Why a high seasonal win rate can still mean little advantage
Two SPY calendar windows both made money in most years. What they added beyond simply holding SPY differs many times over. Here is why, and what to ask instead.
A win rate counts how often a rule made money. It does not say whether the rule did better than simply holding the same instrument for the same time. SPY's 15 sessions from October 31 rose in 18 of 26 years (69.2%); the November window rose in 23 of 27 (85.2%). The win rates are 16 percentage points apart. Their average excess over holding is +0.12 pp against +1.73 pp, about 14 times larger. The second question tells the two apart; the first barely does.
Did it make money, and did it beat holding?
Every calendar claim can be asked two different questions. The first is whether the trade made money. The second is whether it made more than an ordinary position in the same instrument, held for the same number of sessions, would have made. A win rate answers the first. SeasonalEdge reports both, and leads with the second, because a market that rose over most of the sample makes almost any long window look like a winner.
The holding benchmark is the average return of every ordinary hold of that length in the price history before the research cutoff. For SPY over 15 sessions it is +0.56%; over four sessions, +0.15%. The methodology page explains how it is computed.
Four SPY rules, both questions
The October 31 window and the November window are both once-a-year rules, so each year is one trade and the two counts share a denominator. October 31 made money in 18 years and beat holding in 16. November made money in 23 and beat holding in 21. In each case two years rose without clearing the hold: 2010 and 2015 for October 31, 2015 and 2021 for November.
On the counts alone, the four rules look like a ladder of modest differences.
Read only the counts and the rules form a gentle ladder. Read the size of the advantage and they do not:
| Rule | Trades that rose | Years beating hold | Average excess |
|---|---|---|---|
| October 31, 15 sessions | 69.2% | 16 / 26 | +0.122 pp |
| November | 85.2% | 21 / 27 | +1.725 pp |
| Turn of the month | 58.8% | 15 / 26 | +0.090 pp |
| FOMC announcement day | 54.6% | 14 / 26 | +0.028 pp |
November's average excess is many times the others'. The October 31 window, with a win rate of 69.2%, adds +0.12 pp a trade, and drops to −0.21 pp without its best year. The October 31 study and the turn-of-the-month study show every year behind these figures.
Count what, out of what?
A win rate is only comparable when its denominator is stated. For a once-a-year rule, a trade and a year are the same thing. For the turn of the month they are not: 311 trades in 26 years. Its trade win rate is 58.8%; the share of years whose average trade made money is 65.4%; the share of years that beat holding is 57.7%. Three honest percentages for the same rule, each answering a different question.
The FOMC rule has the same issue at a smaller scale: 207 announcement days, 113 of them up (54.6%), and 14 of 26 years beating holding. When a chart shows a single win rate for a rule that trades many times a year, ask which unit it counts.
What a win rate leaves out
The size of wins and losses. A rule can win often and still lose more on its bad trades than it gains on its good ones. October 31's 18 rising windows averaged +3.45%; its 8 falling windows averaged −5.56%. November's losers were also larger than its winners, on average, though there were only 4 of them.
| Rule | Rose / fell | Average rise | Average fall | Payoff ratio |
|---|---|---|---|---|
| October 31, 15 sessions | 18 / 8 | +3.45% | −5.56% | 0.62× |
| November | 23 / 4 | +3.79% | −5.07% | 0.75× |
The benchmark. A win rate is measured against zero. In a market that drifted upward, zero is a low bar; the same-length hold sets a higher one. That is why two October 31 years, 2010 and 2015, count as wins and as misses at the same time.
Time. A single percentage for the whole sample hides when the wins happened. A rule whose wins cluster in one decade reads the same as one whose wins are spread evenly.
Five questions to ask of any seasonal win rate
- Against what? Zero, or an ordinary hold of the same instrument for the same number of sessions?
- Out of what? Trades or years, and how many of each?
- How big? What did the average winning and losing trade return, and what was the worst?
- How fragile? Does the average survive removing the single best year?
- When? Are the good years spread out, or bunched in one stretch?
None of these answers turns a historical pattern into a forecast. They make it possible to tell a rule that mostly tracked the market from one that did something more, in the years measured.
Scope: every figure here is gross, before trading costs and tax, and computed on history through 31 December 2025. A higher win rate or excess in the past does not establish a probability of success in future years.
Read a study with both questions answered
Each research page shows the rule, every year against holding, the weak years and the results after the cutoff.
Browse SeasonalEdge researchSources and research notes
- October 31 and November (SPY): SeasonalEdge MCP
check_rule, retrieved 2026-10-19 (rule idsdd9df50ad4b1eb0f,d7bff62caad48bf0); October 31 re-checked against the public research API on 2026-10-19T20:51:46Z. - Turn of the month and FOMC (SPY): SeasonalEdge public research API, retrieved 2026-10-19T20:51:46Z (rule ids
92688fa1e33982f6,35e591dac79d6da6). - Evaluator run
scan_20261004T193904Z; research cutoff 31 December 2025; prices through 16 October 2026. Rising/falling splits and payoff ratios are computed from the returned trades. - How SeasonalEdge research is computed · Glossary
Join our newsletter
We share our research, product news and upcoming seasonalities. Free, and you can unsubscribe at any time.