Essentials of Time Series for Financial Applications

Table 9.3 shows that all information criteria unanimously select an MS regression in which all the coefficients are time invariant but the standard error of the regression follows an MS process. All attempts to extend such model to three regimes are rejected by the information criteria that increase. In the following, we report estimates of the linear benchmark in which all coefficients are constant and of the selected MS heteroskedastic regression. To provide additional information, we also show estimates of an MS regression in which all coefficients are driven by the same two-state MC (p-values are in parentheses):

$x_{t+1}^{\mathit{Japan}}=\underset{(0.325}{0.312}$