Based on a non-linear asset-pricing theory, we assume a stochastic investment opportunity set and both risk premium and risk loadings follow the first-order Markov process. Therefore, a theoretical time-varying factor model, which allows both risk premium and risk loadings vary through time, is obtained. We then adopt individual stock data from China and US to investigate the regime-dependent risk premium and correspondingly time-varying betas. We identify the states for each market and illustrate the dynamic risk patterns across the two markets.