This study considers a new error correction model (ECM) for stock price and dividend,which accommodates nonlinearities in both long- and short-run relationships. First, time varying coefficient cointegration is adopted to explain the nonlinear long-run relationship between stock price and dividend. Second, the model allows for endogenous regime switching to describe the short-run relationship. The empirical application on the S&P 500 Index and dividend shows that our model fits the data significantly better than existing models and provides estimates with meaningful interpretations. In addition, the linear cointegration is unsuitable to describe the long-run relationship, and the ECM with endogenous regime switching better explains the data than that with conventional Markov switching. An extract latent factor specifically reveals the periods for each regime, and the periods of high-volatility regime include the NBER recession periods and certain periods of financial crisis.