The loss that is an assumption, not a measurement

You have two lenders in front of you, both with gross loan books of about ₹10,000 crore, both lending to small businesses in the same three states. The first reports credit-impaired loans of 4.5% of the book and holds provisions covering 40% of them. The second reports 3.0% and holds 70%. On every screen ever built the second company is the better underwriter and it is not close. Then you open the note that reconciles the movement in the loss allowance, find the line marked write-offs, and discover that the second company removed ₹390 crore of loans from its books during the year against the first company’s ₹120 crore. Nothing was recovered on either set. One company left its problems on display and the other took them off the page, and the ratio you were comparing does not distinguish between those two things at all.

Where the number comes from: an estimate in three stages

A finance company preparing accounts under Indian Accounting Standards does not wait for a loan to fail before providing against it. It carries an expected credit loss on every loan from the day it is written, computed as the chance of default multiplied by the loss if default happens, applied to the amount outstanding at that point. Each loan sits in one of three stages, and the stage decides how much of that future is provided for.

The second set of books, and the table that compares them

The expected credit loss is the company’s own estimate, produced by a model with assumptions in it. The regulator does not rely on it. It maintains its own asset classification and provisioning norms, which are formula-driven — a loan overdue beyond a prescribed period is a non-performing asset, and each category carries a minimum provision. Because a finance company therefore has two loss figures for one book, its notes carry a prescribed table comparing them: the gross carrying amount by classification, the allowance under the accounting standard, the provision the regulatory norms would require, and the difference between the two. Where the regulatory figure is the higher, the excess is appropriated to a separate reserve out of retained earnings rather than being charged through profit. The exact format of that table has been revised more than once, so read the version printed in the accounts in front of you rather than a remembered one.

Adding back the line that was removed

The other numbers, and what each of them can and cannot say

• Provision coverage ratio — provisions held against stage 3 assets, divided by those assets. It answers how much of the recognised problem has already been paid for out of past profits. It says nothing about whether the right loans are in stage 3, and it is not immune to write-off policy either. A loan is normally written off once it is fully provided, so the write-off removes the same rupee amount from the provisions above the line and the impaired loans below it — and taking an equal amount off a smaller numerator and a larger denominator pulls the ratio down. Coverage of 70% on ₹300 crore becomes 55% once ₹100 crore of fully-provided loans leave. A lender that carries its old bad loans for years therefore tends to show both a higher impaired percentage and a higher coverage than one that writes them off promptly, and neither figure on its own says which of them lost more money.
• Collection efficiency — the share of amounts due in a month that actually came in. The fastest-moving indicator a lender publishes, and the least comparable across companies, because the definitions differ. Some include arrears collected from earlier months in the numerator, some include prepayments and foreclosures, some count only current-month dues. A figure above 100% is a definitional artefact rather than an achievement. Use it against the same company’s own history and be careful using it against anybody else’s.
• Restructured loans — accounts whose terms were changed because the borrower could not meet the original ones. These may retain a standard classification while carrying a higher provision, so they do not appear in the bad-loan percentage. The disclosure exists; the number to watch afterwards is how many of them slip anyway within a year of being restructured.
• The upgrade rule — an account that has become non-performing under the regulatory norms returns to standard only once the entire arrears of interest and principal have been paid, rather than when the borrower resumes paying instalments. When that rule was clarified, several lenders reported a one-off jump in reported bad loans with no deterioration whatsoever in the underlying book. A discontinuity in a series is sometimes a change in a definition.