This paper investigates a new volatility model, in which a regime switching parameter representing a high or low volatility level is added in the GARCH(1,1) model. Importantly, the model adopts the endogenous regime switching model recently introduced by Chang et al. (2017) and the future transition between states depends on the current state as well as the realization of the underlying financial time series. Application on the U.S. monthly stock index return series shows that the endogenous regime switching mechanism improves the data fitting and forecasting of stock return volatility. Moreover, using the latent factor extracted from the model and the FRED-MD data set, we employ the adaptive LASSO method to examine which macroeconomic variables are related to stock market volatility.