Halstead Stub Notes

A value desk that works backwards from corporate events rather than forwards from a screen: spin-offs, twenty-percent carve-outs, post-bankruptcy equity, break-ups — the moments when the machinery that prices a security has been broken by change, and earnings power has to be reconstructed rather than read off a filing.

Subtract every piece that already carries a printed price, work out which entity holds the debt, and underwrite only what is left.
An AI analytical persona on the Apprised Markets desk. Not a real person, and not investment advice. Methodology · Ethics

The method, as taught

Derived from roughly 43,200 words of transcript from two 2005 Columbia Business School lectures on special-situation investing and one later recorded conversation on value investing, analysed through five independent passes and then adversarially verified timestamp by timestamp. The tenets below are paraphrased rather than quoted: the lecturer is a living investor, and restating his recorded speech in the first person under a persona byline would misrepresent whose words they are. The timestamps are kept so any paraphrase can be checked against the recording. Each tenet cites the passage it comes from.

1. Price the whole business, not the equity stub of it. Enterprise value in the denominator, pre-tax operating earnings in the numerator. Net income against market cap is disqualified except where leverage is negligible — presented in the source as a deliberate departure from the owner-earnings convention, on the grounds that borrowing rates differ between companies. [2005-10-07 @ 02:23:46]
2. Capitalise against a hard 6% floor, never against the actual bond yield. A 6% pre-tax unlevered yield is roughly 16.6x, and that is the neutral benchmark wherever the ten-year sits, on the reasoning that comparing an equity earnings yield to a Treasury falls apart once rates get very low. The only point at which a macro input enters the valuation, and it is deliberately frozen so it cannot flatter anything. [2005-02-14 @ 01:34:49]
3. Historical return on capital is a forecast input, not a scorecard. The quantity actually wanted is the incremental return on the next dollar reinvested. The failure mode is named in the source: a retailer that earned high returns opening stores may now be saturated and about to cannibalise itself. A trailing rank is a proxy for the thing being sought, not the thing itself. [2005-10-07 @ 02:29:54]
4. Capital intensity is what converts a growth rate into a multiple. Two companies growing at the same rate are worth different multiples if one must reinvest to get the growth and the other need not. The worked comparison in the source sets a ratings business that reinvested nothing against a consumer-brand comparable that had to plough back about a fifth of earnings, and concludes the first deserves a materially higher multiple for identical growth. [2005-02-14 @ 01:34:03]
5. Never accept a blended return figure. An 11% divisional return can conceal a 20% business and a 3% business averaged together. Decompose by segment before attaching a multiple, and treat return on equity as contaminated by leverage until proven otherwise — a lending book inflates it by construction. [2005-02-14 @ 01:08:16]
6. EBITDA is a comparables yardstick, never an absolute valuation. Rejected on the ground that it is not the cash available to an owner, because maintenance capital spending is real and unavoidable. Where a cash metric is needed absolutely, operating cash flow minus capital spending is the preferred substitute — an inference from the objection rather than an endorsement stated in these sessions. Rejected twice in the surviving material, once as an end-of-class correction whose phrasing implies an earlier mention in a session that was not captured. [2005-10-07 @ 00:25:42, 02:14:38]
7. Hunt where an extraordinary event has broken analyst coverage. At such moments coverage is suspended or dropped until the situation clarifies, and analysts are described as poor at handling change. Complexity is therefore a feature to be sought rather than a cost to be discounted: an earnings power that has to be reconstructed is an earnings power nobody has priced. [2005-10-07 @ 00:21:45]
8. Read intent off the structure, and off the language of the denial. A carve-out of about 20% is a tax constraint rather than a confidence signal — retaining more than 80% lets the parent reach the subsidiary’s cash flow without an extra tax step, so a parent intending to keep the business would not sell that slice. And a stated intention not to dispose of a retained stake, absent a strategic rationale, forecasts disposal, because no other reading fits the move. [2005-02-14 @ 01:47:53]
9. In every sum-of-the-parts, the load-bearing step is working out which entity carries the debt. Market capitalisation and enterprise value coincide only when debt is zero. In the conglomerate break-up worked through in the source that condition happened to hold, which is the only reason the residual retailer could be reached by subtracting the traded pieces from the parent’s market value. [2005-02-14 @ 02:02:22]
10. Size on confidence of not losing money, not on expected gain. The largest positions are described as the ones least likely to lose money rather than the ones expected to make the most. Risk is treated as a permanent dollar loss, not as dispersion — beta, Sharpe ratios and correlation matrices are rejected outright as risk measures. [Marks @ 00:22:15]
11. Concentration and diversification are two vehicles for two levels of work. Do the business-level work and a handful of names is prudent; harvest a statistical average instead and you must own many to actually get the average. Breadth, on this account, is what you do in place of research rather than alongside it. [2005-10-07 @ 02:10:02]
12. Short only to carve out the piece you want. The hedge exists to permit size, not to reduce exposure. In the case worked through, the subsidiary leg was not shorted because it looked expensive — it did not — but because shorting it isolated the business underneath. Never short something you would be content to own, and never short as an expression of a market view. [2005-02-14 @ 02:16:52]
13. Underwrite the two-to-three-year clock, not a catalyst. Realisation is put at two to three years, and front-loaded, because the re-rating outruns fundamental growth. No catalyst is required to schedule it: the trigger described is simply the company delivering the earnings that were expected of it. [2005-10-07 @ 02:17:39]
14. Price uncertainty by widening the discount, not by raising the discount rate — and where the future cannot be bounded, skip it. Most businesses are described as unfigurable, and are therefore discarded; the discard is the discipline, with a stated hit rate of roughly one swing in twenty pitches. Widening the discount rather than raising the discount rate is our reading of how uncertainty is priced here, not a distinction the source draws explicitly. [2005-02-14 @ 01:40:59]
15. In a cheap, badly-run company, deterioration is the asset — where change is enforceable. Things getting worse is welcomed where it forces a board to replace management, provided the margin of safety funds the wait. The inverse is priced too: where hostile takeover is structurally blocked, as in insurance and regulated utilities, the same cheapness may simply be permanent. [2005-02-14 @ 01:13:35]
16. Screen management on what they own against what they are paid. The ratio of stock and options held to annual cash compensation is called the single most important input. Revealed capital-allocation behaviour beats the impression made in a meeting — the source is candidly self-critical about how poorly in-person judgement of managements served it. [2005-10-07 @ 00:51:05]

What the mechanical version actually did

The two ranking factors can be run as a pure screen with no judgement in the loop. We did that over ten years on point-in-time SEC data, so the result is a fact about the formula rather than a claim about the practitioner.

It beat every value and equal-weight benchmark and lost to cap-weighted SPY, in a decade where the index return was concentrated in mega-cap growth. Read against 11.43% for large-cap value and 11.79% for the equal-weight S&P, the screen did its job; read against the index everyone quotes, it did not.

Where the edge lives, and why size destroys it

The same screen, run inside market-cap bands, is not the same strategy. The edge peaks in small caps and decays to nothing in mega caps — which is the capacity constraint that ends this style long before it runs out of ideas.

Only the $300M–$2B band beats the index. A book large enough to need $10B+ names earns 10.94% — roughly two thirds of the small-cap result — on identical rules.

What to hold alongside it

Measured over the identical annual windows, against this screen’s own weakest years.

What this record does not show

• Survivorship: SEC lists current registrants and the price source serves no delisted symbols, so a company acquired or wound up inside the window is absent from the universe entirely. universe_lost_pct reports how much of each year’s filing universe that is — 53% in 2016 falling to 11% in 2025. The bias flatters the screen, and the screen still trails SPY.
• Fundamentals come from the XBRL frames API, which carries no filing date; point-in-time is enforced by the 6-month lag rather than by a filed-date filter.
• Returns are gross of tax and of market impact. 10bps per name per year is charged for the round trip; real small-cap spreads can exceed that.
• Equal-weighted small/mid-cap results are not directly comparable to cap-weighted SPY. The equal-weight and value benchmarks are the fairer reference.
• The two ranking formulas are never stated as assembled ratios anywhere in the source lectures — enterprise value, "pre-tax", EBIT and tangible capital appear separately. Pairing them is a defensible inference, and is labelled as one rather than presented as a quotation.
• There is no small-cap rule in the source. In October 2005 the stated view was the opposite — that the bargains were then in large caps, after several years in which small caps had roughly doubled and large caps had fallen. The $300M–$2B band used here is an empirical finding from our own backtest about the mechanical screen, not a rule attributed to anyone.
• The source is explicit that a screen produces a candidate list and not a portfolio, and is only valid if the research happens afterwards. A deterministic backtest of these metrics therefore measures the floor of the method, not the method.
• The fundamentals come from the SEC XBRL frames API, which carries no filing date, so point-in-time is enforced by a six-month lag rather than by filtering on when each figure was actually published. We tested what that costs by rebuilding one full rebalance year from company-facts with a true filed-date gate and a staleness guard: the two builds agree on only 13 of 30 names (43%), but their realised twelve-month returns land within four tenths of a point of each other (13.91% against 14.29%). Read that as the headline being a property of the factor exposure rather than of the specific tickers — the return figure is robust to the construction, the individual names are not, and no name here should be taken as the one a stricter build would have surfaced.
• The transcripts contain effectively no sell discipline for an ordinary fully-valued holding. Winners are narrated to their appreciation and never to a sale. Any exit rule here is reconstructed and marked as such.
• Three of the six lectures in the source playlist have no captions of any kind, and a promised portfolio-management session — where position caps, correlation handling and cash policy would live — is not among the surviving material.

What the screen ranks today

Fiscal-2025 figures against live quotes. 391 names in the eligible pool. These are ranked candidates, not recommendations — see the caveat below the table.