05.02.2014 Views

Static Hedging and Model Risk for Barrier Options

Static Hedging and Model Risk for Barrier Options

Static Hedging and Model Risk for Barrier Options

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong><br />

<strong>Barrier</strong> <strong>Options</strong><br />

Morten Nalholm<br />

Department of Applied Mathematics <strong>and</strong> Statistics<br />

Institute <strong>for</strong> Mathematical Sciences<br />

University of Copenhagen<br />

www.math.ku.dk/∼nalholm<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.1/16


Introduction<br />

Traditional option pricing: Try to replicate whatever with a<br />

dynamically adjusted portfolio in the stock.<br />

Next idea : Replicate exotic options with plain vanilla<br />

options.<br />

Sometimes the hedge may not even require dynamic<br />

trading, ie. it is static.<br />

But the hedges are based on model assumptions so what<br />

about model risk?<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.2/16


<strong>Barrier</strong> <strong>Options</strong><br />

A barrier option is a plain vanilla option where something<br />

happens if the stock price hits the barrier B be<strong>for</strong>e expiry.<br />

A discontinuous case: The up-<strong>and</strong>-out call<br />

F ≡ 0<br />

B<br />

0<br />

❆❑<br />

❆<br />

✁ ✁✕ F(B,t) = 0<br />

❈❖<br />

K<br />

❈<br />

F(x,T) ❈<br />

= (x − K) +<br />

Alive region: F solves B/S-PDE<br />

0 T<br />

<br />

✛<br />

❝<br />

<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.3/16


<strong>Static</strong> <strong>Hedging</strong><br />

What <strong>Hedging</strong> an option by a portfolio bought at initiation,<br />

which has the same payoff as the option <strong>and</strong> requires<br />

no dynamic trading.<br />

Why Less trading necessary with resulting advantages.<br />

How Two major approaches in the literature<br />

Calendar spreads.<br />

Strike spreads.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.4/16


Literature<br />

Derman, various papers from the mid-90’ies<br />

Carr, numerous papers from the mid-90’ies <strong>and</strong> on<br />

Andersen, Andreasen & Eliezer (2002)<br />

Nalholm & Poulsen (2005), “<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong><br />

<strong>for</strong> <strong>Barrier</strong> <strong>Options</strong>”<br />

Poulsen (2004), “Exotic <strong>Options</strong>: Proofs without Formulas”<br />

Nalholm & Poulsen (2005),“Applied <strong>Static</strong> <strong>Hedging</strong> of<br />

<strong>Barrier</strong> <strong>Options</strong>”<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.5/16


<strong>Hedging</strong> Using Calendar Spreads I<br />

Assume that<br />

Call, Put (spot, time | strike, expiry)<br />

are known functions (enter model risk) <strong>and</strong> that time points<br />

0 = t 0 < t 1 ... t n−1 < t n = T are given.<br />

Form a PF of the underlying option <strong>and</strong> α n strike-B expiry-T<br />

calls<br />

α n ∗ Call(B, t n−1 |B, T) + Call(B, t n−1 |K, T) = 0<br />

Value 0 at (B,t n−1 ); same payoff as the barrier option at T<br />

if S(T) < B.<br />

Hedge portfolio by recursion:<br />

α i ∗ Call(B, t i−1 |B; t i ) +<br />

nα j ∗ Call(B, t i−1 |B; t j ) + Call(B, t i−1 |K; T) = 0<br />

j=i+1<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.6/16


<strong>Hedging</strong> Using Calendar Spreads II<br />

(Andersen et al.) This also works <strong>for</strong> any local volatility<br />

model.<br />

(Andersen et al.) Jumps can be hedged using<br />

hedge-strikes beyond the barrier at each point in time.<br />

Effectively, this sets the hedge value to 0 at possible levels<br />

to which the stock can jump.<br />

(Allen & Padovani <strong>and</strong> Fink) Stochastic volatility can<br />

operationally be hedged by hedging different volatility levels<br />

at the barrier by use of extra hedge-strikes.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.7/16


<strong>Hedging</strong> Using Strike Spreads<br />

It can be shown (Carr) that under certain model<br />

assumptions there exists a simple T -claim that is equivalent<br />

to the barrier option along the barrier <strong>and</strong> at expiry. The<br />

payoff is an adjusted version of the original payoff.<br />

This adjusted payoff can be hedged by varying strike,<br />

expiry-T options.<br />

The method exploits symmetry<br />

(Poulsen) provides a survey.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.8/16


<strong>Model</strong> risk<br />

How sensitive are the static hedges to model risk?<br />

Guess I: Not very because ...<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.9/16


<strong>Model</strong> risk<br />

How sensitive are the static hedges to model risk?<br />

Guess I: Not very because ...<br />

... natural hedge instruments are used<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.9/16


<strong>Model</strong> risk<br />

How sensitive are the static hedges to model risk?<br />

Guess I: Not very because ...<br />

... natural hedge instruments are used<br />

... Carr says so in <strong>Risk</strong> 2003: incorporating static or<br />

semi-static positions in options [...] can often reduce<br />

or eliminate model risk<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.9/16


<strong>Model</strong> risk<br />

How sensitive are the static hedges to model risk?<br />

Guess I: Not very because ...<br />

... natural hedge instruments are used<br />

... Carr says so in <strong>Risk</strong> 2003: incorporating static or<br />

semi-static positions in options [...] can often reduce<br />

or eliminate model risk<br />

Guess II: Quite a lot because<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.9/16


<strong>Model</strong> risk<br />

How sensitive are the static hedges to model risk?<br />

Guess I: Not very because ...<br />

... natural hedge instruments are used<br />

... Carr says so in <strong>Risk</strong> 2003: incorporating static or<br />

semi-static positions in options [...] can often reduce<br />

or eliminate model risk<br />

Guess II: Quite a lot because<br />

... OTM options are where model prices differ the<br />

most; these options are used in the hedges<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.9/16


<strong>Model</strong> risk<br />

How sensitive are the static hedges to model risk?<br />

Guess I: Not very because ...<br />

... natural hedge instruments are used<br />

... Carr says so in <strong>Risk</strong> 2003: incorporating static or<br />

semi-static positions in options [...] can often reduce<br />

or eliminate model risk<br />

Guess II: Quite a lot because<br />

... OTM options are where model prices differ the<br />

most; these options are used in the hedges<br />

... strong assumptions about model, parameter<br />

values <strong>and</strong> values of the processes are used<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.9/16


<strong>Model</strong>s<br />

We investigated different implementations of the strategies<br />

against popular non-Black/Scholes models<br />

Constant elasticity of variance (CEV) where σ = σS α t<br />

Heston’s stochastic volatility model where<br />

σ 2 = a Cox/Ingersoll/Ross square-root model<br />

Merton’s jump diffusion where stock returns are hit by<br />

Poisson arrival of (displaced) lognormal shocks.<br />

Madan, Carr <strong>and</strong> Chang’s Variance Gamma model<br />

which is an (infinite intensity) pure jump process, that<br />

can be thought of as a time-changed Geometric<br />

Brownian motion.<br />

All calibrated to generate the same 1Y implied skew.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.10/16


Numerical results<br />

Reported number: St<strong>and</strong>ard deviation of hedge error<br />

relative to true option price in percent.<br />

<strong>Static</strong> hedges use 3 (<strong>for</strong> D&O-call) <strong>and</strong> 11 (<strong>for</strong> U&O-call)<br />

options. Dynamic hedging is done bi-daily.<br />

Mixed ∆: ∆-hedge “barrier - underlying option”.<br />

Smile <strong>and</strong> uni<strong>for</strong>m scaling: “What would traders do”-type<br />

adjustments to Black/Scholes-based static strike-hedges.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.11/16


DOWN-AND-OUT CALL<br />

Hedge method BS CEV SV JD VΓ<br />

∆; pure 10.1 10.1 15.4 32.6 29.8<br />

∆; mixed 2.0 2.8 4.5 6.1 5.1<br />

STR-static; simple 1.9 14.0 13.3 11.2 11.2<br />

STR-static; unif. scaled 6.2 2.5 2.5 2.5<br />

STR-static; smile-scaled 0.8 1.3 2.3 2.3<br />

CAL-static; simple 1.4 1.8 2.9 2.9 2.8<br />

CAL-static; true CEV 1.5<br />

CAL-static; 1 cond. vol. 3.2<br />

CAL-static; 5 vol. levels 3.1<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.12/16


UP-AND-OUT CALL<br />

Hedge method BS CEV SV JD VΓ<br />

∆; pure 95 120 86 133 151<br />

∆; mixed 94 122 93 150 166<br />

STR-static; simple 52 36 38 30 32<br />

STR-static; unif. scaled 55 58 66 70<br />

STR-static; smile-scaled 26 19 31 42<br />

CAL-static; simple 63 75 57 54 57<br />

CAL-static; true CEV 65<br />

CAL-static; 1 cond. vol. 49<br />

CAL-static; 5 vol. levels 32<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.13/16


You don’t need that many options to beat ∆-hedging,<br />

misspecification or not.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.14/16


You don’t need that many options to beat ∆-hedging,<br />

misspecification or not.<br />

Discontinuous options are hard to hedge <strong>and</strong> jumps<br />

<strong>and</strong> barriers are a nasty combination.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.14/16


You don’t need that many options to beat ∆-hedging,<br />

misspecification or not.<br />

Discontinuous options are hard to hedge <strong>and</strong> jumps<br />

<strong>and</strong> barriers are a nasty combination.<br />

∆-hedge-accuracy deteriorates rapidly with<br />

“non-Black/Scholes’ness” ...<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.14/16


You don’t need that many options to beat ∆-hedging,<br />

misspecification or not.<br />

Discontinuous options are hard to hedge <strong>and</strong> jumps<br />

<strong>and</strong> barriers are a nasty combination.<br />

∆-hedge-accuracy deteriorates rapidly with<br />

“non-Black/Scholes’ness” ...<br />

... whereas quality of the static hedges are fairly robust<br />

across models.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.14/16


You don’t need that many options to beat ∆-hedging,<br />

misspecification or not.<br />

Discontinuous options are hard to hedge <strong>and</strong> jumps<br />

<strong>and</strong> barriers are a nasty combination.<br />

∆-hedge-accuracy deteriorates rapidly with<br />

“non-Black/Scholes’ness” ...<br />

... whereas quality of the static hedges are fairly robust<br />

across models.<br />

You need to take account of the smile/skew in static<br />

hedges ...<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.14/16


You don’t need that many options to beat ∆-hedging,<br />

misspecification or not.<br />

Discontinuous options are hard to hedge <strong>and</strong> jumps<br />

<strong>and</strong> barriers are a nasty combination.<br />

∆-hedge-accuracy deteriorates rapidly with<br />

“non-Black/Scholes’ness” ...<br />

... whereas quality of the static hedges are fairly robust<br />

across models.<br />

You need to take account of the smile/skew in static<br />

hedges ...<br />

... <strong>and</strong> not just any ad-hoc adjustment to Black/Scholes<br />

will do.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.14/16


Conclusions <strong>and</strong> ongoing research<br />

<strong>Hedging</strong> exotic options with plain vanilla is possible to<br />

underst<strong>and</strong> <strong>and</strong> do, also in realistic settings.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.15/16


Conclusions <strong>and</strong> ongoing research<br />

<strong>Hedging</strong> exotic options with plain vanilla is possible to<br />

underst<strong>and</strong> <strong>and</strong> do, also in realistic settings.<br />

The reported misspecification analysis was of<br />

“horse-race”.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.15/16


Conclusions <strong>and</strong> ongoing research<br />

<strong>Hedging</strong> exotic options with plain vanilla is possible to<br />

underst<strong>and</strong> <strong>and</strong> do, also in realistic settings.<br />

The reported misspecification analysis was of<br />

“horse-race”.<br />

We are working on a “full Monty” model based<br />

implementation. Many details. Some are not important.<br />

Others are.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.15/16


Conclusions <strong>and</strong> ongoing research<br />

<strong>Hedging</strong> exotic options with plain vanilla is possible to<br />

underst<strong>and</strong> <strong>and</strong> do, also in realistic settings.<br />

The reported misspecification analysis was of<br />

“horse-race”.<br />

We are working on a “full Monty” model based<br />

implementation. Many details. Some are not important.<br />

Others are.<br />

Empirical testing on times-series option data is then a<br />

“no brainer”.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.15/16


Conclusions <strong>and</strong> ongoing research<br />

<strong>Hedging</strong> exotic options with plain vanilla is possible to<br />

underst<strong>and</strong> <strong>and</strong> do, also in realistic settings.<br />

The reported misspecification analysis was of<br />

“horse-race”.<br />

We are working on a “full Monty” model based<br />

implementation. Many details. Some are not important.<br />

Others are.<br />

Empirical testing on times-series option data is then a<br />

“no brainer”.<br />

Single-barrier options aren’t that exotic. OK fine, but<br />

I’ve have never seen anything remotely resembling a<br />

quote. I underst<strong>and</strong> “OTC”, but this is ridiculous.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.15/16


Bibliography<br />

Allen & Padovani (2002). ’<strong>Risk</strong> management using quasi-static hedging’, Economic<br />

Notes 31(2).<br />

Andersen, Andreasen & Eliezer (2002). ’<strong>Static</strong> replication of barrier options: some<br />

general results’, Journal of Computational Finance 5.<br />

Carr & Chou (1997). ’Breaking barriers’, <strong>Risk</strong> Magazine 10.<br />

Derman, Ergener & Kani (1995). ’<strong>Static</strong> <strong>Options</strong> Replication’, Journal of Derivatives 2.<br />

Fink (2003). ’An examination of the effectiveness of static hedging in the presence of<br />

stochastic volatility’, Journal of Futures Markets 23.<br />

Heston (1993). ’A closed-<strong>for</strong>m solution <strong>for</strong> options with stochastic volatility with<br />

applications to bond <strong>and</strong> currency options’, Review of Financial Studies 6.<br />

Madan, Carr & Chang (1998). ’The variance gamma process <strong>and</strong> option pricing’,<br />

European Finance Review 2.<br />

Merton (1976). ’Option pricing when the underlying stock returns are discontinuous’,<br />

Journal of Financial Economics 5.<br />

Nalholm & Poulsen (2005). ’<strong>Static</strong> hedging <strong>and</strong> model risk <strong>for</strong> barrier options’, Journal<br />

of Futures Markets <strong>for</strong>thcoming.<br />

Poulsen (2004). ’Exotic options: proofs without <strong>for</strong>mulas’, working paper.<br />

<strong>Static</strong> <strong>Hedging</strong> <strong>and</strong> <strong>Model</strong> <strong>Risk</strong> <strong>for</strong> <strong>Barrier</strong> <strong>Options</strong> – p.16/16

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!