Four attempts to beat intraday trading costs failed -- the barrier is the fill, not the signal
Retail traders in India do most of their business inside a single session, and that is where the arithmetic is hardest. Strota's own testing has priced it. Charges on a round trip in intraday equity come to 0.0824% of the position, and entering on a stop -- an order triggered as price crosses a level -- costs roughly 0.34% more in slippage. The intraday effects worth hunting measure 2 to 16 basis points, a basis point being one hundredth of a percent. That shortfall is a factor of five, not a tuning gap.
Four studies went at the barrier from four angles: a cheaper instrument, a better entry moment, an unrelated family of signals, and stacking filters for conviction. All four failed, and where they failed is the point. Three constructions built on independent signals converged on one measurement -- the price you actually get on entry -- and no signal work shifted it.
Instruments first. Index futures genuinely are cheaper than intraday stock: brokerage, taxes, exchange fees and stamp duty over a roughly Rs20 lakh index-future position come to 2.91bp for a round trip, against 8.24bp in equity, or about 2.8x less per rupee. Holding a stock portfolio for delivery while shorting an index future against it looked outstanding over the full history: 33.4% a year, a Sharpe ratio of 2.14 (return per unit of volatility) and a worst drawdown of 31.0%, against 12.5%, 0.80 and 38.4% for the index alone.
Splitting by period ended the argument. The hedge added 18.4pp of annual return in the earliest stretch and 11.0pp in the middle, then 0.03pp from mid-2023 on, measured with no trading friction at all. Break-even execution cost in that recent window is 0.00-0.10bp a side, which is another way of saying nothing is left over. The short leg alone made 6.25bp a day gross, but 4.86bp in training against -0.20bp out of sample, eroding 0.84bp a day each calendar year. The drift could shrink 53.5% before break-even. It has already shrunk 57%.
Attack two sounds like prudence: if buying into a violent open is expensive, wait for a calmer price. Across 352 stocks and 10,212 gap days it is simply wrong. A stock that opened 5% or more below its prior close returned 2.393% on average when bought in the opening bar, 0.560% five minutes later, and 0.038% if the entry waited until 10:45. Waiting for price to return to the opening level selects for failure: on the 511 days it came back the gross result was 0.133%, against 3.754% on the 848 days it never did.
A by-product of that study matters more than the study. Over this history the gap book carried a break-even slippage budget of 1.155% a side against realistic execution of 0.10% to 0.25%, clearing the cost barrier by around 5x and positive in every year of every variant. Nothing subtle explains it. The entry lands in the opening call auction, where orders are matched at a single price, rather than chasing a stop through a level. Its limit sits elsewhere: only a handful of stocks gap 5% on a given day, so capacity binds, not cost.
The third attack changed signal families, testing two ideas popular in trading courses -- price pushing through an obvious earlier high or low and then turning, and a gap left by a fast move that the market later trades back into. Five-minute bars, 352 stocks, 1,240 sessions, benchmarked against every other stock entered at the same clock time on the same date. An early pass looked good in both directions at once, 0.62% long and 0.77% short, which is the tell for a broken benchmark. Repaired, the sweep is dead across all 18 configurations before a rupee of cost.
Buried in that failure is the fill penalty again, on signals sharing nothing with the earlier ones. Filled at the breached level, two configurations show 0.091% and 0.199%. Filled at the next bar's open -- the first price genuinely available -- the same two show -0.052% and -0.069%. The difference, 0.143% and 0.269%, widens as the required breach grows, because a bigger sweep implies a bigger snap-back. That is the whole explanation for why this genre of backtest impresses and the trading does not.
One real effect did emerge, and it inverts the folklore: entering as price retraced into an earlier imbalance and continuing with the original move earned 0.110% of excess return over 755,146 trades, steady at 0.111% in training and 0.108% out of sample. It still does not pay -- 0.051% a trade at zero slippage, -0.049% at 5bp a side.
Stacking filters was the fourth attack, and the instinct behind it is retail chart analysis at its most common -- demand more confirmation. Six conditions, all 63 combinations, 12.7M candidate bars. Median results climb as conditions accumulate, which looks like conviction working and is an artifact: a stack inherits its strongest member instead of compounding them. The best achievable result flattens near 0.12% from the second condition onward, then declines, while frequency drops 12x. Out of sample, not one of the 63 cleared the fee.
Rerunning that test on a survivorship-free universe of 40 names, using only what was knowable at each date, produced the sharpest lesson. There the hypothesis appears confirmed: median excess return climbs to 0.24% as conditions accumulate, and four combinations survive a Bonferroni correction -- the standard penalty for testing many ideas at once -- while staying profitable after 5bp of slippage, the best printing 0.239%. On held-out data that combination returned 0.0197%. The statistical corrections passed the curve; only the holdout caught it.
Nothing measured here is deployable, and the limits deserve stating plainly. An earlier Strota result was corrected along the way: re-measured on 330 liquid names, the overnight share of returns rises steeply as stocks get less liquid -- 55.4% a year in the most traded tier, 97.3% in the least -- while the intraday share sits near -25% throughout. That shape is a bid-ask artifact, not a tradeable drift. It also reaches a portfolio whose 22.7% return against the index's 12.5% now reads as survivorship -- a sample holding only companies that lasted -- plus a small-company tilt, not an edge. What survives the four failures is a budget: an intraday design either avoids stop and level fills or pays 0.14% to 0.34% a trade, and needs 0.6% to 0.8% gross in testing for any chance of clearing 0.0824% live.
More money stories
- A rare positive in our backtests: deep gap-down buying cleared the cost wall, at retail scale only
- A Hormuz deal cooled oil, the AI trade stumbled, and Asia's markets split in two
- A 274-point open in New York, a 544-point climb in Mumbai: the morning two markets turned green at once
- Foreign funds kept selling India. The people running the companies kept buying.
- Oil fell 5% because an attack did not happen, and that says a lot about what was in the price
- A 12.3 billion won crypto scam, a group in custody, and a news line that leaves the rest blank