VIX Spillovers and Equity Return Predictability in Central European Markets
Empirical research on whether the CBOE Volatility Index (VIX) predicts next-day equity returns in four Central European markets. Presented at ŠVOČ 2026, Faculty of Management, Comenius University in Bratislava.
When US investor fear spikes, Central European markets fall the next morning. That contemporaneous co-movement is well documented. This project asks a sharper, asymmetric question:
Does the VIX, observed at the close of the US session on day t, predict next-day returns in Central European markets on day t+1, after those markets have already opened at prices reflecting the prior US session?
The four markets studied are the WIG20 (Poland), BUX (Hungary), PX (Czech Republic), and ATX (Austria), using daily data from January 2018 to December 2024 (1,637 aligned trading days).
- VIX changes negatively predict next-day CEE returns in all four markets. The VIX level carries no predictive content; the predictive information is in the daily change.
- The relationship is remarkably persistent. Across 1,384 rolling 252-day windows, the coefficient is negative in 97–99% of windows in every market.
- Predictability is 2–3× stronger in calm markets than in crises, which inverts the standard contagion-amplification view. This is the central and most counterintuitive result.
- In the pre-COVID subperiod, the VIX alone explained ~9.4% of next-day variance in the Czech PX, exceptionally high for daily-frequency return predictability.
- Results strengthen to 1% significance in all four markets after GARCH(1,1) standardization, confirming the effect operates on the conditional mean of returns rather than on their variance.
Five complementary specifications, each addressing a different concern:
| Specification | Purpose |
|---|---|
| Baseline OLS with Newey-West HAC standard errors | Core predictive regression, robust to heteroskedasticity and serial correlation |
| Rolling 252-day window estimation | Temporal stability of the coefficient |
| Regime interaction (VIX > 25 threshold) | Whether predictability strengthens or weakens in crises |
| GARCH(1,1) standardization | Robustness to conditional heteroskedasticity |
| Pre/post-COVID subperiod split | Stability across structurally distinct regimes |
notebooks/
01_data_collection.ipynb Data download, alignment, log returns, volatility proxies
02_diagnostics.ipynb Descriptive stats, ADF stationarity tests, correlations
03_ols_regression.ipynb Baseline OLS with Newey-West HAC standard errors
04_rolling_window.ipynb Rolling 252-day window estimation
05_regime_analysis.ipynb Crisis/calm interaction regressions
06_garch_robustness.ipynb GARCH(1,1) fitting and standardized re-estimation
07_subperiod_analysis.ipynb Pre/post-COVID subperiod regressions
data/ Daily price data (.parquet)
figures/ Generated charts
- VIX: CBOE, via Yahoo Finance
- WIG20, BUX, PX: Stooq
- ATX: Wiener Börse historical archive
git clone https://github.com/clonedfoxx/vix-cee-return-predictability.git
cd vix-cee-return-predictability
pip install -r requirements.txt
jupyter labRun the notebooks in order, 01 through 07. Each writes its outputs to data/ and figures/ for the next stage. Requires Python 3.11+.
Core dependencies: pandas, numpy, statsmodels, arch, scipy, matplotlib, seaborn, pyarrow.
Python · pandas · NumPy · statsmodels · arch · SciPy · matplotlib · seaborn
MIT. See LICENSE.
Herman Kaufman, Faculty of Management, Comenius University in Bratislava.
The full paper is available in the repository. This work was reviewed and presented at ŠVOČ 2026 (Section: Economics and Finance).