21 Sep 2018 5 min read

Consensus forecasts: poor prophets of bank profits?


Bank earnings forecasts from sell-side analysts have consistently been more optimistic than subsequent earnings reports. Is there more to this than meets the eye?

Bull and graph

The discrepancy between forecasts and reality in bank earnings is well known. It’s often attributed to the fact that management look to maximise their share-based incentives with positive guidance, and that sell-side analysts are incentivised to come up with ‘buy’ recommendations. But we believe there is a bigger problem at hand, and one that is getting worse.

International accounting standards are increasingly moving towards requiring banks (and indeed other companies) to account for long-term contracts on a ‘smoothed’ basis over the life of the contract. Under IAS 39, the effective interest rate (EIR) method is used to smooth the expected net revenue over the expected transaction life. Banks have to estimate all the future cashflows associated with each contract, including costs of acquiring the customer relationship as well as payments both to and from the customer.

IFRS 9 now even requires banks to calculate expected losses over the life of the loan. All of these future cashflows have to be modelled over the expected life of the loan, and discounted at the ‘effective interest rate’. This creates an ‘effective interest rate asset’ on the balance sheet, which is amortised to the income statement over the product’s life.

Bank earnings are increasingly a function of models based on customer behaviour in the past

One of the most critical variables that banks have to estimate is the expected life of the contract, particularly if the customer or lender has options to prepay (e.g. a mortgage) or if the terms of the loan are unpredictable (e.g. credit cards).

The effect of EIR accounting is to smooth earnings – effectively pushing out upfront costs and bringing forward revenues. All of this (and quite a lot else) flows through the ‘net interest income’ line reported by banks, and is used by analysts to calculate the ‘net interest margin’ (NIM, which is simply net interest income divided by a bank's average earning assets).

The problem comes when sell-side analysts assume the NIM is based on cash revenue streams that reflect customer behaviour today, and can be forecast simply based on volume growth.

For example, EIR assumptions for 0% balance transfer credit card products include the expected life of the customer relationship, the evolution of card balances, how actively customers will use the card, and stick rate assumptions (the proportion of customers that will stay after the 0% promotional period). These, in turn, depend on other variables, such as how much competition there is from other card lenders, and how constrained borrowers are by other debt.

Banks generally use historical customer data to predict future customer behaviour, believing that “human behaviour is predictable and stable”, but over time consumer behaviour does change (for example due to technological change) and so does the competitive environment banks are operating within (which can lead to greater customer churn). Critically, when the assumptions and behavioural models used by the banks prove to be wrong, hefty write-downs are needed. It is precisely these write-downs that analysts are so bad at predicting, but which prove critical when a bubble bursts.

Even more concerning is that economists, policymakers, regulators and politicians all use reported bank earnings to reach conclusions about what is happening in the economy and predict what might happen next – when in fact bank earnings are increasingly a function of models based on customer behaviour in the past.


LGIM contributors