Skip to content
Kelly Portfolios
Search

The corpus

The research

35 syntheses and 10 decision records, 45 pages in all, sorted by the question each one answers rather than by filename. Several conclude that the instrument could not see the effect, which is a result and is written as one. Nothing has been removed because it was disappointing.

If you want the short version instead, the six pages in the masthead cover the same ground in plain English, and each page below names the one that covers it.

Start here

The four that carry the argument

The rest of the corpus is evidence for these, or a test that one of these rests on.

What to hold

What should be held, and why that construction

The working position, the investor it was derived for, and the tests that scored it against a cheap index.

  • Current portfolio position Load-bearing

    Given the evidence now held, what is the best-supported implementable portfolio for the reference US investor, and which inputs or new evidence would change it?

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/portfolio-recommendation.md

  • The construction tournament Load-bearing

    Compared as whole portfolios rather than sleeve by sleeve, does the proposed stacked, tilted construction beat a cheap cap-weighted control — and can this data tell?

    Evidence status: Unresolved Last changed 2026-08-24 docs/research/construction-tournament.md

  • The final construction, tested as one object

    Three gaps, and they turn out to have one answer. First: every result behind the recommended portfolio is component-level, and the construction tournament scored the investor's original eight-fund proposal…

    Evidence status: Unresolved Last changed 2026-08-24 docs/research/final-construction-test.md

  • Part A — the portfolio for one investor, derived

    as of 2026-08-23. Working analysis behind /portfolio. Not personalised advice; every figure is a function of stated inputs.

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/portfolio-for-one-investor.md

  • Setting the equity share: what is arithmetic, what is preference, and what neither can reach

    The repository has declared net geometric growth as its objective and has refused to set the equity/bond split. Given both, what can actually be said about setting it?

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/setting-the-equity-share.md

  • Valuation and the allocation: what a CAPE of 41 does and does not license

    US CAPE is at the 99th percentile of 145 years. Should that change the equity share, the US/international split, or the rebalancing rule?

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/valuation-and-the-allocation.md

  • Four tilts the recommendation never priced

    The recommended portfolio was assembled by correcting the investor's own list of eight tickers. It was never asked whether better tilts existed on the shelf. Four candidates were never scored against it: AVDV…

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/untested-tilt-candidates.md

  • Rebalancing: a maintenance policy, and how to run one

    Two questions, and they have different answers.

    Evidence status: Rejected Last changed 2026-08-24 docs/research/rebalancing-policy.md

Stacking

Does piling up more good bets help

Breadth, borrowing, and how each new strategy is paid for. This is where most of the money in the proposal was riding.

  • Stacking: does a pile of 55% bets beat the market? Load-bearing

    The proposition under the whole proposed portfolio, in the investor's words: "if you stack a ton of strategies that are 55% likely to beat the market then you will beat the market." Under what conditions is…

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/stacking-and-effective-breadth.md

  • Independent engines: how many exist outside this repository's own data, and what the stack is worth

    Stacking and effective breadth measures an average pairwise correlation of 0.435 on this portfolio's own active positions, caps an unlimited stack of 55% bets at 57.6%, and ends on an instruction: lower the…

    Evidence status: Source-reproduced Last changed 2026-08-24 docs/research/how-many-independent-engines.md

  • Capital efficiency and accessible diversifiers

    How does the funding rule change the value of a diversifier, and which economically distinct return sources can a retail investor currently access at plausible cost?

    Evidence status: Unresolved Last changed 2026-08-24 docs/research/capital-efficiency-and-breadth.md

  • Leverage and the notional budget: what the candidate portfolio already carries, and what it should

    The investor says "I am okay with leverage but it absolutely must be with purpose and must understand market conditions." Their candidate portfolio puts 30% of capital in a stacked ETF. What leverage is that…

    Evidence status: Unresolved Last changed 2026-08-24 docs/research/leverage-and-the-notional-budget.md

  • Marginal value of a sleeve

    What does a candidate add inside an existing portfolio, and how much of that contribution is standalone return versus diversification?

    Evidence status: Unresolved Last changed 2026-08-24 docs/research/marginal-sleeve-value.md

  • Trend: the index, the products, and a clause that was ambiguously specified

    Three warnings that belong before any number.

    Evidence status: Rejected Last changed 2026-08-24 docs/research/trend-marginal-value.md

  • The trend weight under acknowledged ignorance

    The construction tournament nominated this as its number-one next test: how much trend notional should the investor hold, when three instruments return three different answers and the disagreement is about a…

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/trend-weight-under-uncertainty.md

  • Live stacked funds: what the second dollar has actually paid

    Capital efficiency prices the funding rule and loading comparability measures what a stacked wrapper delivers. Neither asks the plainest question a reader will ask: what have these funds actually returned to…

    Evidence status: Source-reproduced Last changed 2026-08-24 docs/research/live-stacked-fund-records.md

  • Live managed futures: what the funds actually paid, and what the vendor index overstated

    Every trend figure in this repository comes from AQR's published TSMOM series, which states no fee, transaction-cost, slippage or financing basis anywhere. What did real managed-futures funds pay their…

    Evidence status: Unresolved Last changed 2026-08-24 docs/research/live-managed-futures.md

  • Loading comparability, and the wrapper exposure this repository said it could not measure

    Two questions that turn out to have the same answer. First: can a stacked fund's delivered trend exposure be measured at all, when no research-grade price feed is licensed? Second: are the factor loadings…

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/loading-comparability-and-wrapper-exposure.md

Factors

Are the factor returns real, and can a fund deliver one

The academic series, what survives publication, and what a fund an investor can buy actually hands over.

  • Factor persistence and decay: what survives publication in the French series

    Do HML, UMD, RMW and CMA retain a positive, economically meaningful premium after publication — and when the United States window turns out to be too short to answer that, do independent regions supply the…

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/factor-persistence.md

  • Investable factor products

    Do listed factor funds deliver their advertised exposure at a cost and stability that makes them useful implementation candidates?

    Evidence status: Unresolved Last changed 2026-08-24 docs/research/factor-products.md

  • The long-only capture fraction: what a tilt delivers of a long-short premium

    Every factor premium measured here is an academic long-short spread — long the high-scoring portfolio, short the low, zero net investment, gross of every cost. A retail investor cannot hold one. What fraction…

    Evidence status: Rejected Last changed 2026-08-24 docs/research/long-only-capture.md

  • Where outperformance can come from, how large it is, and how certain it is

    Where can a real retail investor's net return actually exceed a stated benchmark, how large is each source, and how certain is each one?

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/expected-edge-decomposition.md

  • Fama–French factor reproduction: the Phase 1 ingestion gate

    Does this repository's data path — download, raw-byte cache, zip extraction, multi-table parsing, percent-to-decimal conversion, sample-boundary selection and monthly statistics — reproduce a precisely…

    Evidence status: Source-reproduced Last changed 2026-08-24 docs/research/fama-french-reproduction.md

Fees and tax

What is contractual: fees, taxes and accounts

The part of the answer that follows from a filing or a tax rule rather than from a backtest, and is the largest reliable line on the site.

  • Structural and tax-aware edges: what else is contractual, and how big is it

    The edge decomposition prices a contractual edge of about 89 bp/yr against the investor's own plausible alternative. What else belongs in that class, how large is it, and does it double-count what is already…

    Evidence status: Source-reproduced Last changed 2026-08-24 docs/research/structural-and-tax-edges.md

  • Tax-loss harvesting and direct indexing: what it is worth, and to whom

    The edge budget books 30 bp/yr for tax-loss harvesting and dismisses direct indexing in a sentence — it "needs direct security ownership that funds do not give". This investor holds roughly a third of the…

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/harvesting-and-direct-indexing.md

  • Currency and the international sleeve: the position nobody chose

    The portfolio holds 35% international — VEA, DFIV, IDMO, IEMG, AVES — every one unhedged. That is a long position in a basket of foreign currencies, funded in dollars, sized at roughly a third of the equity…

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/currency-and-the-international-sleeve.md

Rejections

What was tested and did not earn a place

Each rejection scoped to the data, the window and the yardstick that produced it.

Method

How the evidence was made, and how sure it is

The instruments, their resolution, the review that went looking for errors in all of it, and what is next.

  • Adversarial review of the August 2026 construction session Load-bearing

    Nine agents sharing one repository, one data cache and one house style converged on a portfolio recommendation. Which parts of that position are wrong, overconfident, or an artifact of shared method …

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/adversarial-review.md

  • The evidence base: what this repository can measure, and what it cannot

    What sources does this repository hold, what can each resolve, and what new instrument is needed when the current one cannot answer a decision-relevant question?

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/evidence-base.md

  • Portfolio-edge research framework

    What evidence would justify changing a cheap, diversified portfolio for a stated investor?

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/portfolio-edge-research-framework.md

  • Numerical engine specification

    What should this repository's portfolio mathematics compute, with which algorithms, and where should that computation run?

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/portfolio-engine-specification.md

  • Market scan, August 2026

    An outside-in check on what has changed in the real world, and what this repository's audited shelf and research corpus would now get wrong. Written for the public website rewrite.

    Evidence status: Source-reproduced Last changed 2026-08-24 docs/research/market-scan-2026.md

  • Research agenda: where a better test can change the portfolio

    What should be researched next, given the current portfolio decision and the limitations of the instruments already used?

    Evidence status: Exploratory Last changed 2026-08-24 docs/research/search-coverage.md

Decisions

The decision records

Why a durable choice was made, what it rests on, and what would reopen it. The syntheses above cite these about fifty times, so they are rendered here rather than left on GitHub.