This paper examines whether stock excess return predictability is dependent upon the stock market volatility. The paper introduces a two-state regime switching model with endogenous feedback effect for the stock return predictability test. To model regime switching, this paper adopted a new approach proposed by Chang et al. (2017), allowing an endogenous feedback effect channel through which the underlying time series affect the next period volatility regime. This paper shows that modeling such a channel is important in the return predictability context to incorporate the leverage effect. Monte Carlo simulation results demonstrated that additional power gain and bias improvement could be achieved in the endogenous regime swithcing (ERS) model, compared to the conventional Markov switching model. The empirical test results using the ERS model indicate that none of the tested predictors have significant predictive power when stock returns are highly volatile. However, the dividend-price ratio and macro variables such as T-bill rate and term spread had significant predictability, at least in the low volatility regime.