Double Bottom (W-Pattern) Breakout Backtest: Real Edge, or Just Costs?

The strategy isn't a scam. The confidence is. Asked to “build a profitable strategy,” a popular broker's AI assistant produced a clean, well-formatted W-pattern breakout — entry rules, stop-loss, a 1:5 target, position sizing — called it a “significant edge,” and listed stocks to watch. Every part of that looked professional. None of it was tested.

A chart pattern is a hypothesis, not a result. So we turned its description into exact, codeable rules and ran it the only way that settles the question: on real historical prices, with the costs a real trader actually pays.

The strategy, made testable

The assistant's rules were readable but loose — “a W forms, a red candle breaks, you enter.” A backtest can't run on “a W forms.” Pinning it down (the step the assistant skipped) means deciding exactly what counts: two swing lows within 2% of each other, a neckline at least 3% above them, confirmed with no look-ahead; then the first red candle near the neckline arms the trade. Enter on the break of that candle's high, stop below its low, target five times the risk.

The assistant answered a question it had never asked itself: which W, how equal, how deep? Those choices are the strategy. Skip them and you're not testing a rule — you're trusting a vibe.

How we tested it honestly

We ran it across 97 of the NIFTY 100 — every large-cap with continuous data — from 2018 to mid-2026, feeding the engine 1-minute bars so the entry, stop and target fill minute-by-minute instead of at flattering same-bar prices. Then we charged every round trip the full Indian delivery cost stack: STT on both sides, stamp duty, exchange transaction charges, GST and DP charges — the lines a “significant edge” claim always forgets. The full cost stack is here.

Trades

3,065

Win rate

22.4%

After costs

~1.2%/yr

Costs / gross

43%

A 22.4% win rate sounds like failure, but at a 1:5 payoff you only need to win about 20% of the time to break even. The strategy clears that bar — by 2.8 points. That thin sliver, spread across 3,065 trades and 8½ years, is the entire “edge.” After costs it comes to roughly 1.2% a year on deployed capital — less than you'd earn in a savings account, with none of the 78% losing trades to sit through.

The cherry-picked three

Here is the part that should make anyone pause. The assistant didn't just claim an edge — it handed over a watchlist of specific names (with prices that, on checking, were partly made up). Run the strategy on the three real stocks it named, and it looks decent: +30%over the period. Run it on all 97, and the edge nearly evaporates.

The 3 it pickedAll 97 stocks

Trades1073,065

Win rate27.1%22.4%

Break-even needed19.7%19.6%

Cushion over break-even+7.4 pts+2.8 pts

Costs as % of gross22%43%

Net result+30% total+9.9% (~1.2%/yr)

A hand-picked watchlist flatters any strategy — that's what cherry-picking does. On the full universe, 60 of 97 names were positive and 37 negative, with no single stock driving more than 8% of the total. It's a real, tiny, evenly-spread effect — which is almost worse than a fluke, because it looks real right up until costs finish it.

What the confidence hid

“Significant edge” is two words. Here are the four numbers underneath them that decide whether a real person can actually run this — none of which the assistant mentioned:

78%Trades you lose — you sit through ~4 losers for every winner

2.8 ptsCushion over break-even — 22.4% wins vs a 19.6% break-even

43%Gross profit eaten by costs — STT both sides, stamp, exchange, GST, DP

2–4 weeksTypical winning hold — not the '2–10 day swing' it was sold as

None of this makes the pattern worthless — it makes it ordinary, and honestly so. The cost drag alone is the whole story: strip out the STT, stamp, exchange, GST and DP charges and the strategy looks fine; charge them and 43% of the gross is gone. A confident answer that skips the cost line isn't optimistic. It's incomplete in the one place that matters.

The academics already knew this

We're not the first to test it — we're just the first to test it on your market with your costs. The most-cited academic paper on chart patterns, Foundations of Technical Analysis (Lo, Mamaysky & Wang, 2000), found the double bottom does carry some statistical information — and then warned, in the same paper, that patterns “optimal for detecting statistical anomalies need not be optimal for indicating trading profits.” Later surveys that control for data-snooping and transaction costs (Park & Irwin, 2007) find no reliable edge over simply buying and holding in liquid equities.

That gap — between a pattern that's statistically real and one that's actually profitable — is the entire subject of this note. The literature described it in 2000. An AI assistant erased it in 2026 with one confident adjective. A backtest is what puts it back.

How to reproduce this

Everything below is the config we ran. Nothing here is secret — the point is that it's checkable.

double-bottom-nifty100.yaml

universe:  NIFTY 100 (97 with continuous data)
bars:      4H swing structure, built from 1-min fills
pattern:   double bottom — two swing lows within 2%,
           neckline >= 3% above, confirmed (no look-ahead)
trigger:   red candle near the neckline
entry:     break of the trigger candle's high (+0.5%)
stop:      the trigger candle's low (-0.5%)
target:    5 x risk  (1:5)
sizing:    risk 2% per trade, capped at capital
period:    2018-01-01 -> 2026-06-19   (every signal)
costs:     full equity delivery — STT 0.1% x2, stamp,
           exchange, GST, DP, + slippage (applied on top)

Common questions

Does the double bottom (W-pattern) actually work?

In a 3,065-trade backtest across 97 NIFTY 100 stocks (2018–2026, real 1-minute data, full Indian delivery costs), the double-bottom breakout barely cleared break-even: a 22.4% win rate against a 19.6% break-even, netting about 1.2% per year on deployed capital after costs — less than a fixed deposit. A real but marginal edge, not a 'significant' one.

Why did the pattern look better on a few stocks?

On the three names an AI assistant highlighted, the strategy returned about +30% over the period. Run on all ~100 NIFTY constituents, the edge nearly vanished — a reminder that a hand-picked watchlist flatters any strategy. 60 of 97 names were positive, but the aggregate barely beat break-even, and no single stock drove more than 8% of the total.

What killed the edge?

Costs. The full Indian delivery stack — STT on both sides, stamp duty, exchange charges, GST and DP charges — ate 43% of gross profit, nearly half. And you lose about 78% of trades; the entire return rides on the ~22% that reach the 5R target. Most confident 'this pattern works' claims never count these.

Do academic studies say chart patterns work?

The most-cited study (Lo, Mamaysky & Wang, 2000) found the double bottom carries some statistical information — then warned, in the same paper, that patterns 'optimal for detecting statistical anomalies need not be optimal for indicating trading profits.' Surveys that control for data-snooping and transaction costs (Park & Irwin, 2007) find no reliable edge over buy-and-hold in liquid equities. Our backtest is that warning, measured.

The honest frame

Investment in securities market are subject to market risks. Read all the related documents carefully before investing.

Backtested results are hypothetical, do not represent actual trading, and are not indicative of future results. This article is educational and is not investment advice or a recommendation; Algoshastra is a strategy-building and testing tool, not a registered investment adviser or research analyst. Past or backtested performance does not guarantee future returns.

An AI called this pattern a “significant edge.” We tested it on 3,065 trades.