This study develops the first regional inflation forecasting models for the Philippines employing non-linear machine learning approaches for a few representative regions of the country. These regional forecasting models are expected to supplement the BSP’s suite of macroeconomic models used for forecasting and policy analysis. In particular, three machine learning methods are employed: support vector regression (SVR), artificial neural networks
(ANN), and long-short term memory (LSTM) to forecast inflation for the selected regions using univariate and multivariate processes. These models are evaluated based on root mean square error (RMSE) and mean absolute error (MAE) in one-month ahead static forecasting and 12-month ahead dynamic forecasting. The results indicate relatively good performance of the
models for month-ahead forecasting while SVR models dominated in the 12-month ahead dynamic forecasting exercises. Furthermore, the models are evaluated vis-à-vis traditional ARIMA models and this paper finds evidence that machine learning methods do outperform
ARIMA models in forecasting.
