Buying weakness worked, buying strength lost: results from a broad signal sweep
One rule explained almost every result in a broad sweep of trading signals on Indian derivatives-eligible stocks: entries taken into weakness made money, and entries taken into strength did not. Dips, gap-downs, oversold readings and liquidity sweeps carried a positive edge after costs. Breakouts, momentum crossovers and trend-continuation rules lost. That split held across a dozen strategies built on unrelated logic, which is the kind of consistency that points at something structural rather than a quirk of one screen.
The strongest version was blunt. Fading a gap of five percent or more at the open and closing out the same day returned about 1.11% per trade after fees, winning 64% of the time, with 62% of days finishing green. Taking the opposite side — buying the bounce after a gap-down of that size — did better per trade, at roughly 1.28% with a 73% hit rate. Both are descriptions of a finished test window, not of anything on offer.
Here is how it was tested. Every run used the derivatives-eligible universe as it actually stood at each point in time, not today's membership list — a distinction that matters, because testing on today's survivors quietly hands a strategy the winners and erases the names that got dropped along the way. Daily bars, fees deducted, and every result measured against simply holding the same stocks, which returned a median of about 6.2% over the two-year window. Day-consistency was tracked as well, so no strategy could pass on the back of a few explosive sessions.
The losing half is the more useful one. Block-chart trend entries — a method that ignores time and prints a new block only when price moves a fixed amount — lost at every block size tested. Momentum-crossover entries lost, winning 34% of trades. A trend-cycle oscillator was worst of all, shedding 0.41% per trade. A squeeze-momentum rule came out roughly flat and still trailed buy-and-hold, which is its own verdict: breaking even while the market drifts upward is a cost, not a result.
A structure-based family popular with retail traders — the 'smart money' vocabulary of breaks of structure, change of character, fair value gaps and liquidity sweeps — split along exactly the same line. The one variant that buys a down-wick after sellers are flushed came in positive, if thinly: 0.15% per trade, a 49% win rate. Every variant that buys an upside break lost, including the one marketed as a reversal signal. Its name says reversal; its trigger is a break upward. It lost like the rest.
Why do gaps carry an edge when the indicators don't? Earlier work in the same sweep found the overnight move is already fully priced into the opening print — the correlation between the overnight signal and what followed afterwards was 0.05, which is nothing. The edge is therefore not information. It looks mechanical: the opening auction overshoots, then partly unwinds during the session. A gap is a sharp, discrete dislocation. An oscillator cross or a price block is a smoothed restatement of drift, with no dislocation in it to fade.
None of that would matter without the next test, and this is where most published backtests quietly stop. Every winning figure above assumes a fill at the exact opening price — the most contested moment of the day, and worst precisely on the violent gaps that looked best. So both legs of every trade were degraded by a fixed cost per side, ramped from zero up to half a percent, to find where each strategy stops paying.
Deep gaps held up; shallow ones did not. Fading gaps of three percent or more broke even at roughly 0.25% of slippage per side and went negative past that, across 971 trades. The five-percent fade needed around 0.55% per side before it stopped paying, over 222 trades. The gap-down bounce at that same threshold survived to about 0.6%, on 100 trades, still winning 68% of the time and returning ₹79 per trade at a deliberately pessimistic 0.2% per side. Typical slippage on a liquid derivatives-eligible open runs 0.05% to 0.15%.
Now the limits, and they are substantial. The sample spans one to two years of a tape that was mean-reverting with an upward drift. That regime is the entire reason weakness entries worked; a sustained trending or falling market could invert the finding and reward the very momentum tools that failed here. The deep-gap buckets are small — roughly 100 to 230 trades — and they are news-driven, so they are rare.
Clustering is a genuine problem, not a footnote. Gaps arrive together, which means a day's positions are one bet wearing many names. The shallower fade shows this plainly: it won 61% of individual trades while only 44% of days finished green. That divergence between trade-level and day-level results is the signature of a correlated book, and it is what turns a decent-looking average into a lumpy equity curve.
Two further limits deserve naming. The entry is the open and the exit the close, which is a proxy for a real fill rather than a real fill. And the structure, oscillator and block-chart strategies were mechanized reconstructions of methods discretionary traders apply by eye, several of them intraday tools tested on daily bars. A reconstruction that loses is evidence about the reconstruction first. What survived the whole sweep is far narrower than where it began: one threshold, one direction, one holding period.
More money stories
- An 80% win rate collapsed to 31% the first time we tested it forward
- Crypto's momentum formula loses money on Indian stocks. A six-month version doesn't.
- Two of eight sessions made the entire profit — and the short side lost money
- Low-volatility stocks, not momentum, produced a book positive in every completed year
- We tried to make a momentum strategy win every year. The momentum had to go.
- Fading morning gap-ups paid in liquid stocks and lost in illiquid ones