Blocked dates are hindsight
Leaving out the 'event days' after you have seen which days hurt makes any backtest look safer. A fair blocklist uses only dates that were announced in advance, and even then it can get lucky.
The trap
A backtest has a few terrible weeks. Looking at them, they line up with events: an election result, a budget, a crash. “Nobody would trade through that,” so those dates go on a blocklist and the backtest is run again. It now looks much safer. But the list was written by someone who already knew which dates would hurt.
How it fools you
- The worst days choose themselves. A list built after seeing the results removes exactly the losses that made the strategy look risky. Any strategy improves that way, including a bad one.
- It sounds like risk management. Avoiding event days is a sensible idea, so the hindsight hides behind a good reason.
- The next big day won’t be on the list. The worst moves often come from news no calendar shows. Blocking them in the past says nothing about the future.
The honest version blocks only dates that were announced in advance, applies the rule to every year, and accepts whatever result follows.
A real example
Our published 5-delta weekly NIFTY strangle: 277 weekly trades from April 2021 to August 2026, with the same costs as that page. Its worst trade was the week of the 7 April 2025 gap, and its third worst was election-results week in June 2024.
With hindsight. Two versions a reader might be tempted to show:
| Version | Weeks traded | Total | Per year | t | Worst trade | Worst drawdown | Return ÷ drawdown |
|---|---|---|---|---|---|---|---|
| As published | 277 | +917 pts | +172 | 1.49 | −368 | −572 | 0.30 |
| Skip the weeks of 4 Jun 2024 and 7 Apr 2025 | 275 | +1,427 pts | +267 | 3.06 | −157 | −205 | 1.31 |
| Skip the 5 worst weeks | 272 | +1,791 pts | +335 | 4.39 | −98 | −131 | 2.56 |
Points per lot, after costs. Drawdowns measured at trade close.
Leaving out two weeks out of 277 doubles the t-statistic, cuts the worst drawdown by almost two-thirds and turns a strategy that failed our bar into one that passes it easily. Nothing about the rules changed.
Fair: dates known in advance. A week is skipped if a scheduled event falls between entry and expiry: the general election counting day (announced in March 2024), Union Budget days, and RBI policy decisions.
| Blocklist | Weeks skipped | Total | Per year | t | Worst trade | Worst drawdown | Return ÷ drawdown |
|---|---|---|---|---|---|---|---|
| Election + Budget days | 7 | +908 pts | +170 | 1.53 | −368 | −599 | 0.28 |
| Election + Budget + RBI days | 43 | +1,141 pts | +214 | 2.70 | −157 | −188 | 1.14 |
| Same, but had the April 2025 gap come in a non-RBI week | 42 | +773 pts | +145 | 1.37 | −368 | −465 | 0.31 |
- Election and Budget days changed almost nothing. The fair rule did catch election-results week, a real loss. But six of the seven skipped weeks were winners: premiums were richer in event weeks (about 30 points against 18), and the strategy was being paid for that risk.
- The RBI rule looks like a big improvement, but by luck. The RBI policy decision of 9 April 2025 happened to fall on that week’s expiry day, so the rule skipped the week of the 7 April gap. The gap itself came from overseas tariff news over a weekend, nothing to do with the RBI. The other 35 RBI weeks made +135 points: skipping them cost money.
- Without that coincidence, the fair version is a little worse than doing nothing (t 1.37 against 1.49).
Skipping 43 weeks at random, 10,000 times, gave a median t of 1.29; a t of 2.70 or more came up 5% of the time, and the worst trade was removed in 16% of the draws. A rule that removes the worst trade by chance looks like a rule that works.
What no calendar could foresee. The 7 April 2025 gap and the November 2024 Monday gap (the second-worst trade) were not on anyone’s calendar. A fair backtest keeps them in, because the live strategy would have been holding them.
How to spot it
- The excluded dates are the same days as the worst trades.
- The list includes days that were not scheduled in advance: a crash, a surprise announcement, a weekend gap.
- The improvement comes from skipping a handful of trades, and the skipped trades are mostly losers.
- Nobody reports what the skipped weeks would have made, or how often a random skip of the same size does as well.
How we guard against it
Our strategies trade every eligible day unless the rule names, in advance, a calendar of scheduled dates it avoids. When a rule does skip dates, we show what the skipped trades made and compare it with random skips of the same number. See Methodology and Look-ahead bias.
Was this useful?
More to say, or spotted a mistake? Write to [email protected]. We read everything.