Momentum on India's most liquid stocks beat the Nifty — after we found the bug in our own test
Momentum investing has a simple pitch: buy whatever has climbed the most lately, on the theory it keeps climbing a while longer. A detailed version of that recipe reached us from outside, with impressive numbers attached, and we set out to check it. Our first attempt said the claim was inflated. Our second attempt said the first attempt was wrong.
Here is the finding, stated up front. Run on the kind of data the recipe actually called for, the rule produced a Sharpe ratio of 1.10 — Sharpe measures return earned per unit of volatility endured, and anything above 1 is respectable. The Nifty over the same test scored 0.66. So the tilt is real. Its size, however, landed below the 1.46 that was claimed, and the single largest caveat in the whole exercise was never resolved.
The rule itself fits on an index card. Take roughly 120 of the most liquid names on the Indian market. Each month, measure how much each one returned over the previous three months, but stop the measurement five trading days early, so the freshest week of price action is ignored. Rank them. Hold the top 10 in equal weight until the next rebalance. Charge 25 basis points on every trade, entering and exiting, to stand in for brokerage, taxes and slippage. No shorts, no leverage, no judgement calls.
Our first pass produced a flat rejection. Sharpe came out at 0.60, with the in-sample stretch at 0.31 — worse than the index on a risk-adjusted basis and nowhere near the claim. We wrote it up as overstated. That conclusion survived only until we looked at which prices had gone into it. The series were raw closing prices, unadjusted for dividends and stock splits. A price that drops on an ex-dividend date, or gets chopped overnight by a split, is not describing a return to the shareholder. It injects fake moves into precisely the calculation this strategy depends on.
Re-running the identical rule on the same universe, this time with closes adjusted for dividends and splits, flipped the answer. Sharpe: 1.10, split into 0.77 in-sample and 1.64 out-of-sample — the out-of-sample stretch being the stronger of the two, which is unusual and mildly reassuring, since backtests normally look best on the data used to build them. Total return over the test came to 783%. The worst peak-to-trough drawdown, meaning the deepest fall from a high-water mark before recovery, was 49%. That belongs in the same breath as the Sharpe ratio, not in a footnote.
Variations shifted things less than you might expect. Widening the book from 10 names to 15 nudged Sharpe to 1.17. Bolting on a 200-day moving average filter, intended to sidestep crashes by standing aside when the broad trend breaks, cut the worst drawdown from 49% to 44% — while dragging Sharpe down to 0.96. That is a trade, not an upgrade. The original write-up had billed it as the most valuable improvement available. On our numbers it isn't.
One claim came through unambiguously: the short leg is poison. Selling the weakest names alongside buying the strongest produced a Sharpe of minus 0.10, on both versions of the data. Whatever works here works on the long side only. Plain price momentum also beat the more sophisticated variant that strips out market beta first — the residual version degraded rather than improved, matching what earlier work of ours had already turned up.
Several other claims did not survive contact. The headline Sharpe of 1.46, quoted as 1.57 in-sample and 1.28 out, did not reproduce, and neither did the 243% total return. Much of that gap is plausibly down to universe construction: our liquidity screen yielded 112 usable stocks against their 122. A claimed 24% momentum crash in 2025 never showed up either — the adjusted run had 2025 up 2%. And the fine-grained sensitivity findings, that short lookbacks of 21 days or fewer turn negative and that the five-day skip is essential, read as artefacts of one particular setup rather than properties of the strategy.
Now the limit that actually matters, and the original author flagged it too. Both versions of the universe are survivor sets. The list of liquid names was drawn as it stands today, which quietly excludes every company that was liquid once and then delisted, collapsed, or dropped out of the derivatives segment. Survivorship bias of that sort flatters almost any long-only backtest, because the stocks that went to zero were never in the sample to drag it down. A true point-in-time universe, rebuilt month by month from the members who existed at the time, would pull the number lower. We have not run that test. It is the real remaining threat to this result.
The verdict is a qualified yes with a discount attached. Direction: robust. Magnitude: believable at roughly 1.10 on adjusted, survivor-biased data, and worth marking down before anyone treats that as the true level. Everything above describes what a mechanical rule did over a historical window on one specific list of stocks. It is a measurement, not a plan, and not something on offer.
Arguably the most durable thing to come out of the exercise is the correction to our own method. Momentum computed on unadjusted prices is momentum computed wrong, and the error runs in one direction: it makes a genuine signal look dead. Every earlier study of ours that leaned on that raw data is understated for the same reason, and those results now need revisiting.
More money stories
- A short-volatility test on Nifty returned 11.3% a year, inside a 15% drawdown
- Boring stocks held up better than winners in our re-test of Indian factor books
- Dual momentum tripled the Nifty's risk-adjusted return over 19 years, and halved the drawdown
- Five ways to build a 0.5% intraday trade, and the cost wall that stopped all of them
- Strota backtested its own Darvas box screen: no edge once you pay the fees
- Four attempts to beat intraday trading costs failed -- the barrier is the fill, not the signal