How a Position Hits the BooksMedium

The same holding can be worth three different numbers depending on which shelf it was put on the day it was bought — and the shelf, not the market, decides when a loss appears.

11 min read · 1 949 words

The question nobody asks until it matters

A bank buys a ten-year government bond. The market moves against it and the bond is now worth less than it paid. How much has the bank lost?

The honest answer is that it depends on a decision made on the day the bond was bought, before anybody knew which way the market would go. That decision — which measurement category the holding was put into — is invisible from the outside, it is rarely revisited, and it decides whether the loss appears in the profit and loss account, in a reserve nobody reads, or nowhere at all until the bond is sold.

This is the single most useful thing to understand about a financial statement, and it is missing from almost every explanation of one. It is also the mechanism behind Silicon Valley Bank in 2023, and the reason its balance sheet looked unremarkable until the week it did not.

Three shelves, one holding

• Amortised cost — the holding is carried at what was paid, adjusted towards what will be repaid, and interest is recognised evenly over its life. Market prices are not used. A price move produces no accounting entry at all. This is where a loan sits, and where a bond sits when the holder intends to keep it to maturity and the contract pays only principal and interest.
• Fair value through profit or loss — the holding is remarked to a current value every reporting date, and every movement goes straight into reported profit. This is where a trading book sits, and where anything with an embedded derivative usually ends up whether or not anybody wanted it there.
• Fair value through other comprehensive income — the holding is remarked, but the movement is parked in a reserve inside equity rather than run through profit. Reported profit stays smooth; equity moves. The loss is disclosed, and it is disclosed in the part of the statements a reader has to go looking for.

Three treatments, one economic position. Nothing about the bond changed; the number changed. That is not an accounting scandal — it is the framework working as designed, and the design is a compromise between two things that cannot both be true at once.

Why the compromise exists at all

Marking everything to market every day makes reported profit swing with the market, which for a business whose whole model is to hold assets for years describes noise rather than performance. Marking nothing to market lets a holder carry something at a price the market stopped agreeing with years ago. Both are wrong in the same direction: they make the accounts less informative than they look.

So the framework asks what the holder is actually doing with the asset — the business model — and what the contract actually pays. A loan held to collect its payments is measured one way; the identical loan held to be sold next quarter is measured another. The rule is about intention and cash-flow shape, not about the instrument's name, which is why two banks can hold the same bond and report differently without either being wrong.

The invisible loss, and how it becomes visible

• Under amortised cost, a price fall produces nothing. The carrying value marches towards par regardless of what the market thinks. A portfolio bought when rates were low and held while rates rose is carried at a value the market would not pay, and the accounts do not say so on their face.
• It is disclosed, in a note. The fair value of amortised-cost holdings is required to be given. It is a table in the back of the report, without a headline, and reading it is the whole skill — see reading an annual report.
• It becomes real the moment the holding is sold. Selling crystallises the difference between the carrying value and the price, and the entire gap arrives in profit in one line, in one quarter, for a move that happened over years.
• Which is why a forced seller is a different animal from a holder. The intention that justified amortised cost was "we will hold this". A depositor run removes the choice, the classification stops being true, and the accumulated difference lands at once. That sequence — not a credit loss, not a fraud — is what happened in March 2023.

Nothing here is a claim that one treatment is better. The point is narrower and more useful: a reported number is a measurement, and the measurement rule is a separate fact from the market. A reader who does not know which rule is in force cannot interpret the number.

Expected loss: the provision that arrives before the default

A holding measured at amortised cost still has to carry an allowance for credit losses, and the rule for setting it changed shape after 2008. The older approach recognised a loss when there was evidence of one — which meant provisions arrived late, together, and at the worst point in the cycle. The current approach is forward-looking:

• Stage one — no significant increase in credit risk since the loan was made. The allowance covers losses expected from defaults within twelve months.
• Stage two — credit risk has increased significantly. The allowance jumps to cover losses expected over the loan's whole remaining life. Nothing has defaulted; the number moves anyway.
• Stage three — the loan is credit-impaired. Interest is recognised on the net figure rather than the gross one.

The step from stage one to stage two is the one that matters, because it is a cliff: a loan that crosses it takes a full lifetime provision in a single reporting period without anybody missing a payment. It is also the step that depends most on judgement, on a forecast, and on a definition of "significant" that each firm sets for itself and discloses. See a bank's own capital disclosure for where these figures surface, and spread measures for what the market is charging for the same risk in the meantime.

Day-one profit: recognising a gain nobody has been paid

A dealer sells a structured product and prices it with a model. The model says the position is worth more than the client paid. That difference is profit, arithmetically, on the day of the trade — before a single cash flow has happened and before the model has been tested against anything.

• Where the inputs are observable, the difference may be recognised immediately. The market has effectively agreed the price.
• Where a significant input is not observable — a long-dated correlation, a volatility nobody quotes — the difference is deferred and released over the life of the trade. The framework is saying, in effect, that a profit which exists only inside a model is not yet a profit.

The distinction rests entirely on which inputs count as observable, which is why the fair-value hierarchy matters: level one is a quoted price in an active market, level two is a model fed by observable inputs, level three is a model fed by inputs that are not. A firm's level-three balance is a measure of how much of its reported value nobody outside the firm can check. Model risk is the same subject from the other side.

Where the reported number and the economics part company

• A hedge that works economically and not in the accounts. The exposure sits at amortised cost; the hedging derivative is at fair value through profit. Both move, only one is reported, and profit swings for a position that is flat. Hedge accounting exists to reconnect them, and it is granted only against documentation and effectiveness testing — which is why a real hedge is sometimes not accounted for as one. Hedging covers the economic half.
• A liability that gains value when the borrower deteriorates. Debt measured at fair value falls in price when the issuer's credit worsens, which arithmetically is a gain. The framework now routes that particular movement away from profit, because reporting a profit for becoming less creditworthy is the sort of number that discredits an entire statement.
• Realised and unrealised are not a quality ranking. An unrealised mark on a liquid futures position is settled in cash daily and is about as real as money gets; a realised gain from selling into a thin market may have cost more in impact than it recorded. What separates them is the accounting event, not the certainty.

Why this belongs to more seats than the accountants'

Product control signs off the marks that become the trading book's profit. Risk reads the same positions under a completely different measurement — see risk measures — and the two rarely reconcile, which is a feature rather than a defect. Audit tests whether the classification still matches the stated intention. A journalist reading a set of results is reading measurement decisions as often as business outcomes. And anybody preparing for a conversation on a desk will be asked why a bank with unremarkable accounts failed in a week; the answer is on this page rather than in a credit model.

Product control, internal audit and treasury and ALM describe the seats — the third of those is the one that actually owns the amortised-cost bond portfolio at a bank. This is the mechanism all three are looking at.

What to take away

• The measurement category is chosen at recognition and it decides where a price move appears — profit, equity, or nowhere.
• An amortised-cost portfolio can carry a large difference to market value, disclosed only in a note, until something forces a sale.
• Credit provisions move on forecasts, and the stage-one to stage-two step is a cliff rather than a slope.
• A profit that exists only inside a model is deferred, which makes the level-three balance a useful thing to look up.
• Reported profit and economic outcome are two different questions. Knowing which one a number answers is most of financial statement analysis.