Recent papersView all
Reinforcement Learning for Perpetual Futures Market MakingOriginal contributions
Martin Cekal
FFA Working Papers 6:004 (2026)63 
This paper studies market making in BTC and ETH perpetual futures with a simple event-driven reinforcement learning setup. PPO chooses quote skew and quote width, while the environment tracks execution gains and inventory losses separately. A temporary reduction of the adverse-fill penalty makes early training more stable. Across simulated and historical data, PPO quotes more selectively, carries smaller inventory, and usually ends with better portfolio value than a zero-action baseline. BTC is tighter than ETH, so the same control logic requires market-specific calibration, yet the qualitativepolicy behavior remains consistent across both markets....
Natural Gas Storage Valuation Using Deep Reinforcement LearningOriginal contributions
Masood Tadi, Milan Fičura, and Jiří Witzany
FFA Working Papers 6:003 (2026)181
We study natural gas storage valuation under a stochastic futures term structure using deep reinforcement learning (DRL). The storage problem is formulated as a continuous-state, continuous-action Markov Decision Process and solved using the Deep Deterministic Policy Gradient (DDPG) algorithm with Prioritized Experience Replay (PER) buffer and a constraint-aware policy network. We benchmark the approach against intrinsic and rolling intrinsic strategies and find that DRL consistently outperforms intrinsic valuation and achieves competitive performance relative to rolling intrinsic in markets with jumps and seasonality. The results show that DRL provides...
Integrating Flood Risk into House Price Models Using Expected Discounted Loss: Evidence from the Czech Republic in 2024Original contributions
Marek Folprecht
FFA Working Papers 6:002 (2026)256
Prices of houses in flood risk zones are subject to a price discount reflecting the risk of losses caused by floods. First, the article establishes a framework for pricing flood risk using the expected discounted loss approach, based on the capital asset pricing model and the Gumbel mixture model of estimated likelihood and impacts of flood risk events. The measure advantage is dimensionality reduction and interpretability. Second, the resulting measure of flood risk is tested to assess whether it can explain differences in house prices using a large data sample from the Czech Republic in 2024. I show that the flood risk measure, expected discounted...
Measuring Flood Risk in Czechia with Stress Testing and a Gumbel copula‑based VaROriginal contributions
Marek Folprecht
FFA Working Papers 6:001 (2026)502
The study presents a holistic approach to modeling flood risk of real estate properties. The method combines the hydrological flow simulation model and a model of financial losses. Two use cases of the model are discussed. First, a stress testing method, based on historical scenario simulations, is presented. Next, a Value at Risk approach using the Generalized extreme value distribution and the Gumbel copula is discussed. Both methods are then tested on a large sample of Czech house data. The results show that the model can replicate the order of historical flood magnitudes under the historical scenarios. Moreover, the Value at Risk approach can generate...