Estimation of Taiwan’s binary monetary policy reaction function

This paper proposes a probit regression with autocorrelated errors (PAR) to estimate the reaction function of monetary policy in Taiwan using newly constructed binary monetary indicators. We develop a practical sampling scheme via the Gibbs sampling algorithm with data augmentation to make posterior inference of the binary monetary policy reaction function. In contrast to the conventional approach, our method avoids the problem of multiple integrals by directly drawing values of latent variables from the relevant full conditional density along with all the other parameters. Empirical results show that the monetary authority responds to macroeconomic conditions asymmetrically. Specifically, in the high-inflation regime, a contractionary monetary policy is implemented to reduce the inflation rate. Once inflation is under control, that is, in the low-inflation regime, attention is paid to stimulating the growth of the economy.