1. Value a listed operator's ignored private stakes separately — the "venture company inside a large-cap wrapper"
The repeatable method
- When a listed company also owns equity stakes in private businesses, value the two pieces separately — the operating business on its own multiple, and each private stake at its most recent funding-round valuation × the ownership %.
- Cross-check that sell-side price targets and consensus estimates are capturing the stakes. If analysts model only the core business, the stakes are "free" optionality the market is under-counting — that's the mispricing.
- Confirm the balance sheet lets the operator keep scaling without diluting you (net-debt-neutral, or asset-backed debt tied to contracted cash flows beats serial equity raises). Then size for the volatility (a high-beta name gets a smaller position).
Here: NBIS — the NeoCloud core plus ~$8B of private stakes the Street ignores: ClickHouse (28% × ~$15B = ~$4.2B), AV-Ride (83% × ~$4B = ~$3.3B), Toloka (50% × ~$500M), TripleTen (100%). Net-debt-neutral (vs CRWV's ~$40B debt), de-risked via a $775M asset-backed facility; consensus buy $244-265 vs a $177 stock. "Wall Street doesn't focus on this enough." Beta >2 → size small.
Watch for
- Private stakes valued at a recent round; sell-side models that ignore them; a self-funding balance sheet (net-debt-neutral / contracted-cash-flow debt); and high beta demanding a smaller position.
2. When a gauge looks "too calm," map where the shock's risk actually lives instead of calling complacency
The repeatable method
- When a headline volatility gauge stays low through a scary event, don't assume the market is asleep. Ask which specific risk that gauge measures — and whether this shock even threatens that risk.
- Go find where the premium is being paid: check the other vol surfaces (oil vol, rates vol, single-name vol), the term-structure shape (contango/skew), and cross-asset moves. A shock routed "around" the index shows up as high sector/single-name vol with low index correlation.
- Treat repeated episodes as the market "Bayesian updating" — pricing the transmission mechanism, not the headline. Then define the one channel that would break the pattern and monitor that single variable.
Here: the low VIX (15-17) through the Iran restart is not complacency — the war's risk sits in oil (OVX +39%), rates (2-yr to a 15-mo high) and single-name vol (Nasdaq-100 vol 27 vs VIX 16), with one-month implied correlation at a two-year low. The falsifiable edge: oil to ~$110 forcing the Fed hold→hike flips dispersion into correlation. "Watch the two-year yield, not the tanker count."
Watch for
- A low headline gauge amid a real shock; the premium showing up in oil/rates/single-name vol + low implied correlation; a steep contango + bid skew (not "flat and cheap everywhere"); and the one channel (rates/inflation) that would convert dispersion into correlation.
3. Run the Jevons-paradox test before assuming "cheaper models = less chip demand"
The repeatable method
- When a cheaper substitute threatens a value chain, don't stop at "prices fall, revenue falls." Trace the money the buyer saves and ask how much of it gets reinvested into more usage.
- Solve for the reinvestment (reuse) rate at which the key beneficiary breaks even, then compare it to how budgets are actually behaving. If real budgets are growing far faster than the breakeven, the "commodity" shift is net-positive for the picks-and-shovels layer.
- Corroborate with the borrowing tell: if the spenders are funding capex with debt (widening credit spreads), demand is real but the financing risk is the thing to monitor, not the demand.
Here: cheaper open/Chinese AI models look bearish for chips — but the switch turns a developer pass-through into a direct hyperscaler sale, and a 32% reinvestment rate is breakeven while budgets grow 5-7×/yr → more compute + memory demand (a Jevons's paradox; "bananas made of gold vs Chiquita bananas"). The tell to watch: hyperscaler G-spreads >150 as they borrow to spend. Supports NBIS, MU, SNDK.
Watch for
- The buyer's saved dollars and the reuse rate; the beneficiary's breakeven reinvestment threshold vs actual budget growth; open-model share migration (OpenRouter token mix); and widening hyperscaler credit spreads as the financing-risk flag.
4. Merger-arb: read the deal multiple to judge whether a topping bid is coming
The repeatable method
- On an announced bid, don't just measure the premium to the prior close — value the target on its own cash-flow yield to the deal price and its multiple vs its own history and peers.
- If the bid implies a high free-cash-flow yield to the acquirer and a multiple below the target's multi-year average, the deal is cheap — which invites a competing/topping bid rather than a clean close.
- Position for the topping-bid path (upside above the announced price), and sanity-check the financing structure (all-cash vs highly-levered) for deal risk.
Here: PYPL +20% on a Stripe/Advent ~$60 bid — but ~$6B FCF next year = a ~10% yield to the deal and 8× EBITDA is ~1-1.5× below PayPal's 3-yr average → "there could be a topping bid... mid-60s." Contrast the housekeeping-only realized deal: ATAI +33% on Eli Lilly's definitive $6.75 + $2.50-CVR agreement (done, not contested).
Watch for
- The FCF yield to the bid; the deal multiple vs the target's own history + peers; a cheap multiple as the topping-bid trigger; and the financing structure (leverage) as the deal-risk read.
5. Pressure-test a featured thesis with an expert call before you size it up — and downgrade honestly
The repeatable method
- Before scaling a special situation, commission an expert/industry call to stress the parts a model can't see — customer concentration, contract durability, competitive wins/losses, regulatory path.
- If the call surfaces a real negative (e.g. revenue leans on a few customers), revise the upside down and de-rate your conviction even if the headline value gap is still positive — don't anchor to last week's thesis.
- Separate "still cheap to fundamentals" from "big upside": a name can stay a hold (cheap, but capped) rather than a featured buy once the risk is priced. Keep the position small until the binary (here, deal approval) clears.
Here: VISN — an expert call (Zayo's Carlos Treves) mapped the asset sales to ~19.5% stock / ~32% stub / ~45% business-value upside at ~3-4× EBITDA vs 6-7× fair — but found "quite a lot of customer concentration," so "not a tremendous amount of upside." Downgraded from last week's featured long to a cheap-but-capped hold, awaiting government approval.
Watch for
- A featured thesis before you add; an expert call testing concentration/contracts/regulation; a real negative that caps the upside; and the discipline to hold-not-add (small size) until the binary clears.
6. Regulatory-scare test: does the new rule actually bind the existing asset, or only new builds?
The repeatable method
- When a stock drops on a new regulation/moratorium, read the rule's scope precisely: most target new development, not vested, already-operating assets ("grandfathered").
- Separate the layers — a state-level permit freeze may be a non-issue while a local ordinance that explicitly covers "expansions" is the binding constraint; check whether the company's planned upgrades trip the expansion definition, and whether a hardship-exception path exists.
- If existing operations continue and the asset is scarce, treat the selloff as an overreaction — but keep it a small, risk-managed position given the genuine open question.
Here: DGXX — NY's Hochul froze new 50MW+ data-center air permits, but DGXX's ~140MW is grandfathered (existing crypto-mining upgrading to tier-3 AI). The local North Tonawanda moratorium (covers expansions) is the more relevant risk, but ops continue + a hardship path exists → "not a clean all-clear, but not the outright ban some bears portray." 140MW of controllable power stays strategic.
Watch for
- Whether the rule targets new builds vs existing assets; the state-vs-local layer that actually binds; whether planned upgrades trip an "expansion" clause; and a hardship-exception path — with small sizing while it's open.
7. Harvest new-issue baby bonds for double-dividend windows in the income book
The repeatable method
- Track recently-issued baby bonds (small $25-par unsecured notes) from issuers whose common you already understand — a fresh 10% unsecured coupon is attractive in a high-rate tape.
- Time the entry around the ex-dividend calendar: buying before an x-date can lock in two near-term coupon payments in quick succession, front-loading the yield.
- Keep the bond and the common as distinct trades — the bond is the income sleeve (buy near par, clip the coupon), separate from any deeper-discount equity dislocation in the same name.
Here: RWTS — Redwood's new end-May baby bond ~9.8% at ~$24.90; buying before the Aug-14 x-date captures ~60¢ (Aug 14) + ~60¢ (Nov 13) = ~$1.20/share. Kept separate from the RWT common (bought at a 42% discount to book / 17% yield).
Watch for
- A new-issue ~10% unsecured baby bond from a known issuer; an x-date that stacks two near-term coupons; near-par entry; and keeping the bond distinct from the common's own dislocation.
8. Screen speculative growth names by cash-runway, not the chart
The repeatable method
- For a beaten-down story stock, ignore how oversold the chart looks and compute the runway: annual cash burn vs cash on hand, and the year positive operating profit (EBITDA) is even forecast.
- If the company burns through its cash within a few years and profitability is a half-decade out, the selloff isn't a bargain — dilution/financing risk dominates. Prefer real, cash-generative names in the same theme.
- Rank within the theme by quality: own the profitable incumbents over the pre-revenue story, even when both fell together in the momentum crash.
Here: JOBY at ~$7 — burns ~¾B/yr, runs out of excess cash in ~3 years, no positive EBITDA until past 2030 → won't buy the chart. Prefers the big defense primes (LMT) and AVAV among the beaten-down growth names.
Watch for
- Annual burn vs cash on hand; the year positive EBITDA is forecast; dilution risk over "oversold" charts; and profitable incumbents ranked over pre-revenue stories in the same theme.
9. When supply is delayed, model a cycle extension, not a peak — and trade the washout accordingly
The repeatable method
- When a "the cycle is peaking" call lands (e.g. memory), check the supply side: if a large share of planned capacity is delayed (power, equipment, permits), the demand doesn't vanish — it slides into later years and extends the cycle.
- Weigh the disconnect: a peak call that ignores un-built capacity is likely early. Use the resulting selloff as a trading opportunity in cheap-looking names, not a buy-and-hold, since earnings/margins may be temporarily inflated.
- Gate the trade on the macro releases (rates peaking, war de-escalating) that would let the beaten-down names bounce.
Here: Morgan Stanley says memory pricing peaks this year; Singh disagrees because ~70% of 2027 data-center builds are delayed → higher build pace in 2028, so demand "could be much higher going into 2028." He treats memory/AI names (MU, SNDK, NBIS) as short-term trades off the washout — "a trading opportunity, not a buy-and-hold" — gated on rates peaking + the war ending.
Watch for
- The share of planned capacity that's delayed; a peak call that ignores un-built supply; possibly-inflated current margins (trade, don't hold); and the rate/geopolitical gates for the bounce.
Methods distilled from the premium Special Situations Report weekly call (2026-07-19; transcript, report & deck PDFs in this folder; notes in transcript.md) for personal study. Not investment advice. © Special Situations Report for source material.