Trinomial Tree Implementation

Embed Size (px)

Citation preview

  • 8/18/2019 Trinomial Tree Implementation

    1/2

    Stochastic Differential Equations and Interest Rate Models

    Trinomial Tree Implementation of the Hull-White Model

    12 February 2008

    The model is

    drt = (θ(t)− λrt)dt + σdwt,so that

    rt   =   e−λtr0 +

       t0

    e−λ(t−s)θ(s)ds +

       t0

    e−λ(t−s)σdws

    ≡   a(t) + e−λtX t,   (1)

    where X t   is the zero-mean gaussian martingale

    X t =

       t0

    eλsσdws.

    The tree models  X t. The final maturity time is  T  years, and there are  M  time steps per year,

    so the discrete time increment is  δ  = 1/M  with a total of  T M   steps. The spatial step is h. If 

    we take

    h =  σ√ 

    3δ    (2)

    and up and down probabilities equal to 1/6, then the trinomial step matches the first 5 moments

    of the normal distribution  N (0, σ2). Here  X t  does not have stationary increments, so we let  σ2

    be an ‘average’ variance, defined by

    σ2 =  σ2

    2λT 

     (e2λT 

    −1) =

      1

     E [X 2T ]

    ≡σ2b2.

    Thus

    h =  σ 

    3b2δ.

    Now

    E (X t+δ − X t)2 =   σ2   t+δt

    e2λsds

    =  σ2e2λt

    2λ  (e2λδ − 1)

    =   σ2t δ,

    so

    σ2t   = σ2e2λtb1,

    where

    b1 =  1

    2λδ (e2λδ − 1).

    To match the incremental variance at time  t  =  kδ , and recalling that  h   is given by (2), the up

    and down probability  pk  must satisfy 2 pkh2 = σ2t δ , from which we find

     pk  = 1

    6

    b1b2

    e2λδk.

    1

  • 8/18/2019 Trinomial Tree Implementation

    2/2

    Calibration to the Yield Curve for given   σ, λ

    Let us denote the discrete-time process on the tree by (Y k), so that Y k ∈ { jh  : −k ≤  j ≤ k}.We want to ensure that the tree correctly calculates the zero-coupon bond values

    Bk  = p(0, kδ ),   1

    ≤k

    ≤T M,   (3)

    which are today’s market data. Index the nodes at time   k   as (k, j) where −k ≤  j ≤   +k. In(1) we take   a(t) to be piecewise constant:   a(t) =   ak, t ∈   [kδ, (k + 1)δ [. We thus have   T M constraints to determine the   T M   values  a0, . . . , aTM −1. This is done by a forwards induction

    procedure involving the so-called   Arrow-Debreu prices   q k,j . The price  q k,j  is the value at time 0

    of a contract that delivers  $1 if  Y k  =  jh  and zero otherwise. The calibration condition is thus

    Bk  =k

     j=−k

    q k,j , k = 1, . . . , T M .

    (Of course, B0

     =  q 0,0

     = 1.) At node (k, j) the short rate is, from (1),  ak

    + e−λδk jh. We interpret

    this as the period-δ  rate set at time  kδ , so the discount factor for the period [kδ, (k + 1)δ ] when

    Y k  =  jh  is1

    1 + δ (ak + e−λδk jh) ≡   1

    1 + αk + jβ k

    We recursively compute the coefficients   αk   and the Arrow-Debreu (AD) prices, and then we

    throw away the latter. They are not needed subsequently. Note that the  β k  are known.

    At time  k  = 0, the discount factor is 1/(1 + α0), so

    B1 =  1

    1 + α0, α0  =

      1

    B1− 1.

    The AD prices at time  k  = 1 are

    q 1,1 =  q 1,−1 =  p01 + α0

    , q 1,0 = (1 − 2 p0)

    1 + α0.

    Suppose we have determined   α0, . . . , αk−1   and all the AD prices   q k,j , at some time index   k.

    Then the next ZCB price is

    Bk+1 =k

     j=−k

    q k,j1 + αk + jβ k

    .   (4)

    The right-hand side of (4) is a monotone decreasing function of   αk, so we can determine the

    unique value of  αk  satisfying (4), by a binary search procedure.If we consider a node (k+1, j) with −k+1 ≤  j ≤ k−1, then Y k+1 =  j  only if  Y k  = j−1, j , j+1.

    The AD price for (k + 1, j) is therefore

    q k+1,j  =  pkq k,j−1

    1 + αk + ( j − 1)β k+

     (1 − 2 pk)q k,j1 + αk + jβ k

    +  pkq k,j+1

    1 + αk + ( j + 1)β k.

    There are similar expressions for the ‘boundary’ cases   j   = −k − 1,−k,k,k + 1. The order of business is thus to use the above expressions to compute  α0 → q 1,· → α1 → q 2,· → . . . → αTM −1.This completes calibration to market zero-coupon bond prices.

    Calibration of the volatility parameters   σ, λ  must use volatility-sensitive market data, i.e.

    cap or swaption prices.

    2