Methodology

VerifiedBeta is an independent ETF research project. It is a systematic screen across US-listed and Canadian-listed equity ETFs, using industry-standard factor regressions, risk-adjusted performance relative to the market, and dividend-based relative valuation to support better ETF due diligence.

If you want the broader context before diving into the formulas, start with the About VerifiedBeta page. If you want to see the framework applied rather than described, start with the US ETF Leaders, the Canadian ETF Leaders, or the full US and Canadian universe tables. This page explains the logic behind those rankings and how to interpret the public metrics.

Coverage

Coverage begins with the current US ETF universe and the current TMX Canadian ETF universe. A fund enters the regression universe only after it has established at least 24 monthly total-return observations. Funds that do not yet meet that threshold are tracked as excluded for insufficient history and are not mixed into the leaders tables.

To qualify for a Leaders list, a fund must also earn at least one visible non-market factor score beyond market beta alone.

Separately, VerifiedBeta uses an intentionally loose first-pass equity detector to decide which funds merit a broader factor-space search. That routing gate only asks whether the fund has enough market exposure and statistical significance to justify deeper analysis. It is deliberately inclusive and is not itself a publication-quality stamp.

What We Optimize For

1. Factor exposure that has historically delivered better risk-adjusted results than the market after a realistic allowance for implementation drag.
2. Clean, statistically credible capture of the intended factors rather than incidental noise.
3. Dividend-based relative valuation that compares the current cleaned trailing-12-month distribution yield with the fund's own history.

Score Construction

The public Fund Score is a capped ratio rather than a raw percentile. That matches the code used to build the leaders tables:

Fund Score = min(Relative Sharpe / P95(Relative Sharpe of passing funds), 1.0) * 100

In plain terms, the score is each fund's backtested relative Sharpe scaled to the 95th-percentile passing fund and capped at 100. That works better than a simple percentile because a percentile forces equal spacing between ranks even when the underlying Sharpe differences are small. Scaling to the 95th-percentile passing fund preserves more of the real gap between credible funds while still capping extreme outliers at 100.

Dividend-Based Valuation

Relative valuation is based on trailing-12-month distribution yield, which is our most objective and least gameable valuation proxy at the fund level. For every calendar month, we sum the cleaned distributions paid during that month and the preceding 11 months, then divide by the month-end fund price. This calculation is independent of whether a fund pays monthly, quarterly, semiannually, or changes its payment frequency. We use split-adjusted fund distributions so older payouts are comparable with today's share count, and we exclude identified one-off special or capital-gains distributions from the valuation series because they do not represent durable earning power. Those cash payments still remain in total return. We compare the current cleaned trailing-12-month yield with its own historical median and express the result on a 100 = median scale. Values above 100 indicate the fund is rich relative to its own history; values below 100 indicate it is inexpensive.

Definitions

Process

1. Load the current ETF universe for the relevant market.
2. Exclude funds with fewer than 24 monthly returns from regression eligibility.
3. Run the market's default factor regression.
4. For funds that clear the loose equity-candidate gate, search a bounded factor space across supported families and simple regional blends.
5. Accept an alternate factor family only when the best searched model clears the minimum fit threshold.
6. Use the fitted coefficients to estimate a synthetic all-time factor portfolio.
7. Compare the synthetic portfolio's Sharpe ratio with the market's Sharpe ratio.
8. Score the fund and each significant factor exposure relative to the passing universe.
9. Overlay split-adjusted recurring-distribution valuation and publish the resulting leaders tables.

Models

The US leaders page uses the Fama-French 5-factor model plus momentum (UMD). The Canadian leaders page uses the current AQR Canada factor set: market, size, value, quality, and momentum. Because the Canadian model does not expose separate profitability and investment legs, the Canadian public table uses a single quality score instead.

Currency basis. Canadian-fund regressions are run on a native CAD basis: fund total returns are kept in CAD, the risk-free rate is the FRED 3-month Canadian interbank series (IR3TIB01CAM156N), and the Canadian market factor is AQR's USD-denominated MKT_CAN algebraically inverted to CAD using the same end-of-month CAD/USD series applied consistently across VerifiedBeta's return calculations, ensuring same-day alignment. The four AQR long-short factors (size, value, quality, momentum) are first-order FX-neutral and used as published. We previously regressed Canadian funds in USD basis, which inflated R² by 1–13 percentage points (depending on each fund's FX-correlated exposure) because the FX series sat inside both sides of the regression. The CAD-basis numbers are the un-inflated, structural fit. The same minimum R² threshold (0.85) is now applied to the CAD-basis figure.

Model search and public best-fit regressions. For broader factor-family classification, VerifiedBeta runs a bounded search across supported pure factor families first, then Europe/Pacific regional blends only when a hybrid improves adjusted R² by a material margin. When a fund is successfully classified, its Universe-table metrics and fund profile use the best-fit model instead of the listing-market default. TMX-listed funds are compared in USD-normalized space for the fit-search step so cross-region model selection is not distorted by CAD/USD translation. Public Canadian-equity winners remain on the native CAD decomposition basis; TMX-listed non-Canadian-equity winners use the best-fit USD-normalized regression basis publicly.

Machine-Readable Data

VerifiedBeta also publishes a static API for agents and researchers. The API exposes usable-regression funds, leaders tables, ticker aliases, model metadata, and currently published factor-similar peers as JSON and CSV files. Use the API for scripts and structured workflows; use the Universe page URLs when you want to share a human-readable filtered browser view.

Verified Factor Similarity

Who can be compared. Both ETFs must be US-listed equity funds eligible for a fund page, with the same accepted public factor model, USD currency basis, and factor schema. Each must have at least one statistically significant and material non-market factor. They must share at least one such factor with the same sign in the common regression window. Significance means |t| ≥ 1.96; materiality means the absolute beta exceeds the larger of a factor-specific floor and the 60th percentile among significant funds in that model group (when at least 30 peers qualify). These tests define the comparison universe and descriptive factor tags. The funds do not need identical material-factor sets.

Same dates for both funds. For every supported model group, we regress each ETF's adjusted monthly total return minus the US risk-free rate on its market and non-market factors over the same 36 complete months ending at the latest month in the published results and factor data. A fund with a missing cached return in that window is omitted rather than compared using a different-length history. The matched window and fitted market beta are shown on comparison pages. Published adjusted R² and relative Sharpe still describe each fund's separate full-history public regression; relative Sharpe is not part of the similarity score.

How close is close. Let Δβ be the difference between two fitted factor-loading vectors, including market beta. Let Σ be the monthly covariance matrix of those factors, expressed in percentage-point returns. The modeled annualized volatility of their factor-return difference is sqrt(12 × Δβᵀ Σ Δβ). We call this modeled factor tracking error. We calculate it twice: once using factor returns from the same 36 months as the regressions, and once using 120 months of factor returns ending at the same date. A pair qualifies when both estimates are at most 3% a year. The longer-history check reduces dependence on one recent factor-volatility regime. Both estimates appear beside each match. Neither is an average return difference, a direction of outperformance, or a forecast; the 120-month check uses older factor returns, not ten years of both funds' observed returns.

Which fund actually performed better. Pair pages separately show each ETF's adjusted-price total-return compound annual growth rate (CAGR) over the exact same 36 monthly return observations used to fit exposures. The displayed leader and percentage-point-per-year gap compare those two realized CAGRs. For context, we also multiply each fitted factor-loading difference by that factor's average monthly return over the same 36 months and annualize the arithmetic contribution by 12. The market contribution and combined non-market contributions are shown with a sign from the first listed ETF minus the second. This attribution excludes the regression intercept and fund-specific return, and its arithmetic average does not have to reconcile to the observed compounded CAGR gap. None of these backward-looking figures is an expected return or enters the similarity score.

Score and ordering. We rank qualifying pairs by the larger of those two modeled volatility estimates, from lowest to highest. The displayed 0–100 score is max(0, 100 − (20/3) × larger volatility in percentage points). It is 100 for identical fitted factor exposures and 80 at the 3% admission limit. A score of 90 is a very close factor fit, not a probability that the ETFs are interchangeable. Scores under the previous matching formula are not comparable. Fund pages show up to 12 matches; full cluster pages show every qualifying match, and both ends of a qualifying pair receive a cluster entry.

Published comparison pages remain available. Qualification controls whether a pair appears in current recommendations, clusters, and match lists; it does not control whether an already-published URL continues to exist. If a pair later exceeds the similarity limit or cannot be recalculated on a complete common window, its comparison page remains available and clearly reports its current status. This preserves cited links and search history without presenting a stale match as a current recommendation.

What the estimate misses. All model factors enter the calculation, including factors whose individual t-statistic is below 1.96; significance is not treated as proof that a small estimated exposure equals zero. Factor-return covariance captures both market and non-market risk, including offsetting exposures. The estimate omits return differences unexplained by the factor model, holdings concentration, trading costs, taxes, turnover and fund management rules. Similar holdings can corroborate a pair during individual underwriting, but holdings are not part of this site-wide score. This is a screening tool, not a substitution or investment recommendation. The 3% limit is an explicit screening choice, not a statistical equivalence test or a guarantee of future tracking error.

Why this method. We compared continuous exposure measures using three historical 36-month training windows and each following 12 months of realized fund-return differences. Common windows improved subsequent tracking relative to unequal fund histories; full-factor risk distance did better than simple beta-distance scores in that exploratory test. The test used today's surviving eligible funds and only three forward periods, so it does not establish a stable predictive advantage. A more detailed description of Fama–French market, size, value, profitability and investment factors is available from the factor provider.

FAQ

What does Fund Score actually measure?

Fund Score is the fund's Rel. Sharpe / Backtested expressed as a percentage of the 95th-percentile passing fund and capped at 100. It is a normalized way to compare credible funds without letting a few outliers dominate the top of the table.

What is relative Sharpe?

The site uses two relative Sharpe variants. Rel. Sharpe / Backtested is the fixed-loading factor backtest's geometric average excess return divided by the fund's actual realized lifetime volatility, then compared with the market's Sharpe ratio over the same full-history framework. Rel. Sharpe / Actual uses the fund's realized geometric average excess return and realized volatility over its actual analysis period. In both cases, a value above 1.0 means the fund delivered better risk-adjusted results than the market benchmark used in that framework.

Both Sharpe fields are shown only when Adj. R^2 clears the same minimum quality threshold used by the Leaders screen for that market. Below that threshold, the values are hidden because the factor model is not reliable enough to support a meaningful Sharpe estimate.

Why does alpha matter if factor exposure explains most of the return?

Alpha in this context mostly captures implementation drag that the factor model does not explain cleanly. Fees, turnover, trading frictions, index construction choices, and long-only constraints all tend to show up here, which is why persistent negative alpha is an important warning sign.

How should I interpret dividend valuation?

Dividend valuation compares the current cleaned trailing-12-month distribution yield with the fund's own history on a 100 = median scale. Values below 100 imply the fund is inexpensive relative to its own yield history, while values above 100 imply it is rich. It is not a market-timing tool; it is a cross-check on what price you are paying for a given factor sleeve.

Why are live and backtested relative Sharpe different?

The backtested series is a synthetic factor portfolio built from the fund's fitted exposures across a longer factor-history window. The live figure is the fund's actual realized history. The gap between them is expected because factor cycles, fees, construction drag, and timing of the fund launch all affect realized outcomes.