Deep Reinforcement Learning for Dynamic Cryptocurrency Pair Trading
Summary
This study tests deep reinforcement learning as an execution overlay for pair trading in volatile cryptocurrency markets. Its framework first filters and ranks candidate pairs, then uses a Proximal Policy Optimization agent with a Long Short-Term Memory layer to make execution decisions. A fixed-risk, adaptive-mean design and deterministic risk controls constrain the agent, aiming to reduce divergence risk. The evaluation uses hourly Binance USD-M futures data and compares the learned policy with a heuristic baseline.
The reported out-of-sample results favor the reinforcement learning policy, and a stationary circular block bootstrap indicates statistically significant risk-adjusted outperformance at the 10 percent level. The result does not meet the stricter 5 percent significance threshold, which the study relates to high idiosyncratic variance in digital assets. The evidence is specific to the tested data and setup; the description does not provide performance magnitudes, transaction-cost assumptions, or tests across other venues and periods. The hybrid design offers a way to bound learned execution decisions, but broader robustness remains an open question.
Key ideas
- The proposed system combines statistical pair selection with a learned execution overlay.
- A PPO agent with an LSTM makes execution decisions inside deterministic risk boundaries.
- The study evaluates the approach on hourly Binance USD-M futures data against a heuristic baseline.
- The reported risk-adjusted outperformance is significant at the 10 percent level but not at the 5 percent level.
- The results are specific to the tested market, data, and strategy configuration.
Tags
Full text
# Dynamic Multi-Pair Trading Strategy in Cryptocurrency Markets with Deep Reinforcement Learning # Dynamic Multi-Pair Trading Strategy in Cryptocurrency Markets with Deep Reinforcement Learning This study aims to determine whether the application of Deep Reinforcement Learning (DRL) as a specialized execution overlay can enhance pair trading in highly volatile cryptocurrency markets. Although classical implementations of the strategy have proven successful in traditional equities, they frequently exhibit rigidity and suffer from severe divergence risks when applied to high-variance environments. To address this need, this research introduces novel concepts. To construct a robust system, we developed a hierarchical "Filter-then-Rank" pair selection methodology and a proprietary "Fixed Risk, Adaptive Mean" execution model. The system employs a Proximal Policy Optimization (PPO) agent with a Long Short-Term Memory (LSTM) layer to govern execution decisions within strict deterministic risk management boundaries. Evaluated on 1-hour interval data from the Binance USD-M Futures market, the optimized RL policy achieved an out-of-sample performance that substantially outperformed the heuristic baseline. A stationary circular block bootstrap robustness check confirms that the agent's risk-adjusted outperformance is statistically significant at the 10 percent level. Although falling marginally short of the stricter 5 percent threshold, this result highlights the extreme idiosyncratic variance characteristic of digital assets. Ultimately, this thesis contributes to the quantitative finance literature by introducing a hybrid architecture that combines statistical arbitrage with DRL execution policies. Furthermore, it delivers a novel framework for safe reinforcement learning via deterministic shielding, proving that anchoring a neural policy to statistically robust boundaries successfully mitigates severe divergence risks.
Shown in full with attribution under the source's licence. Licence: abstract CC0
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.