Archive / Equities & volatility / E929
Sector leaders' move to 15:30 shows no reliable lead over peers
Ranking sectors by their biggest stocks' move until 15:30 and trading the sector's other stocks from close to next open earned +1.42 bp a day against a 10 bp cost, not distinguishable from zero.
A common idea says news hits the biggest stock in a sector first and reaches its smaller peers later. An earlier test had found peers catching up in the overnight gap between close and next open. So we ranked sectors by their five largest stocks' move until 15:30, early enough to submit a closing-auction order, bought the peers of the three strongest sectors, shorted those of the three weakest, and exited at the next open. Over 722 days large-cap peers earned +1.42 bp a day (a bp is 0.01%) against a 10 bp cost, and small-cap peers +0.01 bp against 30 bp.
News in a sector shows up first in its biggest, most-watched stocks and reaches the sector's smaller names with a delay.
Why people believe it
People argue that investors in less-followed stocks pay less attention and react more slowly, so a move in the sector leaders should be followed by the same direction in their peers. An earlier test of the same idea on daily data found the peers catching up between the close and the next morning's open, a gap that cannot be traded from a signal known only after the close.
Why it would persist The argument, as stated before testing
People argue that the less-followed peers get fewer analysts' updates, so the delay could last long enough to survive until the next open.
Each day we measured how the five most heavily traded large stocks in each of 11 sectors had moved from the previous close until 15:30 New York time, relative to the S&P 500. That is 20 minutes before the deadline for closing-auction orders, so the signal is usable. We bought the other stocks of the three strongest sectors at the close, shorted those of the three weakest, and closed both sides at the next morning's opening auction. Returns are measured relative to the S&P 500 and are charged a assumed trading cost. The same test ran on large-cap peers and on small-cap peers.
Per cell: gross mean at least 3 times cost; Newey-West t (5 lags) at least 2.5; median above 0; net above 0 in both halves; net above 0 under doubled costs; Holm-adjusted p below 0.05 over the 2 cells.
Gross mean at or below cost.
Cell 1: 4 bp per leg on two legs plus 2 bp for borrowing the short side = 10 bp per day. Cell 2: 14 bp per leg on two legs plus 2 bp = 30 bp per day. Stress case: costs doubled.
Descriptive checks on the same days: the same overnight trade using the leaders' full-day return to the close instead of the 15:30 return (the version that cannot be traded), and the followers' return from 15:30 to the close with the 15:30 signal.
Gross return was below cost in both cells (+1.42 bp against 10 bp, +0.01 bp against 30 bp), which meets the pre-set rejection criterion.
- Large-cap peers (cell 1) earned +1.42 bp a day gross over 722 days (t 0.83, a t-statistic, is the average divided by its noise; median -1.3 bp), a net of -8.58 bp after the 10 bp cost. Net is gross minus cost.
- The positive average does not count: a t of 0.83 is consistent with zero, the median day lost 1.3 bp before cost (11.3 bp after), and it is far below the 10 bp cost.
- Small-cap peers (cell 2) earned +0.01 bp a day gross (t 0.00) against a 30 bp cost.
- The first half of the period earned +1.3 bp a day and the second half +1.5 bp a day in cell 1; both are below the 10 bp cost.
- Using the leaders' full-day return to the close, which cannot be traded because it is known only after the closing-auction deadline, the same overnight trade earned +1.1 bp (large caps) and +0.0 bp (small caps).
- The long-short portfolio's return from 15:30 to the close on the 15:30 signal was +1.2 bp (large caps) and +0.6 bp (small caps), gross; far below the 10 and 30 bp costs of the full trade.
- The earlier test on daily data over a longer period had found about +2.5 bp a day overnight for large-cap peers. The comparable figure in this window was +1.1 bp, a difference within this window's noise.
Gross return against cost, both cells
Gross returnCost
Mean gross daily return of the overnight pairs trade, next to the cost of running it. Neither cell reaches its cost; the large-cap bar of +1.42 bp has a t-statistic of 0.83, so it is not separable from zero.
Chart data (bp)
| Gross return | Cost | |
|---|---|---|
| Large-cap peers | 1.42 | 10 |
| Small-cap peers | 0.01 | 30 |
Other versions of the same signal
Large-cap peersSmall-cap peers
Gross daily return of three versions on the same days. The tradable 15:30 signal and the untradable full-day signal give the same near-zero answer overnight, and the 15:30-to-close row is a separate leg, not an increment.
Chart data (bp)
| Large-cap peers | Small-cap peers | |
|---|---|---|
| 15:30 signal, close to open (tested) | 1.42 | 0.01 |
| Full-day signal, close to open | 1.1 | 0 |
| 15:30 signal, 15:30 to close | 1.2 | 0.6 |
Large-cap peers by half of the period
Gross mean per day in each half, against the 10 bp cost. Both halves are positive and both are far below the cost, so the sign does not decide anything.
Chart data (bp)
| Gross return | |
|---|---|
| First half | 1.3 |
| Second half | 1.5 |
Why it failed Interpretation · not established
The earlier daily test measured about +2.5 bp a day overnight in peers. Even if that average were real, it is a quarter of the 10 bp cost of trading a long and a short basket in and out, and in this window the comparable figure was +1.1 bp. Moving the signal to 15:30 made the trade possible but did not make the gross return larger than the cost, which is what decided the result.
Lesson
An effect measured in one test can fail to appear in another, and even where it appeared it was smaller than the cost of trading it.
Scope of this result
Covers current S&P 500 and S&P 600 members, 11 sectors, a 15:30 signal, closing-auction entry and opening-auction exit, December 2023 to October 2026, equal-weighted baskets and the stated costs. Index membership and sector labels are today's, and the 15:30 price is the last trade of the hourly bar rather than a quote.
A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.
1. Test design
Pre-registeredDataUniverseSignalExecutionCostsControlsSample splitPass criterionKill criterionPower2. Result
ObservationsInstrumentsNet mean95% intervalMedianWorstFinding 1Finding 2Finding 3Finding 4Finding 5Finding 6Finding 73. Interpretation
Why it failedLessonScope4. Reproducibility
StepsGrade