Trading bot metrics: equity, drawdown and misleading win rates

Evaluate a trading bot with account equity, drawdown, costs and a complete trade history. Win rate and closed-trade profit are supporting measures. A strategy can close many small winners while an open position loses more than all of them combined.

Begin by asking what the report includes, over which dates, with how much capital and in which mode. A backtest, paper account and exchange statement answer different questions. Do not merge their numbers into one track record or treat a short profitable interval as an annual return forecast.

Establish the account boundary first

Record starting capital, valuation currency, open positions and any deposits or withdrawals. Without external cash flows, a simple period return is ending equity divided by starting equity, minus one. Equity includes the marked value of remaining holdings, not just the balance available for new orders.

For example, an account starts with 5,000 USDT. Closed trades contribute a net gain of 200 USDT, while open holdings show a loss of 450 USDT. With no other flows or costs omitted, ending equity is 4,750 USDT and period return is −5%. Reporting only the 200 USDT gain describes completed trades rather than the account result.

If the owner deposited 1,000 USDT during the interval, the change in balance would no longer equal trading performance. Use a documented cash-flow-adjusted method and retain the timing of transfers. Do not repair the problem by silently excluding inconvenient deposits.

Read win rate alongside payoff size

Imagine ten completed hypothetical trades: eight earn 10 USDT each and two lose 50 USDT each, before fees. Win rate is 80%, but gross gains are 80 USDT and gross losses are 100 USDT. Net trading result before costs is −20 USDT.

Expectancy per trade is 0.8 × 10 − 0.2 × 50 = −2 USDT. Profit factor is 80 / 100 = 0.8. Both calculations explain why a high win rate can coexist with a loss. They describe this sample and do not establish its future distribution.

Define a trade consistently. Partial exits, grid cycles and complete position lifetimes can produce different counts from the same executions. A bot that counts every small sale as a win cannot be compared directly with one that records a whole averaging sequence as one deal.

Measure the depth and duration of drawdowns

Drawdown compares equity with its previous peak. If equity peaks at 5,500 USDT and later reaches 4,400, the decline from that peak is 1,100 / 5,500 = 20%. Returning from 4,400 to 5,500 requires a 25% gain. Equal percentage losses and gains do not cancel each other.

Specify the valuation frequency. Daily closing equity can miss a deep intraday loss. A closed-trade balance curve can miss losses that remain in open inventory. Keep the maximum observed drawdown, its dates and time spent below the previous peak together.

Use drawdown as an observed property of a particular sample, not a guaranteed future ceiling. For an account running several bots, measure their combined equity on common timestamps rather than adding their separately observed worst drawdowns.

Audit costs and the comparison baseline

Separate gross price gains from trading fees, adverse execution, funding or borrowing where applicable, and service costs. Avoid double counting: a net exchange result may already include trading fees. Document what remains outside it, including tax treatment rather than assuming a universal after-tax number.

Choose a baseline that answers the original objective. An unleveraged BTC strategy can be compared with holding BTC over the same interval, provided starting capital and valuation dates match. Also inspect its time in the market and drawdown. A mostly-cash strategy and continuous exposure do not provide identical risk simply because returns are similar.

For return ratios, write the return interval, risk-free assumption if used, annualization method and treatment of missing observations. A ratio without those settings may not be comparable across tools. Start with the backtesting checklist before relying on advanced ratios.

A report checklist worth keeping

• Label historical simulation, paper trading and actual executions separately.
• Preserve dates, strategy version, capital and external transfers.
• Show account equity with open positions included.
• Pair win rate with payoff size, profit factor and trade definition.
• Include drawdown depth, duration and valuation frequency.
• Reconcile costs and compare a suitable baseline over identical dates.

Keep the underlying fills, quantities, timestamps and fees so another person can reproduce the simple totals. Missing fields should produce an explicit limitation, not an inferred precise metric.

Clerix's features overview explains the intended analysis workflow. Prepare transaction history in CSV, or a ZIP containing CSV files, and check which fields are available. Missing data can limit the analysis; the website's example figures are not verified bot results. Testing engines and live execution are in development.

Frequently asked questions

Is a profit factor above one enough?

It indicates gross gains exceeded gross losses under the report's definitions. It does not resolve sample size, open losses, omitted costs or whether the strategy was selected after testing many alternatives.

How many trades prove a bot works?

There is no universal count that proves future profitability. Dependence between trades, market regimes, strategy selection and the size of rare losses all affect what the sample can establish.

Can I annualize a strong week?

You can calculate a mathematical extrapolation, but it assumes repeatable conditions that one week does not demonstrate. Report the observed dates and return without presenting the extrapolation as an expected outcome.