Time Series Momentum
Time series momentum is a trend-following strategy that sets a position from an asset’s own trailing return — the canonical “Sign(R)” rule goes maximally long when the trailing 12-month (252-day) return is positive and maximally short when it is negative, with no reference to a cross-section of peers. It is well documented as a robust, persistent effect across futures, equity indices, currencies and commodities (Moskowitz, Ooi & Pedersen 2012), which is why it serves as the standard baseline that reinforcement-learning trading papers test against. In Zhang Zohren Roberts 2019 the all-contracts Sign(R) momentum portfolio reached an annualised Sharpe of 0.441 — the bar the RL agents (DQN 1.288) were measured against — and the paper’s premise is that RL improves on momentum by mapping market state directly to positions rather than relying on a single fixed trailing-return rule.
Connections
- Zhang Zohren Roberts 2019 — compares_benchmark, Sign(R) momentum Sharpe 0.441 vs RL agents, source: https://arxiv.org/abs/1911.10107
- Reinforcement Learning Trading Policy — relates, RL trading framed as an improvement over fixed momentum rules, source: https://arxiv.org/abs/1911.10107
- Futures Markets — trades_market, momentum tested across 50 liquid futures, source: https://arxiv.org/abs/1911.10107