The annualised rate of return over a multi-year period that accounts for compounding, normalising returns across different backtest lengths so two strategies tested over different spans can be compared.
also: CAGR, annualised return, annual growth rate
Updated
If the account had grown by the same percentage every year and finished exactly where it did, what would that yearly percentage have been? It is a smoothed average, not a description of any single year the strategy actually lived through.
Total return is meaningless until you know how long it took, and backtests are almost never the same length. CAGR is the conversion that makes two results comparable — and the conversion that hides how uneven the path between the two endpoints was.
CAGR = (End balance ÷ Start balance)^(1 ÷ years) − 1 The exponent is the number of years the sample actually covers, including the fraction — a run from 2019-01-01 to 2026-04-01 is 7.25 years, not 7. Using the whole-year count inflates the figure. Both balances must be the same currency and must come from the same run; splicing a backtest onto a forward record produces a number that describes neither.
No single number describes a strategy. These metrics change how this one should be read.
An EA is tested on a $10,000 deposit and finishes the run at $18,000. The window is five years to the day. The question CAGR answers is what constant yearly rate would have produced that same ending balance.
The same $8,000 gain over ten years would be a 6.1% CAGR — the identical profit, describing a very different strategy.
Calculation (18,000 ÷ 10,000)^(1 ÷ 5) − 1 = 1.8^0.2 − 1 = 0.125
Result 12.5% compound annual growth rate
CAGR is a summary of two endpoints and a duration. Everything that happened between the endpoints has to be read off other numbers, so never quote it alone.
| Range | What it means |
|---|---|
| Window under one year | Annualising a partial year magnifies whatever the sample happened to do. A strong four-month run extrapolates to a figure the strategy has never sustained. |
| One to three years | Readable, but likely to span a single market regime. Expect the figure to move once a different regime enters the sample. |
| Five years or more, costs included | The window worth quoting. Long enough to contain more than one regime and to let costs accumulate honestly. |
| Quoted without the drawdown from the same run | Incomplete by construction. A 30% CAGR alongside a 15% drawdown and a 30% CAGR alongside a 55% drawdown are not comparable claims. |
mt5depot EA pages state the tested window with its start and end dates and publish the full trade ledger, so an annualised figure can be re-derived from the source instead of taken from a headline. Re-deriving all fourteen gives 0.48% to 32.08%, median 4.0% — eight sit under 5% and exactly one reaches the 15–35% band above. That one is the basket that holds twenty-four legs open at 0.1 lot, on the same 10,000 USD deposit every other run uses, which makes the band a sizing statement rather than a verdict on the strategies: the percentage tracks how much is open at once, not the quality of the edge. Read the band together with the deposit-to-lot ratio the run used, or it measures money management instead of the edge.
It is a description of one historical path, smoothed. A strategy with a 20% CAGR built from one exceptional year and four flat ones has never once returned 20% in a year, and nothing about the figure implies it will.
Scaling three months up to a year multiplies both the edge and the luck by four. This is the single most common way a modest result becomes a marketing number, and it is invisible unless the window length is stated next to the figure.
Compounding and fixed lots produce different curves from identical trade sequences. Comparing the annualised figure of one against the other rewards the money-management setting rather than the strategy.
A percentage is only meaningful against the balance that generated it. The same EA on a $1,000 deposit at the same lot size takes ten times the relative risk, and its percentage return is not the same claim.