Modelling and forecasting unemployment time series

The purpose of this thesis is to develop machine learning and econometric models to forecast the values of the dependent variable, U.S. unemployment rate. In the beginning, we have 26 independent variables that contain quarterly data with a starting date of the second quarter of 2003 through the third quarter of 2024. The main part of the thesis provides a detailed explanation of the 11 models we developed, which are based on advanced econometric methods and machine learning techniques. The methodologies we used are the following: Multiple Linear Regression, Ridge Regression, Lasso Regression, Elastic Net Regression, Principal Component Analysis (PCA), Partial Least Squares (PLS), Random Forests, Gradient Boosting Machines (GBM), Autoregressive Model (AR), Moving Average Model (MA) and Autoregressive Moving Average Model (ARMA). In order to predict our dependent variable, U.S. unemployment rate, we employed recursive out-of-sample forecasting to assess model performance and found the forecasting values over 2 horizons, 6 - months and 1 - year. The models were assessed using evaluation metrics and relative metrics, such as relative Mean Absolute Error (MAE), relative Mean Squared Forecast Error (MSFE), relative Root Mean Squared Forecast Error (RMSFE) to assess the predictive ability of the models. The results of this study show that the Ridge regression model works well for the short-term horizon (i.e. h=1 and h=2), having lower forecast errors and so, it is more suitable for creating accurate forecasts until 6 – months. For long-term horizons (i.e. h = 4), the Random Forest seems to be the most reliable, demonstrating better effectiveness in reducing forecast error. This research highlights how important is to adapt model selection to the forecasting horizon. This thesis advances the field of data-driven economic modelling and provides important insights into enhancing macroeconomic forecast accuracy by fusing machine learning methods with traditional and advanced econometric methodologies.