A Closed – Form Binomial Interest Rate Model

 

 

 

 

 

 

 

Thomas S. Y. Ho

 

Brownlee O. Currey Visiting Professor

 

Owen School

 

Vanderbilt University

 

and

 

President

 

Thomas Ho Company

 

New York

 

 

212-571-0121

 

tom.ho@thomasho.com

 

 

 

 

 

 

September 2000

 


Abstract

 

 

A  Closed-Form Binomial Interest Rate Model

 

 

This paper proposes a closed form binomial interest rate model.  The model extends the Ho-Lee model to incorporate the term structure of volatility. The model exhibits a mean reversion process. It is a market model that takes the Black volatility as given. The paper also derives the drift of the forward rate process, as proposed by Heath, Jarrow and Morton.  Given the model computational efficiency, the model can be used for Asian options with optimal early exercise option valuation.


A  Closed-Form Binomial Interest Rate Model

 

Ho-Lee (1986) proposes an arbitrage-free binomial interest rate model.  Extending Cox-Ross-Rubinstein arbitrage-free binomial lattice model for stock, Ho-Lee provides a closed form solution to the binomial movement of the yield curve (the discount function).  This model has the following properties: (1) at each node, that denoted by the state (i) and future time (n), the model specifies the yield curve as a function of the initial yield curve and the initial interest rate volatility ( a closed form solution), (2) there is no arbitrage possibility by holding any portfolio of bonds along this yield curve over a one step binomial movement at any node point (binomial model.)  These two properties ensure that the valuation of interest rate contingent claims is arbitrage-free and is efficient in computation.

 

A limitation of Ho-Lee is its assumption of constant interest rate volatility. The model assumes that the one-period interest rate volatility is constant over time and, as a result, the model cannot accept a term structure of volatilities.  The generalization of the model was suggested but not reported in Ho and Lee. Ho (1999) describes a model that takes the term structure of interest rates and volatilities as given and specifies the arbitrage-free yield curve movements in a closed form solution.  The purpose of this paper is to describe the behavior of this model.

 

Specifically, this paper uses the model Ho (1999) to derive the “money market account” model for deriving the pricing kernal (Harrison and Kreps), the forward drift of Heath-Jarrow-Morton approach (HJM) (1990).  In so doing, this paper shows that a closed form binomial model can capture the term structure of volatilities as in Black-Derman-Toy. But unlike Black Derman and Toy model, this model has a closed form solution of the yield curve at each node point and hence a closed form solution for a broad range of interest rate contingent claims at any future time (n) and state (i). This model is similar to Hull – White, being a normal model and has mean reversion of interest rates. But this model is a market model in the sense of Brace, Gatarek and Musiela ( 1997)/ Jamshidian (1997), where we take the market observed  Black volatilities directly as inputs to the model.

 

HJM provides a general approach to interest rate modeling.  In contrast, this paper provides a specific, but restrictive, model.  In so doing, the value of the model is based on its practical applications. The model can efficiently price an early exercise option on a long dated fixed rate instrument over a short expiration period.  For example, consider an American swaption.  The holder of the option can terminate the swap any time up till the expiration date (say one year), where one party pays a fixed rate with long dated tenor (say 10 year) and receives a daily floating rate. To value such a security, a daily step-size binomial model and the recursive valuation procedure over option period are used. At each node point, a closed form solution of the long dated fixed rate bond is calculated.  Other binomial interest rate models ( example Black Derman Toy) requires calibrating first the daily step size binomial model and then recursively calculates a ten year bond value at each node point over the expiration period would required extensive computations.

 

The ability to calculate any long dated interest rate contingent claims at any node point and can apply the recursive methodology to value an American option is important to an interest rate model.   This property is important to determine the cost of convexity of a constant maturity swap, the prepayment options of mortgagors, lapsing behavior of insurance policyholders, withdrawal behavior of the customers of the savings and deposits accounts. More generally, portfolio strategies in asset liability management require optimal short-term multi-period strategies on long-term assets and liabilities.  The proposed model can provide a practical solution.

 

The paper proceeds as follows.  Section A describes the model.  Section B describes the yield curve movements of the model. We show that the yield curve movements are related directly to the shape of the term structure of volatilities, including reversion for a downward sloping volatility curve. Section C analyzes the behavior of the term structure of volatilities. The spot volatilities is related to the forward volatilities. A direct relationship ( and not a calibration) of Black volatilities is given to the input volatilities of the model. Section D derives the money market rate model and the forward drifts along the forward curve. This establishes the relationship of the simulated interest rates to the term structure of interest rates and volatilities.  Section E contains the conclusions, which summarizes the attributes of the model.

.           

A.     The Model

 

The model is a binomial lattice model with the basic assumptions similar to Ho-Lee.  We assume a multi-period discrete time model where the market is perfect and complete.  At each future time, n, and state i,  we define a discount bond with maturity T to be the bond that pays $1 at the end of the Tth period counting from time n, with no other payments to the bondholder.  The price of this discount bond is denoted by Pni (T).  The function Pni (.) where the time to maturity T is the input parameter is called the discount function.  P00(T) is the initial discount function, which is observed and is given to the model.

We note that Pni (0) must equal to 1 since it is the value of $1 at time n and state i.  Pni (1) is the discount factor for one period at time n and state i. At each node of the binomial model, we assume that the discount function can shift up or down. We assume that the risk neutral probability for the upstate and for the downstate is the same, 0.5.  At the initial node of time n and state i, the binomial outcome for the upstate is denoted by the time (n+1) and the state (i + 1).  Similarly, for the down state, the node is denoted by time (n+1) and state (i).

 

B.     Arbitrage-free Term Structure Movements

 

The term structure movement is arbitrage-free if there is no arbitrage opportunity by holding a particular portfolio of the discount bonds at each node point of the binomial lattice.  Such is the case, if the expected return of holding a discount bond of any maturity T over any one period is the one period return.

 

Specifically, according to Harrison and Kreps, we require that:

 

 Pni (T)  =   0.5  Pni (1) ( Pn+1i+1 (T - 1)  +   Pn+1i (T –1 ) )                                                               (1)

 

Equation (1) should be satisfied for any n = 0,1,…. and i = 0, 1, 2,…, n.

 

Pni (1) is the discount factor at the node time n and state i that determines the present value of the expected value of the discount bond with maturity (T – 1) at time (n+1). Pni (T)  is the term structure at the node of time n and state i.  -ln Pni (1) is the “money market rate.” And therefore it ensures that the term structure follows a martingale process, as expressed in equation (1).

 

The model is discrete time and recombining, in the framework of Cox, Ross Rubinstein.  Indeed, CRR derives the stock option model assuming the stock price to follow a binomial movement.  This model assumes that the yield curve follows a binomial movement, with the constraint that the movement is arbitrage free.  From equation (1), we can then specify the model of the term structure at each node point given the initial term structure of interest rates and inputs related to the term structure of volatilities (which will be discussed later.)  This is similar to CRR solution for stock lattice where the stock movements are specified by the stock volatilities and risk free rate. The difference here is that we are not specifying one price (the stock) but a function of prices (the term structure of interest rates) at each node point.

 

Proposition 1.  The Closed Form Solution

 

Let 1 > di  > 0 be some real numbers for i = 1, 2, 3, … and let P(T) be the initial discount function.  Define d n, m = dn  dn-1 dn-2    dm+1 dm .  ( Section C will show that d n, m  are related to the volatilities.) For the purpose of this proposition, di  are input data to the model.

 

The arbitrage-free movement of the discount function is given by:

Pni (T)  =     P(T+n)( 1+dn-1     d1)   (1 + dn-1    d2) …   (1+ dn-1)_____    2__(dT+n-1    dn)i         (2)

                    P ( n) (1 +dT+n-1 d1)(1+dT+n-1 d2) …(1+dT+n-1dn-1)(1+dT+n-1…dn)

 

Alternatively, it can be expressed as

Pni (T)  =    [ P(T+n)/P(n)]  (  P  G-1Tnj ) (dT+n-1    dn)i – 0.5n                                                     (2a)

 

where G Tnj  = (d T+n-1, n –0.5+ d T+n-1, n 0.5 * d n-1, j) /( 1 + d n-1, j  ) 

 

The result shows that that the T- period discount bond price n-period hence is the forward bond price (determined by the initial discount function), adjusted by a factor G n-times the forward  price.  The bond price at the ith state is then a multiple (higher and lower) of this base value.

 

Proof:

 

The proof is straightforward by substituting Equation (2) to Equation (1) to verify the arbitrage free conditions holds.

 

QED.

 

We will show later that d n, m  is related to the term structure of volatilities, which are taken as given in this model.  Equation (2) provides that the complete description of the movements of the discount function.  It shows that any future discount function has three factors.  The first factor is the forward price implied from the initial discount function.  Then there is an adjustment term, which depends on the volatilities of the yield curve movements.  In the limiting case when there is no volatility, (we will show when di  equals 1 for all i) the adjustment factor is one.  Both of these two terms are independent of the state i.  Therefore the uncertainty of the interest rate movement is specified only by the last factor.  

 

The importance of Equation (2) is that the model is closed form. The complete information of the discount function at each node point is given by the input data of the term structure of interest rates and volatilities.  Consider for example Black-Derman- Toy, to derive the value of a long bond at a node, we must first calibrate an extended lattice of one period rate that fits the term structure of interest rates and volatilities. Then a backward substitution model is used to price the zero-coupon bond at a particular node.  This process can be inefficient and may have significant numerical estimation errors.

 

Proposition 2.  Extension to the Ho – Lee Model

 

When  dn   is a constant d for all n, then the model is the same as  Ho – Lee.

 

Proof:

 

Note that when dn are constant, then

 

d n, m = d n-m+1                                                                                                                                 (3)

 

If we substitute Equation (3) to Equation (2), we have

 

Pni (T) =[ P( T+ n)/P(n)] [2( 1 + dn-1) ( 1 + dn-2) ( 1 + dn-3)…( 1 + d)/{ (1+ d T +n-1) ... (1+dT)} ]d Ti         

 

(4)

 

Equation (4) is Ho-Lee model. 

 

QED.

 

The main difference of the model proposed is the introduction of the time dependent d of the Ho- Lee model such that the interest rate movement can be modeled in a binomial lattice.  In the continuous time model, Ho-Lee model can be expressed as

 

dr (t) =   q(t) dt  +   s dz

where r is the instantaneous interest rate,  q(t) a function of time, initial discount function, dz is the wiener process, and s  the constant volatility.  The model proposed in this paper takes the form in continuous time as follows:

 

dr (t) =   q(t) dt  +   s(t) dz                                                                                                            (4)

 

The main difference is that the volatility is time dependent but not state dependent.  An important aspect of the model described by Equation (4) is that the model is completely specified by the term structured of interest rates and volatilities.  The model has no parameters to be estimated or calibrated from observed prices. We will show the Black caplet prices can be used as direct input to the model.

 

C.     Analysis of the Term Structure of Volatilities

 

Section B describes the term structure of interest rate movements entirely by the discount function.  This section will relate the results to the yield curve movements and term structure of volatilities of the bond yields, which are market conventions. At the initial date, we have the discount function P(T). We define the yield of a discount bond with maturity T initially to be:

 

r (T)  = - ln P(T)/T                                                                                                                         (5)

 

More generally, we can define the yield curve at each node point to be

 

rni (T) =  -ln Pni (T)/T                                                                                                                     (6)

 

Equations (5) and (6) assume continuously compounding rate of returns for the bonds.  At the end of the first period, the yield curve is either  r10 (T) or r11(T). We now define spot volatility of term T. That is:

 

s (T)  =  0.5 ( r11 (T)  -  r10 (T)  )                                                                                  (7)

 

Since this is a binomial model, the spread between the two binomial outcomes is twice the standard deviation. Therefore Equation (7) is the standard deviation of the shifts of the yields for a binomial model.

 

We also define the forward volatility at time n and state i for the T period bond yield.  It is defined as:

 

sn i f (T)  =  0.5 ( rni+1 (T)  - rni (T))                                                                                                (8)

 

Therefore rni  =  -ln Pni (1) can be interpreted as the one period rate, analogous to the instantaneous rate in a continuous time interest rate model. Now substitute Equation (1) into Equation (6) and (8), we show that the forward volatility

 

sn  f (T) =  0.5 ln d T+n-1, n / T                                                                                                         (5)

    

We note that the volatility is independent of the state i, since the RHS of equation (5) does not contain any index i.  At any time n, the yield of a discount bond has the same volatility at any state i. For this reason the model is normal. The yields have a binomial distribution ( and becomes a normal distribution in the limit of increasing smaller step size of the lattice.)  Note that ln d T+n-1, n has T terms. Therefore Equation (5) states that the volatilities is related to 0.5 of the log of the geometric means of the di

 

This normal model belongs to the class of models that was first proposed in Vasicek (1977).  The model assumes that the interest rate rises and falls over the next instant (or next period ) is independent of the interest rate level.  But the standard deviation depends on the time n.  The next proposition relates the model to input data  d  to the spot volatilities.

 

Proposition 3  Term Structure of Volatilities

 

The spot volatility of term T is given by

 

s (T) =   - 0.5 (ln dT   dT-1  dT-2    d1 )/T.                                                                                                    (9)

 

Proof:

According to Equation (7), we can compute the value of the discount function for the upstate and the downstate after one period.

 

P11 (T) = 2 [P(T+1)/P(1)][1/( 1 + dT,1)] d T,1 

 

P10(T) = 2 [P(T+1)/P(1)][1/( 1 + dT,1)]

 

Now

 

s (T) = - 0.5 (r10 (T) - r11(T))                                                                                              

 

          = - 0.5 ln d T,1 / T

 

QED

 

The importance of Proposition 3 is that the observed spot volatilities provide the complete information to determine all the input data to the interest rate model.  However, the observed volatilities are the forward volatilities implied by the Black model from the observed prices of the caplets and floorlets.  We therefore first determine the relationship of the forward volatilities and the spot volatilities.

 

Proposition  4.   Forward and Spot Volatilities Relationship

 

The relationship between the term structure of forward volatilities and spot volatilities

 

( T + n) s (T + n)    =  n s (n )   + Tsfn  (T)                                                                                (10)

 

Proof:

 

From Equation (5), we have

 

T.sn  f (T) =  0.5 ln d T+n-1, n

 

From Equation (9), we have

 

n s (n) = 0.5 ln d n,1

 

QED

 

Caplets are call options on a future rate. Consider a caplet that resets at n, on a rate of term T. Then the payoff is the maximum of the realized rate of term T net the strike rate and zero. Specifically, we have max (  rni (T) – r, 0 )  as the payoff of a caplet. But Equation (6) gives 

rni (T) =  -ln Pni (T)/T, and therefore, we have the boundary conditions to price the caplet.  A backward recursive procedure would enable us the price the caplet.  A series of caplet prices can provide us a corresponding series of forward volatilities   sn  f (T) =  0.5 ln d T+n-1, n / T  .

 

These forward volatilities are then used to determine the spot volatilities according to equation (10). A smooth function is then used to fit this series of spot volatilities. From this smooth spot volatilities, we can now determine all the d by equation (9).  Then Equation (1) shows that the entire term structure movement can be specified. 

 

In essence, Black volatilities of caplets can be expressed as the forward volatilities based on the model. We can use proposition 4 to measure the volatilities in terms of the spot volatilities, which is not dependent on the reset rate.  Then the term structure  Pni (T) can be determined at each node point in closed form. Since the spot volatilities are not calibrated as in Black Derman and Toy or estimated from any economic parameters as in Hull and White, this model is similar to the market model of Brace, Gatarek, and Musiela.

                                                                                                                                        

D.    Yield Curve Movements and Term Structure of Volatilities

 

This section analyzes the rates movements of the model. To isolate the movement from the shape of the initial yield curve, we consider the rate distribution around the forward rates.  Then we provide the model for the money market rates and the model for the one period forward rates, both for used in literature in deriving interest rate models. 

 

Proposition 5.    Forward Rate and Expected Interest Rate

 

The expected yield of a T period discount bond over an n period horizon is the forward rate with a drift.  The drift D is given by:

 

D( n, i, T) = (1/T) ln ( Gn1T Gn2T … Gn n-1T)

 

where GniT =  ( a  -0.5 + a 0.5 d n-1,i)/( 1 + d n-1,i),     for a = d T+n-1, n                              

                                                        

Proof:

 

From Equation (2), we have:

 

Pni (T)  =  [ P(T+n)/P(n)]  (  P  G-1Tnj ) (dT+n-1    dn)i – 0.5n                                                     

 

where G Tnj  = (d T+n-1, n –0.5+ d T+n-1, n 0.5 * d n-1, j) /( 1 + d n-1, j  ) 

 

From Equation (6), we have:

 

rni (T) =  -ln Pni (T)/T

 

          =  -(1/T) ln [P(T+n)/P(n)] + (1/T)ln (  P  GTnj ) – (i- 0.5n)ln (dT+n-1    dn)                   (11)

where G Tnj  = (d T+n-1, n –0.5+ d T+n-1, n 0.5 * d n-1, j) /( 1 + d n-1, j  ) 

But the expectation of  (1/T) i ln d T+n-1, n  over the states i  is  (0.5 n/T) ln d T+n-1, n, since the random variable i has a binomial distribution.   The first term of the RHS is the yield of the forward discount bond. Therefore the drift term is given by:

 

D = (1/T) ln ( Gn1T Gn2T … Gn n-1T)                                                                                            (12)

 

Where GniT =  ( a  -0.5 + a 0.5 d n-1,i)/( 1 + d n-1,i),     for a = d T+n-1, n                                                 (13)

 

QED

 

Proposition 5 provides a clear description of the yield curve movements. For each term (T) on the yield curve, we can construct the forward yield curve for that term.  The yield curve is

-(1/T) ln [P(T+n)/P(n)] for each n = 1, 2, ….   The forward rate at on each date is a binomial distribution.  The spread between two nodes is constant  (1/T)ln d T+n-1, n.  That is, the spread is twice the forward volatility for reset term T and option expiration date n, according to Equation (5). 

 

Now, if we have a downward sloping term structure of volatilities, then di is increasing with i. According to the last term of equation (7), the spread of between two nodes of the binomial distribution tightens, as n increases.  In another words, the yield curve movements follow a mean reverting process.  Further, for longer the term T, lower is the volatility ( consider a = d T+n-1, n ) , which is consistent with market observations.

 

The mean of this binomial distribution is not on the forward curve however.  The mean is drifted above the forward curve, which is to be proved in the next proposition.

 

Proposition 6   Convexity Drift

 

The drift is always positive and increases with volatilities. That is, the expected forward rate of any time period of delivery T and future date n is higher than the forward rate implied by the initial yield curve. And this spread increases with volatilities.

 

Proof:

 

To show D > 0, according to equation (12), it suffices to show that GniT is greater than 1 for all i.   

 

Note that 1 >  a 0.5 , d n-1,i > 0

 

Therefore,  it follows that

 

( 1 - d n-1,i a 0.5) ( 1 -  a 0.5 )  > 0                                                                                                  (14)

 

In expanding Equation (14) and simplify, we get

 

1 +   d n-1,i a  > a 0.5  ( 1 + d n-1,i)

 

Then it follows that GniT > 1

 

Note that with higher volatilities,  both  d n-1,i , a  become smaller and Gi becomes larger.

QED.

 

The positive drift can be explained as the convexity effect.  Since the stochastic future rates are distributed as a binomial, the price behavior of a zero coupon bond has a positive convexity.  As a result, the expected return of holding a bond over a period of time would have a higher return than the risk free rate over that same period, if there were no drift in the expected forward rate.  Therefore, to assure the arbitrage-free condition to hold in Equation (1), we must have a positive drift. 

 

Proposition 6 provides a description of the forward curve movements.  To specify an interest rate movement in a one-factor model, in fact it is necessary and sufficient only to specify the one period rate, which is the shortest rate on the forward curve.

 

Vasicek, Cox-Ingersoll- Ross and others have specified interest rate models using the short rate movement.  Black Derman and Toy, Hull and White are examples of using the short rate only to specify the arbitrage-free movements of the yield curve.  In principle, once we have a lattice of one period rate, we can always calculate any bond value at any node point. In practice, this can be computational intensive.

The one period rate is also important for determining the arbitrage-free condition in Equation (1). This rate, often called the “money market rate” is used as a numeraire to specify the movements of other securities in the market.   The following proposition analyzes this money market rate of the model.

 

Proposition 7    The Money Market Rate Model

 

The money market rate is the one period rate of the binomial model.  The process is given by

 

rni   =   - ln [P(n +1)/P(n)] + ln [Gn1 Gn2 … Gn n-1 ]  +  (n/2 -  i ) ln d n                                                            (15)

 

where Gni  is defined as

 

Gni  =  (d n -0.5 +  d n 0.5 d n-1,i )/ ( 1 + d n-1,i )                                                                                    (16)

 

Proof:

 

The proof is quite straightforward.  We let T =1, as we are only concerned with the one period rate, and substitute it in Equation (14).  Equation (16) is the same as Equation (13), except that we suppress the subscript T, since it is always one.

 

QED

 

Equation (5) provides a clear description of the model.  Both the first term and the third term are quite simple to explain.  The first term is the forward one period rate into the future.  The third term is the binomial distribution of the interest rate with mean zero, with the volatility (which is the standard deviation of a binomial distribution) of n0.5 .sn  f (1), where sn  f (1) is the one period forward volatility at time n (see equation (5).)   Equation (15) shows that the factors Gni  are important to the specification of the model.  The following proposition provides insight into these factors.

 

Consider a node of time n and state i. The one period forward rate as seen at time n for a contract maturing at time T (as measured from time n) is denoted as f ( n, i, T).

 

Proposition 8   The Forward Rates Model

 

The movement of one period forward rate of the forward date is given by:

 

 f1i (T)  =   - ln [P(T+1)/P(T)] + ln GT  -  (i – 0.5) ln d T                                    for i = 0 or 1

 

where

 

GT  =  (d T -0.5 +  d T 0.5 d T-1,1)/ ( 1 + d T-1,1 )

 

And

 

 d T-1,1   =    d T-1 d 1

 

Proof: 

 

The one period forward rate with forward delivery date T is given by:

 

f (n, i, T ) =  - ln Pni (T+1)/ Pni (T).

 

For simplicity, we consider the case when we are at the initial position, n = i =0.  The general case can be similarly calculated and the analogous result can be derived.

 

At the following one period,  at state 1, the bond P(T+1)

 

P11 (T) = 2 [P( T+1)/P(1)] (d T-1 d 1)/ ( 1 + d T-1 d 1).

 

Then the forward price at state 1 is given by

 

P11 (T)/ P11 (T-1 ) = [P(T+1)/P(T)][ ( 1 + d T-1 d 1)/( 1 +  d T d 1 )] – ln d T 

 

Similarly, at state 0, we have

P10 (T)/ P10 (T-1 ) = [P(T+1)/P(T)][ (1 + d T-1 d 1)/( 1 +  d T d 1 )]  

 

Then, forward rates are given by – ln Pni.  The drift is given by the forward rate net of the expected forward rates. That is:

 

 Drift = d =  - ln[P(T+1)/P(T)]   - 0.5 (- ln [P11 (T)/ P11 (T-1 )] – ln[ P10 (T)/ P10 (T-1 )])

 

The result follows by a direct computation.

 

QED

 

Heath-Jarrow-Morton approach is to specify the movement of the interest rate by the movement of the short term forward rate.  Proposition 8 shows that the drift of the one period forward rate is simply ln Gni.

It is also interesting to note that previous research has shown that the continuous time version of Ho- Lee model as an instantaneous drift of  sT.  Such is the case when we consider the limiting case with the step size approach zero.  However, the value of the Ho –Lee model is not based on the continuous time formulation, but on its binomial lattice formulation for practical implementation reasons.  For this reason, Proposition 8 provides a useful specification of the drift.

  

E.     Conclusions

 

This paper analyzes a closed form binomial interest rate model.  We can compare the model with other interest rate models in terms of their attributes. 

 

1.      Spot yield curve.

 

Similar to other arbitrage-free model, the model takes the initial yield curve is given.  Unlike Vasicek, Cox-Ingersoll-Ross, and some extent Hull White, where the short term interest rates are postulated to follow a certain process, and the model is calibrated in fit the observed data. Equation (2) shows that the pricing model and the rates movements take the observed spot curve as given.  For this reason, this model can accept any yield curve shape and for this reason, the model can be used to calculate key rate durations or other yield curve analytics without any economic justifications of the yield curve movements.

 

By taking the initial yield curve as given without other constraints have other implications.  In capital markets, liquidity risk and sovereign risk (particularly for developing economies) are also important consideration for the determination of the term structure of interest rates. Since the strengths of the arbitrage-free interest rate model is to enable us to relative value other securities to the term structure of interest rates, we need to take these risks into account, beyond the fundamental time value of money concept.  However, there is no constraint ensuring the forward rates to be positive, given other market technicalities.  Indeed, there were times that the forward rates in the US Treasury market were negative. Or in recent times, the floors of Yen rates with zero strike rate has positive value, since major institutions still have to deposit their significant holdings with banks or capital markets, even the if market rates were zero. In these instances, lognormal model like Black Derman and Toy would not accept the spot curve as given. 

 

2.      Term Structure of Volatilities

 

Similar to Black Derman and Toy, this model takes the term structure of volatilities as given.  The model accepts any shape of the term structure of volatilities, specified by the one period forward volatilities, Equation (5).  There are constraints to the shape of the spot volatilities only because the spot volatilities may infer extremely high or other unacceptable values of the forward volatilities , violating the constraints 1 > di  > 0.

 

Further, the model forward volatility can be specified by the Black volatility.  Black volatilities assume that the forward rate follows a lognormal diffusion process, and the caplet price is assumed to be priced by a stock option model.  These assumptions can be eliminated by re-specifying the Black volatilities in terms of this arbitrage-free interest rate model. In so doing, we can have a more reasonable term structure of volatilities to analyze the market and at the same time a relatively transparent model to translate the volatility value to the caplet price.

 

3.      Binomial,Discrete Time, and Continuous Time Models

     

Recently, Grant and Vora (1999) and Das and Sundaram (1999) have pointed out that for practical implementation, discrete time models (as opposed to continuous time) can provide computational efficiency.  While they derive their models in discrete time, they do not require the paths to recombine.  This model requires the paths to recombine.  Binomial model can be efficient for most American  option valuation.

 

Heath Jarrow Morton continuous time approach offers flexibility and generality in specifying the interest rate model, particularly in simulating the forward movements of interest rates.  But the approach often lacks the computational efficiency and as the result, while the approach may offer a more realistic model in capturing the market behavior, the valuation may have larger model errors.

 

The paper provides a model that can be extended to value Asian options and multi-factor models.  Given its closed form solution, the model can be used for small step sizes and allow more computing time to accept other factors ( like time dependency in Asian options or other risk factors.)  These issues will be left for future research.

 

 


References

 

Black F., E. Derman, and W. Toy  “ A one factor model of interest rates and its application to treasury bond options,” Financial Analysts Journal Jan-Feb, 33-39 1990 

 

Brace A., Gatarek D., Musiela M “ The market model of interest rate dynamics” Department of Statistics, University of New South Wales 1997

 

Cox, Ingersoll and Ross (1985) “ A theory of term structure of interest rates” Econometrica  53 363-384

 

Cox Ross Rubinstein (1979) “ Option pricing: a simplified approach”, Journal of Financial Economics 7, 229- 263

 

Das, Sanjiv and Rangarajan Sundaram “ A discrete-time approach to arbitrage-free pricing of credit derivatives”  Working Paper Finance Department,  New York University 1999.

 

Grant, Dwight and Gautan Vora “ Implementing no-arbitrage term structure of interest rate models in discrete time when interest rates are normally distributed” March 1999 vol 8 #4

 

Harrison  J. M. and  d. M. Kreps “ Martingales and arbitrage in multi-period securities markets,”  Journal of Economic Theory 20, 381-408 1979.

 

Heath, D., R.A. Jarrow, and A. Morton “ Bond pricing and the term structure of interest rates: a discrete time approximation,” Journal of Financial and Quantitative Analysis, v 25 (4), 419-440

 

Ho, Thomas  “Market valuation of liability: transfer pricing, profit release and  credit spread.” Proceedings of Fair Value of Insurance Business Conference 1999

 

Ho, Thomas  and Sang Bin Lee “ Equilibrium term structure movements and pricing interest rate contingent claims” The Journal of Finance, vol XLI, no 5, December 1986

 

Hull J. and White A. “ Pricing interest rate derivative securities”, Rev. Fin Stud., 3 573-592, 1990

 

Rebonato, Riccardo  Interest-rate option models  2nd edition Wiley 1998

 

Vasicek O “ An equilibrium characterization of the term structure”, Journal of Financial Economics 5 177-188, 1977