International CAPM with Regime Switching GARCH Parameters

This paper tests a conditional version of Adler and Dumas'(1983) International CAPM with regime switching GARCH parameters. As benchmark the same model is estimated without state dependent parameters. The switching representation is found to react faster than the benchmark to shocks in stock market returns. This suggests that the non-switching model suffers from spuriously high persistence. In particular, when a financial crisis occurs, the conditional risk exposures appear to be underestimated, while overestimated in the aftermath. The introduction of a regime switching model should hence improve forecasting power. We also find that in periods of financial turmoil, weight is shifting from the GARCH, towards the ARCH termes of the conditional covariance generating process. During such events investors, when formin their (co)variance expectations, seem to put more emphasis on current shocks, at the expense of the current second moments.