Democritus University of Thrace (DUTH)
Study of macroeconomic variables of the euro area using machine learning
Abstract
dc:descriptionThe task of Macroeconomic forecasting is demanding and precarious. Demanding in terms of the data needed to create an accurate forecasting model and precarious in terms of the many dynamic relations that influence Macroeconomic variables. Traditional Econometrics investigate the validity of simple theory-driven forecasting models in real data considering a small fraction of the interwoven Macroeconomical network; cutting-edge Machine Learning models, in contrast, don’t need to be based on theory and can treat as many variables as needed.In this dissertation I train multiple models to forecast three key macroeconomic variables for the euro area using machine learning: Eurocoin index, unemployment rate and output gaps. More specifically, I train models to directionally forecast the Eurocoin index using simple sum and Divisia monetary aggregates. By doing so, I also compare the forecasting ability between the simple sum and Divisia monetary aggregates (Barnett critique) and test whether money affects real economic growth, known as the money neutrality theory. I also train models to directionally forecast the unemployment rate while identifying the variables that affect its predictability. Finally, I forecast the output gaps of the euro area GDP using various combinations of long- and short-term interest rates.The comparison between the simple sum and Divisia monetary aggregates reveals that the latter exhibit superior directional forecasting results. The best model includes 12 lags of the Divisia M2 and 6 lags of the Eurocoin index, achieving an out-of-sample accuracy of 82.05% and an in-sample accuracy of 87.2%. Furthermore, the impact of measures of money supply, including simple sum and Divisia monetary aggregates, on the economic activity of the euro area is significant, as they offer superior forecasting capabilities for the Eurocoin index. These findings challenge the notion of money neutrality, demonstrating that money supply plays a crucial role in economic activity.Additionally, I investigate the effectiveness of various machine learning models for forecasting unemployment rate in the euro area. The empirical findings reveal that the random forest model outperforms other models and achieves a full-dataset accuracy of 88.48% and an out-of-sample accuracy of 85.42%. Additionally, the empirical analysis shows that financial indices and past values (lags) of unemployment significantly influence the accuracy of euro area unemployment forecasts. The incorporation of these variables in the random forest model leads to more accurate and reliable predictions of the unemployment rate in the euro area. Regarding the output gaps forecasting, the most accurate model is an SVM model coupled with the RBF kernel using all 9 interest rates yielding an in-sample accuracy of 85.29% and an out-of-sample accuracy of 94.74%. Furthermore, the analysis of Support Vector Machines (SVM) and the elastic net models suggests that incorporating all nine interest rates yields the best results. This finding provides empirical evidence that short-term and long-term interest rates offer more informative insights for forecasting euro area output gaps, compared to relying solely on the slope of the yield curve.The extant literature on macroeconomic variable forecasting provides numerous studies aimed at predicting the direction of these variables and developing forecasting applications and crisis detection models, among other objectives. However, the literature on forecasting macroeconomic variables for the euro area is relatively limited, with a sparse number of studies focusing on the macroeconomics of the euro area as a whole. The majority of the research has concentrated on specific EU member nations rather than on the supranational level of the euro area. Consequently, there exists a pressing need to optimize current models further and develop novel effective models.The models developed in this dissertation create a toolkit that can be used by governments and central banks, but also by large corporations. This toolkit is expected to assist stakeholders in monitoring changes in the macroeconomic environment of the euro area.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Sofianos, Emmanouil
- Σοφιανός, Εμμανουήλ
Subjects
dc:subject × 17Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/53907