This paper estimates the Taiwan’s monetary policy reaction function using a unique data set of narrative-based monetary policy indicators. In particular, we introduce a probit model with autocorrelated errors to take into account the specific features of the discreteness of the dependent variable as well as the serial dependence in time series data. A practical sampling scheme via the Gibbs sampler with data augmentation algorithm is developed to make posterior inference.Empirical evidence shows that the monetary policy responds ountercyclically to the inflation rate but not to the economic growth rate. In addition, we find that the autoregressive parameters are significantly positive in all cases. This suggests that estimating the binary monetary policy reaction function without considering the serially correlated errors would be inappropriate, if not incorrect.