How ATLAS measures a trading strategy

Closed candles only

ATLAS evaluates a strategy on candles that have finished forming. A signal exists at the close of a candle, never in the middle of one. This removes one of the most common errors in backtesting, which is to act on information that was not yet available at the time of the decision.

The last candle of a chart can still be forming, and the interface says so. Measurements and proposals are always built on closed candles.

One position at a time, with a fixed exit rule

In the signal statistics, every strategy is replayed the way the Signals page and the chart terminal display it: one position at a time, entered at the close of the signal candle. For the strategies that belong to the platform, the stop sits at 1.5 times the average true range and the target at 2 R, where R is the distance between entry and stop.

Strategies published in the community store use their own rules for entries and exits. The same engine replays them, so two publications can be compared on the same footing.

What is measured

For each strategy, time frame and direction, ATLAS counts the settled signals, the share of winners, how many ended at the target and how many at the stop, and the average gain per trade expressed in R. The detail is also given per market.

The average gain in R is the expectancy of the strategy. It is the figure that tells you whether the strategy makes money over many trades, which the win rate alone cannot do.

Results are raw, without costs

Spreads, commissions, swaps and slippage are not deducted from the platform statistics and from the store measurements. Each trader has their own broker and their own conditions, and a common raw basis keeps results comparable. In the strategy lab you can add your own costs to a test.

Real costs always reduce real results. A strategy with a small positive expectancy before costs can be negative after them, which is why the guide on trading costs exists.

A sample size on every figure

A percentage without the number of trades behind it means very little. ATLAS shows the sample size next to each result. On the public pages of this website, a measurement is displayed only when it rests on at least 200 signals.

Small samples are unreliable and even large ones describe the past. The figures are measurements, not forecasts.

A statistical test before any claim

When the statistics compare buying with selling, the difference is judged with a test of two proportions. Below a z value of 2 in absolute terms, the gap can come from chance, and the screen says so instead of naming a better direction.

In the strategy lab, a chronological split separates the data into a training period, a validation period and a test period, by default 60, 20 and 20 percent of the history. A strategy is also compared with simply staying exposed to the market over the same period and with the same costs. An idea that does not beat that benchmark is still shown, with its negative gap, because knowing that an idea adds nothing is useful information.

The community store re-measures every publication

An author of the store does not announce a result. When a strategy is published, the ATLAS server measures it again on a history from the common catalog, with identical costs for everyone, and that measurement is the one displayed. Authors cannot improve their own numbers, and installed strategies can be read in full before you add them.

What these figures cannot tell you

Markets change, and a strategy that worked on past data can stop working at any time. Historical data can contain gaps. Simulations rest on assumptions. ATLAS is decision support: it never places orders and never connects to your broker, and nothing it shows is financial advice.

Frequently asked questions

See it in ATLAS

Terms used in this guide

Expectancy, R multiple, ATR (average true range), Sample size, Out-of-sample test.

More guides