Lead-lag
The lead-lag strategy
In plain words
Some stocks react to news slightly before others. When the leader moves, the follower tends to follow in the coming days. The lead-lag strategy spots these couples and uses the leader's moves to anticipate the follower's.
A lead-lag signal, number by number
This is what a lead-lag signal looks like in PairScanner. The stock and the figures are fictional.
Click a number to learn its role
Expected move+1.4%over 5 sessions, versus market and sector, after costs
My role: tell you what this kind of signal earned on average in the past, once costs are paid. Here +1.4% over 5 sessions, on top of what the market and the sector did.
80% range-1.8% to +4.3%median +1.1%
My role: show the range of the most common outcomes. Eight similar signals out of ten ended inside it. The narrower it is, the more predictable the outcome.
Probability59%of a positive outcome
My role: estimate the chance that this signal ends positive, after costs. 50% would be a coin flip; 59% means about six similar signals out of ten ended positive.
Sample283 / 480positive cases, 2019 to 2026
My role: tell you how many past situations the figures rest on. 480 cases make a solid base: the figures do not hang on a few lucky trades.
Worst decile-2.7%one outcome in ten did worse
My role: give the size of an unfavorable outcome, so you can size the position knowingly.
ValiditySep 29, 2026expires Oct 6, 2026
My role: date the data used and tell you how long the signal counts. An expired signal is no longer tracked.
What PairScanner does
For every stock it follows, the algorithm examines a short list of possible leaders: its sector ETF, the market, its three statistical peers and known relationships. Every month, it trains a statistical model, a so-called ridge regression, that forecasts the follower's return from the leader's last five days, taking into account the follower's past, the market and the sector.
A signal only appears when two conditions are met: the forecast is clearly stronger than usual, beyond one standard deviation, and the leader genuinely improves the forecast compared with a model without it.
The algorithm then computes, for 1, 5 and 20 session horizons, the expected move after costs, the estimated probability of a positive outcome, the range of plausible results and the size of an unfavorable outcome, from replaying the same rule on history.
You code nothing and calculate nothing: you open the Lead-lag tab of the dashboard and read today's result.
The idea in detail
Information does not spread everywhere at once. A large customer reports strong demand: its price reacts immediately. Its suppliers, less covered by analysts, sometimes react one or several days later.
The strategy looks for these leader and follower couples. It does not test every possible combination, which would surface coincidences: it starts from plausible links, such as the sector or statistical peers, and requires the leader to genuinely bring information.
A worked example
A large chipmaker reports record orders and gains 8% in the day. One of its suppliers, which the model identified as a follower, gains only 1%.
| Step | Value |
|---|---|
| Leader, move of the day | +8% |
| Follower, move of the day | +1% |
| Model forecast for the follower, 5 sessions | +2% |
| Usual forecast | about 0.7% either way |
| Does the leader improve the forecast? | Yes |
| Decision | Buy the follower |
The forecast clearly exceeds the usual and the leader brings information: a buy signal appears on the follower.
Fictional example for teaching purposes. It is neither a signal nor a recommendation.
Why it would work
In 1990, Andrew Lo and Craig MacKinlay showed that large companies' returns lead those of small ones. In 2007, Kewei Hou showed that the effect plays out mainly within the same industry: information reaches large companies first, then spreads slowly.
In 2008, Lauren Cohen and Andrea Frazzini showed that customers' returns help predict their suppliers'. A long-short strategy exploiting this delay earned over 1.5% per month in risk-adjusted returns in their study.
What research says today
In 2010, Lior Menzly and Oguzhan Ozbas confirmed that returns can be predicted from an industry to its suppliers and customers, along economic chains.
More recently, less visible links, such as technological proximity between companies, turned out to carry the same information: that is why PairScanner includes statistical peers among possible leaders.
Sources
Studies cited on this page:
- Andrew W. Lo and A. Craig MacKinlay (1990). When Are Contrarian Profits Due to Stock Market Overreaction?. The Review of Financial Studies, 3(2), 175-205.
- Kewei Hou (2007). Industry Information Diffusion and the Lead-lag Effect in Stock Returns. The Review of Financial Studies, 20(4), 1113-1138.
- Lauren Cohen and Andrea Frazzini (2008). Economic Links and Predictable Returns. The Journal of Finance, 63(4), 1977-2011.
- Lior Menzly and Oguzhan Ozbas (2010). Market Segmentation and Cross-predictability of Returns. The Journal of Finance, 65(4), 1555-1580.
- Charles M. C. Lee, Stephen Teng Sun, Rongfei Wang and Ran Zhang (2019). Technological Links and Predictable Returns. Journal of Financial Economics, 132(3), 76-96.
PairScanner promises no results: the figures describe what the strategies did in the past. Educational content, not investment advice; PairScanner places no orders.
See today's lead-lag signals