Portfolio analysis methodology

Data provenanceCopy Anchor Link

Every strategy exists as a pair of trade list and equity curve; the SPY daily series is added as a market reference. Every number in the analysis is derivable from exactly those files. An automated test recomputes samples with an independent parser directly from the raw files.

Portfolio constructionCopy Anchor Link

The portfolio is built as a union fusion across the full history of all selected strategies: each enters at its own backtest start (staggered entry) and is rebalanced daily under the chosen scheme across the strategies active that day.

All portfolio metrics come from this one real daily curve. There is no second version assembled from individual metrics.

Calculation conventionsCopy Anchor Link

All curves start at 100 and compute additively in base-capital points. One point therefore equals one percentage point on starting capital, without compounding.

Drawdowns are relative to base capital and reported as positive depth.

The card of a single strategy shows two common conventions side by side, each labelled. Max drawdown is given as the main value from the start (as a percentage of starting capital, the way the NFA prescribes it for programs with fixed position size, which is how the strategies trade) and next to it from the peak (as a percentage of the account high, the way QuantDesk and the CFTC compute it). It is the same decline in currency; only the denominator differs, and the difference grows with the account. The card's Calmar follows the QuantDesk definition (mean trade return times trades per year, divided by the drawdown from the peak). The card's Sharpe is computed on daily account results; the per-trade Sharpe of the QuantDesk card stands next to it as its own value. Sortino divides by all periods, not by losing days alone.

In every table where a strategy stands next to the portfolio or the SPY, one row has one convention: drawdown and Calmar for all columns come from the daily closing curve, relative to the start, the way the portfolio itself is computed. The card values from the peak and per QuantDesk do not appear in those tables.

Amounts per trade (expectancy) are stated in the account currency of the respective strategy and are not converted to a common currency.

Rounding happens only at the output boundary of the calculation service, never inside the calculation.

ConfidenceCopy Anchor Link

The PSR (Bailey and López de Prado) estimates the probability that the observed Sharpe is genuinely positive. It corrects for skew, heavy tails, track length and autocorrelation.

Beside every backtest result sits the number of configurations tested. Trying many variants finds a good one even without an edge, and without that number a result cannot be placed.

LimitsCopy Anchor Link

Everything in the portfolio analysis is an in-sample backtest on historical data and not a forecast.

The equity curves are realised daily P&L at trade exit, not intraday marking. For strategies with multi-day positions the path therefore differs from an intraday curve.

SPY comparisons only hold where the convention is identical, and are marked as such otherwise.

What this analysis is not: not investment advice, not a return forecast and not a guarantee of future results. Historical simulations cannot anticipate future market phases.