Dynamics of Idiosyncratic Volatility and Market Volatility: An Emerging Market Perspective

Abstract

Estimating idiosyncratic volatility (IVOL) using various model-dependent and model-independent measures, we investigate the characteristics of aggregate IVOL in Malaysia over the period 1990–2008. The IVOL estimated in all models have similar patterns and has no trend over the sample period. There is evidence of episodic phenomenon. During financial crisis periods, market volatility is relatively higher than IVOL – a plausible reason is high correlation between firms' returns. Small firms and low-priced stocks appear to influence IVOL more than large firms and high-priced stocks. In Malaysia, market volatility and IVOL may predict GDP growth.

Key Words:

• Idiosyncratic volatility
• firm-specific risk
• market volatility
• emerging markets

Jel Classification:

• G19

Notes

1. A typical investor in the US holds on average only 3.41 stocks (Friend & Blume, 1975). Later studies reveal a decline of this average. For example, Barber and Odean (2000) show that the average number of stocks held by a household investor is 2.61, and Goetzmann and Kumar (2008) report an average of 3 stocks per household. An investor who holds only 3 or 4 individuals’ stocks in the portfolio may not be able to diversify away the firm-specific risk in their portfolios (Campbell et al., 2001).

2. A study of volatility in an international context is available in Brooks (2007) where 26 emerging markets are investigated.

3. In the past decade, with the emergence of modelling volatility as a general autoregressive conditional heteroscedastic (GARCH) process, researchers started to use various extensions of the GARCH model in conjunction with the Fama–French model to estimate IVOL. In this study, we estimate IVOL using the residuals of the Fama–French three-factor model and do not model as a GARCH. Campbell et al. (2001) argue that a parametric model is more suitable for volatility forecasting, and such a model is not needed if the main interest is only on historical movement.

4. Previous studies also have computed IVOL using sum of squares of residuals. For example, Campbell et al. (2001) compute industry-specific volatility as the sum of the residuals estimated in a simple linear regression of industry return and market return. The difference between our measure of IVOL given in Equation (3) and that of Campbell et al. (2001) in the case of industry-specific residuals is in the estimation. Campbell et al. (2001) impose restrictions equivalent to α_i = 0 and β_i = 1 in Equation (1). Implications of these assumptions are that (1) the empirically observed possibility that the CAPM may not hold is ignored and (2) the residual becomes the difference between the stock return and the market return.

5. There is a price limit in Bursa Malaysia, such that the price can increase up to a maximum of 69% and decrease by a maximum of 51% in each trading session.

6. This is common practise in empirical studies. For example, Fu (2009) considers 15 days as the minimum requirement.

7. Giving greater weights to large stocks is arguably a way of mitigating the problem of the bid-ask bounce problem associated with small stocks (Bali et al., 2005).

8. In each month, we define the 30% of the firms with the largest market capitalization as large firms and the 30% of the firms with the lowest market capitalization as small firms.

9. There is no publicly available business cycle recession indicator for Malaysia. Talib (2009) considers the leading and coincident indices to identify business cycles for Malaysia. The recessions identified in Talib (2009) are 1998–1999, following the Asian financial crisis, 2000–2001 and 2008, the pivotal period of the global financial crisis.

10. The results are available from the corresponding author upon request.

11. Campbell et al. (2001) use this procedure to test for existence of a deterministic trend in the US market. When a volatility series is highly persistent, the standard trend tests are inappropriate. Vogelsang (1998) robust trend test is robust to this property.

12. Since the price per share depends on the number of outstanding shares, price per share can be altered by stock split. We do not account for this possibility.

13. This measure is based on the suggestion by Campbell et al. (2001) that volatility within a year may be used as a forecast for the next period volatility.