Economic Policy Uncertainty Index for Germany

Replication of Baker, Bloom & Davis (2016) methodology using Handelsblatt articles with validation analysis

Seminar thesis project - Big Data in Finance, Department of Financial Engineering and Derivatives, KIT

Overview

Economic Policy Uncertainty (EPU) measures the level of uncertainty about government economic policy decisions. High EPU can cause businesses to delay investments and hiring, while consumers may postpone major purchases, potentially slowing economic growth. This index quantifies such uncertainty by analyzing newspaper coverage.

This project replicates the methodology from Baker, Bloom & Davis (2016) "Measuring Economic Policy Uncertainty" for the German market using 671,782 Handelsblatt articles from 2000-2020.

Note: This repository presents a cleaned-up implementation of the methodology described in the seminar thesis. The actual thesis analysis was conducted using Jupyter notebooks. This code serves as a reusable framework for EPU index construction and validation.

Key Findings

• Strong correlation with original Baker et al. index validates methodology
• Successfully identifies major uncertainty periods: 9/11, 2008 financial crisis, European debt crisis
• Index shows meaningful correlation with German market volatility (VDAX-NEW)
• Post-2016 index quality affected by declining article counts in dataset

EPU Index for Germany showing major political and economic events

Methodology

The EPU index construction follows these steps:

1. Text Analysis: Articles must contain at least one word from each category:

   • Economic: wirtschaft, wirtschaftlich
   • Policy: steuer, wirtschaftspolitik, regulierung, ausgaben, bundesbank, ezb, defizit, haushalt, haushaltsdefizit, zentralbank
   • Uncertainty: unsicher, unsicherheit

2. Monthly Aggregation: Count EPU articles per month ÷ total monthly articles

3. Normalization: Standardize to σ=1, then scale to mean=100

Comparison of our replication with the original Baker et al. (2016) index

Validation Results

Correlation Analysis

• Strong alignment with original EPU during major crises
• Rolling correlations show time-varying relationship
• Post-2016 deterioration attributed to declining article counts in dataset

Arcticle count in the newspaper dataset

Market Validation

• Meaningful correlation with VDAX-NEW (German volatility index)
• Both indices spike during major uncertainty events
• Confirms EPU captures genuine market sentiment

EPU Index vs German market volatility (VDAX-NEW)

Topic-Specific Indices

Extended the methodology to create specialized uncertainty indices:

Equity Uncertainty Index

• Newspaper-based stock market uncertainty measure
• Smoother version of general EPU
• Strong visual correlation with VDAX-NEW

Data

• Source: Handelsblatt newspaper articles
• Period: January 2000 - December 2020
• Articles: 671,782 total
• Coverage: Daily German business newspaper, authorized journal for Frankfurt Stock Exchange announcements

Key Results & Insights

1. Methodology Validation: Successfully replicated Baker et al. approach with 0.66 overall correlation
2. Event Detection: Index correctly identifies major uncertainty periods (9/11, financial crisis, European debt crisis)
3. Market Relevance: Strong correlation with market volatility confirms economic significance
4. Data Quality Impact: Post-2016 deterioration linked to declining article counts in single-newspaper dataset
5. Topic Flexibility: Framework successfully extended to domain-specific uncertainty measures

Limitations & Future Work

Current Limitations:

• Single newspaper source (vs. multiple newspapers in original)
• Declining article counts affect index quality post-2016
• Potential editorial bias from single source

Future Improvements:

• Incorporate multiple German newspapers (FAZ, SZ, etc.)
• Apply machine learning techniques beyond word counting
• Extend analysis to policy impact studies
• Develop real-time updating capabilities

Academic Context

This work contributes to the growing literature on text-based uncertainty measurement in finance. The successful replication validates the robustness of the Baker et al. methodology while highlighting the importance of diverse data sources for reliable uncertainty quantification.

Usage

from src.epu_index import EPUIndex
import pandas as pd

# Load data
news_df = pd.read_csv('data/handelsblatt.csv')

# Create EPU index
epu = EPUIndex(news_df)
index_results = epu.calculate_monthly_index()

# Plot results
epu.plot_index(title="German EPU Index")

# Get validation statistics  
stats = epu.get_summary_stats()
print(f"Correlation with Baker: {stats['correlation']:.3f}")

Author

Adrian Wiedemann
Master's Student, Industrial Engineering and Management
Karlsruhe Institute of Technology (KIT)

Seminar Thesis: Big Data in Finance, SS 2023
Supervisors: Prof. Dr. Marliese Uhrig-Homburg, M. Sc. Caroline Grauer
Department of Financial Engineering and Derivatives

References

Baker, S. R., Bloom, N., & Davis, S. J. (2016). Measuring economic policy uncertainty. The Quarterly Journal of Economics, 131(4), 1593-1636.

---

Keywords: Economic Policy Uncertainty, Text Mining, Financial Markets, Germany, Sentiment Analysis, Time Series