Actionable insights — Optimistic On AI, Calm About Yields, But Own Land

11:20 1. Judge a "high" multiple against the earnings growth rate

1. Take the current P/E and the company's stated forward earnings-growth expectation.
2. Ask what the multiple becomes on next year's earnings if the guide is hit (P/E ÷ (1 + growth)).
3. Compare with the historical market multiple, not the stock's own history — a high trailing P/E can be a low forward one.
4. Extend it to the index: if adoption lifts earnings broadly, "high" aggregate valuations may be too low.

• Growth guides being cut — the method only holds while the stated growth is delivered.

7:01 2. Separate rate level exposure from rate spread exposure

1. For a financial, find how it prices client cash and loans: a fixed spread to the policy rate, or a margin that widens with rates?
2. If both sides are pegged to the same benchmark (lend benchmark + x, pay benchmark − x), net interest income tracks balances, not the rate level.
3. Then the rate call is secondary; the drivers are client balances, margin loans and securities-lending demand.

• Client cash and margin-loan balances; any change to the spread policy; securities-lending income.

9:34 3. Worry about the speed of a yield rise, not the level

1. Track long yields on a rate-of-change basis (e.g. weekly/monthly change), not just against thresholds like 5%.
2. Treat a gradual grind higher as absorbable; treat a very sharp sudden jump as the trigger for "all kinds of problems."
3. Pair it with the leverage map: sharp moves hurt where OTC leverage is opaque and positions are offside.

• An outsized one-week move in the 10- or 30-year; stress at a bank or fund with large OTC books.

13:36 4. In a winner-take-most arms race, own the users, not the buyers

1. Identify an input every competitor must over-buy to stay in the race for "one or two winners."
2. Expect a supply glut and falling input prices once the race peaks.
3. Expect the buyers to write down the over-bought capacity; the beneficiaries are downstream users who get it cheap.

• Falling GPU rental/compute prices; hyperscaler impairments or useful-life changes on hardware.

40:55 5. Pick tail-risk assets by what survives seizure and gets returned

1. Name the tail risk and assign it odds (he: 20% for a US democratic-socialist turn).
2. Use a historical analog to ask which assets were destroyed and which were restituted afterward.
3. Prefer assets that can't be ruined or moved (undeveloped/agricultural land); reject assets that can be declared illegal or must be surrendered (gold).
4. Size it as a multi-generational holding for heirs, not a trade.

• Election outcomes that move the odds; wealth-tax or property-rights proposals.

20:24 6. Use event markets to price the environment around a stock

1. List the macro, regional, political and climate questions that determine a company's future.
2. Pull the market-implied odds for those questions from prediction markets where they are liquid (political markets already are).
3. Prefer those odds over polls and over AI forecasts, which can't learn new information after training.

• Liquidity growing in economic contracts (rates, GDP, weather) enough to trust them beyond politics.