16.30 - 17.15:  Songrun He (Washington University in St. Louis) presents “Interpretable Systematic Risk around the Clock”.

This paper combines high-frequency market data with real-time news analytics to identify the causes of market jumps across the full 24-hour cycle. Using a large language model, each jump is linked to a specific economic narrative (e.g., macroeconomic releases, policy announcements, or geopolitical events), allowing the construction of “pure-play” risk factors. The results show that different types of news carry very different risk premia, with macro-driven jumps earning the highest and most persistent compensation. These insights translate into implementable trading strategies that achieve strong out-of-sample Sharpe ratios and significant alpha relative to standard factor models.

17.15 - 18.00: Vitor Azevedo (RPTU Kaiserslautern-Landau) presents “The Expected Returns on Machine-Learning Strategies”, joint work with Christopher Hoegner and Mihail Velikov.

This paper assesses how machine learning trading strategies perform in realistic conditions by incorporating transaction costs, post-publication decay of signals, and today’s high-liquidity market environment. The analysis shows that these frictions reduce strategy performance by more than half relative to standard backtests. However, more sophisticated models—particularly those designed to capture complex, nonlinear patterns—remain profitable, delivering positive alpha and strong performance in certain market conditions such as recessions. The findings highlight that while competition erodes simple signals, advanced data processing remains a key source of return.