Futures Markets
Futures markets trade standardised, exchange-listed contracts to buy or sell an asset at a fixed future date, spanning commodities (energy, metals, agriculture), equity indices, fixed income and foreign exchange. They are organised on venues such as CME Group, ICE, Eurex and SGX, with near-continuous electronic trading and transparent central limit order books per contract. In this vault, futures markets appear as the test universe for the strongest positive reinforcement-learning trading evidence cited anywhere here — Zhang Zohren Roberts 2019, the Oxford-Man Institute study that trained DQN, Policy Gradient and Advantage Actor-Critic agents on the 50 most liquid futures contracts and reported a costed, out-of-sample risk-adjusted edge over classical momentum baselines.
Futures are an attractive — arguably the most favourable — venue for systematic-trading backtests, for four structural reasons. First, deep liquidity and low cost: the most active contracts trade deep order books, and per-contract transaction costs are low and well defined (exchange + clearing + brokerage fees plus spread). Second, symmetric long/short positioning: going short a futures contract is exactly as cheap and easy as going long, removing the shorting frictions and borrow costs that distort cash-equity backtests — a genuine advantage for any model, like a Markov Decision Process Trading Model, whose action space includes negative positions. Third, built-in leverage via margin, so a strategy can target a chosen volatility level without large capital outlay. Fourth, trend persistence: futures exhibit documented time-series momentum and clear volatility regimes, giving regime-aware and trend-following models real structure to exploit. These properties are why the managed-futures / CTA industry has built diversified futures portfolios as its standard universe — and why a futures backtest faces a relatively easy cost and structure hurdle.
Futures Markets [supports] Markov Decision Process Trading Model Futures Markets [relates] Time Series Momentum
Zhang Zohren Roberts 2019 exploits exactly this favourable structure. Its data are 50 ratio-adjusted continuous futures contracts from the Pinnacle Data Corp CLC Database — 25 commodity, 11 equity-index, 5 fixed-income and 9 FX contracts — spanning 2005-2019, with an expanding-window retrain every five years that produces a genuine 2011-2019 out-of-sample window. On the all-contracts portfolio with portfolio-level volatility targeting, the RL agents posted annualised Sharpe ratios of DQN 1.288, A2C 1.050 and PG 0.754, against 0.441 for Sign(R) time-series momentum, 0.091 for MACD and 0.058 for a long-only benchmark — and a transaction-cost robustness check showed DQN and A2C staying profitable up to a realistic 25 basis-point cost rate. This is the highest-graded positive RL backtest in the vault: a moderate profitability grade, earned because it clears the out-of-sample, transaction-cost and benchmark bars that most RL-trading papers fail.
Zhang Zohren Roberts 2019 [trades_market] Futures Markets Zhang Zohren Roberts 2019 [uses_dataset] Pinnacle Data Corp CLC Database Zhang Zohren Roberts 2019 [reports_profitability] Reinforcement Learning Trading Policy
But the same paper supplies the decisive caveat, and it is a property of futures markets as a test venue, not just of one model. The RL edge is regime-dependent: on the equity-index sub-portfolio a simple long-only strategy beat the RL agents, because the 2011-2019 test window was a sustained equity bull market in which the agents’ ability to go short or flat was a liability. The RL advantage concentrated in commodities and FX, where two-sided positioning had room to add value. That is direct evidence that a futures RL result is conditional on the asset mix and the macro regime of one historical window — favourable liquidity and low costs do not make the edge robust, they only make it cheap to express. The paper also models a flat per-trade cost rate with no slippage or market impact, publishes no maximum-drawdown table, releases no replication code, and reports no multiple-testing control (Probability of Backtest Overfitting / Deflated Sharpe Ratio) despite an undisclosed hyperparameter search.
Buy-and-Hold Benchmark [contradicts] Zhang Zohren Roberts 2019 Zhang Zohren Roberts 2019 [excludes_costs] Transaction Costs and Slippage Zhang Zohren Roberts 2019 [suffers_overfitting_risk] Overfitting in Quantitative Trading
Futures and FX are closely linked: the 9 FX contracts in the Pinnacle universe include the euro currency future, the exchange-traded analogue of the EUR USD Currency Pair OTC spot market. Both venues share the qualities that make them favourable test grounds — deep liquidity, low and well-measured costs, leverage, two-sided positioning and trend structure — and both share the same limitation in this vault’s evidence base: no result tracked here, in futures or FX, rises above a costed academic backtest with no live or paper-trading record. The verdict is that futures markets give Markov/RL models their best shot, and the best shot to date is one moderate-grade, regime-dependent, un-replicated backtest.
Futures Markets [relates] EUR USD Currency Pair Sim-to-Real Gap [opposes] Zhang Zohren Roberts 2019
Connections
- Zhang Zohren Roberts 2019 — trades_market, 50 futures contracts, 2011-2019 OOS, DQN Sharpe 1.288 net of 25bp, source: https://arxiv.org/abs/1911.10107
- Reinforcement Learning Trading Policy — trades_market, strongest positive RL backtest in the vault run on futures, source: https://arxiv.org/abs/1911.10107
- Markov Decision Process Trading Model — supports, symmetric long/short futures suit an MDP action space, source: https://arxiv.org/abs/1911.10107
- Pinnacle Data Corp CLC Database — uses_dataset, 50 ratio-adjusted continuous futures contracts, source: https://arxiv.org/abs/1911.10107
- Buy-and-Hold Benchmark — compares_benchmark, long-only beat RL on equity-index futures — edge is regime-dependent, source: https://arxiv.org/abs/1911.10107
- Time Series Momentum — compares_benchmark, RL beat Sign(R) momentum (Sharpe 1.288 vs 0.441), source: https://arxiv.org/abs/1911.10107
- Transaction Costs and Slippage — excludes_costs, flat per-trade cost only, no slippage/market impact, source: https://arxiv.org/abs/1911.10107
- Overfitting in Quantitative Trading — suffers_overfitting_risk, undisclosed hyperparameter search, no PBO/DSR, source: https://arxiv.org/abs/1911.10107
- Sim-to-Real Gap — lacks_live_evidence, academic backtest only, no live record, source: https://arxiv.org/abs/1911.10107
- EUR USD Currency Pair — relates, FX contracts are part of the futures test universe, source: https://www.bis.org/statistics/rpfx22_fx.pdf