Case study: Micron, the cyclical

Some businesses are boats and some are tides. A cyclical company's results are dominated by an industry cycle it does not control, and analyzing one with growth-company tools produces confident, systematically wrong answers. Micron, one of three companies on earth making DRAM memory chips, is the purest specimen: a genuinely well-run company whose earnings can swing from record profit to loss and back inside three years, on schedule.

The short version

Cyclicals invert the P/E rule: they look CHEAPEST at the peak (huge trailing E about to collapse) and most EXPENSIVE at the trough (tiny or negative E about to explode), so buy-low-P/E logic buys tops. Professionals value cyclicals on mid-cycle earnings and on price-to-book against history, track SUPPLY (capex cuts, inventory, competitor discipline) rather than demand headlines, and buy when conditions are terrible but improving at the margin. The risk that never leaves: telling an ordinary downcycle apart from a structural break, and leverage decides who survives long enough to find out.

Why this case

DRAM is a commodity: one maker's gigabyte substitutes for another's, so price is set by industry supply against demand, and supply arrives in billion-dollar fab-sized lumps that take two years to build. That combination (commodity product, lumpy capacity, long lead times) is the recipe for violent cycles, and it makes Micron the ideal laboratory: everything cyclical about airlines, steel, shipping, housing and oil appears here in concentrated form, with the added modern twist of an AI-driven boom (HBM memory) testing whether the cycle itself has changed.

Anatomy of the memory cycle

The loop, which has run for four decades

1. demand firms -> prices rise -> margins explode
       (fixed costs + commodity price = extreme operating leverage)
2. record profits -> every producer expands capacity at once
3. new fabs arrive together, 18-24 months later -> supply overshoots
4. prices collapse (DRAM can fall 50%+ in a year) -> losses
5. capex slashed, weak players exit or consolidate -> supply tightens
6. demand catches up -> return to 1.

the analyst's edge is knowing WHERE IN THE LOOP the industry stands,
which is readable from capex announcements, inventory, and pricing.

The cyclical's signature: earnings (solid) swing far more than revenue (dashed) around the cycle, because costs are fixed while price is a commodity. Stylized, not to scale.

Three cycles, one script

Cycle	What happened	The tell at the turn
2016-2019	Cloud-datacenter demand spike tripled the stock into 2018; then a supply glut and inventory correction roughly halved earnings	2018 peak: P/E ~4 and headlines about 'this time is different'; capex plans everywhere
2020-2023	Pandemic device boom, then the sharpest downturn in a decade: DRAM prices collapsed, Micron posted losses in 2023	The 2022-23 trough: industry-wide capex cuts and unprecedented supply discipline announcements: the classic bottom signature
2023-2026	AI servers demand high-bandwidth memory (HBM); HBM capacity sold out years forward; earnings rebound to records	The live question: is HBM a structural premium segment or the same cycle wearing a new costume?

The inverted P/E rule, demonstrated

Why cheap is expensive and expensive is cheap

AT THE PEAK (e.g. mid-2018):
  price $60, trailing EPS ~$12  ->  P/E ~5   "cheapest stock in tech!"
  ...but peak EPS embeds peak DRAM prices. Normalized EPS ~$6.
  real multiple on mid-cycle power: ~10x, and the cycle is ABOUT TO TURN.

AT THE TROUGH (e.g. 2023):
  price $60, trailing EPS negative ->  P/E meaningless / "uninvestable"
  ...but trough losses embed trough prices. Same mid-cycle EPS ~$6-8.
  the market prices the RECOVERY long before the income statement shows it.

the rule of thumb, a century old: buy cyclicals when the P/E is
enormous or absent, sell them when it looks like a bargain.

This is the single most counterintuitive lesson in equity analysis, and screens fail it constantly: a low-P/E screen loads up on cyclical peaks by construction. The fix is mid-cycle earnings: average the margin across a full cycle (or take normalized industry pricing), apply it to current capacity, and value THAT. The Analyzing a Stock guide's method-weighting table points cyclicals here for exactly this reason.

The cyclical toolkit, piece by piece

Mid-cycle valuation. Estimate through-cycle EPS or EBITDA and multiple THAT, never the spot number. A cyclical DCF uses a normalized base year, or it is a peak (or trough) extended forever.
Price-to-book against its own history. For capital-intensive cyclicals, P/B is the through-cycle anchor: Micron has historically troughed near or below ~1x book and peaked at 2.5-3x+. Book is imperfect (write-downs cluster at troughs) but it does not swing with DRAM spot, which is the point.
Supply-side intelligence. The demand narrative is loud and mostly priced; the SUPPLY side is where turns are made: aggregate the three producers' capex guidance, watch inventory days across the chain, and treat coordinated capex cuts as the bottom's clearest signature. (This is the commodities guide's balance-sheet method, applied to chips.)
Balance-sheet-first underwriting. The trough is where cyclicals die. Net cash or low leverage means the company reaches the next upcycle; leverage means the equity may not. The credit guide's maturity-wall check is mandatory here, and it is why the same cycle bankrupted some memory makers historically while Micron compounded.
Buy on terrible-but-improving. The tradable bottom is not when news turns good; it is when news is awful but the SECOND DERIVATIVE turns: price declines decelerating, inventory peaking, capex cuts announced. By the time results look good, the stock has doubled.

The four most expensive words: this time is different

At every peak, a structural story explains why the cycle ended: consolidation to three players (2018), or AI (2024-25). Sometimes it is even partly true: consolidation DID dampen the amplitude, and HBM's long-term contracts DO change revenue visibility. The discipline is not to dismiss structural claims but to price them as a scenario while underwriting the cycle as the base case, because the cycle has won that argument for forty years, and one eventual win for "different" does not refund the drawdowns of believing it early.

When a cyclical genuinely changes character

The honest complication: industries DO sometimes graduate. Consolidation can turn a commodity into an oligopoly with pricing discipline; a premium segment can carry contract revenue (HBM's multi-year agreements); a byproduct can become a moat. The analyst's test is falsifiable: does pricing hold through the NEXT demand wobble? Do margins trough materially higher than last cycle? Does capex discipline survive record profits, the hardest test of all? Watch the behavior at the moment of maximum temptation, not the investor-day slides, and grant multiple expansion only after the evidence arrives, because paying a structural multiple for a cyclical is how the largest permanent losses in this sector happen.

The transferable lessons for any cyclical

Identify the boat-vs-tide split first. How much of results is company skill versus industry price? Revenue per unit against the commodity price answers it in one chart.
Invert the P/E instinct, and normalize before valuing anything.
Track supply, not headlines. Capacity, inventory, discipline: the turn lives there.
Underwrite survival before upside. The balance sheet decides who sees the recovery.
Size for the amplitude. A correct cyclical thesis can still traverse a 40% drawdown on the way; the position must be able to hold through the loop's bottom half.
Respect structural claims, price them as scenarios. The cycle is the base case until pricing survives a downturn.

Glossary for this guide

Mid-cycle earnings: A cyclical company's earnings averaged across the cycle rather than taken at the current peak or trough. Valuing cyclicals on spot earnings buys high and sells low by construction.
Operating leverage: How much profits amplify a change in revenue because costs are fixed. High operating leverage makes good years great and bad years terrible; it is a magnitude, not a virtue.
Book value: Shareholders' equity as the balance sheet states it: assets minus liabilities. Meaningful where assets are financial or tangible (banks, insurers), nearly meaningless where the real assets are brands and code.
Supply-demand balance: The commodity analyst's model: production plus inventory change must equal consumption, quarter by quarter. Price is the negotiator that keeps the identity true when the balance tightens.
Cost curve: Every producer ranked from cheapest to most expensive. Long-run price gravitates to the marginal producer's cost, making the curve the closest thing commodities have to intrinsic value.
Maturity wall: A cluster of debt coming due in a narrow window. Refinancing risk has a calendar, published in the debt footnote, and a fine company with everything due in a shut market is a default candidate.
Regime: A market era with its own rules: the inflation seventies, the QE 2010s. Backtests assume tomorrow is drawn from the sample's regime; report performance per named regime instead of one blended number.
Capex: Cash spent on long-lived assets: plants, equipment, software. Maintenance capex keeps the current business running; growth capex expands it. Both reduce free cash flow now in exchange for cash flows later.
Falsifier: The observable fact, chosen in advance, that would prove the thesis wrong and trigger exit. Deciding it while calm is cheap; deciding it mid-drawdown is expensive, which is why professionals write it down first.
Margin of safety: Buying below your estimate of value by enough to survive being partly wrong. The working admission that every valuation is an estimate.