Gamma and the Limits of Delta Hedging

Delta hedging promised that a small move in the stock leaves a hedged book unchanged. That promise is exact only in an idealized limit: infinitesimal price moves, rebalanced continuously. A real desk rebalances discretely — once an hour, once a day, whenever a trader has time to look up — and in between rebalances, the stock can move a real amount. The gap between what the option is actually worth after that move and what the linear, delta-only hedge predicted it would be worth is exactly what gamma measures.

Delta is a tangent line, not the curve

Expand the option value in a Taylor series around today's spot S_0:

Notice the error term is proportional to (\delta S)^2 — it doesn't care which direction the stock moved, only how far. A book that is short gamma (the usual state for a desk that has sold options and delta-hedged them) loses a little to this term on every move, up or down, large or small; a book that is long gamma gains from it. Either way, the bigger the move between rebalances, the bigger the slip — which is exactly why volatile stocks and long gaps between rehedges are dangerous for a delta-only hedge.

Worked example — pricing the slip

A desk's short-option book has \Gamma = -0.08 per share (i.e. per underlying share equivalent, aggregated over the whole book of 10{,}000 share-equivalents from the delta-hedging example). Overnight the stock jumps \delta S = \$2 before the desk can rebalance. The hedging error is:

\tfrac12\,\Gamma\,(\delta S)^2 = \tfrac12 \times (-0.08) \times 2^2 = -0.16 \text{ per share.}

Multiplied across the 10{,}000 share-equivalents in the book, that's a loss of \$1{,}600 — from a stock move the delta hedge, by construction, had already been "corrected" for. Had the stock moved \$4 instead of \$2, the loss would be four times as large (\$6{,}400), not twice — the error grows with the square of the gap, which is why desks with large gamma exposure rebalance far more often than the schedule this arithmetic might otherwise suggest is "good enough."

Seeing the gap

Below, the curved line is the true Black–Scholes call value C(S); the straight line is the linear approximation a delta-only hedge is implicitly betting on — C(S_0) + \Delta\,(S - S_0), the tangent at S_0 = \$50. Near S_0 the two are nearly indistinguishable — that's why delta hedging works well for small moves. Move away from S_0 in either direction and the true curve pulls above the tangent line (option value is convex in S) — that vertical gap, at whatever S the stock lands on, is the gamma P&L the delta hedge alone missed. Shorten T and watch the curve bend more sharply away from its tangent — less time to expiry concentrates gamma near the strike.

Gamma-neutral hedging: bringing in a second option

Delta can be hedged with the stock alone because the stock itself has delta 1 — trading it moves your linear exposure exactly where you want it. But the stock's payoff is a straight line: its gamma is 0. No amount of buying or selling shares can add or remove curvature from a portfolio, so the stock is useless for controlling gamma. The only way to offset the gamma of one option is with another option (or something else with optionality — a warrant, a convertible bond).

The recipe mirrors delta hedging, one level up: find a liquid option with gamma \Gamma_2, and trade n_2 units of it so the portfolio's total gamma is zero:

n_2 = -\frac{\Gamma_{\text{portfolio}}}{\Gamma_2}.

Suppose the book above has \Gamma_{\text{portfolio}} = -50 (per share, aggregated) and a liquid at-the-money option on the same stock has \Gamma_2 = 0.10 per share. Then n_2 = -(-50)/0.10 = 500 share-equivalents (five 100-share contracts) of that option neutralize the book's gamma. But that new option position carries its own delta — say \Delta_2 = 0.45 — which adds 500 \times 0.45 = 225 shares of fresh delta exposure. The last step is always the same: re-hedge delta with the stock, exactly as before. Managing a real book is this loop, run again and again as the market moves: flatten gamma with options, then flatten whatever delta that introduced with stock.

It's a fair question, since a share of stock also "moves with the market." The difference is curvature. Plot the payoff of a share of stock against the stock price: it's a perfectly straight 45° line. A straight line has a constant slope everywhere — its second derivative, gamma, is identically zero. Trading more or less stock shifts your position up and down that same straight line; it never bends the line itself.

An option's payoff, by contrast, is curved — that curvature is precisely what someone is paying (or being paid) for when the option changes hands. Only an instrument that is itself curved can add or cancel curvature elsewhere in a book. This is the fundamental reason a derivatives desk can never reduce its risk management to "just trade the underlying" — beyond delta, every other Greek needs other options.

It's tempting to read "our book is delta-neutral" as "our book cannot lose money from a stock move." The worked example above is the counterexample: a book with \Delta = 0 exactly at S_0 still lost \$1{,}600 when the stock jumped, purely from gamma. Delta-neutrality only guarantees the position is insensitive to the first, linear effect of a move — it says nothing about the second-order (gamma) effect, let alone larger moves where even the quadratic approximation itself starts to break down.

A related trap: gamma-neutrality achieved by matching \Gamma at today's spot is also only a local guarantee — it flattens the curvature exactly at S_0, but two different options rarely have matching gamma everywhere, so a book that's gamma-neutral today can develop gamma exposure again after a big move. "Neutral" always means neutral right now, to this order of approximation — never a permanent, unconditional guarantee.