A bank's internal models produce far lower risk-weighted assets than the standardised approach. Apply an output floor, recompute the CET1 ratio and explain what the floor protects against.

Harnavi reports RWA of Rs 36,250 crore and its CET1 ratio falls from 14.0% to 11.6%. The floor, 72.5% of Rs 50,000 crore, is above the modelled Rs 30,000 crore, so it binds. Keeping a 14.0% ratio would need Rs 875 crore more capital. The floor exists because models can understate risk, and it caps how far any bank's own numbers can cut its capital.

Think of a car insurer that gives careful drivers a discount based on a tracking device, but never lets the premium fall below 72.5% of the standard price, however good the data looks. The discount is real until it hits that line, and then it stops. The floored RWA is the higher of the modelled figure and 72.5% of the standardised figure. For Harnavi that is the higher of Rs 30,000 crore and Rs 36,250 crore, so the floor wins by Rs 6,250 crore.

Hold the ratio the bank showed investors, 14.0%, and ask what capital it now needs. 14.0% of Rs 36,250 crore is Rs 5,075 crore, so Harnavi is Rs 875 crore short of its old ratio, about 21% more equity for the same loans. It could raise equity, retain profit, or shrink the exposures where its models and the standardised weights differ most, since those are the ones the floor now prices at the standardised rate.

Harnavi's models say its book is 40% less risky than the standard weights assume. That may be true. But risk-weighted assetsExposures scaled by a weight for how risky each one is, so that a secured home loan needs less capital than an unsecured business loan. built from a bank's own default and loss estimates depend on short data histories, few defaults in safe portfolios, and choices the bank makes about its own capital. Studies of banks holding similar portfolios found their modelled RWA differed widely, which is hard to explain by risk alone. The floor does not say the model is wrong. It says the cut a model can deliver is capped at 27.5%, so an optimistic model, an honest error or a gamed input can only do limited damage.

Say the limitation as well. The floor is blunt: a bank whose low RWA is genuinely earned, say from well-collateralised lending, is charged as if it were not. And because it is set against the standardised approach, the floor inherits that approach's crude weights. A good answer calls the floor a backstop against model risk, not a better measure of risk.