Delta Hedging and Hedging Error

Motivation: why this matters in quant finance

Black-Scholes is not principally a pricing formula — it is a replication argument. Its central claim is: an option's payoff can be replicated by dynamically trading the underlying and cash, at zero net cost, if you follow a specific strategy ("delta hedging"). The price of the option is the initial cost of that replicating portfolio. If dynamic replication works exactly, any other price admits arbitrage.

In practice, delta hedging produces residual hedging error — P&L that wasn't supposed to happen if the Black-Scholes assumptions held perfectly. That error has a precise decomposition:

Discretisation error: you rebalance at $n$ discrete times, not continuously. Error vanishes as $n \to \infty$ but the variance scales as $O(1/n)$ .
Volatility mis-specification: you hedge using a Black-Scholes model with hedge vol $\sigma_h$ , but the underlying realises volatility $\sigma_r$ . Error is proportional to $(\sigma_r^2 - \sigma_h^2)$ integrated against gamma.
Jump risk: Black-Scholes assumes continuous paths; if the underlying jumps, delta hedging captures no gamma benefit and the hedger's P&L includes the unhedged jump.
Transaction costs: each rebalance costs a bid-ask spread; the hedger trades off rebalancing frequency against transaction drag.

Every options desk lives inside this error budget. This note derives the hedging-error decomposition from Itô's lemma, makes the volatility-mis-specification term explicit, and discusses the frequency/variance trade-off.

The informal idea

A perfect delta hedge means holding $\Delta_t := \partial V/\partial S$ units of the underlying at every moment. Then the portfolio $\Pi_t = V_t - \Delta_t S_t$ is locally insensitive to $S$ : $d\Pi$ has no $dS$ term. Under Black-Scholes assumptions (continuous paths, known $\sigma$ , no transaction costs), the remaining pieces cancel and the hedger's P&L is exactly zero: the option was replicated.

Break any assumption and an error term appears. The error has a structure that makes delta hedging interpretable as a trade on realised vs implied volatility: a long-option holder earns

\tfrac12\sigma_r^2 S^2 \Gamma \, dt

every instant from realised volatility and pays

\tfrac12\sigma_h^2 S^2 \Gamma\, dt

as time decay. If

\sigma_r > \sigma_h

, the hedger (long gamma) profits.

Formal analysis

Continuous hedging

Let

V(S, t)

be the option value in a Black-Scholes world with volatility

\sigma_h

(the "hedge vol"). Let the actual underlying follow

dS_t = \mu_t S_t\,dt + \sigma_r S_t\,dW_t,

with realised volatility $\sigma_r$ (possibly different from $\sigma_h$ , possibly time-varying).

The hedger's portfolio is $\Pi_t = V_t - \Delta_t S_t$ , with the cash balance invested at rate $r$ . Applying Itô's lemma to $V(S_t, t)$ :

dV = \frac{\partial V}{\partial t}dt + \frac{\partial V}{\partial S}dS + \frac{1}{2}\frac{\partial^2 V}{\partial S^2}\sigma_r^2 S^2\, dt.

The crucial point is that the

dS^2 = \sigma_r^2 S^2 dt

term uses the realised volatility, not the hedge volatility that

V

was priced with.

Over an instant $dt$ , the hedger's P&L is

d\Pi - r\Pi\,dt = dV - \Delta\,dS - r(V - \Delta S)\,dt.

Substituting from Itô and using the Black-Scholes PDE $\frac{\partial V}{\partial t} + \tfrac12\sigma_h^2 S^2\Gamma + rS\Delta - rV = 0$ (which $V$ satisfies under hedge vol):

d\Pi - r\Pi\,dt = \tfrac{1}{2}\Gamma S^2 (\sigma_r^2 - \sigma_h^2)\,dt.

This is the hedging-error differential. Integrating from

0

T

P\&L = \int_0^T \tfrac{1}{2}\Gamma_t S_t^2 (\sigma_{r,t}^2 - \sigma_{h}^2)\,dt.

Interpretation. The hedger's total P&L over the life of the option is a gamma-weighted integral of the squared-volatility difference. When realised vol exceeds hedge vol, long-gamma (long option, delta hedged) is profitable. When realised is lower, you lose.

Discrete rebalancing

If the hedger rebalances at times $0 = t_0 < t_1 < \ldots < t_n = T$ instead of continuously, an additional error appears. To leading order, the discretisation error at each step is a zero-mean random variable proportional to $\tfrac12\Gamma S^2 (Z_i^2 - 1)\,\Delta t$ where $Z_i$ is a standard normal over the step. Summing over $n$ steps:

\text{Discretisation error} \approx \sum_{i=1}^n \tfrac12 \Gamma_i S_i^2 (Z_i^2 - 1)\sigma_h^2 \Delta t.

Variance of discretisation error:

\text{Var} \propto \sum \Gamma_i^2 S_i^4 \sigma_h^4 (\Delta t)^2

. For equal steps

\Delta t = T/n

, this is

O(T^2/n) \cdot \text{(constant)}

. So discretisation-error standard deviation is

O(1/\sqrt{n})

: four times more rebalances halves the error standard deviation.

Transaction costs

Each rebalance requires trading $\Delta_{t_i} - \Delta_{t_{i-1}}$ units of underlying. The expected absolute rebalancing over each step scales as $O(\sqrt{\Delta t})$ (from Brownian volatility of delta), so total rebalancing cost scales as $n \cdot O(\sqrt{\Delta t}) = n \cdot O(\sqrt{T/n}) = O(\sqrt{nT})$ .

Transaction cost grows with

n

. Discretisation error shrinks with

n

. The optimal rebalancing frequency minimises the sum — a classic bias-variance trade-off.

Leland's formula (1985) gives an analytical solution: the optimal hedge vol is

\sigma_h^{Leland} = \sigma \sqrt{1 + \sqrt{\tfrac{8}{\pi}}\tfrac{k}{\sigma\sqrt{\Delta t}}}

where

k

is proportional bid-ask spread. Higher transaction costs inflate the effective hedge vol, reducing gamma exposure.

Key properties

Path-independent total error (continuous hedging, fixed $\sigma_r$ ). If realised volatility is constant, $\int_0^T \tfrac12\Gamma S^2 (\sigma_r^2 - \sigma_h^2) dt$ depends on the path only through $\int \Gamma S^2 dt$ . Different paths produce different total P&L.
Volatility-of-volatility. If realised vol is stochastic, hedging error correlates with vol changes. An option is effectively a leveraged bet on realised vol; vol-of-vol is a second-order sensitivity (vanna/volga) that matters.
Gamma magnitude and hedging error variance. For short-dated ATM options, gamma is huge and hedging error is loud. For deep-OTM or deep-ITM options, gamma is small and delta-hedging approximation holds nearly exactly.
Jump risk. A sudden jump $\Delta S$ produces hedging error $V(S + \Delta S) - V(S) - \Delta \cdot \Delta S$ , which is not a second-order correction — it's a first-order miss. Gamma hedging (buying/selling other options to zero out gamma) partly addresses this.
Asymmetric risk of short-gamma positions. Being short an option and delta-hedging means negative gamma. Large market moves cost, small moves earn (theta). If realised vol spikes, the short-gamma hedger gets hurt twice: once by realised vol, once by the vol-of-vol widening his vega exposure.

Worked example — hedging a short ATM call

Sell one call at hedge vol $\sigma_h = 20\%$ : $S_0 = 100$ , $K = 100$ , $T = 1/4$ (3 months), $r = 0$ . Receive premium $\approx \$ 4.01$.

Realised vol scenario 1:

\sigma_r = 20\%

throughout. Continuous delta hedging: expected P&L =

0

(zero-mean random variable from discretisation only). Discrete hedging at

n = 63

trading days: discretisation error standard deviation

\approx \sigma_h^2 S_0 \sqrt{T/n}\cdot \text{(const)} \approx \

0.15

. So total P&L is mean zero, SD

$0.15

on a premium of

$4.01

— about

3.7%$ CV.

Realised vol scenario 2:

\sigma_r = 25\%

throughout.

\sigma_r^2 - \sigma_h^2 = 0.0625 - 0.04 = 0.0225

. Integrated gamma-weighted error:

\text{Expected P&L} = \tfrac12 \int_0^T \mathbb{E}\left[\Gamma_t S_t^2\right]\left(\sigma_r^2 - \sigma_h^2\right)dt = \tfrac{1}{2} \cdot 0.0225 \cdot \mathbb{E}\left[\text{vega-like integral}\right].

A back-of-envelope: for an ATM call with parameters above, vega $\approx 19.8$ , so sensitivity of price to $\sigma$ is $19.8 \cdot \Delta\sigma$ . From $20\%$ to $25\%$ : premium would have been $\approx 19.8 \cdot 0.05 = \$ 0.99 $higher. A short hedger under-priced by$ $0.99$ and now bleeds that amount over the hedge period.

**So the short hedger loses approximately $\$ 1 $on each call sold, not from catastrophic events but from steady leakage through excess realised vol.** This is the core "sell vol" trade: collect premium hoping realised$ <$ implied, hedge the directional risk, pocket the difference. And lose when realised exceeds implied.

Common confusions and pitfalls

Delta-hedged ≠ risk-free. The hedge zeros out first-order price risk. Second-order (gamma, realised vs implied vol), volatility exposure (vega), rate exposure (rho), time decay (theta) — all remain.
"Just hedge continuously" isn't operationally possible. Exchanges have minimum tick sizes; trading is expensive; overnight and weekend you cannot rebalance.
Historical vs implied. Pricing inputs are risk-neutral. Realised vol is observed $\mathbb{P}$ -measure vol. The hedger's P&L depends on both — the option is priced on implied, realised determines the gamma P&L.
Gamma scalping. The explicit strategy of "buy options, hedge delta, collect gamma when market moves" is what the formula describes. Market-makers scalping gamma pay theta; they win if realised exceeds implied.
Hedging with the wrong Greek. For path-dependent options, $\Delta$ computed from Black-Scholes may not match the true replication strategy. Numerical Greeks from the actual pricing model are essential.

Where this goes next

Black-Scholes PDE — the PDE route to the same conclusion.
Theta and Rho — theta as the payment for gamma.
Stochastic Volatility Models — when realised vol is itself stochastic.