QTSurfer beta
Learn

Equity curve

Learn what an equity curve records during a backtest, how to normalise and read it, and which shapes signal a fragile strategy rather than a robust one.

An equity curve is the account value of a strategy plotted through time. In a backtest it starts at the initial capital and moves as simulated trades realise profits, losses, and costs. It is the most information-dense artefact a backtest produces: every summary metric, from total return to maximum drawdown, is a function of this one series.

A final return compresses the whole experiment into a number. The curve keeps the path, and the path is where most of the useful questions live.

What the curve records

Each point is a timestamp and an equity value. Equity is the account balance in quote currency, initial capital + cumulative net profit, not a percentage and not a price.

How often points are recorded matters. A curve sampled on every trade shows the realised results at the moments the strategy closed positions, but it says nothing about unrealised swings between those moments. A curve marked to market on every tick shows the full path, including open-position risk, at a much larger size. Know which one you are looking at before reading it.

Normalising for comparison

Two curves with different starting capital cannot be compared by raw value. Normalise each point to its starting equity:

return%(t) = (E(t) / E(0) − 1) · 100

An account that starts at 100 and moves to 110.5 then 90.25 reads +10.5 % and −9.75 % on the normalised scale, whatever the currency amounts were.

For long or strongly compounding curves, a logarithmic vertical axis keeps equal percentage moves the same visual size. On a linear axis, the same 10 % loss looks small early and enormous late.

Reading the shape

  • Slope and consistency. A steady slope with small fluctuations is what a repeatable edge looks like. A flat line with a few vertical jumps is a strategy whose return depends on a handful of trades.
  • Drawdowns. Every dip below a previous high is a drawdown. Their depth, frequency, and how long they take to recover are the strategy’s risk profile in graphical form.
  • Flat periods. Long stretches without movement mean the strategy was not trading. That may be intended, or it may mean the entry condition was never met in a regime the strategy was not built for.
  • Stair steps. Regular steps indicate infrequent trading at a fixed size. They are not a problem by themselves, but they warn that the trade count, not the date range, sets the sample size.
  • Regime dependence. A curve that rises only during one identifiable period and is flat or falling elsewhere has likely fitted that period rather than the market.
  • Late acceleration. Strong gains concentrated at the very end of the window are the shape most often produced by luck or by a parameter tuned to recent data.

Common mistakes

  • Judging by the endpoint. The final value is one point. Two curves can share it while one is a straight line and the other a cliff followed by a recovery.
  • Reading a downsampled curve as the raw one. A curve reduced for display keeps its overall shape, but individual peaks and troughs between preserved points may be smoothed away.
  • Comparing curves from different fee assumptions. Costs act on every trade, so two runs that differ only in fees can have visibly different curves. Keep configuration with the chart.
  • Mistaking a smooth in-sample curve for robustness. A curve can be made arbitrarily smooth on historical data by adding rules. Smoothness is only evidence when it appears on data the strategy was not tuned on.

Equity curves in QTSurfer

A completed backtest returns its equity curve with the yield metrics once the strategy has emitted at least one trade. The first point is an anchor at the backtest start with the initial capital; every later point is one sample per emitted yield, so the curve records realised results at those moments. Equity is the account value in quote currency, and the normalisation above converts it to percentage return.

The curve passes through a fixed transform pipeline, resample → differential → outMode:

  • resample limits the result to a chosen number of points while preserving the exact first and last points and the global extrema.
  • differential delta-encodes timestamps and equity from the second point onward to shrink the payload; the first point stays absolute and each later point is reconstructed by adding its delta to the previous value.
  • outMode selects objects (ARRAY) or parallel arrays (SHORT).

The response metadata reports what actually happened, including inputPointCount, outputPointCount, and whether resampling or delta encoding ran. A server size guard may force a compact representation, so the metadata, not the request, is the source of truth for the shape.

For a plain backtest the transform is fixed when the run is submitted and the curve is returned inline. In a parameter sweep, leaderboard rows are aggregate outcomes and carry curves only for retained trials, selected by mode (auto, topN, topPct, or none); those rows carry a pointer fetched separately, with the transform chosen at read time, and some views include the points inline as well. The indicator values and buy or sell markers behind the aggregate curve are available as stored signals when the run requests them.