This study utilized different machine learning algorithms to nowcast the domestic liquidity growth in the Philippines. In particular, different types of regularization (i.e., Ridge Regression, Least Absolute Shrinkage and Selection Operator (LASSO), Elastic Net) and treebased (i.e., Random Forest, Gradient Boosted Trees) methods were employed to support the BSP’s suite of macroeconomic models used to forecast and analyze the said monetary indicator. These models are then evaluated and compared against the traditional time series models (e.g., Autoregressive Models, Dynamic Factor Model) using an expanding window process.
The results indicate that machine learning algorithms relatively provide better estimates than the traditional time series models utilized in this study due to their consistent month-ahead nowcasts with low Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). In addition, the models provide nowcasts with low forecast errors on the months where domestic liquidity suddenly expanded due to the impact of Coronavirus Disease 2019 (COVID-19) in the Philippines. This study also established that machine learning algorithms filter out or identify important indicators to stipulate parsimonious nowcasting models.