When each ratio stops making sense

Two companies both trade at a price-to-earnings multiple of 6, and in each case that is the lowest multiple in their sector. One is a consumer business that has grown its earnings steadily for a decade. The other is a steel maker in the final quarter of a once-in-a-cycle pricing boom. A stock screener ranks them side by side and labels both "cheap," and a buyer who trusts the screener will reach for both. That buyer will be completely right about one and completely wrong about the other. illustrative

The P/E has not been miscalculated. It has been calculated perfectly. The problem is that, for the steel maker, the ratio has quietly stopped meaning what the buyer thinks it means: a low multiple sitting on peak earnings is not a bargain but a warning. This module is about that exact moment - when a ratio goes dark. It keeps producing a number, but it has stopped answering the question you are actually asking, and the skill is seeing that before you act on the number.

Every ratio is a question dressed up to look like an answer. ROCE asks how well a business turns capital into operating profit - a superb question for a manufacturer, a meaningless one for a bank, whose deposits are raw material rather than employed capital. EV/EBITDA asks what cash operating earnings are worth setting aside financing - sensible for a cable company, absurd for a lender, for whom the interest is the business, not an item to add back. Price-to-book is a solid anchor for a bank and wildly misleading for an asset-light software firm with almost no tangible assets. So watch a single ratio survive some contexts and die in others.

A ratio breaks in one of two ways. The sector disqualifies it - the business is built so differently that the underlying question does not apply. For a lender, EBITDA is meaningless (interest is core revenue, not a financing cost), ROCE has no clean capital-employed base, interest coverage treats the cost of its own goods as a burden, and ten-to-one leverage is the model, not distress - so you fall back on book value and return on equity. A life insurer's profit and book are both distorted by reserving, so the industry uses embedded value instead. An asset-light firm's book value is tiny, so price-to-book comes out enormous and ROE is inflated by the same small denominator. A holding company's consolidated earnings double-count the subsidiaries it owns, so the honest tool is the discount to net asset value, not the P/E.

The situation disqualifies it - even where a ratio usually works, a moment in the company's life disables it. A cyclical at its peak shows a low P/E sitting on inflated earnings, so it inverts: low means expensive. The same cyclical at its trough shows a high or negative P/E at exactly the point the stock is cheapest. A company mid-capex carries construction-in-progress that swells capital employed while earning nothing, dragging ROCE down mechanically. A loss-maker has no P/E and a broken EV/EBITDA. A firm fresh from a merger carries goodwill that inflates book and drags down return ratios, and one that changed its year-end reports a stub period that makes every growth rate and trailing ratio non-comparable.

The discipline is a two-step reflex. Before you read any ratio, ask: what sector is this, and what situation is the company in? Then grey out, in your head, everything that just went dark.

Here is a panel of six ratios shown across four contexts. A filled dot means the ratio still answers its question; a struck-out dot means it has gone dark. Notice how few survive for a lender and a holding company - and that the survivors are not the same from one to the next.

Two of the four do more than disable a ratio - they invert the naive reading. For the cyclical at its peak, the earnings multiples point the wrong way, so the entry that screens "cheapest" is the most dangerous. For the holding company, almost the whole standard toolkit goes dark and the one honest tool - the discount to net asset value - is one the screener never computes. The lender and insurer show sector failure in its purest form: ratios that are the backbone of manufacturing analysis simply do not apply, and importing them anyway produces confident nonsense. The lesson is not that these businesses cannot be analysed; it is that each has its own valid set of instruments, and the first skill is knowing which tools on the standard dashboard to ignore.

A composite specialty-sugar maker shows up with a P/E of 5, an ROCE of 24%, and an EV/EBITDA of 3.5. The whole screen is green, and the naive read is obvious: outstanding returns, trivially cheap. Now apply the two-step reflex. illustrative

What sector is this? Sugar is a hard cyclical - output and selling prices swing with cane pricing, government policy and the harvest. What situation is the company in? The trailing twelve months happen to have captured a policy-driven price spike, so realisations sit near a cycle high. That single fact disarms the whole screen. The P/E of 5 is sitting on peak earnings; normalise realisations to mid-cycle and the "E" roughly halves, so the real multiple is closer to 10 and climbing as the cycle turns. The 24% ROCE is a peak-cycle number too - through mid-cycle it might be 10–12%. The EV/EBITDA of 3.5 is the same illusion in a different costume.

What would overturn this? If you genuinely believed the higher prices were structural - a durable policy shift or a permanent supply shortage - the ratios would regain their meaning, because the "peak" would simply be the new normal. Short of that, the honest instruments are the through-cycle ones: price-to-book against the company's own history, mid-cycle margins, and whether the balance sheet can survive the next trough. Same green screen; opposite conclusion; only the cycle context tells the two apart. Work it yourself below - pick a sector and a situation, and watch every disqualified ratio grey out with its reason. A loss-making lender loses two overlapping sets at once, the ones its sector disqualifies and the ones its situation does.

When a stock screens cheap for a reason the screener cannot see, a good analyst tests whether management is leaning on the flattering ratio: "You've pointed to the low P/E and EV/EBITDA - but aren't those flattered by where we are in the cycle? What do mid-cycle returns look like?" The question is really asking management to admit which ratios have gone dark.

A candid answer names the distortion itself:

"You're right not to take the headline multiple at face value. Realisations this year are well above what we'd underwrite through the cycle. On mid-cycle pricing, EBITDA margin is closer to 15% than this year's 24%, and ROCE settles around 11–12%, not the 24% you see now. We manage the balance sheet for the trough - net debt to mid-cycle EBITDA is under 2x - so we'd point you to book value and mid-cycle returns rather than the trailing P/E."

The evasive one stacks the cycle-inflated multiples up as if they were durable:

"We're the cheapest stock in the sector on almost every metric - P/E, EV/EBITDA, you name it. The market simply hasn't recognised the value yet. Demand is robust and we're very confident in the outlook. At these multiples the risk-reward is clearly attractive."

Not one mid-cycle number, and an appeal to a market that "hasn't recognised it" - it is selling the exact ratio that has stopped making sense. The follow-up that settles it is the one nobody asks: "What are your mid-cycle EBITDA margin and ROCE, and where does net debt sit against mid-cycle EBITDA, not peak?" Watch what the silence means. If the room lets "cheapest in the sector" stand on peak numbers, either the mid-cycle figures are unflattering enough that not asking has become the kindness, or the analysts who would have normalised the cycle have already stopped covering the name. A cyclical sold on its trailing multiple, on a call where nobody normalises it, is the precise setup this module exists to catch.

1. Trusting the screener's ranking across unlike businesses. A screen sorts a bank, a cyclical and an asset-light firm into the same P/E column as if the number meant the same thing in each. It does not; the ranking is valid only within a set of genuinely comparable companies.

2. Reading a cyclical's low multiple as cheap. At the peak, a low P/E and EV/EBITDA sit on inflated earnings and invert - cheap-looking is expensive. The anchors that work through the cycle are book value and mid-cycle returns.

3. Applying manufacturing ratios to a lender. ROCE, EV/EBITDA and interest coverage all break for a bank or NBFC, where interest is the business and heavy leverage is the model. Importing them produces confident, wrong conclusions.

4. Admiring an asset-light ROE without checking the denominator. A 35–40% ROE on a near-zero equity base can be pure arithmetic, not excellence. What tells you which is the margins, the growth and the cash conversion - not the ratio.

Ruling a ratio in is only the start. A price-to-book that survives for a lender still cannot tell you whether the book is honest or the loans are quietly going bad; a through-cycle P/E still depends on your own estimate of mid-cycle earnings; embedded value rests on actuarial assumptions you cannot see. And this list is not exhaustive - a novel business model or an odd accounting policy can disable a ratio in a way no checklist anticipates. So the reflex matters more than any fixed set of cases: always ask what the ratio assumes, then ask whether this business honours the assumption.