A research measure developed by AlgoTrader
AlgoTrader Score is our proprietary quantitative research measure for comparing hypothetical strategies. It evaluates return, downside risk, profit quality and the strength of the observed record together, across the full available history.
The purpose is to support a more disciplined comparison than sorting by the largest return. The score summarizes selected aspects of the observed record into one versioned 0–100 assessment, while the underlying reports remain available for your own analysis. It does not replace due diligence, portfolio construction or an assessment of suitability.
How the quantitative framework works
AlgoTrader Score combines several dimensions of performance rather than ranking accounts by return alone. Its quantitative framework uses conservative treatment of limited evidence, bounded treatment of extreme inputs and source-history checks. When complete source periods are available, it also examines dependence on exceptional months, prolonged periods below a previous peak and recent deterioration. The formula, component weights, calibration and detailed decision rules are proprietary.
The framework combines performance characteristics with an adjustment for the amount of supporting evidence. Extreme inputs have bounded influence, and a brief record receives a more cautious assessment. The result is a descriptive composite: it summarizes recorded behavior under an explicit model, rather than estimating a guaranteed future outcome.
Return in the context of time
We assess the recorded capital growth alongside the length of the observed history. A short burst of profit and a sustained record are different evidence. The displayed cumulative or compounded annual return retains its own label and source cost basis; it is separate from the score.
Downside risk and recovery burden
Reported full-history drawdown contributes to the assessment of the losses borne to achieve the return. Deeper losses require a larger subsequent gain to recover. A low observed drawdown over a short or incomplete record does not establish low future risk.
Profit quality
Reported profit factor contributes evidence about the balance of aggregate trading gains and losses. It is evaluated with return, drawdown and record length. An unusually large ratio or an isolated favorable statistic cannot substitute for a substantial observed record.
Concentration and the path of results
Where complete comparable monthly source history is available, we examine whether performance depends heavily on an exceptional month, how persistently the record remains below a previous peak, and whether recent results have deteriorated. Accounts with similar total returns can therefore receive different assessments.
Evidence strength
Elapsed observation time and available closed-trade records inform a conservative adjustment for limited evidence. High trading frequency does not replace elapsed history. The evidence labels describe the amount of support available; they are not statistical confidence levels or certifications of robustness.
Source integrity and coverage
Required inputs must be current, valid and consistent with the selected account history. Longer records require complete comparable source-period coverage. Missing values are not filled with zero, stale observations do not earn a current score-based rank, and unavailable records remain discoverable.
How to read a score
Higher values indicate a stronger assessment under the same methodology version. The scale is not a return percentage, a percentile of the visible field or a probability of making money. A score twice as large does not imply twice the edge, and a small rounded difference should not be treated as economically significant.
The score and the rank are distinct. Filters select which accounts are compared and can change their ranks, while each account keeps its score for the same accepted source observation and methodology version. Sorting by drawdown or history changes display order without rewriting the official score-based ranks. New accepted observations or a methodology revision can change the score.
One account history, with its source intact
The current score evaluates the selected Strategy history. A published Historical backtest is separate research evidence and does not supply the current score. Different account epochs, capital bases or strategy versions must not be joined simply to obtain a longer record.
Source-reported statistics remain visible alongside the score. Capital-based return requires its accepted source evidence; trading costs retain their reported treatment. Subscription, platform, data and hosting expenses are not assumed to be included. Payments, subscription status and popularity are not score inputs.
Evidence labels and unavailable scores
Limited, Developing and Established describe progressively more supporting observations and history. Established is not a guarantee of statistical significance, independent trades, robustness across market regimes or future profitability. A short record can be scored when its required inputs are valid, with a more cautious assessment.
A dash means the required current evidence is unavailable, not that the account scored zero. A longer but incomplete record can remain unscored. All preserves published accounts, including losing and unscored histories; the absence of a rank must not erase inconvenient observations.
Public principles, proprietary implementation
AlgoTrader develops and maintains this measure. We describe its purpose, the dimensions it evaluates, the meaning of its scale and the limits of the evidence. We do not publish the formula, exact transformations, component weights, calibration, internal thresholds or detailed decision rules.
The method is versioned. Compare scores within the same methodology version and review the source observation time. Versioning makes a methodological change distinguishable from a change in the underlying trading record; confidentiality is not a claim of independent validation.
What the score does not establish
Version 1 is a proprietary descriptive assessment, not an empirically validated predictor of future returns. It does not estimate the probability that a strategy has a persistent edge, correct for all selection bias, measure every tail risk or prove that a backtest was not overfit.
The score is not a Probabilistic Sharpe Ratio or Deflated Sharpe Ratio. Those methods answer specific statistical questions and require evidence and assumptions of their own. A composite score should not be presented as having performed those tests when it has not.
All strategy results remain hypothetical. Simulation, fill assumptions, costs, limited observations and changing market conditions can materially affect interpretation. Review the report, losses, source coverage and subsequent observations rather than relying on a single number.
Further quantitative context
Research on track-record uncertainty and selection bias helps frame the questions a trader should ask of any performance assessment. These papers describe their authors’ methods and do not validate or endorse AlgoTrader Score.
- Bailey & López de Prado — The Sharpe Ratio Efficient Frontier: track-record length and uncertainty in estimated performance.
- Bailey & López de Prado — The Deflated Sharpe Ratio: selection bias, multiple testing and non-normal returns.