Long-run growth and volatility: which source really matters?

Abstract

The aim of the article is to analyse the relationship between long-run growth and business cycle volatility. In particular, the main purpose of this article is to identify which source of volatility is most detrimental to growth. Using cross-country data from 1970 to 2000, and several indicators of volatility (such as inflation, exchange rate, government expenditure, output and investment volatility) this article shows that although, all these measures of volatility are remarkably harmful for growth, business cycle investment volatility is the main source that hampers long-run growth. This relation is robust to different measures of business cycle, and to different sub-samples of countries.

Acknowledgements

I would like to thank Alicia Adsera, Sara Borelli, Georgios Karras, Lawrence Officer, Paul Pieper, two anonymous referees and Mark Taylor for the useful comments. I, alone, am responsible for any errors.

Notes

¹ See, for example, Martin and Rogers (1997, 2000).

² For example, in models that assume investment irreversibility, Bernanke (1983), Pindyck (1991), Pindyck and Solimano (1993) and Price (1995, 1996) show that uncertainty and volatility can lead to lower investments, and thus to a lower economic growth. Jovanovic (2006) shows pre-commitment to a risky technology determines a negative relation between volatility and growth. More recently, Aghion et al. (2005) and Blackburn and Varvarigos (2005) find that credit constraints and market imperfections can exacerbate uncertainty, leading to lower and more pro-cyclical investments, and thereby to lower growth.

³ This result is largely confirmed in Dejuan and Gurr (2004), who analyse Canadian provinces, in Martin and Rogers (2000), who use European regional data, and in Imbs (2007), that also presents empirical evidence suggesting that at industry level this relation is reversed (in particular, volatile sectors command high investment rates, as they would in a mean–variance framework).

⁴ There are several channels through which inflation uncertainty can affect macroeconomic performances. First, greater inflation uncertainty will tend to hamper the efficient allocation of resources as risk-averse agents will reduce the length of their contracts, to the extent that more frequent contract negotiations involve higher negotiation costs. Second, and perhaps more important, under the assumption of investment irreversibility, increased price uncertainty can lead to lower investments, and thus to a lower growth (Pindyck and Solimano, 1993; Feldstein, 1996).

⁵ See these works for a more detailed theoretical discussion.

⁶ It is worthwhile to mention that our estimates of volatility are in fact proxies for the true underlying volatility measures. This means that our measures will be affected by some quantifiable classical measurement error.

⁷ This robustness will be formally confirmed by the findings of the next section.

⁸ See Data Appendix, for a detailed description of the variables.

⁹ The correlation coefficient is −0.33.

¹⁰ See Table 1 for average volatility measures.

¹¹ The average effect is computed by multiplying the estimated coefficient for the average measure of volatility. Very similar results are obtained when we use IMF's World Economic Outlook (2006) data on real GDP for the same period. In particular, in a dataset including the same (116) countries considered in the Heston et al. (2002) dataset, the decrease in GDP growth rate due to business cycle volatility is (on average) 1.21 percentage points. Considering a larger dataset of 152 countries, the magnitude of the decrease in the growth rate is (on average) 1.08 percentage points.

¹² In fact, using the HP filter with a smoothness parameter equal to 100, the BP filter and the simple differencing, a 1% increase in the SD of the business cycle measure across countries translates, respectively, into a decrease of 0.83, 0.81 and 0.99 percentage points in the long-run rate of growth.

¹³ Same results are obtained even if we drop the human capital variable (that we did not find significant in our previous regressions), allowing the full sample to contain 116 observations.

¹⁴ See Table 3.

¹⁵ We include the real government expenditure to control for the size of governments when we test the effect of government volatility. However, similar results are obtained when we do not include this variable as control.