Forecasting Korean Macroeconomic Variables with Autoregressions and Vector Autoregressions (in Korean)

We compare out-of-sample forecasting performance across 94 forecasting models for Korean interest rate, growth rate, and inflation rate. They are 80 standard autoregressive and vector autoregressive models and 14 average forecast models based on the 80 standard models. The standard models differ in five aspects: (i) in the form of included variables (level/difference), (ii) in the way to set up estimation samples (recursive/rolling), (iii) in the timing and rule of lag order selection (once at the beginning/every period, AIC/BIC), (iv) in the estimation method (OLS/Bayesian) and (v) in the construction of multi-step ahead forecasts (iterative/direct projection). For the forecast of interest rate, Bayesian vector autoregressions using the variables in level and average forecast models are superior. For the forecast of growth rate, average forecast models are superior overall while autoregressive models tend to be better than vector autoregressive models. For the forecast of inflation rate, models with a recursive window showbetter performance.