Graveyard / Calendar effects
The turn-of-month effect
Do stocks rise around the turn of the month, and does it still work?
The idea
Stocks are said to earn most of their return in a few days around month end, when salaries and pension money arrive. Hold the market from the close before the last trading day through the third trading day of the next month, and sit in T-bills otherwise.
How we tested it
| Data | Kenneth French daily US market returns (1926 to 2026) as the main series and SPY (1993 to 2026) as a replication. One rule, frozen before the first run: four sessions held each month, T-bills otherwise. |
|---|---|
| Period | July 1926 to September 2026. Split fixed at 1 January 2000, with 2015 onward as a recent window. |
| Costs | 14 bp per monthly round trip, also 21 bp. Applied to the whole history, which is conservative for early decades. |
| Holdout | 2000 onward, pre-registered. Pass needed: t above 2 after costs, positive since 2015, and Sharpe at least equal to buy-and-hold since 2000, on both series. |
| Control | Buy-and-hold is the comparison. No random-entry control was run for this test. |
Result
| Full history, net per turn-of-month day | 9.2 bp (t 6.2), 1,202 months, 61% positive |
|---|---|
| Before 2000, net | 11.7 bp (t 6.84 by month) |
| 2000 onward, net | 2.5 bp (t 0.93 by month) |
| 2015 onward, net | 1.8 bp |
| 2000 onward, Sharpe strategy vs buy-and-hold | 0.14 vs 0.42 |
| SPY 1993 onward, net | 3.4 bp a day (t 1.28 by month) |
Why it failed
The effect is real in history and then decays. After 2000 it is statistically indistinguishable from zero, and the strategy spends most of the year in cash, so it earns less than buy-and-hold with a lower risk. A prior test of the same calendar idea on gold came close, with p 0.062, but did not pass either.
What would make it worth retesting
Not as a standalone strategy. It would need evidence that the post-2000 decay reversed.
Test id: tom. Past results of a backtest say nothing certain about the future.