We performed the following procedures to calculate stock market volatility (e.g., Folta and O’Brien
2004). First, we computed the monthly value-weighted market returns for each industry using data from the Center for Research in Security Prices. Second, we specified a Fama and French (1993) three-factor model to forecast monthly industry returns recursively from 1950 to 2005. We further specified a generalized autoregressive conditional heteroskedasticity (GARCH)
11 process to model the variance of the error term (e.g., Bollerslev et al. 1992); likelihood ratio tests showed that the GARCH (11) model outperformed alternative GARCH and other ARCH models. Fama and French (1993) classified the economy into 49 industries that reflect the fundamentals of each industry, and we recorded industries at the three-digit SIC level (e.g., Davis and Duhaime 1992).Third, for each estimated monthly return, we obtained the conditional variance from the GARCH model. We then annualized the variance measure by averaging the conditional variance for the past 12 months for an investment that occurred in the current year, and took the log of the average variance.