restructuring Program’ during 1997–2001. During the execution period, the total armed forces reduced from 450,000 to 385,000. As mentioned by Kuo (2003), this program was proposed as early as 1983.

Yildirim and Sezgin (2003) explained the impact of defence spending on employment with respect to reasons related to aggregate demand and GDP. However, this study discusses this issue under controlling the level of GDP. Therefore, any explanation related to aggregate demand or GDP will not be used as the explanation of the impact of defence spending on employment.

According to Yildirim and Sezgin (2003), it is simply assumed that defence spending DS is a proportion of Y; that is, DS_t  = gY_t , where 0 < g < 1; they further apply this assumption into the labor demand equation. However, as pointed out by the referee, the coefficient of defence says nothing about the effect of defence on employment, only about the effect of output. Therefore, this study does not assume any relationship between defence spending and output based upon the suggestion proposed by the anonymous referee and includes both the output and defence spending variables in the empirical model. The authors deeply appreciate the referee's constructive suggestion on this point.

Adams and Gold (1987) also argued that defence spending creates jobs only in some regions, but not in others, at certain levels of specification or skill, in particular branches or industries, and for certain firms.

A VAR model also allows us to identify the long‐run and short‐run dynamics of defence spending on employment. However, when the number of variables in the system is large, a VAR model is hard to implement due to the consideration of the degrees of freedom.

However, as argued by Fatai et al. (2003), the possible disadvantage of the ARDL is the low number of degrees of freedom when estimating a regression with a small sample size.

The tests are distributed according to a non‐standard F‐statistic irrespective of whether the explanatory variables are stationary or non‐stationary. The critical value bounds for these tests are computed by Pesaran et al. (2001).

The values of this variable prior to 1979 are estimated figures and are provided by the Directorate‐General of Budget, Accounting and Statistics, Executive Yuan, Taiwan.

The industrial sector includes mining & quarrying, manufacturing, construction, and electricity, gas, & water industries.

The regression result is MS_t  = 71.80–45.36 × T + 1.11 × MSM_t , t = 1979 to 2002, and adjusted R²  = 0.999. MS_t is monthly salary of employment including both industrial and service sectors, MSM_t is monthly salary of employment in the manufacturing sector, and T represents a time trend. The estimated coefficient of SM_t is significantly different from zero at the 1% significant level. The detailed results are available upon request.

According to a referee's comment that ‘the AIC tends to select too many variables, it is an inconsistent estimator of the lag length’, this study modifies the empirical model and uses both AIC and SBC as the methods for selecting optimal lag length. The ARDL (3, 4, 4, 1) model is determined by using both criteria and most second‐ and third‐order terms are significant at the 5% level and even at the 1% level. The authors are thankful for the referee's comment on this point which makes this paper better than its previous version in determining the optimal lag length.

Due to a lack of consecutive yearly data of personnel cost, this study is unable to use personnel cost instead of total defence spending as the independent variable in the model and further supports the hypothesis that a decrease in personnel cost would lead to an increase in employment in the private sector.

Yildirim and Sezgin (2003) pointed out that the political conjecture of the period 1974–1979 made trade unions powerful enough to resist an offer for wage decreases in spite of rising unemployment in Turkey.

This study adds a constant term in the long‐run dynamics when considering other factors that might affect employment. However, since the time‐trend variable is included in the Error Correction ARDL model, it is not included in the long‐run dynamic model. The authors appreciate the referee's valuable suggestion on this point.

The test belongs to the class of asymptotic (large sample) tests known as Lagrange multiplier (LM) tests. For a further discussion about this method, please refer to Godfrey (1988).

The ARCH LM Test is a Lagrange multiplier (LM) test for autoregressive conditional heteroskedasticity (ARCH) in the residuals.

As mentioned in Greene (2000), Jarque–Bera is a test statistic for testing whether the series is normally distributed. The test‐statistic measures the difference of the skewness and kurtosis of the series with those from the normal distribution. Under the null hypothesis of a normal distribution, the Jarque–Bera statistic is distributed as χ² with 2 degrees of freedom.

RESET is a general test for the following types of specification errors: (1) omitted variables: the empirical model does not include all relevant independent variables. (2) Incorrect functional form: some or all of the dependent and/or independent variables are transformed to logs, powers, reciprocals, or in some other way. (3) Correlation between independent variable(s) and ϵ, which may be caused by the measurement error in the independent variable, simultaneous equation considerations, combination of the lagged of dependent variable values and serially‐correlated disturbances.

Additional information

Notes on contributors

JR‐TSUNG HUANG †

‡ E‐mail: [email protected] † E‐mail: [email protected].