Actionable insights — Haymaker Daily: Mean Reversion, Mean Indeed

1. Decompose the "E" before you use any multiple built on it

1. Take the reported profit and split it into cash generated by selling things versus gains recognised on holding things — investment mark-ups, warrant revaluations, fair-value adjustments, one-off legal/tax items.
2. For every gain in the second bucket, ask three questions: was cash received? Is the asset publicly priced or marked to a private round? And who set that price — an arm's-length buyer, or a counterparty inside the same commercial loop?
3. Compute the second bucket as a percentage of total profit. That single ratio is the earnings-quality score, and it is comparable across companies and across quarters.
4. Track the ratio's direction over time, not just its level — a rising share of non-operating profit means the operating engine is doing proportionally less of the work each quarter.
5. Re-strike the P/E on the operating bucket alone. If the multiple on that number is unrecognisable, the stock is not cheap; the accounting is.
6. Remember the symmetry: a mark-up booked as income becomes a mark-down booked as loss at the next down-round. The line item is a two-way exposure to somebody else's valuation.

• The next 10-Q's "other income (expense), net" line and the fair-value footnote — specifically a down-round at a major private holding, which converts the same line from the profit engine into the loss engine. Also watch whether management includes or excludes each gain from its own adjusted EPS, which reveals what they think is durable.

2. Cross-check an accounting decomposition with an independent sector strip

1. Having decomposed profit by type (operating vs mark-to-market), decompose the index's growth by sector — a completely different cut of the same data.
2. Remove the two or three sectors carrying the theme and the commodity cycle, and compute what growth remains for everything else.
3. Prefer a named outside analyst's calculation to your own here: it removes the suspicion that the strip was reverse-engineered to reach the conclusion.
4. Compare the residual growth rate to the market's implied growth expectation. The gap between "4%" and what a 20× multiple assumes is the actual risk being run.
5. Treat agreement between the two methods as conviction, and disagreement as a signal you have mis-specified one of them — not as a reason to pick the more convenient answer.

• The ex-energy, ex-AI growth rate published each quarter. It rising toward the headline would mean the boom is genuinely broadening; it staying near 4% while the headline stays spectacular means the index's earnings are still resting on two legs.

3. Treat an extreme print on a mean-reverting series as a base rate, not a trend

1. Establish first that the series actually mean-reverts — profit margins, earnings growth, crack spreads, freight rates — because the same reasoning applied to a trending series (revenue, population, price levels) is wrong.
2. Find the longest available history and locate the current print within it. "Strongest since 1955" is not a bullish datapoint; it is a position on a distribution.
3. Look at what happened after every prior comparable print, and state the base rate plainly rather than forecasting a cause. "Spikes have been followed by equally dramatic declines" needs no theory of what breaks.
4. Do not require a catalyst. Mean reversion is the null hypothesis; a catalyst is only needed to explain the timing, not the direction.
5. Position for it as a probability weight on the payoff, not as a dated call — reducing exposure to the names most levered to the extreme reading rather than shorting the index.

• The first quarter in which year-over-year S&P earnings growth decelerates while the multiple holds — the classic sequencing of a profit peak, and the point at which the "cheap on forward earnings" argument starts working against holders.

4. Use the language people reach for at extremes as a dating device

1. Collect the specific phrases that recur at cycle peaks — "this time is different," "a permanently higher plateau," "a new paradigm," "the old valuation rules don't apply to this business model."
2. When you catch yourself or a consensus write-up using one to dismiss a mean-reversion argument, treat it as evidence for the argument rather than against it.
3. Anchor each phrase to its dated historical instance so the analogy carries a cost: Fisher's plateau was nine days before the 1929 crash, which is what makes it a warning rather than a witticism.
4. Keep the rebuttal narrow. The claim is not that this time can't be different — it is that the burden of proof sits with the person asserting it, and that the assertion itself has a poor track record.
5. Pair the rhetorical tell with a hard number (the earnings decomposition, the ex-sector growth rate) so the argument does not rest on tone alone.

• Sell-side notes and CEO calls explaining why margins or earnings growth are now structurally higher — the modern form of the plateau claim, and the most reliable qualitative marker that the quantitative extreme is late-stage.