Modeling Crypto Trading Features as Expected Returns
Summary
The article explains how to express trading signals as expected returns, giving a common scale for comparing features and combining them with risk estimates and trading costs. Its example uses Binance perpetual futures and considers carry, short-term cross-sectional momentum, and a breakout measure based on time since a recent high. The author recommends inspecting how each feature relates to forward returns across its range and through time before choosing a model.
In the reported analysis, carry appears roughly linear, while momentum looks weak in the combined sample but has different, potentially changing relationships in different years. The article favors simple models where justified and describes rolling coefficient estimation as a way to adapt to changing relationships. It also cautions that model choice, estimation-window length, and refit frequency can introduce data snooping or future-peeking biases. The provided text omits much of the implementation and results section, so it does not support a detailed assessment of the strategy’s performance.
Key ideas
- Expected-return estimates put different features on a common scale for comparison and portfolio decisions.
- Inspect feature-return relationships across signal values and over time before selecting a model.
- A simple linear model can limit overfitting when a feature’s relationship with returns is approximately linear.
- Rolling estimation may adapt to changing relationships, but choices about model form and refitting introduce bias.
- Expected returns can be considered alongside risk, costs, turnover, and trading constraints.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.