A novel dynamic ordered probit model with time-varying parameters is proposed to estimate a monetary policy reaction function with narrative-based monetary indicators. The estimation and inference are carried out using the Bayesian simulation-based approach. Empirically, these are the following findings. First, there is strong evidence in support that the Central Bank in Taiwan responds counter-cyclically to inflation but weaker, if any, evidence to economic growth. Secondly, the persistence and consistence in policy-making of the monetary authority is confirmed by the significance of the positive autoregressive coefficient. Although not all, the estimates of the TVP-DOP model provide, at least, partial support of time-varying parameters. Finally, the results indicate that studies of the discrete monetary policy reaction functions without explicitly considering the possible dynamics inherent in the time series data and time-variations in model parameters may be inappropriate, if not incorrect.