Research
Research Interests: Machine learning in asset pricing and macroeconomic time series forecasting.
Working Papers
Proner, R. A Multi-Task Deep Learning Model for Inflation Forecasting: Dynamic Phillips Curve Neural Network Managing inflation is vital for a stable economy, but forecasting remains challenging. In recent years, off-the-shelf machine learning (ML) methods such as random forests have been shown to outperform traditional benchmarks. I introduce the Dynamic Phillips Curve Neural Network (DPCNet), a deep multi-task learning model, to jointly forecast inflation and unemployment. DPCNet incorporates economic structure and dynamics, leading to significant gains in out-of-sample forecast accuracy compared to traditional benchmarks and state-of-the-art ML models. Download working paper here
Proner, R. and Liu, F. Factor Timing with Deep Learning Factors in the cross-section of stocks are persistent sources of risk premia (Fama, E. F. and French, K. R. 1993), forming the basis for factor investing. Factors such as size and value deliver high returns over the long run, but can underperform in the short run, creating the scope for investors to engage in factor timing. We develop a dynamic multi-task neural network (DMT) for the simultaneous timing of several factors, while imposing several economically motivated constraints with the model architecture. We show that DMT is more accurate on average and provides significant gains in excess returns over the benchmark buy-and-hold strategy and other state-of-the-art machine learning methods.