What Is It?
A signal is a rule, model, or transformation that takes inputs from the world and turns them into a forecast about future returns, risk, or market state. In a systematic strategy, the signal is the atomic unit of belief. It says, in effect, “given these observations, this asset now looks slightly better or worse than baseline.”
Good signals are rarely dramatic. Most real alpha is weak, noisy, and probabilistic. A 52-week momentum rank, a funding-rate dislocation, a prediction-market mispricing versus a base-rate model, or an order-book imbalance can all be signals. What matters is not story quality but whether the mapping produces a positive expected edge after costs.
How It Actually Works
The basic pipeline is: observe data, transform it, score assets, convert scores into expected returns, then size positions subject to risk and execution constraints. A simple momentum signal might be past 6-month return minus the most recent month. A microstructure signal might be the imbalance between aggressive buyers and sellers over the last minute.
Mathematically, signals are often standardized so they can be compared and combined. If signal s_i assigns each asset a z-score, then higher values imply a stronger long preference and lower values a stronger short preference. You test whether the ranking implied by s_i is associated with future returns. If it is, the signal has predictive content. If it is only predictive before fees, it is research, not edge.
In practice, systematic shops maintain a library of signals rather than a single master indicator. Some are slow and fundamental, some are fast and microstructural, and some are regime filters that decide when other signals should be trusted less. The job of quant research is to decide which signals add independent information and which are just alternate packaging for the same bet.
The Jargon Decoded
- Alpha — Return above the passive or risk-adjusted baseline.
- Feature — An input variable used to construct a signal.
- Forecast horizon — The time window over which a signal is expected to predict returns.
- Cross-sectional — Comparing many assets against each other at one time.
- Time-series — Forecasting the future of one asset from its own history.
- Z-score — A standardized value showing how far an observation is from the mean in standard deviations.
Why This Matters
If you do not understand signals, systematic trading collapses into cargo cult. Everything else, backtests, portfolio construction, execution, depends on having some measurable forecast to act on. Prediction markets make this especially clear: your edge is not “being smart” in the abstract, but encoding information into a better probability estimate than the market price implies.
What This Unlocks
Once you can think in signals, you can decompose a strategy into researchable pieces. You can ask whether the signal is real, whether it survives costs, whether it overlaps with other signals, and whether it scales. That is the difference between engineering a process and improvising trades.
What Still Breaks
Signals decay as they are crowded, arbitraged, or pushed outside their valid regime. Many “signals” are just leakage, bad timestamping, or disguised exposure to market beta. Others fail because the underlying data process changes, for example when exchange microstructure, regulation, or market participants change.
Sources
- The Math Behind Combining 50 Weak Signals Into One Winning Trade — Roan / RohOnChain — Modern intuition for weak alpha combination, IC/IR, and prediction-market adaptation.
- 151 Trading Strategies — Zura Kakushadze and Juan Andrés Serur — Broad taxonomy of strategies across equities, futures, options, FX, fixed income, and more.
- The Fundamental Law of Active Management — Grinold — Classic result linking skill, breadth, and expected information ratio.
- Advances in Financial Machine Learning — Marcos López de Prado — Practical guidance on features, backtests, leakage, and overfitting.