Does Your Trading Algorithm Work? Read the Evidence

“Works” can mean that code follows its specification, that a measured sample is profitable, or that an account arrangement is permitted. Those are separate claims. Define which claim you are testing and what observation would contradict it before treating a positive chart as an answer.

Check behaviour before interpreting profit

Match orders and logs to the exact program, preset, symbol and account conditions. Verify signal timing, quantities, rejected orders, restarts and attempted exits. A profitable sample can contain an implementation error; correctly executed code can implement an unprofitable idea.

Separate historical development results, genuinely unused historical evaluation, later demo observations and any actual deployment records. The MT5 Forward tester option is a held-out historical segment, not necessarily a live demo. Demo fills and infrastructure do not establish identical execution elsewhere.

Reconcile the record and the comparison

Use the complete stated period, including losses, open exposure, costs, deposits, withdrawals and missing records. Name the instrument, volume, currency, time basis and price/execution source. A deposit is not trading profit and a displayed balance is not automatically a paid Reward.

Compare with a stated baseline on the same period, cost basis and exposure assumptions. Report what was selected after trying alternatives. A smooth chart, fixed number of months or hundreds of trades does not by itself establish statistical significance or future profitability.

Use the metrics' actual definitions

Profit factor is gross profit divided by the magnitude of gross loss when that denominator is non-zero. With no losses it is not an ordinary finite ratio. Define balance versus equity drawdown, sampling and currency versus percentage. A provider's initial-balance or trailing limit is a separate calculation.

Recovery factor is profit divided by maximum drawdown on the stated basis, not recovery speed. Time to regain a prior peak is a separate observation, possibly unfinished. Sharpe needs a defined differential-return series, reference, frequency and annualisation; Sortino additionally needs the target and downside-deviation convention. Scores on incompatible bases should not be ranked as if comparable.

Work through an explicit numerical example

Fictional arithmetic, not client results: assume gains of €1,200 and losses of €1,000 on the same stated basis. Profit factor is 1.2 and their difference is €200 on that basis. If the chosen profit measure is €200 and maximum drawdown is €400 in the same currency, recovery factor is 0.5. None of these values tells you the number of recovery days or proves an edge.

A second fictional assumption: 40% winning trades averaging +€30, 60% losing trades averaging −€10 and €1 additional cost on every trade give 0.4 × €30 − 0.6 × €10 − €1 = €5 per trade. This illustrates the calculation, not a future expectation. Do not subtract costs again if the average outcomes already include them. A win rate or an ambiguous risk/reward label alone cannot answer the question.

Investigate uncertainty rather than retune every loss

Keep evaluation observations outside the parameter-selection process. Repeatedly inspecting a holdout and selecting a new winner makes it development data. Dependence among trades, sample selection and changing conditions affect uncertainty; there is no universal PF, Sharpe or win-rate threshold, 99.9% label or fixed test duration that certifies a system.

When results differ, investigate data, code, costs, fills, exposure and rule events before assigning a cause. Define review and suspension conditions and retain old versions. A losing sequence alone proves neither overfitting nor inevitable recovery. Changing parameters or increasing capital is a separate decision requiring its own evidence and permission.

Keep provider permission and product evidence separate

Check the actual purchased programme, phase, size, platform, add-ons, permitted strategy and rule date. An unmet consistency or Reward condition is not automatically the same as a loss-limit breach. A profitable backtest and a vendor's account record do not grant permission to deploy a particular EA or copying arrangement.

For JPTC EA Hub downloads, request the actual file/version report and test conditions. The separate hosted JPTC Algo requires no customer installation or VPS. Its results page publishes monthly figures from its own broker statement: it is not an account-linking analytics service, independent audit, client earnings dataset or proof that another strategy will pass or pay out.

Sources checked 16 October 2026: Strategy Tester · MT5 test report · EA inputs and permissions · Sharpe (1994) · Data leakage.

Sortino (CFA Institute)

Algo Backtesting: A Reproducible Forex and Futures Workflow · Test a Trading Algorithm: The Deployment Readiness Checklist · JPTC EA Hub · JPTC Algo

Broker-statement images of the separately hosted JPTC Algo

These service images do not establish another algorithm’s results, execution or provider permission.

@jptradingcapital on Instagram →

FAQ