Fig. 3 plots the correlation estimated by modified model. Considering the dynamic change of m,the long- and short-term components follow a similar trend, which indicates that the modified model is more efficient than the original model.Although the coefficients of dummy variables are not significant, we keep the specification with dummy variables for theintegrity of the economic implication of the model. The AIC, BIC, and loglikelihood in Tables 2 and 4 show that the fit of the modifiedmodel is improved, which means that incorporating the dummy variables can be helpful in capturing the structural changes.In addition, we also perform an alternative study by using the first investor sentiment index proposed by Baker and Wurgler (2006).The results justify the robustness of the above findings. All the results can be found in Appendix A2.