BookLev
Long low BookLev · rebalanced monthly · cap-weighted long-short
Total assets (at) divided by book value of equity plus deferred taxes (txditc) and preferred stock. Equity is shareholder equity (seq) if available, or book equity (ceq) plus preferred stock (pstk, if missing pstkrv, if missing pstkl), or total assets minus total liabilities (lt).
Leverage/financing: capital-structure tilts that proxy for risk and financing pressure have historically forecast the cross-section of returns.
Growth of $1 — long-short
Financial Services only · cap-weighted · 331 months
The bottom-bucket figures are that basket’s ownreturn (a long position in what the strategy shorts) — not the short side’s P&L. Long-short return = long leg − bottom bucket, so a falling bottom bucket widens the spread.
’00–’12 · ’13–’18 · ’19–’26windows
Recomputes the split Sharpes and test alpha. Full Sharpe, ann. return, max drawdown and OAP correlation span the whole sample and don’t change.
Worked example
Live inputs for a real holding
| Input | Value |
|---|---|
| assets | 371,082,000,000$ |
| equity | 106,491,000,000$ |
assets / equitysigned_value is the internal ranking value; only its cross-sectional rank matters. $-vs-$M unit mix means ratio magnitudes carry a constant offset and are not comparable to textbook levels.
Replication
How closely this rebuild tracks the published research
The OSAP comparison is only computed for the all-stocks variant — switch to “All stocks” above to see how the rebuild tracks the published series.
Style tilt
Size × value map of the long leg
Share of long-leg capital by market cap × book-to-market, using terciles of the 554-name universe at July 2026. Rows: Large ≥ $4.3B, Small < $753M. Columns: Value ≥ 0.61, Growth < 0.39 B/M. The dot marks the capital-weighted centroid of the book — 98% toward Large, 95% toward Growth. +1% unclassified (4 missing size/value). Hypothetical holdings — descriptive, not realized P&L.
Capital-weighted centroid of the long leg, December 1998 → July 2026; the right-hand end is the same value as the box’s dot. 110 names now. Hypothetical holdings — descriptive, not P&L.
Since publication
Did the edge survive the paper coming out?
There is no “before” to compare against — the predictor was already public when our data begins, so every month shown is out-of-sample relative to the original study.
Split at January of the year after the factor was published. The post-publication stretch IS genuinely out-of-sample relative to the original study -- the predictor was public by then -- so it speaks to whether the effect survived being known. Nothing is fitted here, so this is performance SINCE PUBLICATION, not validation of a model. Note the 'pre' side is our data before publication (our panel starts ~1999), NOT the study's original in-sample period, which usually ran decades earlier; a decay figure compares before-vs-after within our sample and is not a comparison against the published result.
Derivatives risk (Rule 18f-4)
Relative VaR against the designated reference portfolio — as implemented here
This assessment describes the research long-short construction — long the top bucket, short the bottom.
This fund does NOT qualify for the limited-derivatives-user exception (100% of net assets vs <= 10% of net assets), so the relative VaR test governs. Note the inversion: a dollar-neutral long-short book suppresses the market risk the VaR ratio measures and so tends to pass it, while being barred from the exception by construction — what disqualifies the strategy is the exposure threshold, not the risk limit.
Rule 18f-4(c)(4): a fund whose derivatives exposure (gross notional, including the value of assets sold short) is <= 10% of net assets is excepted from the VaR tests and the full derivatives risk management program.
No confidence interval: this follows from portfolio construction, not from an estimate.
Rule 18f-4(c)(2)(i): fund VaR at 99% over 20 trading days must not exceed 200% of the designated reference portfolio's VaR on the same basis.
Within the limit as implementedhistorical method · 331 monthly observations · about 3.3 in the 99% tail · paired percentile bootstrap over months
The estimators agree, which is mild evidence the number isn’t an artefact of one method.
Engineering approximation of SEC Rule 18f-4 for research display. Not a compliance opinion and not a determination that any fund is compliant. Hypothetical model portfolio, not a registered fund.
Consistency across eras
Is this record broadly durable, or one regime?
3-year rolling windows, stepping 1 year (26 windows). Dispersion is the spread of window Sharpes — higher means the record depends more on which era you look at.
Rules-based factors fit no parameters, so these windows are not out-of-sample tests and do not validate a fitted model. They show whether the factor's record is consistent across eras or driven by one regime.
Returns
Long-short is the research line; the long leg is what a long-only fund could hold
| Period | Long-short | Long leg |
|---|---|---|
| 1 yearcum. | -20.8% | -1.0% |
| 3 yearsp.a. | -16.3% | 12.2% |
| 5 yearsp.a. | -8.9% | 8.4% |
| 10 yearsp.a. | -4.2% | 14.1% |
| Since inceptionp.a. | -0.9% | 9.7% |
| Year | Long-short | Long leg |
|---|---|---|
| 20267 mo | -11.6% | -1.4% |
| 2025 | -17.7% | 8.7% |
| 2024 | -12.0% | 23.7% |
| 2023 | -1.3% | 21.3% |
| 2022 | 5.9% | -7.1% |
| 2021 | -13.7% | 16.9% |
| 2020 | 15.6% | 16.2% |
| 2019 | -16.1% | 24.3% |
| 2018 | 36.3% | 5.3% |
| 2017 | 5.1% | 29.7% |
| 2016 | -13.1% | 14.0% |
| 2015 | 0.9% | -1.1% |
| 2014 | 2.5% | 14.1% |
| 2013 | -8.3% | 41.3% |
| 2012 | -7.6% | 25.9% |
| 2011 | 34.3% | 2.4% |
| 2010 | -2.8% | 10.8% |
| 2009 | -10.8% | 23.3% |
| 2008 | 30.6% | -37.8% |
| 2007 | 48.7% | 12.9% |
| 2006 | -2.1% | 22.8% |
| 2005 | -1.7% | 9.6% |
| 2004 | -1.8% | 11.1% |
| 2003 | -11.0% | 24.7% |
| 2002 | -4.4% | -16.1% |
| 2001 | -0.1% | -2.4% |
| 2000 | -19.2% | 4.1% |
| 1999 | -14.5% | 3.5% |
Long-short is the research line (long the top bucket, short the bottom); the long leg alone is what a long-only fund could actually hold. Trailing figures run through July 2026: 1 year is a plain cumulative 12-month return (cum.), 3 years and longer are annualised (p.a.). Years marked with a month count are partial. These match the fact sheet’s tables by construction. Hypothetical backtest, gross of the placeholder expense ratio and trading costs — not investor results.
Methodology
Signal computed monthly from Sharadar point-in-time data, signed so higher = higher expected return. Restricted to the Financial Services sector, then sorted into quintiles within that sector. The virtual ETF is long the top bucket, short the bottom, cap-weighted, rebalanced monthly. FF5 alpha regresses the long-short on the Fama-French 5 factors.
Hypothetical research backtest. Not an offer, recommendation, or investment advice.