Macroeconomic risk forecasting, real-time evaluation and model uncertainty.
5 projects
01
Working paper draft · 2025–present
High-Dimensional Distributional VAR (HiDVAR)
Behind the Curve: How We Missed Inflation Risks Using a High-Dimensional Distributional VAR (HiDVAR)
I develop a high-dimensional distributional VAR for macroeconomic risk forecasting. The framework constructs path-consistent, nonparametric predictive densities and evaluates them out of sample against benchmark models.
Question: What can large real-time datasets tell us about risks that a point forecast misses?
The project extends the Outlook-at-Risk approach using nonparametric distributional estimation and path-consistent forecast construction. Evaluation compares the resulting predictive distributions with benchmark forecasts.
Presented in 2026
International Symposium on Forecasting Montréal
European Meeting of the Econometric Society Dublin
September 2021 inflation forecast. One-year-ahead CPI inflation densities: HiDVAR (blue) and FRBNY (orange dashed).Real-time macro-risk scenarios. Shares of 10,000 simulated paths assigned to mutually exclusive macroeconomic scenarios.Policy-rule comparison. Actual rate and FRBNY- and HiDVAR-based policy-rule paths, with 50% and 80% intervals shown.
1 / 3
02
Work in progress · 2025–present
Many Models, Little Certainty
Lessons from 25 Years of Real-Time Macroeconomic Forecasting in the US
I compare 50 DSGE, time series, machine learning, neural network and foundation models using modified FRED-QD data across forecast horizons and data scenarios.
Question: How does predictive performance change across models, economic conditions and information sets?
Forecasts must be evaluated against the information available when they were made. Data releases and revisions can change conclusions about which models perform well.
03
Work in progress · with Volker Wieland · 2024–present
Forecasting with an Ensemble of DSGE Models
Evidence from 25 Years of US Data
We evaluate DSGE and Bayesian VAR ensembles in real time across crisis episodes and forecast horizons, comparing their performance with institutional forecasts and time series benchmarks.
Question: Can combining models improve forecasts across changing economic conditions?
Research motivation
Individual models embody different assumptions about the economy. Ensembles allow us to study whether their information is complementary and useful across forecast horizons.
04
Research project · DVD
History Does Not Repeat Itself, but It Rhymes
A VAR Dirichlet Process Mixture Approach to Macroeconomic Forecasting
I combine a Dirichlet process mixture with a VAR and dynamic volatility modelling to capture changing macroeconomic regimes. Applied to US macro-financial data, the DVD framework examines whether flexible regime identification improves forecasts across structural breaks and periods of shifting volatility.
Question: Can learning from recurring economic regimes improve forecasts when the economy changes?
05
Master’s thesis · MSc Quantitative Economics · Goethe University Frankfurt
A Macroeconomic Uncertainty Index for Euro Area Countries
Construction and Applications to Tail Risk Forecasting
My master’s thesis constructs forecast-based macroeconomic uncertainty indices for ten euro area countries using large macroeconomic datasets and stochastic volatility models. I assess their value for forecasting inflation, GDP growth and unemployment tail risks beyond financial-stress indicators, and develop a composite policy-risk index for country-level risk monitoring.