Predictive Density Simulation of the Korean Yield Curve: Pooling Method Approach (in Korean)

One of the various causes which interrupt accurate prediction is model misspecification. Generally, almost every prediction model is misspecified. The pooling method is recently used to improve forecasting performance with the linear combination of predictive densities of future random variables made by some interesting prediction models which are possibly misspecified. In this paper, we utilize the pooling method to forecast Korea government bond yields with six alternative prediction models four dynamic Nelson-Siegel models, a first-order autoregressive model, and a random walk model. As a result of our out-of-sample forecasting, the constant mixture model of AR(1) and RW model, AR(1)-RW, shows the best forecasting performance overall. In that sense, we find that the pooling method is useful to describe more precisely the dynamics of Korea bond yields by maturity.