Signals & Alpha

At the core of any strategy is the signal, the rule or indicator that suggests when to buy or sell. The signal is the engine meant to generate alpha, the excess return from genuine edge that you met in Unit 6. But the hard truth of quantitative trading, which this lesson takes on directly, is that telling a real predictive signal from random noise is very hard, and most signals that look real are not.

What a signal is

A signal is an indicator or rule that provides input to a trading decision, suggesting when conditions favor buying or selling. Signals can be drawn from many sources: price and technical patterns from Unit 4, fundamental data from Units 2 and 3, statistical relationships from Unit 5, or alternative data of many kinds. A predictive signal aims to forecast future returns better than chance would, even slightly. The collection of signals that drives a strategy's decisions is often called its alpha model, because it is meant to be the source of the strategy's edge.

Real signals are weak

A humbling reality is that genuine predictive signals in financial markets are usually very weak. A real edge rarely means being right most of the time, more often it means being right slightly more than half the time, or earning a fraction more than the market on average. This thin edge is easily lost to transaction costs and is dominated by randomness in any small sample. As the law of large numbers from Unit 5 implies, such a weak edge can only be reliably harvested over a large number of trades, where the small advantage accumulates while the noise washes out. Expecting a strong, obvious, consistently winning signal is a sign of naivety or of having been fooled.

Combining signals

Because individual signals are weak, quants often combine multiple signals into a stronger composite. If several signals each carry a little genuine predictive value and are not too correlated with one another, combining them can produce a more reliable forecast than any single one, in much the same way that diversification combines imperfectly correlated assets to reduce risk. The caution, familiar from Unit 5, is that combining many signals also multiplies the opportunity to overfit, so the combination must be approached with the same rigor and skepticism as everything else in quantitative trading.

Most signals that look predictive are just noise. The central skill of a quant is telling the rare real edge from the many false ones.

The central danger: noise masquerading as signal

This is the crux of the whole field. Most signals that appear predictive in historical data are not real, they are noise, patterns that arose by chance and will not persist. This connects directly to the p-hacking and overfitting warnings of Unit 5: if you test enough rules against historical data, some will appear to work splendidly purely by luck, with no genuine predictive power. The fundamental challenge of quantitative trading is to distinguish a true signal, reflecting a real and persistent market effect, from the vast sea of spurious patterns that history inevitably contains. This is far harder than it sounds, and the failure to do it honestly is the single most common reason quant strategies fail.

Alpha is rare and fleeting

Even when a genuine signal exists, it tends to be rare and to decay over time. Real alpha is competitive: once a profitable signal becomes known and exploited by enough participants, the very act of trading on it erodes the edge, a phenomenon called alpha decay or crowding that echoes the efficient-market pressures of Unit 4. A signal that worked in the past may stop working as others discover it or as market conditions change. The realistic view, consistent with the skepticism running through this curriculum, is that genuine predictive signals are weak, rare, hard to distinguish from noise, and prone to fading, which is exactly why the rigorous testing and honest self-skepticism developed in the rest of this unit matter so much.