Cost of risk: what loan-loss provisions take away from net interest margin
· 6 min · beginner
Automated material · TradeAlmanac editorial deskDraft prepared by a language model from our stored data; not reviewed by an editor.
Cost of risk is the amount a bank sets aside in provisions for credit losses, expressed as an annualised percentage of its loan book. In essence, it is the price a lender pays for the fact that some borrowers will not repay: it is deducted from interest income before any profit appears. That is why, for an analyst, cost of risk (CoR) belongs next to the margin rather than under "other expenses": taken together, the two show how much of the spread the bank actually keeps once credit risk has materialised in the figures of its financial statements.
How the calculation works
The numerator is the provisioning charge for the period, meaning the movement in provisions that passes through the income statement, not the balance of the provision on the balance sheet. The denominator is the average loan book over the same period, usually before provisions are deducted. The result is annualised so that quarterly and full-year figures can be compared with each other.
Two details decide whether the metric is comparable or meaningless. The first is what exactly goes into the numerator: impairment on loans only, or also provisions against guarantees, credit-related commitments and other assets. The second is whether the loan book is taken gross or net. A bank that calculates CoR on its net loan book will mechanically report a higher figure, all else being equal. When comparing lenders with one another, the methodology has to be aligned by hand, and the disclosures in the notes to the financial statements matter more here than the headline line in a press release.
Why it is not the same as overdue debt
The share of overdue loans and NPLs describes the stock of problems that has built up in the loan book. Cost of risk describes a flow: how many new losses have been recognised during the period. The two figures diverge regularly. A bank may write off an old problem debt: the share of overdue loans falls, while cost of risk stays high because the write-off is backed by a provision created earlier. The reverse is possible too: releasing excess provisions turns CoR negative, and this is not a sign of loan quality but a reversal of charges booked in the past.
Hence a practical rule: cost of risk can only be read together with the coverage of problem loans by provisions and with the dynamics of the loan book itself. Rapid loan growth temporarily depresses the metric, because young loans have not yet reached the age at which defaults start to show. This seasoning effect distorts a year-on-year comparison of a bank with itself more than any methodological difference does.
Stages of expected credit loss
Modern accounting requires banks to recognise not the loss already incurred but the loss expected. A loan moves through stages: as long as its credit quality has not deteriorated significantly, the provision covers the losses expected over the next year; after a significant deterioration, it covers the losses expected over the entire life of the loan; on default, the asset moves to the final stage, where recovery is assessed individually.
The main consequence for an analyst is that the provision appears before the missed payment. A large borrower moving between stages can push up a quarter's cost of risk while not a single payment on that loan has yet been missed. A spike in CoR therefore calls for the question "what happened to the stages?", not just "who stopped paying?". The model parameters — probability of default, loss given default and exposure at default — also depend on the macroeconomic forecast the bank builds into its model. A change of scenario changes the provision without a single event occurring in the loan book itself.
What is left of the margin
The real analytical work begins when cost of risk is subtracted from net interest margin: the result is the risk-adjusted yield on the loan book. A bank that charges high lending rates and is aggressive in selecting borrowers, and a bank that lends cautiously, can end up with similar results: the first earns the spread and hands it over to provisions, the second neither earns it nor gives it up. The mechanics of the spread itself are covered separately: net interest margin.
The metric is also driven by the external environment. More expensive funding squeezes the margin, while rising debt burdens among borrowers push provisions up: both forces come from the same source, and both weigh on the same bottom line. How a decision by the regulator feeds through to a specific loan book is explained in our article on the key rate.
From cost of risk to the value of the bank
For valuing the company, what matters is not a one-off figure but normalised cost of risk through the cycle: the average level of losses that the business model generates across good years and bad. The peak of a crisis year and the trough of a recovery year are equally unsuitable as a base for a forecast. It is normalised CoR that feeds into the model of sustainable earnings, from which the terminal value is then derived.
From there, the question becomes one of value creation: does risk-adjusted profit cover the required return on the capital that the regulator obliges the bank to hold against this loan book? The logic of that analysis is set out in our article on economic value added. The market prices its own version of the normalised loss level into the valuation multiple: the current earnings-based valuation can be seen on the SBER page and in the value of 3,92, while the sections on Russian stocks and issuer reports provide the raw data for recalculating it. The notion of the compensation required for bearing risk is defined more precisely in the glossary: risk rate.
What this article does not cover
This article does not give CoR figures for specific banks or industry benchmarks: such values depend on the disclosure methodology and on the phase of the cycle, and quoting them descriptively, without reference to a source and a date, would mislead the reader. Comparable figures come from issuers' financial statements and from the regulator's statistics, and those are the sources to turn to, after first checking how the numerator and the denominator are defined.
Related instruments
Draft prepared by a language model from our stored data; not reviewed by an editor.
Model: claude-opus-5
How we use language modelsSimilar articles
- ARPU: who counts as a user, and why the debate is about the denominatorARPU is a service's revenue for a period divided by the average number of users over the same period.
- Netback in oil production: what a company keeps from each barrel after logistics and taxesNetback is the price of a barrel worked back to the point where the company produced it.
- AISC versus cost of sales: why depreciation comes out and capex goes inAISC (all-in sustaining costs) is the cash cost of sustaining production at its current level, expressed per ounce of gold sold.
- Churn: how customer attrition translates into revenue lifetime and valuationChurn is the share of its customer base or recurring revenue that a company loses over a period.