Taiwan's financial holding companies: an empirical investigation based on Markov regime-switching model

Abstract

Though the Financial Holding Company Act in Taiwan permits banks, securities firms and insurance companies to affiliate, identifying the influence on Taiwan's financial holding companies hasn’t been discussed rigorously yet. This paper presents a formal methodology, using two-state Markov regime switching approach, to allow for the uncertainty event-date of financial holding companies’ stock return and risk. This study serves as one of the first studies that adopt a Markov regime-switching model to estimate financial holding companies’ stock behaviour. The evidence shows that 12 of 13 financial holding companies have regime-switching and one has no regime-switching in Taiwan. Therefore, the stock behaviours of Taiwan's financial holding companies follow two regimes and the traditional linear model cannot be descriptive. However, the levels of 12 financial holding companies’ risk are significantly low of state 1 and stock returns are indifferent between two states. Hence, there are diversification benefits of Taiwan's financial holding companies. Summarily, to assess the influence on Taiwan's financial holding companies, it is recognized that this methodology developed by the model is meaningful for research.

Acknowledgments

The authors thank anonymous reviewers and the editor Mark P. Taylor and Anita Phillips for helpful recommendations. Financial support from the National Science Council (NSC91-2416-H-110-029) is gratefully acknowledged.

Notes

 In 1933 in the aftermath of the Great Depression and its widespread bank failures, Congress enacted the Banking Act of 1933. Four sections of the Banking Act of 1933 are referred to as the Glass-Steagall Act. Many critics believe that banks had engaged in inappropriate securities activities that harmed investors. In particular, critics charged that if a bank had a bad loan to a failing company or even a country, it would sell securities to pay off the loan and leave investors with the poor investment.

The Bank Holding Company Act of 1956 was implemented in response to banks forming bank holding companies in order to own both banking and non-banking businesses. This Act, among other things, generally prohibited a bank holding company from engaging in most non-banking activities or acquiring voting securities of certain companies that are not banks.

In addition, the finance-ministry in Taiwan endorse the Internal-Auditing Act, the Internal-Control Management for the Financial Holding Company Act, the Record of Annual Report for the Financial Holding Company, and the Management of Capital Adequacy for the Financial Holding Company, to supervise financial conglomerates.

Diebold (1986) and Lamoureux and Lastrapes (1990) find that structural changes could effect estimations of financial assets. Hence, time-varying parameters are assumed to capture structural changes. In order to avoid computational difficulty, a Markov regime-switching model is assumed to be the estimate time-varying parameter.

Hansen (1992) proposed a bound test under LR principle, to circumvent these problems, but the method has computational difficulty. Then, Garcia (1998) exploited the fact that the likelihood ratio test statistic for the null hypothesis of one state only is the supremum over all admissible values of the nuisance parameters (the transition probability), and derived analytically the asymptotical distribution of the test statistics and simulated critical values tables.

See Hamilton (1990) for detail.

See Hamilton (1994) for detail.

There were 14 financial holding companies in Taiwan on 31 December 2003. To avoid less observational problems, the study does not consider First financial holdings, which is listed on 2 January 2003.

See Appendix for detail.

To avoid spurious regression results in Equation (1), we test unit root for daily stock return (R_kt ) and Taiwan stock market index (R_mt ).

Since Chinatrust financial holding company has no regime-switching, its filtered and smoothed probability is not considered in this section.

Since the sum of probability is equal to unity on state 0 and state 1, the study only considers probability on state 1 in Figs 2 and 3.

The general approach to specification testing developed by Newey (1985), Tauchen (1985), and White (1987). Hamilton (1996) proposed Lagrange Multiplier (LM) tests under the null hypothesis for specification tests of Markov regime-switching time series model, which is more powerful than Newey-Tauchen-White test.