Rebalance Timing Luck
Two versions of the same strategy that differ only in which day of the month they rebalance on can show meaningfully different Sharpe ratios, purely from where the calendar happened to land relative to market moves. That dispersion is luck, not skill, and it's easy to mistake for one.
Prerequisites: Turnover and Rebalancing
A monthly-rebalanced strategy has to pick a day: first trading day of the month, fifteenth, last day, whatever's convenient. That choice feels arbitrary and mostly is — but "arbitrary" doesn't mean "harmless." Two implementations of the exact same signal and the exact same rules, differing only in which day of the month they trade, hold different positions at different times relative to whatever the market happened to do that month. Over a long enough sample the dispersion this creates is real and can be as large as differences researchers spend months trying to explain with skill-based stories.
Worked example: the same strategy, two rebalance dates, two different Sharpes
A cross-sectional momentum strategy is rebalanced monthly. Backtested two ways, identical signal and universe, differing only in the calendar day of the trade:
| Rebalance date | Annualized return | Volatility | Sharpe |
|---|---|---|---|
| 1st trading day of month | 11.2% | 8.0% | 1.40 |
| 15th of month | 9.8% | 8.1% | 1.21 |
| Last trading day of month | 7.9% | 8.7% | 0.91 |
Same signal, same universe, same holding period length — the only difference is which day inside the month the portfolio turns over. The gap between the best and worst of these (Sharpe 1.40 vs 0.91) is large enough to look like a meaningfully better strategy design, when it is entirely a byproduct of which day-of-month happened to catch favorable versus unfavorable intra-month drift in this particular sample. Average across many possible rebalance dates and the differences mostly wash out; picking the single best one after the fact and reporting only that number is a subtle form of overfitting to the calendar.
When a strategy involves a periodic rebalance, test it across several different rebalance dates (or, better, average results across many offsets) rather than reporting the number from one arbitrary calendar choice — the spread across dates tells you how much of the headline Sharpe is timing luck.
This is easiest to miss precisely because the rebalance date feels like an implementation detail rather than a research decision. If a strategy's reported Sharpe was chosen from among several rebalance-date variants that were tried, it is a form of backtest overfitting even if no one explicitly went looking for the best one.
A practical robustness check: run the same strategy on every offset within the rebalance period (all 20-odd trading days of a month, for a monthly strategy), and report the distribution of Sharpes, not just the one your production system happens to use.
Related concepts
Practice in interviews
Further reading
- AQR, Rebalance Timing Luck: The Difference between Hired and Fired