Combining Signals

Individual trading signals are usually weak and noisy. Combining several signals that capture different information can produce a much more reliable overall forecast, in the same way that diversifying across assets reduces portfolio risk. Quant firms build models from dozens or hundreds of signals; even discretionary traders combine trend, valuation, sentiment and timing. How signals are standardised, weighted and combined makes a large difference to results.

Why combining works#

If two signals each have some predictive power and are not highly correlated, their combination has a higher expected information coefficient relative to its noise than either alone.

IC of equal weight combination ≈ (IC1 + IC2) / √(2 × (1 + ρ))

for two signals with similar IC, where ρ is the correlation between the signals. Lower correlation means a bigger benefit.

Step 1: standardise#

Signals come in different units: P/E ratios, percentage returns, sentiment scores. Before combining, put them on the same scale:

Step 2: choose weights#

Research on forecast combination in economics has repeatedly found that simple equal weighting is hard to beat out of sample, because estimated optimal weights are noisy. This "forecast combination puzzle" is relevant to trading signals too.

Signal correlation#

Check the correlation between signals before combining. Highly correlated signals add little diversification and can double count the same information. A new signal is most valuable if it predicts returns and has low correlation with existing ones. See Covariance and Correlation.

Combining at different stages#

Signal level combination is usually more efficient because offsetting trades are netted, reducing turnover and costs. See Signal Turnover, Breadth and Neutralization.

Pitfalls#

• Overfitting weights to past data.
• Adding signals until the backtest looks good. See P-Hacking and Multiple Testing.
• Ignoring different decay horizons: mixing fast and slow signals needs care. See Signal and Alpha Decay.
• Double counting factors under different names.

Frequently asked questions#

Why combine trading signals?#

Because combining several weak, not highly correlated signals produces a more reliable forecast and smoother returns than relying on one signal.

How should signals be weighted?#

Equal weighting is a strong, robust default; more complex weighting can help but risks overfitting.

What is signal neutralisation?#

Removing unwanted exposures, such as sector or size, from a signal so it captures only the intended information.

Next, learn how much money a strategy can handle in Alpha Capacity and Crowding.