Double Tops and Double Bottoms: Where the Definition Breaks

9 min readQuantParadox research

A double top is obvious once it has completed and genuinely difficult to identify while it is forming. Almost everything interesting about the pattern lives in that gap.

The short answer

A double top is two peaks at a similar price separated by a trough, confirmed only when price closes below that trough, and the pattern's apparent reliability collapses once the height tolerance and confirmation rule are made explicit.

What defines a double top?

A double top is two peaks at approximately the same price with a trough between them, completed when price closes below the trough — and the word 'approximately' is where every detector differs.

How close is close enough? Half a per cent apart? Within a quarter of an ATR? Exactly equal to the tick? A tolerance of two per cent on a volatile instrument admits pairs of peaks that no trader would call a double top, while a tolerance of a tenth of a per cent finds almost nothing.

The trough matters as much as the peaks. A shallow dip between two highs is not the same structure as a deep retracement, and without a minimum depth requirement the detector will match every minor wobble at a resistance area.

The third choice is time. Peaks a week apart and peaks two bars apart are both technically two peaks. A maximum and minimum separation is what stops the definition from matching things nobody would trade.

What did we find when we audited a loose detector?

We audited a double-top detector in our own tooling and found it was firing on roughly a third of all bars, which meant it had stopped being a pattern detector and become a description of ordinary price behaviour.

That is a finding about detector design rather than about markets, and it is worth publishing precisely because it is unglamorous. Any statistic that detector produced would have looked like a statement about double tops and would actually have been a statement about how often price makes two similar highs, which is very often.

The lesson generalises to every pattern rule. Before you measure what happens after a pattern, measure how often the pattern occurs. A detector firing on a large fraction of bars cannot possibly be selecting a rare structure, and no amount of downstream statistical care fixes an input that is not selective.

It is also a warning about tooling you did not write. If a platform marks patterns on your chart and does not document its thresholds or its firing rate, you cannot know whether you are looking at a pattern or at noise with a label.

When is a double top actually confirmed?

The pattern is conventionally confirmed only when price closes below the trough between the two peaks, and treating the second peak as the signal is the most common way traders get the timing wrong.

The distinction matters because those two moments can be a long way apart in both time and price. Entering at the second peak means acting on an incomplete pattern that frequently resolves upward instead; entering at the neckline break means accepting a worse price in exchange for the pattern having actually happened.

Both are testable and they are different strategies. The early entry has better reward-to-risk arithmetic and a much lower hit rate; the confirmed entry has the reverse. Which is better on your instrument is an empirical question, and the answer will not be the same on every timeframe.

What is not defensible is measuring the confirmed version's hit rate and then trading the early version's entries. That combination — the reliability of one and the pricing of the other — is how backtests end up describing a strategy nobody actually ran.

How should you test a double top strategy?

Test it with a detector whose firing rate you have checked, a three-way grade including expired patterns, and a comparison against what happens after arbitrary pairs of similar highs.

The arbitrary-pairs comparison is the null this pattern most needs. Two similar highs with a dip between them is a common configuration; if the detector's outcomes match what follows any such configuration, the pattern label was decoration.

Report the count. Strict definitions on higher timeframes yield few instances over a decade, and a promising rate on sixty occurrences is not something to size positions from.

Finally, resist the urge to loosen the tolerance when the count comes back low. That instinct is understandable and it is exactly the mechanism that turns a rare pattern into a detector that fires on a third of the chart.

Questions people actually ask

How similar do the two peaks have to be?

There is no standard tolerance, which is why it must be stated explicitly and expressed relative to volatility rather than in absolute points. A quarter of an average true range is a defensible starting point on many instruments. What matters far more than the specific value is that it is fixed before results are examined and not widened afterwards to increase the sample.

Is a double bottom just a double top upside down?

Geometrically yes, and most detectors implement it as the mirror image. Whether the two behave identically is an empirical question worth checking separately, since many markets have asymmetric behaviour between advances and declines. Testing both directions independently costs almost nothing and occasionally reveals that the pooled result was hiding a difference.

What target should a double top trade use?

The conventional projection is the height of the pattern measured from the peaks to the trough, extended downward from the neckline break. That convention is a testable parameter rather than a rule, and it is worth measuring how often the projected level is reached before the trade is invalidated, reporting the patterns that expired without reaching either.

The only backtest that settles it is yours.

Build a strategy from a sentence, paste your own Python, or import your live trade history and have it graded. Five full backtests free, no card, and we'll tell you plainly when the result is indistinguishable from luck.

We publish research and tooling, not trading advice, and we make no claim about future returns. Everything above describes how to test an idea — not a reason to trade one.