Total factor productivity and macroeconomic instability

Abstract

Total factor productivity (TFP) is an important component of growth for most countries. This article assesses the role of macroeconomic instability on TFP growth. We consider volatility in inflation, openness of an economy and financial market deepness as measures of macroeconomic instability. Empirical evidence provided from Turkey suggests that volatility of openness and financial market deepness reduce TFP growth, whereas volatility of inflation increases TFP growth.

Keywords:

• macroeconomic instability
• TFP growth
• VAR-GARCH models

JEL Classifications:

• D24
• E22

Acknowledgments

The authors thank Anita Akkas and Rana Nelson for their helpful comments. The views presented here are those of the authors; they do not necessarily reflect the official position of the State Planning Organization of Turkey or of The World Bank.

Notes

1. Although later work has indicated a lesser role of TFP in growth, it still remains the major factor accounting for growth (see, for example, Kendrick 1961; Denison 1985; Jorgenson, Gollop and Fraumeni, 1987; Maddison 1995; Mankiw, Phelps and Romer 1995; Klenow and Rodriguez-Clare 1997; Jones 1997; Abramovitz and David 2000).

2. Schultz (1961); Becker (1962); Becker Murphy and Tamura (1990); Black and Lynch (1996); Miller and Upadhyay (2000); Aiyar and Feyrer (2002).

3. Edwards (1998); Harris (1999); Cororatan and Zingapan (1999); Miller and Upadhyay (2000); Alcala and Ciccone (2004).

4. Miller and Upadhyay (2000); Clark (1982).

5. Kugler and Neusser (1998); Levine (2003); Tadesse (2005); Jeong and Townsend (2004); Beck, Levine and Loayza (2000).

6. Harris 1999.

7. Volatility measures are the proxy of second moments such as moving variance or conditional variance of macroeconomic variables.

8. Type-II error is defined as not rejecting the null although it is false. The way to reduce Type-II errors would be to either increase the level of significance denoted by (a) or decrease the confidence coefficient denoted by (1 − a). Apart from these conventional methods, another suggestion is to increase the dispersion of the collected data as per Netter, Wasserman and Kutner (1985), p. 71. They argue that increasing spacing results in an increase in the t-statistics of a given estimated parameter, the sample size and the variance of the errors. In addition, increasing spacing also decreases the standard errors of the parameters of interest. Concerning the fact that Turkey is the only country experiencing high and sustainable inflation, the result is that spacing is higher and Type-II error is lower (see also Berument, Akdi and Atakan 2005 for further discussion).

9. This article assesses the determinants of TFP growth; however, we will call this TFP in the text.

10. See Hamilton (1994, pp. 668–70) and Berument, Coskun and Sahin (2007) for the advantages of EGARCH specifications against other types of ARCH models and estimation using general error distribution.

11. The estimated coefficients of the Cobb-Douglas production function for capital is 0.42 and for labor is 0.58.

12. The lag order of 4 is determined by the final prediction error (FPE) criteria that set the lag length such that the residuals are no longer autocorrelated. Casimano and Jansen (1988) suggest that autocorrelated errors imply the presence of the ARCH effect even if the ARCH effect is not present.

13. The level of significance is at 5% unless otherwise mentioned.

14. We used various ARCH and GARCH specifications but the basic evidence on the mean effect is robust.

15. The lag order of 2 is selected by the final prediction error criteria when we consider a class of models with the same lag order across equations.

16. We also estimate the model with the EGARCH specification. The estimated coefficient for the leverage effect and the estimated coefficient for the conditional variances were not statistically significant (possibly due to over-parameterization), thus we did not elaborate on them here.

17. In this article, we explore the behavior of TFP growth by using three-equation VAR-GARCH specifications. We also estimate a model that incorporated deepness, TFP, openness and inflation simultaneously: a four-variable VAR-GARCH specification. The estimated coefficients for inflation volatility, openness volatility, and deepness volatility were too small, even if they were statistically significant, and the results were sensitive to initial values. This might be due to the highly nonlinear nature of the specification and the high collinearity of deepness, TFP and openness measures. We did not report and elaborate on these estimates here, but they are available for interested readers on request.

18. Conducting specification tests on the multivariate GARCH models is not an easy task, and we tried various classes of them. Since the tests are so extensive, we did not report them all; however, sign-biased, Ljung-Box-Q and ARCH-LM tests mostly passed for the tfp _t specification equation (3) of Table 4.

19. Introducing one-standard deviation shocks is a common exercise in empirical simulation. See for example, Sims and Zha (1999).

20. Since increasing inflation volatility may have other adverse effects, we did not pursue this avenue further.