Chapter 18 Option Pricing Without Perfect Replication Framework On the Edges of Arbitrage One-period Good-deal Bounds Multiple Periods and Continuous Time Extensions, Other Approaches, and Bibliography 2015/4/8 Asset Pricing 2 On the Edges of Arbitrage No Arbitrage Principle is Obvious, But… Not Trade Continuously, and Transactions Costs Stochastic Interest Rate or Stochastic Stock Volatility Underlying Asset Not Traded, Real Option for Example, or Short Selling Forbidden Internal Inconsistency of Arbitrage Portfolio So Unavoidable Basis Risk Arises. Many Authors simply add market price of risk assumptions. But this leaves the questions, How Sensitive are the Results to MPR Assumption? What are the reasonable Values for MPR 2015/4/8 Asset Pricing 3 On the Edges of Arbitrage We are still willing to take as given the prices of lots of assets in determining the price of an option, and in particular assets that will be used to hedge the option. We Form an “Approximate” hedge or portfolio of basis asset closest to the focus payoff, hedge most of the option’s risk. The uncertainty about the option value is reduced only to figuring out the price of the residual. Good-Deal Option Price Bounds: Systematically Searching over all Possible Assignments of the MPR to a reasonable Value, and Impose No Arbitrage Opportunities and Find Upper and Lower Bounds on the Option Price 2015/4/8 Asset Pricing 4 One-period Good-deal Bounds We want to price the payoff xc, where xc = Max(ST - K ,0) for a call option. Given an N-dimensional vector of basis payoffs x, whose prices p we can observe. The goodc deal bound finds the minimum and maximum value of x by search over all positive discount factors that price the basis assets and have limited volatility C min E (mx c ) {m} s.t. p E (mx); m 0; ( m ) h / R f E ( m 2 ) A2 , 2015/4/8 Asset Pricing 1 h2 A f2 R 2 5 One-period Good-deal Bounds In order of calculation, all the combinations of binding and nonbinding constrains: - - Volatility Constraint Binds, Positivity Constrain Slack Positivity Constraint Binds, Volatility Constraint Slack Volatility and Positivity Constraints both Bind Volatility Constraint Binds, Positivity Constrain Slack - C min E (mx c ) s.t. p E (mx), E (m2 ) A2 {m} - - Lagrange Multipliers Orthogonal Decomposition 2015/4/8 Asset Pricing 6 Volatility Constraint Binds, Positivity Constrain Slack 1, Given X , find x* c c 2, $x = proj ( x c | X ) = E ( x c x ') E ( xx ')- 1 x, w = x c - $x 3, p = E (mx ) = E (( x* + m* ) x ) = E ( x* x ) + E (m* x ) Þ E ( m* x ) = 0 Þ E ( m* x * ) = 0 4, m > 0, so triangular region 5, E (m 2 ) £ A2 , so circle m c c 6, E (mx c ) = E[( x* + m* )($x + w)] = E ( x* $x ) + E (m* w) * m m Þ m = x* + vw, m = x* - vw = x* - m Þ v= 2015/4/8 A2 - E ( x*2 ) w E ( w2 ) A2 - E ( x*2 ) E ( w2 ) Asset Pricing 7 Volatility Constraint Binds, Positivity Constrain Slack So the call price bound: C = E(mxc ) = E( x* xc ) - vE(w2 ) Interpretation: - The first term: approximate hedge portfolio E( x* xc ) E( x* xˆ c ) E(mxˆ c ) - The second term: Residual Consistent with Volatility Bound vE(w2 ) E(vww) E[( x* vw)w] E(mw) Explicit Option Pricing Formula: C / C = p ' E ( xx ')- 1 E ( xx c ) m A2 - p ' E ( xx ')- 1 p E ( x c 2 ) - E ( x c x ') E ( xx ')- 1 E ( xx c ) 2015/4/8 Asset Pricing 8 Volatility Constraint Binds, Positivity Constrain Slack Algebraic Expressions Any discount factor m that satisfies p = decomposed as: E (mx) can be m = x* + vw + e, x* = p ' E( xx ')- 1 x where E( x*w) = E( x*e) = E(we) So minimization problem is then: Min E (mx c ) s.t. E (m 2 ) £ A2 , {v ,e} c Min E[( x* + vw + e )($x + w) s.t. E ( x*2 ) + v 2 E ( w2 ) + E (e 2 ) £ A2 , {v ,e} c Min E ( x* $x ) + vE ( w2 ) s.t. E ( x*2 ) + v 2 E ( w2 ) + E (e 2 ) £ A2 {v ,e} Þ e = 0, v = 2015/4/8 ( A2 - E ( x*2 )) / E ( w2 ) Asset Pricing 9 Volatility and Positivity Constraints both Bind Direct Approach Introducing Lagrange multipliers, the problem is: C min max E (mx ) '[ E (mx) p] [ E (m 2 ) A2 ] {m 0} { , 0} 2 c Introducing Kuhn-Tucker multiplier p (s)v(s) on m( s) > 0, and taking partial derivatives wrt. m in each state C = Min å {m} s dé + êå 2 êë s é ù ê p ( s ) m ( s ) x ( s ) + l ' å p ( s ) m( s ) x ( s ) - p ú êë s ú û ù 2 2 p ( s )m( s ) - A ú+ å p ( s )v( s )m( s ) ú û s c 1 ¶ : x c ( s) + l ' x( s) + dm( s) + v( s) = 0 p ( s) ¶ s 2015/4/8 Asset Pricing 10 Volatility and Positivity Constraints both Bind Direct Approach Kuhn Multiplier Discussion - If the positivity constraint is slack, v(s) = :0 - x c ( s ) + l ' x( s ) m( s ) = d If the positivity constraint binds: m( s ) = 0 So the solution: + æ xc + l ' x ÷ ö é xc + l ' x ù ú m = Max çç, 0÷ = ê÷ çè ú ÷ d d ø êë û And Maximize: 2 x 'x 2 C Max E{( } ' p A { , 0} 2 2 2015/4/8 Asset Pricing 11 Volatility and Positivity Constraints both Bind Dual Approach Hansen, Heaton and Luttmer (1995) Interchanging Min and Max, C max min E (mx c ) '[ E (mx) p] [ E (m 2 ) A2 ] { , 0} {m 0} 2 We may obtain: 2 x 'x 2 C Max E{( } ' p A { , 0} 2 2 Search numerically over (l , d) to find the solution to this problem. 2015/4/8 Asset Pricing 12 Positivity Constraint Binds, Volatility Constraint Slack The problem: C Min E (mxc ), s.t. p E (mx), m 0 m These are the Arbitrage Bounds, Denote the lower arbitrage bound by Cl . The minimum-variance discount factor that generates the arbitrage bound Cl solves: 2 E (m )min 2015/4/8 p x Min E ( x ), s.t. E (m c ), m 0 {m} x Cl 2 Asset Pricing 13 Positivity Constraint Binds, Volatility Constraint Slack Introducing the Lagrange Multiplier: é ù ê L = å p ( s )m( s ) + 2l å p ( s )m( s ) x ( s ) - p ú êë s ú s û é ù + 2d êå p ( s) m( s ) x c ( s ) - Cl ú+ å p ( s )v( s )m( s ) êë s ú û s 2 Using the same conjugate method, the optimal m : + c é ù m = êë- (dx + l ' x)úû This problem is equivalent to: E (m2 )min max{ E{[( x c ' x)]2 2 ' p 2 Cl } { , } We search numerically for (l , d) to solve this problem. 2015/4/8 Asset Pricing 14 Good-Deal Comparisons with BS and Arbitrage Bounds We can obtain closer bounds on prices with more information about the discount factor. In particular, if we know the correlation of discount factor with xc, we could price the option better. 2015/4/8 Asset Pricing 15 Multiple Periods and Continuous Time The central fact that makes good-deal bounds tractable in dynamic environments that bounds are recursive. Consider two periods version: C 0 Min E0 (m1m2 x2c ) s.t. {m1 , m2 } pt Et (mt 1 pt 1 ), Et (mt21 ) At2 , mt 1 0, t 0,1 Be equivalent to a series of one-period problems: C1 Min E1 (m2 x2c ) {m2 } C 0 Min E0 (m1 C 1 ) { m1 } E0 (m1m2 x2c ) must optimize So the two-period problem {Min m ,m } c Min E1 (m2 x2 ) in each state of nature at 1 {m } The recursive property only holds if we impose m > 0 1 2 2 2015/4/8 Asset Pricing 16 Basis Risk and Real Options (One Dimension) Consider pricing a European call option on asset V that is not a traded asset, but correlated with a traded asset S that can be used as an approximate hedge Terminal Payoff: xTc Max(VT K ,0) dS Basis Asset: dz S S S dV V dt Vz dz Vw dw V Underling Asset: The dw risk cannot be hedged by the basis asset, so the market price of dw risk will matter to the option price Goal: Discount factor to price S and r f , has instantaneous volatility A 2015/4/8 Asset Pricing 17 Basis Risk and Real Options (One Dimension) SDF solution form: * dL d L = * ± L L dL * L* * m- r dL A - h dw, * = - rdt - hS dz, hS = S sS L 2 2 S éd L 2 ù Et ê 2 ú= A2 êL ú ë û Interpretation: , So the Good-deal bound is given by: And the Result: 1 éL T ù ê C t = Et Max(VT - K , 0)ú êL t ú ë û 1 sV T ) 2 2 ln(V0 / K ) + (h + r )T dV 2 2 2 s V = Et 2 = s Vz2 + s Vw ,d = , V sV T C or C = V0 ehT F (d + s V T ) - K - rT F (d - é æ öù A2 çç 2÷ ê ús , h = mS - r , h = mV - r ÷ h = êhV - hS çr - a 1 1 r ÷ V úV S ÷ hS2 sS sV ÷ çè êë øú û ìïï + 1 Upper Bound ædV dS ö s Vz r = corr çç , ÷ = , a = ÷ í ÷ s çè V S ø ïïî - 1 Lower Bound V 2015/4/8 Asset Pricing 18 Basis Risk and Real Options (One Dimension) Market Prices of Risk: if an asset has a price process P that loads on a shock s dw , then its expected return: Et With Sharpe ratio: l = æd L ÷ ö dP - r f dt = - s Et çç dw÷ ÷ çè L ø P Et dP - r f dt æd L ÷ ö dP P = - Et çç dw÷ Þ E - r f dt = l s t ÷ ç èL ø s P In the above Real Option Example: lz= 2015/4/8 mS - r ,l w = sS A2 - hS2 Asset Pricing 19 Continuous time (High Dimensions) Basis Assets: nS -dimension, Dividends rate D( S ,V , t )dt dS S ( S ,V , t )dt S ( S ,V , t )dz , E (dzdz ') I S Underlying Assets: nV -dimension dV V ( S , V , t )dt Vz ( S , V , t )dz Vw dw, V E (dwdw ') I , E (dwdz ') 0 Goal: Value an asset that pays continuous dividends at rate xc (S ,V , t )dt , and terminal payment xTc (S ,V , T ) s c xs ds Et ( T xTc ) s t t t C t min Et { s ,t s T } 2015/4/8 T Asset Pricing 20 Continuous time (High Dimensions) For small time intervals, discretize: C t t Min Et {x c t t (C t C )( t )} { } xtc E[d (C )] t 0 0 dt Min {d } C C It can also be expressed: d dC dC xtc Et dt r f dt Min Et { d } C C C Discount Factor Form: * * dL d L dL ° ' S - 1s dz , = * - vdw, * = - rdt - m S S S L L L ° = m + D - r, S = s s ' m S S S S S S 2015/4/8 Asset Pricing 21 Continuous time (High Dimensions) Volatility Constraint: 1 d 2 d d * 2 Et 2 A , and * vdw dt 1 d *2 2 vv ' A Et *2 A2 s' s 1s dt Market price of risk:v is the vector of market prices of risks of the dw shocks v 1 d Et dw dt Solutions: Guess the lower bound C follows: dC = mC ( S ,V , t )dt + s Cz ( S ,V , t )dz + s Cw ( S ,V , t )dw C 2015/4/8 Asset Pricing 22 Continuous time (High Dimensions) The Optimal SDF Theorem: The lower bound discount factor L t follows: d d * * vdw And C , Cz , Cw satisfy the restriction: x 1 d * ' C r Et ( * Cz dz ) v Cw C dt Where: Cw 1 d *2 v A ( Et *2 ) ' dt Cw Cw 2015/4/8 Asset Pricing 23 Continuous time (High Dimensions) Proof: xtc E[d (C )] 0 dt Min { d } C C d d * vdw * d ( * C ) xtc dC 0 dt Et * Min vE dw t C C C {v } 1 d *2 2 s.t. vv ' A Et *2 dt And Guess: dC = mC ( S ,V , t )dt + s Cz ( S ,V , t )dz + s Cw ( S ,V , t )dw C So Problem: 2015/4/8 d ( * C ) xtc 1 ' 0 Et Min v Cw * C dt C {v} d *2 1 2 s.t. vv ' A Et *2 dt Asset Pricing 24 Continuous time (High Dimensions) This is a linear objective in v with a quadratic constraint. Therefore, as long as s Cw ¹ 0 , the constraint binds and the optimal v : Cw 1 d *2 v A ( Et *2 ) ' dt Cw Cw So plugging the optimal value: xtc 1 d (* C ) ' 0 Et * v Cw C dt C For clarity, and exploiting the fact that d*does not load on dw , write the term as x 1 d * ' C r Et ( * Cz dz) v Cw C dt 2015/4/8 Asset Pricing 25 Continuous time (High Dimensions) The Optimal SDF: Plug in the definition of * , we may obtain: d rdt s' s 1 s dz A2 s' s 1s Cw Cw ' Cw dw xc ' C r s' s 1 s Cz A2 s' s 1s Cw Cw C The Good-Deal Option Price Bounds Theorem: The lower bound C(S ,V , t ) is the solution to the partial differential equation 2015/4/8 Asset Pricing 26 Continuous time (High Dimensions) ¶C 1 ¶ 2C x - rC + + å Si S j s Si s S' j ¶ t 2 i , j ¶ Si ¶ S j c 1 ¶ 2C + å (s Vz s V' z + s Vw s V' w ) + i j i j 2 i , j ¶ Vi¶ V j æD = çç çè S å i, j ¶ 2C Sis Si s V' z j ¶ Si ¶ V j ' ö÷ ' -1 ' -1 ' ' 2 ° ° ° C' s s ' C r ÷÷ ( SC S ) + (mS S S s S s Vz - mV )CV + A - mS S S m V Vw Vw V S ø Subject to the boundary conditions provided by the focus c x asset payoff T . Replacing + with – before the square root gives the partial differential equation satisfied by the upper bound 2015/4/8 Asset Pricing 27 Continuous time (High Dimensions) In general, the process depends on the parameters s Cw , so without solving the PDE we don’t not know how to spread the loading of d L across the multiple sources of risk dw whose risk prices we do not observe. Thus, we cannot use an integration approach to find the bound; we cannot characterize enough to calculate: s c xs ds E ( t xTc ) s t t t Ct E 2015/4/8 T Asset Pricing 28 Continuous time (High Dimensions) However, if there is only shock dw, then we don’t have to worry about how loading of d spreads across multiple sources of risk. V can be determined simply by the volatility constraint. in this special casedw and Cw are scalars. Theorem: In the special case that there is only one extra noise dw driving the V process, we can find the lower bound discount factor L directly from ' -1 dL ° = - rdt - mS S S s S dz L 2015/4/8 Asset Pricing ' -1 ° ° dw A - mS S S m S 2 29 Extensions, Other Approaches, and Bibliography The one-period good-deal bound is the dual to the HJ bound with positivity --- Hansen and Jagannathan (1991) study the minimum variance of positive discount factors that correctly price a given set of assets. The Good-deal bound interchanges the position of the option pricing equation and the variance of the discount factor. The techniques for solving the bound, are exactly those of the HJ bound in this one-period setup. This kind of problem needs weak but credible discount factor restrictions that lead to tractable and usefully tight bounds 2015/4/8 Asset Pricing 30 Extensions, Other Approaches, and Bibliography Several other similar restrictions: Monotonicity Restrictions: Levy (1985), Constantinide (1998), the discount factor declines monotonically with a state variable; marginal utility should decline with wealth. The good-deal bounds allow the worst case that marginal utility growth is perfectly correlated with a portfolio of basis and focus assets. So more credible correlation setup obtains tighter bounds Gain-Loss Restriction: Bernardo and Ledoit (2000) use the restriction a ³ m ³ b to sharpen the no-arbitrage restriction ¥ ³ m > 0. They show that this restriction has a beautiful portfolio inter-pretation -- a < m < b corresponds to limited gain-loss 2015/4/8 Asset Pricing 31 Extensions, Other Approaches, and Bibliography Define[Re ]+ = Max(Re ,0) and [Re ]- = - Min(- Re ,0) as the gains and losses of an excess return R e , then: [ Re ]+ sup(m) Max = Min {m:0= E ( mRe )} inf(m) {Re Î Re } [ Re ]- Bernardo and Ledoit also suggest a ³ m / y ³ b , where y is an explicit discount factor, such as the consumption based model or CAPM There alternatives are really not competitors The continuous-time treatment has not considered jumps and if with jumps, both positivity and volatility constraints will bind 2015/4/8 Asset Pricing 32 Thanks Your suggestion is welcome! 2015/4/8 Asset Pricing 33