The Macroeconomic Effects of Fiscal Policy: Evidence from Computational Text Analysis of the Debates in the Bundestag (2021-2024)
This cooperation project between the Chair of Monetary Economics (Peter Tillmann) and the Chair of Statistics and Econometrics (Peter Winker) is funded by the DFG (Deutsche Forschungsgesellschaft).
Macroeconomic models suggest that fiscal policy, changes in government spending and tax revenues, is an important determinant of business cycles. However, the empirical impact of fiscal policy is controversial, both in terms of the sign of the effect and the magnitude. Therefore, the aim of this project is to improve our knowledge about the effects of fiscal policy in Germany. Since a change in fiscal policy is often the result of a long and controversial parliamentary process, there are no contemporaneous indicators that allow to capture fiscal policy shocks and to measure their effects on the real economy. The core hypothesis of this project is that the nature of the parliamentary process itself contains valuable information to understand the driving forces and the consequences of fiscal policy. We also acknowledge that national fiscal policy made in parliament is accompanied by public discourse. To account for both aspects of fiscal policy, we use two novel data sets on parliamentary speeches and newspaper reporting, respectively. We exploit a large data set that contains the digitized speeches in the German Bundestag since September 9th, 1949. We augment this parliamentary data with newspaper data from the digitized archive of the Frankfurter Allgemeine Zeitung (FAZ) to cover the public debate about fiscal policy. Based on a number of keywords, we will obtain about 220,000 articles on German fiscal policy for this analysis.
To extract useful insights from the collected text data, we apply methods from computational text analysis which – to the best of our knowledge - have not been used to study fiscal policy before. In doing so, we aim to understand (i) the evolution of the debate of fiscal policy over time and (ii) the extent of disagreement about fiscal policy. Two machine learning approaches are considered to systematically analyze the data set:
- Topic Modelling approaches that allow to reduce the dimensionality of high-dimensional text data and uncover latent topics in the data.
- Word Embeddings approaches that allow to represent word or/and documents in a shared vector space that captures semantic characteristics of natural language.
After constructing text-based indicators that capture fiscal policy sentiment over time, we use these indices within econometric models to examine the macroeconomic impact of fiscal policy.
The figure below illustrates the project pipeline.