How to Normalize Earnings for a Cyclical Business

Nucor Steel's (NUE) diluted earnings per share, as reported in their annual 10-K filings, swung from $2.36 in fiscal 2020 to $28.79 in fiscal 2022 — a twelve-fold difference in two years from the same manufacturing footprint, the same capital base, roughly the same management team.

Run a simple P/E-based valuation on those numbers and you get radically different pictures. Apply a 15× earnings multiple to the trough figure and you arrive at a "fair value" of roughly $35 per share. Apply it to the peak and you get $432. Neither estimate is useful, because neither year's earnings reflects the business's durable earning power — which is what intrinsic value actually depends on. Using the trough, you'll sell a high-quality business for a fraction of its worth. Using the peak, you'll hold a position that looks expensive when the cycle turns — and in steel, the cycle always turns.

This is the core difficulty with cyclical businesses: the earnings aren't fictional. They reflect the economics of a specific moment in the cycle, and they're entirely real. But fair value depends on the sustainable earning power of the business across a full cycle, not on what it earns at any given extreme. The practical solution is normalization — adjusting reported earnings to reflect mid-cycle conditions before feeding them into any DCF model or owner earnings estimate. Two methods do this reliably. Each has strengths and failure modes worth naming before you rely on the output.

The Distortion That Raw Earnings Create

The problem with cyclical businesses is structural. In commodity-driven industries — steel, energy, chemicals, paper — revenues and margins move with the underlying commodity price or capacity utilization cycle. When prices spike, margins expand dramatically. When they collapse, margins compress toward zero or below. Not because of operational failure. Because of how the economics work.

A DCF model built on peak earnings will massively overstate intrinsic value because it implicitly assumes those margins persist. Discount $28 in peak EPS at 8% and you'll estimate a fair value that no patient long-term owner would pay — because those earnings reflect extraordinary conditions that are, by definition, temporary. The same reasoning works in reverse at the trough: a DCF anchored to $2.36 assigns almost no value to the business's earning power in a normal environment, to say nothing of a recovery.

The P/E ratio distorts cyclicals in the opposite direction from what most investors expect. A cyclical at a low 5× P/E at cycle peak appears cheap but is usually expensive — earnings are about to fall. That same business at a seemingly expensive 25× P/E near the trough can be a genuine bargain, because the denominator is temporarily depressed. Investors who bought commodity producers at low headline multiples at the peak of the 2007 supercycle got a hard education in this. But the fix isn't a different multiple. It's a better earnings figure. Normalized earnings — cleaned of cycle noise — belong in the denominator, not the reported number that reflects wherever the cycle happens to be today.

Method 1 — The Multi-Year Earnings Average

The most straightforward normalization approach averages historical earnings over a period long enough to span at least one complete cycle.

Normalized EPS (Average Method) = Sum of annual EPS over N years ÷ N

For Nucor, here's what that looks like using five years of diluted EPS from their annual 10-K filings:

• Fiscal 2019: $4.14
• Fiscal 2020: $2.36 (COVID-driven demand collapse)
• Fiscal 2021: $23.16 (post-pandemic steel supercycle begins)
• Fiscal 2022: $28.79 (supercycle continues; Russia-Ukraine war disrupts European supply)
• Fiscal 2023: $18.05 (steel prices normalizing, still above historical averages)

Five-year average: ($4.14 + $2.36 + $23.16 + $28.79 + $18.05) ÷ 5 = $15.30 per share. Apply a 15× multiple to that figure and you get a fair value estimate around $230 — a usable starting point and far more defensible than applying that multiple directly to either extreme year. Whether 15× is the right multiple for Nucor at mid-cycle is a separate judgment; what matters here is that you're starting from an earnings base that doesn't depend on which year of an extraordinary cycle you happen to be observing.

But look at the composition of that average and the weakness surfaces immediately. Two of the five years — 2021 and 2022 — reflect conditions that were genuinely extraordinary. Post-pandemic supply chain disruptions, China production curbs, and the Russia-Ukraine war's combined impact on European steel capacity produced a market environment that most industry veterans had never seen. Extend the window back to include fiscal 2015, 2016, and 2017 — years when hot-rolled coil prices ran between roughly $400 and $500 per ton and Nucor's EPS ranged from approximately $1 to $3 — and the normalized figure drops well below $10. Which window you choose shapes the answer more than the methodology itself.

For the multi-year average to work, you need a period that spans a complete peak-to-trough-to-peak sequence. Five years rarely accomplishes this in commodity industries, where full cycles commonly run 7–12 years. I've found that starting with at least a decade is the right default — and even then, verifying that the window doesn't accidentally anchor to an unusually good or bad period. That's the method's honest limitation: it's intuitive, but period-dependent in ways that aren't always obvious until you see how much the answer shifts with a different start date.

Method 2 — Mid-Cycle Margin on Trend Revenue

The second method is more deliberate and more assumption-laden — which, in valuation work, is usually a virtue rather than a flaw. Named assumptions can be stress-tested; assumptions embedded inside a historical average cannot.

Normalized EPS = (Trend Revenue × Mid-Cycle EBIT Margin − Trend Interest Expense) × (1 − Normalized Tax Rate) ÷ Diluted Shares

Three inputs. Three explicit judgments.

Trend revenue: smooth revenue over 7–10 years using a linear regression or geometric average. This strips the commodity price level at any given moment while preserving the organic growth in volumes and product mix. For a steel producer, trend revenue captures growth in tons shipped and the mix shift toward higher-margin downstream products — not the price spike of 2021 or the collapse of 2009.

Mid-cycle EBIT margin: identify the operating margin the business earns in years that resemble neither a supercycle peak nor a demand trough. For Nucor, the decade spanning roughly fiscal 2010 through fiscal 2019 — excluding the Section 232 tariff-boosted years of 2018 — saw average EBIT margins in the 5–9% range. That band is the reasonable starting point for a mid-cycle estimate. I'm less confident in where exactly that mid-cycle margin sits than I'd like to be: the post-2020 period may have durably shifted Nucor's cost position in ways that are difficult to separate from the supercycle itself. That uncertainty is an argument for using a range — a conservative margin estimate (say 6%) and a base case (say 8%) — rather than a single number.

Trend interest expense: for a conservatively capitalized business like Nucor, this is relatively stable year to year. For a cyclical company that took on debt during a peak — expanding capacity while commodity prices were elevated — you may need to normalize interest expense down to what the balance sheet would carry at mid-cycle EBITDA, not the debt load it happens to carry today. More on this below.

Apply the mid-cycle margin to trend revenue, subtract trend interest expense, apply the normalized tax rate, divide by diluted shares. The resulting figure is anchored to what the business earns in a typical environment, not any particular year's commodity price. This approach also makes each assumption explicit — which is precisely what calculating fair value should do. A fair value figure functions as a compass, not a GPS coordinate. Showing your work on the assumptions is how the compass stays useful when the assumptions prove imprecise.

Operating Normalization vs. Leverage Normalization

Normalizing operating earnings — adjusting margins for cycle position — solves most of the problem. But it leaves one dimension unaddressed: leverage.

A cyclical business with significant debt has two sources of earnings volatility, not one. Operating margins swing with the commodity cycle — that's what the methods above address. Interest expense doesn't swing with the cycle; it stays fixed regardless of where revenue lands. In a trough year, a highly leveraged cyclical company doesn't just earn less because margins compressed. It earns dramatically less because a fixed interest burden consumes a much larger fraction of a smaller revenue base. The two effects compound, amplifying trough earnings toward zero — or below it — far more than operating cyclicality alone would produce.

This means that for debt-heavy cyclicals, you may need two normalizations in sequence. First, normalize operating earnings: strip the cycle distortion from EBIT margins. Second, normalize for leverage: recalculate interest expense using the debt load the business would carry at mid-cycle EBITDA, not the debt it happens to carry today. For a producer that borrowed aggressively to expand capacity during the peak — common in energy and metals — the interest expense in a trough year can be three times what it would be in a prudently managed capital structure. Using actual trough interest expense in a normalized earnings estimate imports the leverage problem directly into the normalization.

Nucor is a relatively clean example precisely because it manages its balance sheet conservatively — one of the more disciplined operators in North American steel. The leverage normalization step matters less there than at a peer carrying heavier debt. But for a contract driller, a chemical producer at the end of a long capex cycle, or a paper company that refinanced during peak prices, separating operating normalization from leverage normalization is the difference between a useful intrinsic value estimate and a misleading one. The margin of safety you require should be wider for businesses where the leverage step involves substantial judgment — thin interest coverage at mid-cycle is a compounding risk that deserves a larger buffer.

Key Takeaways

Valuing cyclical businesses requires one additional step before any multiple or DCF model is applied: normalizing the earnings. The choice isn't between normalizing and not normalizing — it's between doing it deliberately with named assumptions and doing it implicitly with whatever year's earnings the current P/E happens to present.

• The multi-year average method is intuitive but period-dependent. Use at least 7–10 years to span multiple cycles, and verify that the window doesn't inadvertently anchor to extraordinary conditions. A five-year window that includes 2021 and 2022 will significantly overstate mid-cycle earnings for most commodity businesses.
• The mid-cycle margin method is more robust but requires three named assumptions: trend revenue, mid-cycle EBIT margin, and trend interest expense. Name each explicitly so they can be stress-tested. A sensitivity table showing the impact of a ±2 percentage-point shift in the margin assumption is rarely wasted.
• Leverage amplifies cyclicality. A normalized operating earnings figure is incomplete if fixed interest expense represents most of those earnings at mid-cycle revenue. Normalize both the margin and the capital structure when a company carries significant debt.
• Fair value for a cyclical business is always a range with wide confidence intervals. Require a larger margin of safety than you would for a stable compounder — and don't mistake the current P/E, which reflects cycle position, for a reliable signal of valuation.