Methodology

Data sources

13F holdings come from public filings on the US SEC's EDGAR; quantitative strategies from published academic research and public methods; prices and dividends from professional market-data services, including the history of delisted stocks. All returns are total returns including dividends.

How 13F replication works

We replicate only the US long common-stock positions in a filing (no options or other derivative positions), weighted in proportion to each position's reported market value.

Entry happens on the first trading day after the filing became public on the SEC — a new holding enters the curve only once it was actually knowable (typically ~45 days after quarter-end). No information is ever used before its publication date.

Three replication scopes: top-5 only, top-10 only, and all positions with a reported weight of at least 1% — each renormalized over the positions it keeps.

Between filings the book is buy-and-hold: weights drift with prices, with no interim rebalancing. Names that cannot be priced (e.g., delisted without data) are renormalized away across the remaining priceable holdings.

The quant stock pool

Single-stock factor strategies (momentum, value, quality, low volatility, …) pick from a "market-cap Top 500" pool: at each month-end, the 500 largest US companies by market capitalisation, drawn from all US common stocks — including delisted ones, to avoid survivorship bias. Where a company has several share classes, only the largest one is used.

The pool is rebuilt monthly from information knowable at the time (point-in-time); each strategy then selects within it by its published rule and rebalances at month-end. Macro-allocation strategies (Permanent Portfolio, risk parity, …) use asset-class ETFs.

How the composer overlays strategies

The composer looks through to individual stocks and rebuilds one portfolio, rather than averaging several performance curves.

Each fund’s disclosed holdings are first converted to percentages so that every book equals 100%. They are then added together using the fund-level weights you choose (equal, risk parity, mean-variance), producing one combined holdings table. Finally the scope you pick keeps the largest 10, the largest 20, or every name at 1% or more of the combined book, renormalised to 100%.

Normalising before combining matters: without it, a fund that discloses more positions or a larger book would automatically dominate, and equal weight would no longer be equal.

Risk-parity and mean-variance weights are estimated only from history before the month being measured, never from data unknown at the time; the earliest months fall back to equal weight while the sample is too short.

Each fund contributes its largest 30 positions at 0.4% or more. For a concentrated manager that is effectively the whole book; for an institution holding several hundred names it replicates the core positions only.

Composer curves are computed from monthly data and may differ slightly from the daily curves on a strategy page.

How costs are charged

Every rebalance is charged a flat 10 basis points (0.10%) per side on turnover; the cross-asset trend-following family is charged 15 basis points per side, reflecting the wider spreads of commodity and currency ETFs. Turnover is measured against the book as it actually stands before the trade (including price drift) and charged once on the rebalance day.

No slippage, market impact, taxes or borrowing costs are modeled — for large capital or less liquid names, real costs would be higher. This is a known limitation of the backtest convention, stated plainly.

Reading the metrics

CAGR is the compound annual return over the sample; volatility measures the size of swings; return/risk = annualized return ÷ annualized volatility; "excess significance" is a t-statistic — how statistically reliable the excess return is, higher meaning more trustworthy. Max drawdown and longest underwater stretch show the hardest period.

Limitations (please read)

A 13F discloses US long equity positions only: shorts, futures, derivatives and non-US positions are absent, and the quarterly snapshot lags — a filing is not the institution's full portfolio, and replication returns are not the institution's actual returns.

All performance figures are historical backtests, not anyone's real trading record. Past performance does not predict future results.

We and people associated with us may hold securities that appear in the content. The content is generated from public data under fixed rules and is the same for every reader — it is not shaped by our own holdings, and we do not trade on unpublished content.

Every figure is computed by an automated pipeline under the methods above. Despite checks, errors in calculation or presentation are possible; when we find one we correct it and update the curves. Treat everything here as research reference, not a basis for trading.

Update cadence

New filings are checked and curves extended every trading day; portfolio configurations recompute at each month-end. Every card shows its data as-of date and last backtest date.