Vanna-Volga Method

A scale with only three known reference weights — a 1kg, a 5kg, a 10kg block — can still calibrate an unknown weight in between, by seeing how the known blocks bend the spring compared to theory. Vanna-volga does the same for an option's smile: instead of fitting a full model to every strike, it uses just three liquid benchmark instruments to work out how far a plain Black-Scholes price is from the truth, and applies that correction to any other strike.

Why Black-Scholes alone isn't enough

Black-Scholes assumes one flat vol for every strike, which is why it can't reproduce a smile. A smile exists because the market prices in risks flat Black-Scholes ignores: sensitivity of delta to a change in vol (vanna) and sensitivity of vega to a change in vol (volga). Vanna-volga's idea: measure what it costs to hedge away the vanna and volga risk of a target option using three market-standard instruments, and add that cost to the flat Black-Scholes price.

The correction, defined

$$C_{\text{VV}} = C_{\text{BS}} + w_{\text{vega}} \times X_{\text{ATM}} + w_{\text{vanna}} \times X_{\text{RR}} + w_{\text{volga}} \times X_{\text{BF}}$$

In words: the vanna-volga price $C_{\text{VV}}$ starts from the plain Black-Scholes price $C_{\text{BS}}$ and adds three correction terms. Each $X$ is the market-observed hedging cost of one Greek — $X_{\text{ATM}}$ from the ATM straddle (pure vega), $X_{\text{RR}}$ from a risk reversal (pure vanna), $X_{\text{BF}}$ from a butterfly (pure volga). Each $w$ is a weight — how much of that benchmark's Greek the target option shares, from ratios of the target's own vanna and volga to the benchmark's. In plain terms: measure how vanna- and volga-exposed the odd-strike option is relative to the three liquid reference trades, and charge it that same proportional slice of what those trades cost above flat Black-Scholes.

Worked example 1 — pricing a 10-delta option from three benchmarks

Suppose a desk has backed out that hedging pure volga risk (via the butterfly) costs an extra $0.008 per unit vega, and pure vanna risk (via the risk reversal) costs an extra $0.004 per unit vega, relative to flat Black-Scholes. A 10-delta call, further out than the 25-delta benchmarks, carries 1.4x the volga exposure and 1.1x the vanna exposure of those benchmarks, per unit vega. Its correction: $1.4 \times 0.008 + 1.1 \times 0.004 = 0.0112 + 0.0044 = 0.0156$ vega-equivalent, which converts (times the option's actual dollar vega) into a concrete add-on to the flat Black-Scholes price.

Worked example 2 — why the ATM correction is roughly zero

At the ATM strike, vanna and volga are both close to zero — ATM options are, to first order, pure vega — so both correction terms shrink toward zero and $C_{\text{VV}} \approx C_{\text{BS}}$. This is a useful sanity check: the method reproduces the ATM quote almost exactly since that's one of its own calibration inputs, and deviates more moving away from the money, exactly where vanna and volga exposure grow.

What this means in practice

Vanna-volga is popular on FX desks because it needs only the three quotes every desk already has (ATM, RR, BF) and no separate model calibration — fast, transparent, and exact at the three inputs by construction. It's most often used to price odd-strike or odd-tenor options that don't trade liquidly enough to have their own quoted vol.

Practice

1. Why does the vanna-volga correction shrink to nearly zero exactly at the ATM strike, and what does that imply about how well the method should reproduce the ATM market quote?
2. A far-OTM option has 2x the volga exposure of the 25-delta butterfly benchmark and negligible vanna exposure. If the butterfly's volga-hedging cost is $0.01 per unit vega, roughly what correction (in vega-equivalent terms) should be added to that option's flat Black-Scholes price?