Forecasting macroeconomic series using advanced econometric models

Abstract

This thesis aims to model, forecast, and compare economic activity across Greece, Italy, and Germany, with a focus on evaluating common forecasting methodologies used in time series predictions. Despite numerous studies in this area, determining the superiority of one model over another remains challenging due to various influencing factors such as data type, transformations, data frequency, and variable inclusion. To address this challenge, the research incorporates multiple datasets, recognizing that many studies focus solely on predicting macroeconomic variables for specific countries, potentially leading to conclusions that are not universally applicable. By applying diverse models to different countries with unique historical contexts, this study seeks to derive more broadly applicable and robust conclusions across varied economic landscapes. The thesis employs thirty-two statistical and econometric univariate and multivariate models, including Autoregressive models, classical Vector Autoregressive models, Bayesian variants, and machine learning techniques. The determination of hyperparameters and variable selection emerge as critical decisions that influence the results. Hyperparameter selection relies on data-driven approaches based on an out-of-sample measure. Moreover, further results are drawn for fixed-order specifications. Additionally, the study investigates the estimation of models capable of handling numerous covariates using three different datasets, each containing varying numbers of covariates. To achieve comprehensive analysis, the research focuses on forecasting inflation, examining macroeconomic theory, existing forecasting methods, model formalization, dataset selection, and methodology. The thesis concludes with insights drawn from the results presented across various chapters, offering a deeper understanding of forecasting economic activity and highlighting implications for future research and policy considerations.