Normal vs. Lognormal Greeks for Negative Rates Options

Question

My understanding is that for some of the G10 currencies with negative rates (CHF, EUR), Swaption and Cap / Floor prices are quoted in terms of BOTH, normal and log-normal Vols. That in itself is not controversial, because these vols are self-consistent (you plug the quoted log-normal vol into the (shifted) log-normal formula: you get a specific price. You plug the normal vol into the normal option pricing formula: you must get the same price. Otherwise an identical option would have two different prices and someone would exploit the arbitrage.

What about the arising Greeks through and thereby hedging? I have even come across an implementation of the Libor market model which could switch between normal and shifted log-normal diffusion: don't these different models produce different Greeks? Intuitively, they shouldn't, but just looking at the basic Normal (Bachelier) vs. Log-normal (Black-Scholes) option pricing formulas, the Greeks will be different. Doesn't that imply that two different banks using two different models would calculate the risk differently, with one bank inherently miss-hedging?

Answer 1

Yes, different banks using different models will get different Greeks. Some of them will be right and some of them will be wrong. What do we mean by right and wrong? ‘Right’ means that when the market moves, your Greeks closely predict how the market values of the options are moving. There are multiple examples in all asset classes (rates, equities, fx) where models and their associated Greeks have proven to be wrong , resulting in losses (or sometimes gains) at market makers.

Answer 2

Greeks represent rates of price change. Obviously they vary across models, if they didn't, the models would agree on prices across all market scenarios and thus the models have to be the exact same, which is a contradiction.

As far as hedging goes, nobody has the right hedge. For a Greek to be 'true' the model should:

1. Exactly capture the underlying dynamics, including exact calibration to the true dynamics. This is not possible.

2. If the option market dynamics are inconsistent with the underlying (but consistent enough to avoid arbitrage), the model should take that into account. This too is not possible.

In practice, losses from discretized hedging overpower the errors that creep in during calculation of greeks due to the above.