Forecasting Inflation in Vietnam with Univariate and Vector Autoregressive Models

In this paper, I apply univariate and vector autoregressive (VAR) models to forecast inflation in Vietnam. To investigate the forecasting performance of the models, two naïve benchmark models (one is a variant of a random walk and the other is an autoregressive model) are first built based on Atkeson-Ohanian (2001), Gosselin-Tkacz (2001) and the specific properties of inflation in Vietnam. Then, I compute the pseudo out-of-sample root mean square error (RMSE) as a measure of forecast accuracy for the candidate models and benchmarks, using rolling window and expanding window forecasting evaluation strategies. The process is applied to both monthly and quarterly data from Vietnam for the period from 2000 through the first half of 2015. I also apply the forecast-encompassing Diebold-Mariano test to support choosing statistically better forecasting models from among the different candidates. I find that VAR_m2 is the best monthly model to forecast inflation in Vietnam, whereas AR(6) is the best of the quarterly forecasting models, although it provides a statistically insignificantly better forecast than the benchmark BM2_q.