ChInv
Long low ChInv · rebalanced monthly · cap-weighted long-short
12 month change in inventory (invt) divided by average total assets.
A investment signal from the cross-sectional asset-pricing literature (Open Source Asset Pricing).
Growth of $1 — long-short
Consumer Cyclical only · cap-weighted · 331 months
The bottom-bucket figures are that basket’s ownreturn (a long position in what the strategy shorts) — not the short side’s P&L. Long-short return = long leg − bottom bucket, so a falling bottom bucket widens the spread.
’00–’12 · ’13–’18 · ’19–’26windows
Recomputes the split Sharpes and test alpha. Full Sharpe, ann. return, max drawdown and OAP correlation span the whole sample and don’t change.
Source study
Thomas and Zhang · 2002
Review of Accounting Studies
Worked example
Live inputs for a real holding
| Input | Value |
|---|---|
| inventory | 6,747,000,000$ |
| inventory (t−12) | 6,269,000,000$ |
| assets | 371,082,000,000$ |
(inventory - inventory_l12) / assetssigned_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 310-name universe at July 2026. Rows: Large ≥ $5.9B, Small < $1.3B. Columns: Value ≥ 0.62, Growth < 0.38 B/M. The dot marks the capital-weighted centroid of the book — 86% toward Large, 32% toward Growth. +7% unclassified (9 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. 61 names now. Hypothetical holdings — descriptive, not P&L.
Since publication
Did the edge survive the paper coming out?
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.
Not settled by the dataThe interval spans the limit. The point estimate is 1.39x but the 95% interval runs 0.75–2.22x, which includes 2.00x. This test does not settle which side of the line the fund is on.
historical 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. | -5.9% | -1.9% |
| 3 yearsp.a. | -4.7% | -3.9% |
| 5 yearsp.a. | -7.2% | -0.6% |
| 10 yearsp.a. | -9.4% | 6.6% |
| Since inceptionp.a. | -4.1% | 9.3% |
| Year | Long-short | Long leg |
|---|---|---|
| 20267 mo | -7.4% | -8.2% |
| 2025 | -21.0% | -15.0% |
| 2024 | 9.5% | 6.4% |
| 2023 | -9.5% | 54.2% |
| 2022 | 0.9% | -23.9% |
| 2021 | 11.0% | 28.0% |
| 2020 | -39.6% | 6.3% |
| 2019 | -9.8% | 27.2% |
| 2018 | -8.7% | -9.2% |
| 2017 | -6.5% | 20.9% |
| 2016 | -9.2% | 13.6% |
| 2015 | -6.0% | -6.9% |
| 2014 | -8.6% | 5.7% |
| 2013 | -9.8% | 42.9% |
| 2012 | -3.0% | 25.3% |
| 2011 | -0.5% | 6.0% |
| 2010 | 7.0% | 33.3% |
| 2009 | -8.9% | 60.3% |
| 2008 | 16.9% | -35.2% |
| 2007 | -21.4% | -24.1% |
| 2006 | 15.8% | 20.3% |
| 2005 | 7.4% | 5.8% |
| 2004 | -3.4% | 14.3% |
| 2003 | -10.8% | 40.1% |
| 2002 | 19.8% | -1.1% |
| 2001 | 12.9% | 32.0% |
| 2000 | 6.6% | -7.7% |
| 1999 | -10.5% | 15.2% |
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 Consumer Cyclical 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.