ABSTRACT
This paper employs a wavelet approach to investigate the relationship between economic growth and military spending in a time-frequency domain for the case of Turkey. Turkey presents an interesting case for analysis of military spending and economic growth, as its geopolitical position and history of insurgencies and separatist violence oblige the country to devote an unusually large share of the central government budget to national defence. Timescale regression analysis reveals that military expenditures have significant negative effects on growth in per capita GDP at business cycles of 16 years and longer. Timescale Granger causality analysis indicates that per capita GDP growth responds to movements in military expenditures at business cycles of eight years and above and that this result is very significant. Wavelet coherency analysis corroborates these findings, indicating a significant negative long-run co-movement at business cycles of 16 years and longer. Thus, the neoclassical prediction that military spending may promote growth does not hold in the case of Turkey, at least in the long run. Furthermore, the analysis reveals that, in the long run, military spending has been leading rather than lagging economic growth.
JEL CLASSIFICATION:
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1. For a discussion on the drawbacks of the Hodrick-Prescott filter or Fourier transformation see Andersson (Citation2011).
2. When using wavelet decomposition, one-off events such as crises do not affect the decomposition at other points in time. In contrast, with traditional smoothing techniques, such as the moving average, the impact of a one-off event spreads over several periods.
3. These conditions determine what happens when the filtering algorithm reaches one end (boundary) of the series to compute the remaining coefficients. This reflecting boundary condition indicates that when one end of the series is reached, the actual series is extended to obtain a new series, which is reflected to twice the length of the original series. Based on the extended series, the wavelet and scaling coefficients are then computed using a periodic boundary condition (see, Habimana Citation2018).
4. We employ the Morlet wavelet, which allows us to achieve an optimal joint time-frequency concentration in the Heisenberg sense (see, Aguiar-Conraria and Soares Citation2011).
5. Formally, the data is regressed on a series of cosine waves as described in Müller and Watson (2015, Citation2018), and the fitted values are kept as the smoothed data.
6. The estimates are bias-corrected following Liu, San Liang, and Weisberg (Citation2007) and Veleda, Montagne, and Araujo (Citation2012).