Methodology
Last updated: Nov 10, 2026
This page explains how PortLens computes every number it shows you: where the data comes from, how often it refreshes, the formulas behind each metric, and, just as important, the known limitations. If a metric can mislead in an edge case, we say so here and, wherever possible, in the product itself.
Data sources
Under the current licensed source, prices and price histories are read for each calculation and held only briefly in memory; nothing price-bearing is stored durably or displayed. A price counts as current only while it is still the latest its market has produced: for shares and ETFs, until a further weekday session has closed since it was read, so a Friday close still serves across a weekend. Results use end-of-day data rather than live quotes, and saved scores keep the sources, source version and dates used when they were calculated; a change of source invalidates saved results. Benchmark fallbacks that exclude dividends are labelled price-only, because excess return and alpha can appear higher than against a total-return benchmark. Fetch time is distinct from the last exchange close.
Analysis weights use each holding's share count multiplied by its current cached quote. Inside the restricted brokerage rollout, connected positions use the broker's quantity × price instead. Minor quote units are converted to their major currency (GBp/GBX to GBP, ZAc to ZAR and ILA to ILS), then valued in US dollars. Exchange rates are sourced from the European Central Bank (ECB) euro reference rates retained by PortLens and refreshed daily. Cross rates such as GBP to USD are calculated by PortLens through the euro; they are not separately published ECB rates. An exchange rate is used only if it is at most one ECB publishing day old; weekends and ECB holidays don't count. Saved results show when they were last priced. Holdings without shares, a quote or a usable exchange rate are excluded from analysis and named with their reasons. Cash and unsupported investments in combined portfolios are disclosed outside these security calculations. Weights and scores describe the valued holdings only; if none can be valued, no analysis is shown.When competitions are open, entries require every holding to be valued and are refused with the per-holding reasons otherwise, except for Build Quality inside the brokerage rollout: its entries and refreshes score the supported, valued securities and record exclusions. Display currency does not change these US-dollar analysis weights. These are reference valuations for information, not executable exchange prices or investment advice.
- Prices & price histories: Tiingo end-of-day data for US-listed stocks and funds (US-listed global funds qualify), used in calculations only: no quotes, charts or values are displayed. Native non-US listings and crypto are kept as inputs and excluded from price-based analysis in this release, with the reason shown. Exact market caps and valuation ratios are not displayed. Industry and style exposure and style drift are planned for PortLens Pro. Sector exposure and Simulate's diversification and market cap drift are free where data exists.
- Sector from SEC industry codes: a versioned four-digit SIC crosswalk and reviewed company exceptions map SEC business descriptions to sector groups. Unreviewed or ambiguous codes remain outside the classified coverage. The Diversification Score uses SEC SIC groups for its two industry points during this stage. These broad groups may differ from other services. Fund sectors come from underlying company profiles where available; published holdings and sector coverage can be incomplete.
- Company size: market cap drift uses reported SEC share counts and the licensed raw closing price, calculated on the server. Only Large (over $10 billion), Mid ($2 billion through $10 billion), Small (under $2 billion) and Unknown groups are returned. Numeric market caps are never stored or displayed during this stage. Share counts must be no more than 150 days old and align with known splits. Supported share classes use a company total; unsupported classes and ADRs remain Unknown. The distribution covers directly held stocks only, excluding funds and crypto, and normalizes over those stock weights.
- Fund constituents: for a small set of funds the issuer's own holdings file is read and checked daily; for all others SEC N-PORT filings are checked weekly and can describe holdings several months earlier. Each source's holdings date is distinct from the day we fetched it. Funds whose constituent lists cannot be resolved contribute no underlying companies to scan look-through.
- Benchmark: the SPY ETF's total-return series (dividends reinvested), an ETF proxy for the S&P 500 rather than the index itself. This page is where that is disclosed. A price-only benchmark, if ever served, is labelled as such and understates total return by roughly 1–2% per year.
- Risk-free rate: the 1-Year US Treasury constant-maturity yield from the US Treasury daily par yield curve, refreshed daily. If Treasury is unreachable, we use the most recent stored value.
- Factor data: daily Fama-French three-factor series (market, size, value) from the Ken French Data Library at Dartmouth, updated roughly monthly at the source.
- Fund expense ratios: the SEC's Risk/Return Summary Data Sets, tagged from each fund's own prospectus and refreshed daily. We read the net figure (after any fee waiver the fund files) and fall back to the gross figure when a fund files no waiver. Funds that file no prospectus risk/return summary (unit investment trusts, commodity and crypto trusts, closed-end funds, business development companies and non-US listings) have no ratio here at all. We identify those from the SEC's own registers of fund share classes, so a fund is still recognised as a fund when a price feed calls it a stock.
- 13F filings: featured-investor portfolios are sourced from SEC Form 13F disclosures from SEC EDGAR and refreshed quarterly, the day after each 13F filing deadline.
True Exposure (ETF look-through)
Each resolved ETF is expanded into its available underlying constituents. A constituent's effective weight is the ETF's weight in your portfolio multiplied by the stock's weight inside the ETF. Exposures to the same security identifier, held directly and through one or more funds, are summed into a single position. Share classes that carry their own US ticker remain separate; the share classes of a company that resolves to no US ticker are summed under its legal entity identifier (LEI). This surfaces concentration that a plain holdings list can't show. You can run the look-through on your own holdings.
From the looked-through portfolio we compute effective holdings (the inverse Herfindahl index, 1 / Σw²: a portfolio of one 50% and two 25% positions behaves like ~2.7 equal positions, not 3), top-10 concentration, per-company fund counts, and, where sector data exists, the dominant sector.
Bonds and cash inside funds: when a fund's filing lists bonds, cash or derivatives, they appear by type (US Treasury, US government agencies, corporate bonds, foreign governments, municipal bonds, cash and equivalents, derivatives and a few rarer types) and bonds count in geography by their own country. A fund made mostly of other funds, such as a target-date fund, has those funds opened up one level, so their shares, bonds and cash show the same way; a fund inside them shows as funds. A corporate bond whose issuer code matches a listed company counts with that company, shares and bonds added together; the rest show as corporate bonds. A filing's negative lines, such as shorts, are left out. These groups are not companies, so top-ten concentration, the largest single company and effective holdings leave the bond and cash groups out, and bonds do not yet enter the portfolio score. There is no fixed-income analysis such as yield or duration.
Attention thresholds: orange emphasis appears when one company reaches 10%, one sector reaches 40%, the top ten positions reach 60%, or effective holdings fall to 10 or fewer. The 10% company threshold borrows a regulatory reference point: Article 52(2) of the UCITS Directive lets Member States raise the ordinary 5% same-issuer limit to no more than 10%, subject to a separate 40% aggregate rule. That rule governs UCITS funds, not personal portfolios. The 60% top-ten threshold is intentionally well above the S&P 500's 36.4% top-ten weight (accessed July 29, 2026). The 40% sector and 10 effective-holdings cutoffs are transparent, round-number PortLens design choices rather than regulatory limits or fitted optima. Crossing any threshold changes visual emphasis only; it is not a risk grade, suitability assessment, or recommendation to trade.
Limitation: when the fund's SEC N-PORT filing or the issuer's own holdings file resolves, look-through uses every constituent that real source publishes. Coverage is still bounded by what each fund publishes: funds we can't resolve contribute nothing, crypto remains its own terminal position. The look-through coverage ratio counts a fund as covered when a constituent list is available, even if that list is partial. It is an upper bound on coverage, not proof of completeness. Top-ten weights use the whole portfolio as their denominator; omitted exposure can understate them. Effective holdings measures each holding against the money outside the bond and cash groups (never less than those holdings' own total), so an unreported tail contributes nothing to the sum of squared weights and can make the portfolio appear more diversified. A historical article that normalizes its mapped weights states that different denominator explicitly.
Pairwise ETF overlap
Overlap by weight is the sum of the smaller of the two fund weights for every shared security identifier: Σ min(wA, wB). Weights are fractions; the result is displayed as a percentage. Published weights are not scaled to make an incomplete list add up to 100%. The shared-holdings count covers all matches; the table shows only the ten largest by overlapping weight.
Each fund keeps its own holdings date. The pair is anchored to the older date only when both are known, and that date is stated. Thin lists, unresolved data and incompatible cross-source identifiers can prevent a numerical comparison. Two confirmed share classes of the same fund are identified before arithmetic. The qualitative overlap bands are PortLens labels, not a recommendation about holding either fund.
Curated pair pages show beta versus SPY from daily total-return-adjusted prices over the trailing year, using shared dates with at least 240 daily returns. Shorter histories are unavailable. Beta measures historical sensitivity and does not predict future results.
Weighted expense ratio
Each fund's available annual expense ratio is weighted by that position's value and averaged across the portfolio. Shares you hold directly carry no fund fee, so they enter the average at zero; that is their real cost, not a missing number.
Funds with unavailable expense ratios, including published fees withheld pending verification, are left out of the average entirely and named on the card, never counted as free: a zero there would understate what you pay. The average is therefore computed over the covered part of your portfolio, and the scan states what share of your money that is. Where the uncovered part is large, the figure describes less of your portfolio than it appears to, so the card says so.
The weighted figure, the holdings it covers, the ones it excludes and the per-holding split (what each fund you hold costs on its own) are free, with no account needed.
Risk & return metrics
- Portfolio return: buy-and-hold over the trailing lookback window (one year by default), ending at each holding's last cached close. Position weights derived from your share counts and current prices describe the portfolio as it stands today, so we roll each holding back by its own price change to recover what it was worth at the start of the window, then measure the return on those starting amounts. Weighting by today's values instead would let a holding that doubled dominate a window it only finished large in (look-ahead bias). We require at least ~180 days of price history.
- Beta: the weighted average of per-holding betas. Each per-holding beta is computed from the holding's dividend- and split-adjusted price history as covariance with the SPY total-return proxy divided by the proxy's variance, over the daily sessions the two share in up to the last 400 days and requiring at least 30 return observations; it is a measured figure, not one published by a data provider. Holdings whose beta can't be established by any of these routes are excluded and the remaining weights renormalized, not silently defaulted. If no holding has an establishable beta, the portfolio calculation falls back to the market default beta of 1.0 rather than failing entirely.
- Low-beta lists: low-beta S&P 500 stocks ranks every share line in State Street's daily holdings file for SPY, each share class on its own. Low-beta ETFs lists the 50 largest US-listed stock ETFs. A fund's size is what it last reported to the SEC on Form N-CEN in the four most recent quarterly data sets (monthly average net assets, or total assets for a unit investment trust), or the net assets in its latest Form N-PORT when it has re-registered since its last N-CEN and that filing put it among the 100 largest. Both forms report a whole fund, so an ETF that is one share class of a larger fund (most Vanguard ETFs) is sized with its mutual-fund classes included and can rank above an ETF whose own assets are larger. A stock ETF holds at least 80% of its net assets in common or preferred stock by its latest N-PORT, not counting shares of other funds; leveraged and inverse funds are left out. Each beta is the per-holding beta above, from the same calculation the beta calculator runs, with every holding in a list measured on the same day against the same SPY window. A holding is ranked when it shares at least 90% of SPY's daily returns in that window; one with fewer is named with its count, and one with no beta is named with the reason. The lists are measured on the 2nd of each month and replaced only when a run measures at least 95% of each list; a list more than 70 days old is not shown.
- Correlation: how closely two holdings' daily returns have moved together, from -1 (opposite) to 1 (in step). Each figure is the Pearson correlation of daily returns from dividend- and split-adjusted closes over up to the last 500 calendar days. Each pair is compared on the trading days both have prices for; a move made on a day only one of them has is added into its next shared day, so no part of either path is dropped. A pair needs at least 30 shared returns, otherwise it shows a dash, and figures are rounded to two decimals. Insights computes it for a saved portfolio; the correlation calculator computes it for up to five tickers without an account (20 signed in), names any ticker it leaves out and states the dates each result covers. For each pair of funds, up to 10 pairs per calculation, the calculator also shows their holdings overlap by weight, from the same holdings lists and calculation as the ETF overlap pages, dated by the older fund's holdings date.
- Treynor ratio: excess return over the risk-free rate divided by beta. Below a beta of 0.1 the ratio is mathematically unstable, so we treat it as undefined and score that component neutrally. For those below-threshold betas, the scan resultsshow a dash instead of a figure, while Insights displays the stored sentinel as 0.000.
- Jensen's alpha: the return above what CAPM predicts for your beta, Rp − Rf − β(Rm − Rf). Well defined at any beta, which is why the Alpha Cup uses it for ranking (see below).
- Diversification score (0–10): five weighted components: holdings count, sector spread, industry spread, look-through concentration, and geography. Spread components use weight-based effective counts (1 / Σw²) computed on the looked-through portfolio, so ten tech stocks don't count as ten sources of diversification. A holding counts only for the part your larger holdings don't already own, and a company held directly and through funds counts once, so splitting money across funds that own the same companies doesn't raise the score.
- Overall score: 40% risk-adjusted return (normalized Treynor) + 40% diversification + 20% excess return versus the benchmark. These are transparent product-design weights, not weights fitted to historical returns or claimed to be optimal. Risk-adjusted return and portfolio structure receive equal weight so neither recent performance nor diversification dominates the result. Raw excess return receives half as much weight because short-term market exposure and luck can influence it, and performance is already represented in the Treynor component. Beta is shown as context rather than scored because higher or lower market sensitivity is not inherently better. The same weights apply to every portfolio; they are not tailored to your circumstances or objectives.
- Value at Risk: daily VaR is the historical 5th percentile of your portfolio's actual daily returns; the annual figure comes from a lognormal model calibrated to the average and spread of those returns, so a strong run of past gains lowers it. When either outcome is still a gain, the card shows No loss rather than a figure. Both need at least 30 days of aligned history.
- Max drawdown: the deepest peak-to-trough fall of your portfolio's combined buy-and-hold path over the trailing year, on calendar dates the included holdings share. Today's weights are rolled back to the start of the measured window from each holding's price change, then allowed to drift rather than being reset daily. The figure is measured on the portfolio, not averaged from your holdings' individual falls, which happen on different days. Exclusion of holdings without enough shared price history is bounded at half the measurable weight; every exclusion is named on the card and the remaining weights are renormalized. Beyond that bound, the card states the shorter period it covers rather than calling it a one-year figure, and below roughly six months of shared history no figure is shown: a maximum measured over a short window can only understate the real fall.
Return projections
Return projections use five independent models rather than one blended forecast, so you can see where they disagree. CAPM and Historical run on every scan and Monte Carlo with a free account; Fama-French and Macro are planned for PortLens Pro:
- CAPM: risk-free rate plus your portfolio beta times a 5.5% equity risk premium.
- Historical average: the trailing geometric annualized return of your actual holdings, over up to the last 500 days.
- Monte Carlo: 10,000 simulations built by block-bootstrapping your portfolio's real daily return history in 20-day blocks, resampling what actually happened, including fat tails, rather than assuming a normal distribution.
- Fama-French three-factor: an OLS regression of your daily excess returns on the market, size, and value factors (minimum 60 overlapping days), with expected return built from the estimated loadings and live factor premia. When factor data or history is insufficient, a clearly labeled heuristic approximation is used instead.
- Macro: an earnings-yield build-up (E/P plus expected inflation from the US Treasury 10-year par yield minus the real yield on the same date), applied to the equity sleeve of your portfolio and levered by its beta. BEA quarterly real GDP growth is shown as context, not added to the projection.
This product uses the Bureau of Economic Analysis (BEA) Data API but is not endorsed or certified by BEA.
Uncertainty bands on all models are 95% lognormal intervals. Projections are estimates with wide error bars, not predictions; see Disclosures for how to read them.
Competition scoring
Competitions coming soon. We’re updating the Alpha Cup and Build Quality. Check back after launch. Entries and daily scoring are paused. The rules below describe scoring when competitions are open.
The two competitions rank on different metrics. The Alpha Cup ranks by Jensen's alpha (Treynor is displayed alongside it, but alpha is the sort key because it stays well defined for low-beta portfolios). Your holdings and weights freeze the moment you register, so edits afterwards can't game the result. Every entry is scored over the same date window, using the same benchmark and risk-free rate, and the ranking is frozen when the Cup ends.
Portfolio Rankings ranks by portfolio construction quality on a 0–100 scale: concentration, the effective number of holdings, sector and geographic spread, and hidden look-through overlap. It is computed from how a portfolio is built and reads no price history, so it needs no measurement window; there is nothing to wait out and nothing a late reshuffle can flatter. The board is permanent: it never resets, never ends, and nothing is archived. No performance figure appears on it (no alpha, no Treynor, no return), and it awards no prize; every award belongs to the Alpha Cup. You appear only after you join it, and leaving takes you off it.
Featured investors (13F portfolios)
Featured portfolios (Berkshire Hathaway, Pershing Square, Appaloosa, Himalaya Capital, TCI, Tiger Global and the Gates Foundation Trust) are rebuilt quarterly from each manager's latest SEC Form 13F filing. A 13F reports long positions in securities on the SEC's official 13(f) list, mainly US-listed stocks and ETFs plus listed options; short positions, cash, most fixed income and shares listed only outside the US are not reported. PortLens counts the filing's common and ordinary share positions. Options, depositary receipts (ADRs) and fund positions are set aside, and positions we cannot price against current market data are skipped. Values, where shown, apply current prices to disclosed share counts, so they differ from filing-date values.
Known limitations
- International ETFs resolve their holdings about as completely as US funds do, but their sector coverage is lower: sectors come from SEC industry codes, and most companies inside an international fund are not SEC registrants. Such a fund contributes the share it can classify, and the sector chart states how much of your holdings it shows. The coverage ratio shown on every scan is the honest denominator for look-through itself.
- Metrics need price history: new listings and thinly traded assets can fall below the minimum-observation thresholds and are excluded from the affected metric rather than guessed.
- All analytics are computed from publicly available market data. Nothing here accounts for your taxes, fees, or personal circumstances.
Questions
PortLens is built by Finsight Labs Ltd. If you think a number is wrong, or a limitation isn't disclosed clearly enough, email support@finsightlabs.co. See also About and Disclosures.
