Connors' Double Seven on NIFTY: underperformer even with a high win rate
In short It wins about two trades in three, but the average trade is too small to tell apart from luck, and since 2018 it has made nothing. Over 17½ years it earned under 2% a year, against about 12.6% for simply holding NIFTY.
The idea
Larry Connors’ Double Seven is one of the best-known short-term mean-reversion rules. In a market that is trending up, buy when it closes at a fresh 7-day low, and sell when it closes at a fresh 7-day high. The idea is that short pullbacks inside a long-term uptrend tend to snap back. The rule was popularised on US index ETFs. We wanted to see whether it carries over to NIFTY.
The rules
- Instrument: NIFTY 50 index, daily bars. Long only.
- Trend filter (classic version): only enter when the close is above the 200-day simple moving average.
- Entry: today’s close is the lowest close of the last 7 days.
- Exit: today’s close is the highest close of the last 7 days. No stop, no target.
- Fills: tested two ways: at the signal close (the classic market-on-close assumption) and at the next day’s open (the realistic version).
- Cost: 5 bps per side.
We also ran it without the 200-day filter.
The backtest
NIFTY daily data, trading from 1 January 2009 to 3 July 2026 (17½ years), 5 bps per side. Drawdowns are measured on the daily value of the position, not just at trade exits.
| Variant | Trades | In the market | Win % | Avg / trade | t | CAGR | Worst drawdown | Calmar |
|---|---|---|---|---|---|---|---|---|
| Close fill + 200-DMA (classic) | 176 | 37% | 66% | +0.21% | 0.93 | 1.6% | −44.0% | 0.04 |
| Next open + 200-DMA | 173 | 37% | 70% | +0.25% | 1.07 | 2.0% | −41.3% | 0.05 |
| Close fill, no 200-DMA | 229 | 48% | 66% | +0.30% | 1.52 | 3.3% | −37.2% | 0.09 |
| Next open, no 200-DMA | 226 | 49% | 70% | +0.42% | 2.08 | 4.9% | −37.6% | 0.13 |
| Buy & hold NIFTY | — | 100% | — | — | — | 12.6% | −38.4% | 0.33 |
(t measures whether the average trade is distinguishable from zero; around 2 or more is needed before we call a result more than luck. “In the market” is the share of trading days with a position open. Calmar is CAGR divided by the worst drawdown.)
A high win rate, but no measurable edge. Two trades in three are winners, with a typical (median) gain of about 0.7%. But the losers are bigger, because there is no stop, and the average trade is only +0.2% with a t-statistic below 1. Only one variant, next-day-open fills without the 200-day filter, just clears 2 (t 2.08).
Too few signals. The classic version trades about 10 times a year and has money in the market only 37% of the time. The rest of the time the capital sits idle, which is why even a positive average trade adds up to so little.
It lost to holding the index on every measure: about 1.6–4.9% a year against 12.6%, and a worst drawdown as deep or deeper (−37% to −44% against −38%), despite being out of the market most of the time. The deep drawdown comes from holding through sharp falls with no stop, as in March 2020.
Robustness tests (classic version)
| Test | CAGR | Worst drawdown |
|---|---|---|
| As tested, 5 bps per side | 1.6% | −44.0% |
| Double cost, 10 bps per side | 0.6% | −46.5% |
| Triple cost, 15 bps per side | −0.4% | −48.9% |
| First half, Jan 2009 – Dec 2017 | 4.1% | −16.8% |
| Second half, Jan 2018 – Jul 2026 | −0.9% | −42.0% |
- It stopped working. Whatever it earned came before 2018; over the last eight and a half years it lost slightly.
- Costs decide it. Doubling a very modest cost assumption takes away most of the return; tripling it turns the strategy negative.
- Profitable in only 52% of quarters (57% of the quarters in which it held a position). The worst quarter was −26%.
- Fragile to a few periods. Its single best quarter accounts for about 36% of all the growth, and removing its five best trades cuts the 17½-year total gain from +33% to +4%.
Risks
- No stop. Positions are held until a 7-day-high close, however far price falls first. That is how a strategy that is out of the market most of the time still suffered a −44% drawdown.
- Drawdowns can hide between trades. Measured only at trade exits, the worst drawdown looks like −35%; marked to market every day, it was −44%. See Landmines for more on how measurement choices flatter results.
- Filter transfer. The 200-day filter was designed on US ETFs; on NIFTY it reduced returns with both fill methods (1.6% against 3.3% with close fills, 2.0% against 4.9% with next-day-open fills).
- Opportunity cost. With money in the market only about a third of the time, long periods in cash during a rising market are the main drag on returns.
Pros & cons
Pros
- Very simple, fully mechanical rules.
- A high win rate, with a typical winning bounce of under 1%.
Cons
- The average trade is not statistically distinguishable from zero.
- Few signals: about 10 trades a year, capital idle about two-thirds of the time.
- Nothing earned since 2018.
- About 1.6–4.9% a year against 12.6% for buy and hold, with a similar or deeper worst drawdown.
- Very sensitive to costs.
Our verdict
Double Seven is a good example of how a pleasing win rate can disguise the absence of an edge. On 17½ years of NIFTY data, net of modest costs, it traded only about 10 times a year, its average trade could not be told apart from luck, it made nothing in the second half of the period, and it lost to simply holding the index by about 11% a year, with a deeper worst drawdown. It failed seven of our eight checks.
Was this useful?
More to say, or spotted a mistake? Write to [email protected]. We read everything.
Hypothetical, backtested results on historical data at least three months old. They include modelled costs but can't capture every real-world effect, and past performance does not predict future results. This is educational research, not a recommendation. Full disclaimer.