← Analysis page  ·  David Hay hub  ·  Research hub

Actionable insights — Friday POW!: Vistra Corp (VST)

The repeatable analysis behind the pick: not what was bought, but how to tell a hedge-accounting distortion from a business problem, how to value a company on guidance that deliberately excludes its own announced catalysts, how to demand observed data instead of projections when underwriting a demand thesis, how to test a bear case on joint probability, how to use free-cash-flow yield on a normalised base as the entry metric, and how to tell an evolving thesis from a broken one — written so each step can be rerun on the next name where the reported number and the economic reality disagree.
2026-AUG-14 · Haymaker (Substack newsletter, paid) · The Haymaker Team / David Hay · ↗ Read · full analysis · article text
How to read this page: each insight is a method — the reasoning chain that took Haymaker from a $1.4 billion revenue shortfall to "we are buyers of VST at today's price." The boxed line shows how it played out in this pick. (Written newsletter — the "read" link opens the source post, and there are no timestamps.) Note the structural rhyme with the Aug-7 GILD insights: the same accounting-artifact discipline and the same all-legs-must-hit bear test, applied to a completely different distortion.

1. Separate a hedge-accounting distortion from an operating problem — and check whether the distortion's cause is itself a strength

The repeatable method
  1. When a company with a forward-selling business model (power generators, miners, airlines, commodity processors, insurers) misses on a reported line, ask first whether the miss sits in unrealised mark-to-market on contracts that have not settled. Those swings are revaluations of promises, not cash.
  2. State the distinction in both directions so you cannot be accused of hand-waving: the number is real in the GAAP sense and irrelevant in the economic sense. GAAP is correctly reporting a revaluation; the revaluation is not information about the business.
  3. Find the undistorted line in the same release and use it instead. Segment or generation-level EBITDA, volumes, and physical utilisation sit below the mark-to-market layer.
  4. Now run the step most analysts skip: ask what created the distortion. If the answer is a hedging program, the same program is delivering earnings certainty — the optical cost and the fundamental benefit are the same fact. That converts an apparent negative into a positive without any new information.
  5. Quantify the certainty: get the hedge coverage percentage by year. Near-100% coverage in the current year means the reported guidance is close to arithmetic; declining coverage in outer years is where the real risk (and the unhedged upside) lives.
  6. Confirm the losses actually reverse: unrealised marks on a hedge that will be physically delivered into settle out over the contract's life. If instead the hedge is speculative or the underlying may not be produced, the loss is real and this method does not apply.
Here: a Q2 revenue miss of $4.02B versus $5.46B consensus — "It is real in the GAAP sense and irrelevant in the economic sense… Vistra's revenue figure includes unrealized mark-to-market losses on forward hedges that will settle favorably in future periods." The reversal of the argument: "the hedge program creating the apparent miss is the same driver that gives the company near-complete 2026 earnings certainty" — ~100% of 2026 generation hedged, 94% in 2027, 72% in 2028. The undistorted line: generation EBITDA $994M, +68% YoY, on a fleet at 97% commercial availability.
Watch for

2. Value the company on guidance that excludes its own announced catalysts — the "not in the numbers" gap

The repeatable method
  1. Read the guidance footnotes and exclusions before the guidance itself. Management routinely omits pending acquisitions, unsigned or not-yet-effective contracts, new ventures and pending tax credits, because accounting conservatism forbids including them until they close.
  2. List each exclusion with its size, mechanism and timing — the acquisition's price and megawatts, the contract's capacity and effective date, the venture's committed capital. Vague optionality is not an exclusion; a signed, dated, sized item is.
  3. Size each one's earnings contribution separately and be explicit about the assumption used (a comparable multiple on the purchase price, a per-unit margin on contracted volume). Show the range, not a point.
  4. Build the forward base by adding them to the guided midpoint at their expected effective dates, then state the resulting outer-year number as a single figure you can be judged on.
  5. Compare that base to the current market capitalisation and ask the diagnostic question: is any of it in the price? A multiple struck on excluding guidance is, by construction, the most conservative multiple available.
  6. The whole method only works when the exclusions are announced and contractual. Applying it to pipeline, ambitions or "TAM" is how the same reasoning becomes a trap.
Here: FY26 guidance of $6.8-7.6B EBITDA "explicitly excludes: Cogentrix acquisition ($4.7B, ~5,500MW, pending H2 2026 close); Meta nuclear PPAs (~2,600MW PJM); Helix Platform; nuclear production tax credits." Sized individually — "Cogentrix at $4.7 billion likely adds $470 to $590 million in annual EBITDA at close… the Meta PPA uplift begins flowing through as contracts reach effective dates… Helix generates contracted revenue as the partnership deploys." Conclusion: "the 2026 guidance midpoint of $7.2 billion… is the starting point, not the destination… the 2028 trajectory almost certainly exceeds $9 billion annualized. At a $58 billion cap, the market is pricing none of it."
Watch for

3. Underwrite a demand thesis on observed data, not forward projections

The repeatable method
  1. For any thesis that depends on a demand boom, ask what would count as evidence it is already happening — as opposed to evidence that people expect it to happen. Order books, forecasts and capex announcements are the latter.
  2. Find the industry's real-time operational series: grid peak load for power, throughput for pipelines and ports, utilisation for fabs and fleets, occupancy for property. These are published, high-frequency, and cannot be talked up.
  3. Prefer a series that sets a record — a new all-time high in a physically-constrained metric is unambiguous in a way a growth rate is not.
  4. Pair the demand series with a supply-side series the company controls (availability, uptime, capacity factor). Record demand plus proven delivery is the pair that gets paid; record demand alone might accrue to someone else.
  5. Say explicitly which projection you are declining to rely on, so the thesis is not quietly resting on it anyway.
Here: "PJM and ERCOT set all-time peak loads of, respectively, 168 and 91 gigawatts in July. That is the structural AI demand proof the thesis requires, visible in real-time load data rather than forward projections." Paired with the delivery series: a fleet that "maintained 97% commercial availability during record summer heat in both ERCOT and PJM."
Watch for

4. Test whether a thesis is evolving or breaking — relocate the moat when the original one is regulated away

The repeatable method
  1. When a credible bear piece appears, concede its correct premise explicitly and in its own terms. Refusing the concession costs credibility and, more practically, prevents you from finding out what survives it.
  2. Name the mechanism the bear is attacking. Here it is scarcity rents: the excess price an owner of scarce capacity extracts when demand outruns supply — the leg regulators can and do cap.
  3. Ask what the regulation cannot change. Physical lead times, permits already granted, interconnections already built, and existing customer relationships are not policy variables.
  4. Restate the thesis in the new terms as a before-and-after sentence. If you can write "the market is shifting from X to Y" and the company still wins under Y, the thesis evolved. If you cannot, it broke — and the honest move is to say so.
  5. Check whether the regulatory response creates a substitute revenue channel. Rules that cap merchant pricing frequently push value into bilateral contracts, uprates and brownfield expansion — which favour incumbents with existing sites.
  6. Verify the company is already moving that way, using announced initiatives as evidence. A thesis evolution that management has not begun executing is your own hypothesis, not theirs.
Here: against the WSJ's "Heard on the Street" piece — "the author is right that the easy version of the power thesis has weakened to a degree… regulators in ERCOT and PJM are trying to prevent existing plant owners from capturing unlimited scarcity rents." What regulation cannot fix: "the bottleneck is more than just policy, it is the physical ability to deliver reliable megawatts. New gas plants, nuclear capacity, transmission, transformers, and interconnections still take years to build." The restatement: the market shifts from "own existing power and enjoy higher prices" to "own existing power and be one of the few companies capable of expanding reliable supply" — "the thesis is evolving, not breaking." The evidence it is already under way: "Vistra's Meta, AWS, Cogentrix, and Helix initiatives are already moving it in that direction."
Watch for

5. Read who is partnering with the asset owner as an independent signal of scarcity

The repeatable method
  1. When a venture is announced, ignore the press release framing and ask what each partner is contributing and what each one lacks. The lack is the information.
  2. A structure in which capital, technology and a sovereign mandate all assemble around one physical asset tells you that asset is the binding constraint — otherwise the money would have gone straight to building a substitute.
  3. Distinguish a partnership from an offtake contract. A PPA buys output at a price; a capital-committing platform makes the generator a developer with an equity interest in future capacity — a different and larger claim on the opportunity.
  4. Weigh the committed capital against the company's own size to gauge how much of the thesis is execution risk versus already-earned position.
  5. Check whether the venture's economics are in guidance. If they are not, this insight compounds with the "not in the numbers" method above rather than duplicating it.
Here: the Helix Platform — "NVIDIA provides compute expertise, KKR provides capital and deal sourcing, and KIA provides a long-duration investment mandate. Vistra supplies the one thing none of them can build quickly: 24/7 dispatchable, carbon-friendly nuclear and gas generation in the markets where AI data centers are being built." Vistra commits up to $1B as "preferred power partner," and "none of the revenue from Helix is in any current guidance." Alongside pure offtake — an AWS PPA ~3,800MW at Comanche Peak and ~2,600MW to Meta — the platform is the structurally different item.
Watch for

6. Use free-cash-flow yield on a normalised base as the entry metric — and show the conversion assumption

The repeatable method
  1. For a capital-intensive business, prefer FCF yield to an earnings multiple as the entry metric: it nets out the maintenance capex that an EBITDA multiple hides.
  2. Compute the current-year yield first off guided FCF and today's market cap — that is the return if nothing improves, and it is the floor of the argument.
  3. Then compute the normalised yield: take management's own stated EBITDA-to-FCF conversion target, apply it to the outer-year EBITDA base built from the excluded catalysts, and divide by today's capitalisation. Using today's cap is the point — it states the return on the price you can pay now.
  4. Say the conversion rate out loud ("over 60% of adjusted EBITDA to FCF") so the reader can substitute their own and see how sensitive the answer is.
  5. Cross-check the yield against the EV/EBITDA multiple and against a close comparable — two independent measures pointing the same way is what makes the discount credible.
  6. Treat the sell-side target as a sentiment reading rather than a valuation, and quote the count of analysts behind it.
Here: "FCF yield is the most compelling entry-point metric: 2026 adjusted FCF guidance midpoint of $4.325 billion… is approximately 9.1%." Then normalised: "management's medium-term target of converting over 60% of adjusted EBITDA to FCF, applied to a 2028 EBITDA base approaching $9 billion… implies normalized annual FCF of approximately $5.4 billion. That would represent an 11.4% yield on current market cap before any re-rating." Cross-checked at 10.6× forward EV/EBITDA vs Constellation's 13.6× trailing, with 20 analysts at Buy, avg target ~$217. (Note the post's own inconsistency: the yield arithmetic uses a ~$47.5B cap while the highlights cite ~$58B — always re-derive the denominator yourself.)
Watch for

7. Bound a bear case by requiring every leg to hit — then say what would make you wrong

The repeatable method
  1. Enumerate the bear case as discrete named risks and concede each is "credible to some degree." A bear case you have not stated cleanly is one you have not tested.
  2. Ask whether the current price is only justified if all of them occur, or whether one alone suffices. If all three are required, the price is being set off a joint probability materially lower than any single leg.
  3. Express the conclusion as a threshold claim — "we think that threshold is too high" — rather than a forecast; it survives being early and does not require predicting outcomes.
  4. Check independence before relying on the argument. Legs driven by the same underlying cause are not separate risks, and the joint-probability framing quietly fails.
  5. Handle the category risk (the one that invalidates the whole sector, not just the name) separately and quantitatively: not "will it happen" but "how much slower, and does the thesis survive that rate?"
  6. Do not let the argument absolve the technical picture. State the chart honestly even when it disagrees with the fundamentals, and attach a mechanical exit so the position does not depend on re-arguing the thesis under stress.
Here: three legs — ERCOT forward-curve softness compressing unhedged 2027-28 margins, Helix's $1B exposed to milestones "in an unfamiliar infrastructure-development business," and PJM IRAS capacity-framework uncertainty — with the verdict "the bear case is justifiable only if all three occur simultaneously… we think that threshold is too high." The category risk is answered by rate, not by denial: the build-out "could slow meaningfully, but it is almost certain to continue at a rapid clip, just not as fast as is now projected — the hyperscalers view not keeping pace with their peers as an existential risk." And the chart is conceded against the pick: "VST has been trading below the 200-moving average for over a year. This isn't ideal… to be on the safe side, you could put in a stop at 130."
Watch for

Methods distilled from the paid Haymaker newsletter (text in transcript.txt). For personal study. Not investment advice. © Haymaker / David Hay for source material.