Stock Valuation Methods, Compared

Why the same stock has ten different valuations

Take Axcelis Technologies (ACLS), a mid-cap semiconductor equipment maker. In late 2023, three sell-side analysts each published a price target within a few weeks of each other: $125, $162, and $198. All three were analyzing the same public filings, the same earnings call transcript, the same backlog data. The gap between the lowest and highest target was 58%. None of them were wrong in the sense of making a math error — they were answering different questions with different tools.

The $125 analyst used EV/EBITDA on a peer group that included mature equipment names trading at compressed multiples. The $162 analyst ran a P/E multiple on forward earnings anchored to the sector's historical range. The $198 analyst built a DCF using Axcelis's own long-cycle ion implant opportunity and modeled elevated revenue through 2027. Same company. Three defensible numbers.

This isn't a flaw in the system. It's the system working as designed. Every valuation method is a model of a specific mechanism — it asks one question about a business and answers it precisely. A P/E ratio asks: how much are investors paying for each dollar of current earnings? A DCF asks: what are the future cash flows worth at today's discount rate? Price-to-book asks: how much above the net asset value is the market willing to pay? These are genuinely different questions. They produce different answers. That's expected.

The mistake most investors make is treating these methods as interchangeable — picking whichever one produces the number they like, or averaging them together as if the average is somehow more accurate than any individual estimate. The real skill is knowing which question actually matters for the specific business you're analyzing. That question is almost always determined by the business model, not by preference.

Three families, not ten methods

The intimidating list — P/E, EV/EBITDA, EV/EBIT, P/S, P/B, P/TBV, FCF yield, DCF, dividend discount model, sum-of-the-parts — collapses into three families once you understand what each group is actually measuring. Each family asks a fundamentally different question about value.

Earnings-based methods (P/E, EV/EBITDA, EV/EBIT, EV/EBITA) ask: how much are investors paying relative to the business's current or near-term profitability? These methods anchor value to reported earnings or operating profit, which makes them fast to compute and easy to compare across peers. They dominate sell-side research because they're legible to portfolio managers who need to triage dozens of names. The underlying assumption is that current profitability is a reasonable proxy for normalized earning power — which is sometimes true and sometimes deeply misleading.

Cash-flow-based methods (DCF, FCF yield, dividend discount model) ask: what is this stream of future cash worth at today's cost of capital? These methods are more complete in theory because they explicitly model the timing of cash generation rather than anchoring to a single period's accounting profit. They're also the methods most sensitive to the assumptions you bring to them. A DCF is only as good as its inputs — discount rate, growth rates, and the terminal value that typically accounts for 60–80% of the total output.

Asset-based methods (price-to-book, price-to-tangible-book, price-to-NAV) ask: how much above the liquidation value of the net assets is the market pricing in? These methods originate from Graham-era value investing, where the question was literally whether you could buy a business for less than its parts were worth. They remain highly relevant for businesses where the assets themselves are the primary driver of value — banks, insurers, real estate — and nearly useless for businesses where value lives in intangibles, customer relationships, or proprietary software.

That mental model matters because choosing the wrong family doesn't just give you a less accurate number — it gives you a number that measures something that isn't the main driver of value. Applying P/B to a software company tells you almost nothing about what that business is worth. Applying P/E to a bank at a credit-cycle trough makes it look expensive when the earnings are temporarily depressed — or cheap when they're inflated by reserve releases. The method produces a number; the number just doesn't mean what you think it means.

What breaks each method

Most valuation explainers tell you what each method is. Almost none tell you when it stops working. That's the information that actually matters for avoiding bad decisions.

P/E breaks on cyclically compressed or elevated earnings. The ratio is only useful if the current earnings figure is representative of the business's normalized earning power. For cyclical companies — steel producers, chemical companies, mining stocks, homebuilders — earnings swing dramatically through the business cycle. At the top of a steel cycle, a company might earn $12 per share; at the trough, $1.50. Paying 8x earnings sounds cheap at the top of the cycle. It isn't — you're paying 8x peak earnings, which on a normalized basis is actually expensive. This is exactly the pattern that leads investors to buy cyclicals just before the downturn, drawn in by P/E multiples that look low because the E is temporarily high.

P/E also breaks for companies reinvesting heavily at the expense of current earnings. Amazon looked expensive on P/E for most of its growth phase — not because the stock was overvalued, but because management was deliberately sacrificing reported profit to build logistics infrastructure and AWS. Applying a P/E lens and concluding Amazon was expensive in 2015 missed the mechanism of value creation entirely. The value wasn't in current earnings; it was in the compounding of reinvested capital at high returns.

DCF is dominated by its terminal value. This is the most under-discussed fact in retail investing. In a standard DCF, the terminal value — the perpetuity representing all cash flows beyond the explicit forecast period — accounts for roughly 60–75% of the total estimated value for a mature company. For a high-growth company, it can reach 85–90%. That means a small change in the terminal growth rate assumption produces an enormous swing in the output. Raise the terminal growth rate from 2.5% to 3.0% on a company with a 9% discount rate, and the fair value estimate moves up roughly 15–20%. That's not rounding error. That's the difference between "cheap" and "fairly valued" depending entirely on what you typed into one cell.

EV/EBITDA loses its edge when CapEx is the story. EBITDA adds back depreciation as though it's a non-cash accounting fiction — which it is for some businesses, and emphatically isn't for others. A pipeline company or a semiconductor manufacturer depreciates real assets that actually wear out and require real cash to replace. For capital-intensive businesses, EV/EBITDA systematically understates the true cost of maintaining the asset base. EV/EBIT or EV/EBITA (which adds back only amortization of acquired intangibles, not depreciation) is more appropriate in those cases.

P/B is structurally useless for asset-light businesses. Microsoft's book value per share bears almost no relationship to what Microsoft is worth, because the value lives in software, customer relationships, and brand — none of which appear on the balance sheet at economic value. Applying P/B to Salesforce in any year of its existence would have told you it was wildly overvalued, while the stock compounded at over 20% annually for a decade. The method wasn't surfacing a buy signal you missed. It was measuring the wrong thing with confidence.

Match the method to the business

The selector below is this page's core teaching moment. Pick a business archetype and see which valuation methods are signal, which are noisy, and which are actively misleading — with a one-line reason for each rating. The "Misleading" labels are the payload. A noisy method just has wide error bars. A misleading method gives you a confident, plausible number that points in the wrong direction.

After working through a few archetypes, the pattern becomes clear. The businesses where most methods work are mature, stable, profitable companies with predictable cash flows and balance sheets that reflect real economic value. Every step away from that baseline — toward high growth, cyclicality, asset-heaviness, or financial intermediation — narrows the set of methods that tell you something useful. For banks, you're down to two or three. For early-stage unprofitable tech, you're building a thesis, not running a model.

A worked example: why P/E is the wrong lens for a bank

Consider a hypothetical regional bank — call it Harwick Bancshares, a $4 billion asset community bank in the Southeast. In mid-2023, Harwick traded at 12x trailing earnings with a dividend yield of 3.2%. That P/E looks reasonable at first — below the S&P 500 average, in line with the regional bank peer group. Nothing alarming.

But P/E is the wrong lens for a bank, and applying it here produces a number that obscures more than it reveals. Bank earnings include loan loss provisions — charges taken today against loans that might default in the future. Those provisions are inherently cyclical. In a strong credit environment, a bank releases reserves it previously set aside, which boosts reported earnings. In a weak environment, it builds reserves, which depresses earnings — often sharply, and often at exactly the moment the business is under the most stress.

In Harwick's case, its 2022 provision expense was $18 million — well below its long-run average of $34 million — because credit quality was exceptional and the bank was releasing reserves built during COVID. That reserve release added roughly $16 million to pre-tax income, which at a 25% tax rate means approximately $12 million of the bank's reported earnings were a credit-cycle artifact, not a reflection of underlying earning power. Strip that out and the trailing P/E expands from 12x to something closer to 16–17x. The stock was not as cheap as the headline multiple suggested.

The right tools for valuing a bank are price-to-tangible-book value (P/TBV) and return on tangible common equity (ROTCE). P/TBV asks how much you're paying relative to the bank's hard net assets — loans, securities, and capital — stripped of goodwill from past acquisitions. ROTCE measures how efficiently the bank earns on that tangible equity base. The combination tells you: are you paying a reasonable premium over asset value, and does the bank generate returns that justify it?

Harwick traded at 1.1x tangible book, with a trailing ROTCE of 11.4%. Regional bank peers with similar credit profiles and efficiency ratios traded at 1.3–1.6x TBV. On that basis, Harwick was modestly cheap — but the discount was mostly explained by below-average loan growth and an elevated efficiency ratio. The P/E ratio had told a cleaner story than the reality warranted. P/TBV and ROTCE told the actual story.

The lesson generalizes to any business where accounting earnings include cyclical reserve adjustments or mark-to-market swings — banks, insurers, specialty finance companies. Earnings-based multiples introduce noise at the exact moment in the cycle when you most need clarity. Asset-based methods anchored to tangible book cut through that noise.

When analysts mix methods on purpose

Experienced analysts rarely run one model. They run two or three and compare the outputs — not to average them, but to understand the gap between them. If a DCF produces a fair value of $40 and a trading comps analysis produces $28, that $12 spread is itself information. It almost always traces to one of three things: the DCF is assuming growth the market doesn't believe, the comparable companies aren't actually comparable, or the market is discounting a risk the DCF didn't model.

Two methods that converge on the same number from different angles provide genuine confirmation. Two methods that diverge sharply force you to ask why — and the answer is usually more useful than either number in isolation. A stock where DCF says $40 and comps say $28 is either a mispriced opportunity (if the DCF assumptions are defensible and the comps are wrong) or a sign that the DCF is too optimistic (if the market is right that the growth won't materialize). Knowing which requires engaging with the gap, not dismissing it.

For retail investors, the practical version is simpler: pick two methods that fit the business model, run both, and take seriously any divergence above 20%. Use the DCF for intrinsic value and EV/EBITDA comps for a sanity check on where the market is currently pricing the peer group. If your intrinsic value estimate sits well above where peers trade, you need a thesis for why this company deserves a premium — or you need to revisit your growth assumptions.