Industry Momentum and Lead-Lag Effects

Drop a stone in a pond and the ripple that reaches the shore right in front of you arrives before the ripple that has to travel around a rock. News about a component supplier hits the supplier's stock first; the ripple reaches the manufacturer that buys those components a few days later, once analysts and investors get around to connecting the two. Lead-lag is the name for this delay, and it shows up wherever information takes time to travel from one related stock to another — within an industry, along a supply chain, or between a large, closely-watched stock and small, thinly-covered peers.

Splitting momentum into two pieces

Moskowitz and Grinblatt (1999) asked a sharp question: is ordinary momentum really about individual stocks continuing their own trend, or is it mostly industries continuing theirs, with individual stocks just riding along? They decompose a stock's momentum return into an industry component and a stock-specific component,

$r_{i,t} = \underbrace{\hat\rho_{i,I} \, r_{I,t}}_{\text{industry piece}} + \underbrace{\varepsilon_{i,t}}_{\text{stock-specific piece}}$

In words: a stock's return this period is its correlation with its industry's return times how the industry actually moved, plus whatever is left over that's specific to the stock. They found that buying past-winner industries and shorting past-loser industries captured most of individual-stock momentum's profitability — industry momentum alone, ignoring which stock within the industry you picked, delivered nearly as much return as full stock-level momentum, and controlling for industry membership made stock-level momentum much weaker.

Worked example 1. Suppose semiconductor stocks as a group returned 8% over the past six months ($r_{I} = 8\%$), and an individual chipmaker has an estimated correlation of 0.7 with that industry return. The industry component of its momentum is $0.7 \times 8\% = 5.6\%$ — more than two-thirds of a typical stock's momentum signal, before you've said anything specific about that company. A momentum strategy that sorts on raw past return is implicitly betting heavily on industry rotation even when the trader thinks they're picking individual stocks.

This scatter shows a lead-lag pair at moderate correlation — a follower stock's return plotted against a leader's return from the prior period. The visible upward slope is the exploitable part: knowing the leader's last move gives real information about the follower's next move, which is exactly what lead-lag strategies trade.

Trading the delay directly

Hou (2007) went further, ranking stocks within an industry by size or analyst coverage and testing whether small, less-covered stocks' returns could be predicted from the large, well-covered stocks' returns in the same industry from the prior week. The lag regression looks like

$r_{\text{small},t} = \alpha + \gamma \, r_{\text{large},t-1} + \varepsilon_t$

In words: this week's return for the small, slow-reacting stock is predicted by last week's return of the large, fast-reacting stock in its industry, scaled by a coefficient $\gamma$ — a genuine forecast, not just a contemporaneous correlation, because it uses information that was already public a week earlier.

Worked example 2. Say a large, heavily-covered oil producer rallies 6% this week on a crude price surge ($r_{\text{large},t-1} = 6\%$), and historical estimation puts $\gamma = 0.15$ for its small-cap, thinly-covered peers. The predicted follow-through next week is $0.15 \times 6\% = 0.9\%$ — a modest but real, tradeable edge if it holds up out of sample, and one that a stock-level momentum sort using only each stock's own past return would never see, because the small stock's own price hasn't moved yet.

What this means in practice

Lead-lag signals have visibly decayed as data and computing have gotten cheaper: index funds, ETF arbitrage, and algorithmic scanners now connect a supplier's earnings miss to its customer's stock within minutes rather than days for large-cap pairs, though the effect persists longer in less-covered corners of the market — small caps, less-liquid industries, and non-US markets with sparser analyst coverage. Building a lead-lag book today means going further out on the coverage curve than Moskowitz-Grinblatt or Hou needed to in the 1990s and early 2000s to find a delay that hasn't already been arbitraged away.