Butterfly Spreads and the Vol Surface
A butterfly spread is a local curvature trade. In its familiar equal-strike-spacing form, buy one low-strike option, sell two centre options, and buy one high-strike option of the same expiry. The payoff is bounded, but its value is governed by the shape of the implied-volatility smile, the realized distribution around the centre strike, and the distinction between price convexity and implied-volatility curvature.
The payoff and its finite-difference meaning
For calls with strikes K-h, K, and K+h,
Butterfly = C(K-h) - 2*C(K) + C(K+h)
As h becomes small, this is a finite-difference approximation to the second strike derivative. Under no-arbitrage conditions,
partial^2 C / partial K^2 = exp(-rT) * q(K)
where q(K) is the risk-neutral terminal density. A non-negative butterfly price is therefore a basic test of strike convexity. This connection makes butterflies useful both as trades and as diagnostics for an implied volatility surface.
| Shape | Position | Distribution view | Main risk |
|---|---|---|---|
| Long call butterfly | long wings, short centre | terminal spot near centre | spot escapes range |
| Short butterfly | short wings, long centre | terminal distribution too concentrated | tail move |
| Iron butterfly | short ATM call/put, long wings | collect central premium | two-sided gap |
| Broken-wing butterfly | unequal wing spacing | asymmetric tail view | embedded directional exposure |
Volatility butterflies are quoted differently
Desks often define a 25-delta butterfly in volatility space:
BF_25 = 0.5 * [IV(25d put) + IV(25d call)] - IV(ATM)
It measures wing richness relative to ATM volatility, not the dollar price of an equal-strike butterfly. The two are related but not interchangeable: changing strike spacing, vegas, forwards, or delta conventions changes the comparison. Equity downside skew can make a symmetric strike butterfly strongly asymmetric in vega and in surface risk.
def vol_butterfly(atm_iv, put25_iv, call25_iv):
return 0.5 * (put25_iv + call25_iv) - atm_iv
def call_butterfly(price_fn, k, wing):
return price_fn(k-wing) - 2 * price_fn(k) + price_fn(k+wing)
Use volbutterfly for standardized surface monitoring and callbutterfly for executable package valuation. Never feed IVs directly into the second calculation.
Why a butterfly can lose despite being “long curvature”
At entry, a long butterfly is usually positive gamma near the centre and negative gamma in portions of the wings. Its net vega may be small, but not zero. A parallel vol move can matter, while a smile reshaping can matter more. If spot drifts through a wing, the option Greeks change and the short centre options can dominate. The payoff at expiry is bounded; mark-to-market loss before expiry need not be small.
The centre strike should follow the question. An ATM butterfly asks whether the central distribution is overpriced. A downside butterfly tests a local part of the left tail. In a post-crash market, a cheap-looking ATM butterfly may still sell an expensive downside wing through the centre/wing combination. Examine the full smile, not one curvature statistic.
Surface construction and arbitrage checks
Fit option prices or total variance with constraints, then derive quoted butterflies. Independent spline fits to IV can yield negative price butterflies even if every input quote appears reasonable. Check:
| Check | Condition | Failure implies |
|---|---|---|
| Vertical spread | -exp(-rT) <= dC/dK <= 0 | monotonicity violation |
| Butterfly | d2C/dK2 >= 0 | negative implied density |
| Calendar | total variance broadly nondecreasing | time arbitrage |
| Wing extrapolation | finite, plausible prices | tail model dominates |
For sparse chains, interpolate in delta or log-moneyness only after deriving the appropriate forward. SABR may provide a compact fit, but its approximation must still be checked for density positivity in the traded range.
Managing the strategy
Define maximum loss, exit time, and hedge rule before entry. The bounded expiration payoff does not remove liquidity risk: wide wing markets can make a package impossible to close near stress. Use limit orders at package prices, preserve each leg's exchange multiplier, and account for early exercise in American equity options around dividends.
Backtests must use historical option surfaces or conservative bid/ask fills. Reconstructing a butterfly from end-of-day mids exaggerates attainable curvature alpha because the signal and fill use the same noisy quote. Attribute results into theta, realized gamma, parallel vega, skew, curvature, and residual model P&L.
Key takeaways
- A strike butterfly is a local price-convexity and risk-neutral-density trade.
- A quoted volatility butterfly is a different, delta-based measure of wing richness.
- Small net vega does not mean small surface or spot-path risk.
- Non-negative butterflies are essential static-arbitrage checks for a surface.
- Trade executable packages and stress liquidity, skew, dividends, and pre-expiry marks.