Same rule, three ETFs: how one extra year reorders a November comparison

Three funds, one rule, each on its own history

The November rule is the same for every fund: buy at the open of November's first session, sell at the open of December's first. Each fund's figures come from the evaluator on that fund's full price history, and the histories start in different years.

Read naively, QQQ's November did more than SPY's, by +0.15 percentage points a year. But the three rows do not cover the same years.

The samples are not the same length

SPY and QQQ have a November 1999 trade; IWM's evaluated history starts with November 2000. A comparison of the three averages therefore compares SPY and QQQ over one more year than IWM. When the extra year is ordinary, that hardly matters. When it is extreme, it can decide the order.

November 1999 was that kind of year for QQQ: the trade returned +13.00% against a hold of +0.98%, an excess of +12.02 pp. SPY's November 1999 excess was +1.29 pp.

Put every fund on the same years

The fix is to keep only the years every fund has: 2000 to 2025, 26 Novembers each. For IWM nothing changes. SPY loses one year and moves from +1.73 pp to +1.74 pp. QQQ loses one year and moves from +1.87 pp to +1.48 pp.

What matching leaves alone

Matching changes which years are counted. It does not change how any single year was measured. Each trade is still compared with the benchmark the evaluator gave it: the average return of an ordinary hold of the same fund, for the same number of sessions, across that fund's own price history before the research cutoff. A November of 19 sessions has one benchmark, a November of 21 another, and each fund has its own three.

It would be tempting to go further and re-estimate every benchmark over the shared years only. That is a different calculation, which the evaluator has not made; these pages do not invent it. The comparison above changes the sample and nothing else.

What matching cannot fix

Three funds on the same years are still three different funds. QQQ moves more than SPY in most months, so a larger November excess can partly reflect larger moves in general; excess over a fund's own hold is not a risk-adjusted figure. The three are also not independent evidence: in 23 of the 26 shared years all three rose in November or all three fell, so agreement between them is largely one market's November seen three times, not three confirmations.

Before you rank two backtests

• Same years? Compare first and last years, and count the observations.
• Same rule? Same entry, exit and price convention on every instrument.
• Each against its own benchmark? Compare excess over each instrument's own hold, not raw returns.
• Does one year decide it? Drop the extra years, then the best year, and see whether the order holds.
• Is the difference larger than the year-to-year spread? A gap of a fraction of a point next to swings of several points is not a ranking.

The November research shows every year behind these figures. The mean and median guide covers what one extreme year does to an average.