How Many Stocks Are Driving the Market?

Each faint point is one 63-session window; the dark line is its 21-session average. The single-window reading jumps when a large daily move by a big stock enters or leaves the window, so read the average for the trend. The thin line at 1 is the weight-based reference; the shaded band is the middle half of the average’s own history (0.60 to 0.85). Readings below 1 are the usual state, and only the band’s edges are unusual.

The largest positive contributions and the largest drags over the same 63 sessions. Each contribution is the member’s starting weight times its return, and together they add up to the basket’s +2.85% (SPY +2.85%). The two shares answer different questions. Share of net is how much of the basket’s gain or loss one stock supplied, shown only when gains and losses do not nearly cancel; share of gross is how much of all the movement, gains and losses counted without sign, it accounts for, and the ratio above is built on the gross. 235 members added to the basket and 259 subtracted.

63 sessions to October 16, 2026, 494 of 503 members with a price and share count. Weights are today’s share count times the closing price at the start of the window: plain market value, without the index provider’s float adjustment. Across the record the basket’s 63-session return sits a median 0.44 points from SPY’s. The Nasdaq-100 version is rougher because the index caps its largest weights: over this window the basket returned +3.81% against QQQ’s +1.97%, so it appears only as a rough estimate.

The same measure on Ken French’s 49 US industries, weighted by market value, dividends included, every session since 1926. It has no survivorship gap, and it is the only record the study below uses. French publishes monthly, so this record ends on August 31, 2026; the stock estimate on that same day is beside it.

Over that window the largest industry contributions were pharmaceuticals (+0.90 pts) and banks (+0.72 pts), and the largest drag semiconductors and electronics (−1.19 pts).

Two ways of counting the same history give different answers. Counting every day, the broadest tenth of industry readings, each ranked only against earlier days, was followed by a median next-quarter market return of +4.2% against +2.5% after the most concentrated tenth. Neighbouring days share most of their next quarter, though, so a few long stretches dominate that count. Counting each episode once, the medians were +1.8% and +2.4%, a difference of −0.6 points with a 95% range of −3.4 to +2.6, which cannot be told apart from chance. A 10% fall followed 25% of broad episodes and 23% of concentrated ones. The tool describes what moved the index; on this record it offers no dependable forecasting advantage.

CRSP US market with dividends. Tenths are point-in-time, each day against every earlier day once 1,000 readings exist. An episode is the first day in a tenth after 63 sessions without one. Ranges in brackets are 95% bootstrap intervals over episodes.

• Effective contributors are one over the sum of squared shares of the gross contribution; the weight-based count is one over the sum of squared starting weights. The ratio of the two is the reading.
• The S&P 500 and Nasdaq-100 readings are estimates: today’s members, weighted by today’s share count times each past close. The index providers’ float and capping adjustments are not applied.
• Carried back to 2010, today’s member list misses companies that were removed, so the stock history describes today’s index rather than the one that existed. It feeds no study.
• The study is on 49 industry portfolios. Concentration among industries and among stocks can behave differently, and nothing here establishes what the stock reading predicts.