Abstract
This study decomposes the unconditional stock return volatility into two categories: systematic versus idiosyncratic risk, to re-examine the link between size and risk in the banking industry. The feasibility of the model is tested using data for US banks from 1998 to 2007. The evidence uncovered suggests that the practice of size-related diversification obtained with large banks reduces the firm-specific risk, and thus weakens stock return variances. However, rather than eliminating firm-specific risk, it is being transformed into systematic risk. Additionally, our empirical findings can potentially explain why a bank's size-related diversification does not result in a reduction in its unconditional stock return volatility reported in Demsetz and Straha [Historical patterns and recent changes in the relationship between bank size and risk. Federal Reserve Bank of New York Economic Policy Review, 1(2), 13–26 (1995); Diversification, size, and risk at bank holding companies. Journal of Money, Credit, and Banking, 29, 300–313 (1997)].
Notes
In practice, in the regression with a large number of parameters, it becomes somewhat difficult to obtain significant parameter estimates when the data observations are limited. Equation (4) in which the individual-unit-varying intercept terms (i.e. a i , i = 1, 2, …, N) are excluded can successfully mitigate the over-parameter problem involved in Equation (3).
See Coles, Daniel, and Naveen (Citation2006) and John (Citation1993) for the related discussions about the two control variables.
The specified financial data of some BHCs are not available for the entire 10-year period. For instance, since the databases we selected missed the data of FREMONT GENERAL CORP for the 2007 year, we sampled the data for this bank over the 9-year period from 1998 to 2006. In addition, some banks have not existed for 10 consecutive years. In particular, the BHC of ASSURANT INC was established in 2004 and thus the data for this bank just cover the period from 2004 to 2007. Therefore, the panel data used in this work are unbalanced, and thus the number of available observations over the 10-year period are fewer than 1540 (=154 × 10).
The authors are thankful for this insightful point from the anonymous reviewer.