Why a hit share is not profit

Across all seven models the site counted 49 occurrences: 17 hits, 27 misses and 5 undecided. The table of this page puts each model's share of hits beside its average result in ATR, and the two do not always agree.

Two columns that do not move together

Compare the models in the table. Bullish engulfing hit 4 of 8 times (50.0 percent) with an average result of +0.57 ATR. Double bottom hit 3 of 7 (42.9 percent) with an average of +0.63 ATR: a lower share, yet a higher average. Bearish engulfing hit only 36.8 percent and still averaged +0.25 ATR, while head and shoulders hit 16.7 percent and averaged -0.97.

The share counts only whether the 2 ATR target came before the 1 ATR cancel. The average result measures how far, in either direction, the deciding close was. A model can hit rarely and lose small, or hit often and lose big. The share alone cannot tell which.

The costliest miss in the count

The chart under the heading is the worst miss among the occurrences counted: a bearish engulfing pair on Bharti Airtel, completed on 2 September 2024 at a close of 1,571.35 with an ATR of 30.14. It was decided on 12 September by a close 2.50 ATR above C. The definition calls any close 1 ATR or more against a miss, so the first close past the cancel level counted, but that close was 2.5 ATR away, not 1.

The overshoot is the lesson. A definition with a cancel level of 1 ATR suggests that a loser costs one ATR. The count shows that the close that triggers the cancel can sit far beyond it, and in a real position the loss would be set by where the position could be closed, not by the definition.

What the count leaves out

The count prices nothing. No spread, commission, financing charge or slippage is deducted, and none is modelled, so every average result in the table is before costs. Its unit is the ATR of the completing session, so results from a quiet stock and a volatile one are comparable in ATR but not in rupees. And the yardstick matters: across the 49 counted occurrences, about 15.6 hits would be expected if each had started from a random session in its own direction, against 17 actually found.

That is close. Taken together, the written rules produced hits at about the rate that starting on any day would have produced, and an average result of about +0.06 ATR per occurrence, before costs.

Reading the table without promising anything

The table's average column says how the deciding closes sat relative to the completing close, in ATR, over a handful of occurrences per model. It is not a return, it is not repeatable by construction, and it was measured on ten years of 11 instruments. A model with two occurrences has an average of two numbers.

No figure on this site is a forecast and none is advice. The page exists to show that the number most often quoted for a pattern, how often it works, leaves out the numbers that decide whether anyone would have made money.

A row-by-row reading of the table

Take the seven rows in order. Head and shoulders: 6 counted, 1 hit (16.7 percent), average -0.97 ATR, against a yardstick of 27.8 percent down. Double top: 3 counted, no hit, average -1.38. Double bottom: 7 counted, 3 hits (42.9 percent), average +0.63, against 38.0 percent up. Ascending triangle: 4 counted, 2 hits (50.0 percent), average +0.50. Descending triangle: 2 counted, no hit, average -1.38. Bullish engulfing: 8 counted, 4 hits (50.0 percent), average +0.57. Bearish engulfing: 19 counted, 7 hits (36.8 percent), average +0.25.

Two things stand out when the rows are read together. First, every model whose share is above its yardstick also has a positive average, and every model with no or one hit has a negative one, so on this sample share and average moved the same way. Second, the samples are so small that the match is nearly guaranteed: with two or three occurrences, a share of zero forces a negative average. The apparent agreement is therefore not evidence that share predicts profit; it is what tiny samples do.

Why overlapping occurrences overstate the evidence

The 49 counted occurrences were found on 11 instruments over the same ten years. Large Indian stocks move with the same market, so several of them can complete the same kind of figure in the same weeks, and the results of those occurrences are not independent: they largely come from the same market weather. Forty-nine occurrences may therefore carry the weight of fewer separate cases.

The bearish engulfing pair shows it best. It has the largest count, 19, and several of its occurrences cluster on the same dates across stocks. A reader who treats 19 as 19 independent trials will overstate how much it proves. The table prints the count so the reader can see the size of the sample, and this page adds the warning that the size is an upper bound on the evidence, not a measure of it.

What a profit calculation would need

To say whether a pattern made money, a measurement would need everything this one leaves out: an entry rule and an exit rule that can be run without hindsight, the size of each position, spreads and commission on both sides, financing for any position held overnight, and the tax and currency effects of the account. It would need a large number of independent occurrences and a record of how the rule did on data it had not been tuned on.

This site does none of that and says so. The count of hits is a narrower thing: how often, from a completing close, a close 2 ATR away in the textbook direction came before a close 1 ATR away against it. It is a useful description of how hard the textbook claim is to meet. It is not, and is not offered as, a statement about profit.

A reader who wants to go further can reproduce the count: the rules are quoted in full on each model page, the definition of a hit is in numbers on the first page, and the history is a public series from a named source with the date it was read. Nothing here depends on a private judgement. That is deliberate, because a measurement that cannot be repeated by someone else is only an assertion, and a hit rate is exactly the kind of figure that tends to be asserted. Where a count is small the page says small, where an average is before costs it says before costs, and where the yardstick is close to the share it says the figure shows nothing beyond what any starting day would have shown.