Creating a Large Data Set for Forecasting Inflation in the European Context using Machine Learning Techniques

Inflation forecasting plays a critical role in macroeconomic analysis due to its impact on economic policies and decision-making, which in turn could lead to economic growth. This dissertation contributes to the field by creating and introducing the European Union Economic Data (EUED) dataset, a macroeconomic dataset built upon aggregating multiple Eurostat sources, tailored for the European context, specifically for the regions European Union of 27, Euro Area of 20 and Germany. This dissertation uses the newly built EUED dataset to explore the application of machine learning techniques to forecast inflation across the regions through the macroeconomic indicator Harmonized Index of Consumer Prices (HICP) for the component All-items. The findings demonstrate that machine learning models, in particular LASSO, consistently outperform the traditional benchmark Random Walk and Autoregressive model in terms of predictive accuracy, as it delivers the smallest predictions errors across the three metrics considered – root mean squared error, mean absolute error and median absolute deviation. Additionally, the results highlight that forecasting inflation measured in monthly rate of change yields the most reliable results across models when compared to when it is measured as an index or an annual rate of change.