Interest rate derivatives in a CTMC setting: pricing, replication

and Ross recovery

Tim Leung ∗ Matthew Lorig †
arXiv:2409.14193v1 [q-fin.MF] 21 Sep 2024

This version: September 24, 2024

Abstract

We consider a financial market in which the short rate is modeled by a continuous time Markov chain (CTMC)
with a finite state space. In this setting, we show how to price any financial derivative whose payoff is a function
of the state of the underlying CTMC at the maturity date. We also show how to replicate such claims by trading
only a money market account and zero-coupon bonds. Finally, using an extension of Ross’ Recovery Theorem due
to Qin and Linetsky, we deduce the real-world dynamics of the CTMC.

1 Introduction
In order to model the yield curve, banks and researches often assume the short-rate is a Markov diffusion – typically
an affine term structure (ATS) or quadratic term-structure (QTS) model; see, e.g., (Filipovic, 2009, Chapter 5) and
(Ahn et al., 2002, Section 2) for an overview of such models. The Markov diffusion framework is appealing because it
allows for explicit computation of bond prices, yields, as well as widely-traded interest rate derivatives such as caplets
and floorlets.

However, if one looks at time series data of interest rates, they do not appear to have diffusion-like dynamics. Indeed,
we see in Figure 1 that the federal Funds rate is constant for long periods and then jumps. As such, to the extent that
the federal funds rate is a proxy for the short-rate, it would be better to model the short-rate as a continuous-time
Markov chain (CTMC). The analytic tractability of CTMC short-rate models was established by Elliott and Mamon
(2003), who show how to explicitly compute bond prices, yields, and forward rates. However, they do not consider
the problem of replicating interest rate derivatives with liquidly traded assets nor to they comment on the relation
between risk-neutral and real-world dynamics of interest rates, both of which are addressed in the present paper.

The rest of this paper proceeds as follows In Section 2, we introduce a model for the short-rate driven by a CTMC.
Next, in Section 3, we derive the prices of claims written on the CTMC. Examples of such claims include zero-coupon
bonds, caplets and floorlets. In Section 4, we show how to replicate all claims by dynamically trading a portfolio
consisting of bonds with distinct maturity dates and a money market account. Next, in Section 5, we use an extension
of Ross’ Recovery Theorem Ross (2015), due to Qin and Linetsky (2016), to deduce the real-world dynamics of the
CTMC that drives short rate dynamics. In a related study, Carr and Yu (2012) show how to recover real-world
∗ Department of Applied Mathematics, University of Washington. e-mail: timleung@uw.edu
† Department of Applied Mathematics, University of Washington. e-mail: mlorig@uw.edu

1
transition probabilities in a bounded time-homogeneous diffusion setup with restrictions on the numéraire. portfolio.
Lastly, in Section 6, we perform explicit computations in a simple two-state CTMC setting.

2 A CTMC short-rate model and assumptions

Throughout this paper, we work in the same setting as Elliott and Mamon (2003). Specifically, we fix a time horizon
T < ∞ and consider a continuous-time financial market, defined on a filtered probability space (Ω, F, F, Q) with no
arbitrages and no transaction costs. The probability measure Q represents the market’s chosen pricing measure taking
the money market account M = (Mt )t ∈[0,T] as numéraire. The filtration F = (Ft )t ∈[0,T] represents the history of
the market.

We suppose that the money market account M is strictly positive, continuous and non-decreasing. As such, there
exists a non-negative F-adapted short-rate process R = (Rt )t ∈[0,T] such that

dMt = Rt Mt dt , M0 > 0. (1)

We will focus on the case in which the dynamics of the short-rate R are described by an irreducible positive recurrent
CTMC J = (Jt )t ∈[0,T] with state space S = {1, 2, . . . , n}. Specifically, we have

Rt = r (JT ), r : S → [0, ∞).

Let G denote the generator matrix of J

 
g1,1 g1,2 . . . g1,n
 
 g2,1 g2,2 . . . g2,n  X
G= . .. .. .. , gi ,j = 0 ∀ i , gi ,j ≥ 0 ∀ i ̸= j .
 
 .. . . . 
  j
gn,1 gn,2 . . . gn,n

We can express the dynamics of J as a state-dependent Lévy-type process

Z X
dJt = z N(dt , Jt – , dz ), Et – N(dt , Jt – , dz ) = ν(Jt – , dz )dt , ν(i , dz ) = gi ,j δj –i (z )dz , (2)
j ̸=i

where Et – · := E[ · |Ft – ] denotes conditional expectation under Q.

3 Derivative pricing
Consider a financial derivative that pays φ(JT ) at time T ≤ T where φ : S → (–∞, ∞). Using risk-neutral pricing, the
value of this derivative at time t ≤ T is u(t , Jt ; T) where the function u( · , · ; T) : [0, T] × S → (–∞, ∞) is defined by
 RT 
– r (Js )ds
u(t , i ; T) := E e t φ(JT ) Jt = i . (3)

In the following proposition, we provide an explicit expression for u(t , i ; T).

Proposition 1. Define n × 1 vectors UT

t and Φ and an n × n diagonal matrix R by

UT
t := (u(t , i ; T))i ∈S , Φ := (φ(j ))j ∈S , R := (δi (j )r (j ))i ,j ∈S , (4)

2
Then we have

UT
t =e
(T–t )(G–R)
Φ, (5)

where e(T–t )(G–R) is the matrix exponential of (T – t )(G – R).

Proof. The function u defined in (3) satisfies the Kolmogorov Backward Equation (KBE)

0 = (∂t + G(i ) – r (i ))u(t , i ; T), u(T, i ; T) = φ(i ), (6)

where the operator G is given by

Z   X  
G(i ) = ν(i , dz ) ez ∂ – 1 = gi ,j e(j –i )∂ – 1 . (7)
j ̸=i

Here, we have used the notation ez ∂ to denote the shift operator : ez ∂ f (j ) = f (j + z ). Using (6) and (7), we obtain
X  
0 = ∂t u(t , i ; T) + gi ,j u(t , j ; T) – u(t , i ; T) – r (i )u(t , i ; T)
j ̸=i
X X
= ∂t u(t , i ; T) + gi ,j u(t , j ; T) – δi (j )r (j )u(t , j ; T)
j j
X 
= ∂t u(t , i ; T) + gi ,j – δi (j )r (j ) u(t , j ; T), (8)
j

where, in the second equality, we have used the fact that gi ,j = –gi ,i . Using (4), we can write the system of
P
j ̸=i
coupled ODEs (8) more compactly as

0 = (∂t + G – R)UT
t , UT
T = Φ,

from which expression (5) for UT

t directly follows.

Remark 2. For the purposes of computation, it may be convenient to express u(t , i ; T) as follows
n
⟨a, b⟩ := a ⊤ b =
X
u(t , i ; T) = ⟨Ei , UT
t ⟩, ai bi , Ei = (δi (j ))j ∈S .
i =1

Let us look at a few examples, which will be important in subsequent sections.

Example 3 (Zero-coupon bond prices and Yields). A T-maturity zero-coupon bond is a derivative that pays
φ(JT ) = 1 at time T ≤ T. Denote by B(t , i ; T) the price of a zero-coupon bond assuming Jt = i . We have
 RT 
– r (Js )ds
B(t , i ; T) := E e t Jt = i .

Defining n × 1 vectors BT
t and 1 as follows

BT
t := (B(t , i ; T))i ∈S , 1 := (1)j ∈S ,

we have from Proposition 1 and Remark 2 that

BT
t =e
(T–t )(G–R)
1, B(t , i ; T) = ⟨Ei , BT
t ⟩, (9)

which is equivalent to the (Elliott and Mamon, 2003, Equation (2)). We can also define the Yield of a zero-coupon
bond as
–1
Y(t , i ; T) := log B(t , i ; T). (10)
T–t

3
Example 4 (Caplets and Floorlets). The simple forward rate from T to T is defined as follows
1  B(t , Jt ; T) 
F(t , Jt ; T, T) := –1 .
T – T B(t , Jt ; T)

A forward rate option with reset date T and settlement date T is a derivative that pays h(F(T, JT ; T, T)) at time
T for some function h : R+ → R. As the payoff to be made at time T is known at time T, the value of the forward
rate option at time T is B(T, JT ; T)h(F(T, JT ; T, T)). Thus, the value of v (t , i ; T, T) of a forward rate option at time
t ≤ T given Jt = i is
 RT 
– r (Js )ds
v (t , i ; T, T) = E e t ψ(JT ; T) Jt = i , ψ(JT ; T) := B(T, JT ; T)h(F(T, JT ; T, T)).

Defining n × 1 vectors VT,T

t and ΨT as follows

VT,T
t := (v (t , i ; T, T))i ∈S , ΨT := (ψ(j ; T))j ∈S ,

we have from Proposition 1 and Remark 2 that

VT,T
t = e(T–t )(G–R) ΨT , v (t , i ; T, T) = ⟨Ei , VT,T
t ⟩.

In the case of a caplet or floorlet with strike K we have h(F) = (F – K)+ and h(F) = (K – F)+ , respectively.

Example 5 (Arrow-Debreu securities). An Arrow-Debreu security is a derivative that pays φ(JT ) = δj (JT ) at
time T ≤ T for some j ∈ S. Denote by A(t , i ; T; j ) the price of the j th Arrow-Debreu security assuming Jt = i . We
have from Proposition 1 and Remark 2 that
 RT 
– r (Js )ds
A(t , i ; T, j ) := E e t δj (JT ) Jt = i = ⟨Ei , e(T–t )(G–R) Ej ⟩.

Defining the n × n matrix AT

t := (A(t , i ; T, j ))i ,j ∈S , we have

AT
t =e
(T–t )(G–R)
. (11)

4 Replication
In this section, we will show how to replicate an Arrow-Debreu security with a payoff δk (JT ) at time T by trading a
portfolio of assets consisting of the money market account M and (n – 1) zero-coupon bonds with maturity dates
T1 , T2 , . . . , Tn–1 where Ti ∈ (T, T) and Ti ̸= Tj . Once we establish how to replicate an Arrow-Debreu security with
payoff δk (JT ), we can replicate a claim with general payoff φ(JT ) by noting that φ(JT ) can be written as a linear
combination of Arrow-Debreu payoffs: φ(JT ) = k φ(k )δk (JT ).
P

Proposition 6. Denote by X = (Xt )t ∈[0,T] the value of a self-financing portfolio with dynamics of the form

n–1  n–1 
(i ) (i )
X X
dXt = Δt – dBt (t , Jt ; Ti ) + Xt – Δt – Bt (t , Jt ; Ti ) r (Jt )dt , (12)
i =1 i =1

(i )
where Δ(i ) = (Δt )t ∈[0,T] denotes the number of Ti -maturity bonds in the portfolio. Define an (n – 1) × (n – 1)
matrix
 
δB(t , Jt – ) := B(t , j , Ti ) – B(t , Jt – , Ti ) , i = 1, 2, . . . , n – 1, j = 1, . . . , Jt – – 1, Jt – + 1, . . . n,
i ,j

4
and (n – 1) × 1 vectors

(i )
Dt – = (Δt – )i , i = 1, 2, . . . , n – 1,
δA(t , Jt – ; T, k ) := (A(t , j ; T, k ) – A(t , Jt – ; T, k ))j , j := 1, . . . , Jt – – 1, Jt – + 1, . . . n.

Suppose X0 = A(0, J0 ; T, k ) and Dt ≡ D(t , Jt ; T, k ) satisfies

δB(t , Jt )D(t , Jt ; T, k ) = δA(t , Jt ; T, k ). (13)

Then Xt = A(t , Jt ; T, k ) for all t ∈ [0, T].

Proof. We will show that, if X0 = A(0, J0 ; T, k ) and Dt – satisfies (13), then d(Xt /Mt ) = d(A(t , Jt ; T, k )/Mt ) from
which it follows that Xt = A(t , Jt ; T, k ) for all t ∈ [0, T]. We begin by computing the dynamics of B(t , Jt ; Ti ) and
A(t , Jt ; T, k ). Defining the compensated state-dependent Poisson random measure

N(dt
e , i , dz ) := N(dt , i , dz ) – ν(i , dz )dt ,

and using (2) as well as Itô’s rule for Lévy-Itô processes, we obtain
Z  
dB(t , Jt ; Ti ) = (. . .)dt + B(t , Jt – + z , Ti ) – B(t , Jt – , Ti ) N(dt
e , Jt – , dz ), (14)
Z  
dA(t , Jt ; T, k ) = (. . .)dt + A(t , Jt – + z , T, k ) – A(t , Jt – , T, k ) N(dt
e , Jt – , dz ), (15)

where the dt terms will not be important. Next, using (1), (14) an (15), as well as the product rule for Lévy-Itô
processes, we find
 B(t , J ; T )  1
Z  
t i
d = B(t , Jt – + z , Ti ) – B(t , Jt – , Ti ) N(dt
e , Jt – , dz )
Mt Mt
1
= dB(t , Jt ; Ti ) + (. . .)dt , (16)
Mt
 A(t , J ; T, k )  1
Z  
t
d = A(t , Jt – + z , T, k ) – A(t , Jt – , T, k ) N(dt
e , Jt – , dz ), (17)
Mt Mt

Similarly, using (1) and (12) as well as the product rule for Lévy-Itô processes, we obtain
X  1  1  h 1i
t
d = dXt + Xt d + d X,
Mt Mt Mt M t
1 X 
t
h 1i
= dXt – r (Jt ) dt (as d X, = 0)
Mt Mt M t
n–1 n–1
1 X (i ) 1  X (i )
 X 
t
= Δt – dBt (t , Jt ; Ti ) + Xt – Δt – Bt (t , Jt ; Ti ) r (Jt )dt – r (Jt ) dt (by eq. (12))
Mt Mt Mt
i =1 i =1
n–1 n–1
1 X (i ) 1 X (i ) r (Jt )Xt
= Δt – dBt (t , Jt ; Ti ) – Δt – Bt (t , Jt ; Ti )r (Jt )dt (canceling dt )
Mt Mt Mt
i =1 i =1
n–1   B (t , J ; T )  n–1
X (i ) t t i 1  1 X (i )
= Δt – d – (. . .)dt – Δt – Bt (t , Jt ; Ti )r (Jt )dt (by eq. (16))
Mt Mt Mt
i =1 i =1
n–1
1 X (i ) 
Z 
= Δt – B(t , Jt – + z , Ti ) – B(t , Jt – , Ti ) N(dt
e , Jt – , dz ), (18)
Mt
i =1

5
where, in the last equality, we have used (16) and the fact that the dt terms must cancel, as X/M is a (Q, F) martingale.
Comparing (17) an (18), we see that in order for d(Xt /Mt ) = d(A(t , Jt ; T, k )/Mt ), we must have
n–1 Z  
(i )
X
Δt – B(t , Jt – + z , Ti ) – B(t , Jt – , Ti ) N(dt
e , Jt – , dz )
i =1
Z  
= A(t , Jt – + z , T, k ) – A(t , Jt – , T, k ) N(dt
e , Jt – , dz ).

Noting from (2) that J can only jump to a state j ̸= Jt – , we can write the above equality as
n–1  
(i )
X
Δt – B(t , j , Ti ) – B(t , Jt – , Ti ) = A(t , j ; T, k ) – A(t , Jt – ; T, k ), ∀ j ̸= Jt – .
i =1

which, in matrix form, is (13).

Remark 7 (Completeness of the market). Observe that, if zero-coupon bonds trade at (n – 1) distinct maturities,
then the market described in Section 2 is complete (at least, up until the time of the shortest maturity).

Remark 8 (Replication with fewer than (n – 1) bonds). If, for all i ∈ S, the CTMC J can only jump to
m – 1 ≤ n – 1 states, then replication of any European-style claim can be achieved by trading m – 1 bonds and the
money market account. For example if, for any jump time τ, we have Jτ – Jτ– ∈ {–1, 1}, then any European-style
claims can be replicated by trading two bonds on the money market account.

5 Ross recovery
In this section, we use an extension of the Ross Recovery Theorem Ross (2015), due to Qin and Linetsky (2016), in
order to determine the dynamics of J under the physical (i.e., real-world) probability measure P. While the dynamics
of J under P are not important for the purposes of pricing and replicating financial derivatives, real-world dynamics
are important for designed optimal investment strategies.

Proposition 9. Suppose that, for some constant ρ < 0, we have

(G – R)π = ρπ,

where π(i ) > 0 for all i ∈ S. Then the generator matrix of J under the real-world probability measure P is
given by
 
π
g1,1 π
g1,2 ... π
g1,n
 π π π

 g
 2,1 g2,2 ... g2,n  π(j ) X
G = .
π
, giπ,j = g , i ̸= j , gi ,i = – giπ,j . (19)

 .. .. .. .. π(i ) i ,j
 . . . 
 j ̸=i
π
gn,1 π
gn,2 ... π
gn,n

Proof. Defining

ΠT
t := e
ρ(T–t )
π, Π(t , i ; T) = ⟨Ei , ΠT
t ⟩=e
ρ(T–t )
π(i ),

it is easy to see that ΠT

t satisfies the pricing equation

0 = (∂t + G – R)ΠT
t , ΠT
T = π.

6
Thus, we have that Π(t , Jt ; T) is the value at time t ≤ T of a derivative that pays π(JT ) at time T. It follows
that Π(t , Jt ; T)/Mt is a (Q, F)-martingale, and we can define a new probability measure Pπ , whose relation to Q is
characterized by the following Radon-Nikodym derivative process
dPπ Π(T, JT ; T)/MT r (Js )ds–ρT π(JT )
RT  
–
≡ ZT := =e 0 , Zt = Et ZT .
dQ Π(0, J0 ; T)/M0 π(J0 )
Thus, denoting by Eπ expectation under Pπ , the value u(t , Jt ; T) of a financial derivative that pays φ(JT ) at time T
satisfies
u(t , Jt ; T)  φ(J )
T
  Z φ(J )
t T
  M Π(t , J ; T) φ(J )
T t T

=E Ft = Eπ Ft = Eπ Ft .
Mt MT ZT MT Mt Π(T, JT ; T) MT
Canceling common factors of Mt and MT and using the Markov property of J, we find
 φ(J )   φ(J ) 
T T
u(t , Jt ; T) = Π(t , Jt ; T)Eπ Ft = π(Jt )eρ(T–t ) Eπ Jt .
Π(T, JT ; T) π(JT )
Thus, we have a transition independent pricing kernel
 φ(J ) 
T
PT–t φ(i ) := π(i )eρ(T–t ) Eπ Jt = i . (20)
π(JT )
Under the Assumptions of Section 2, when one has a pricing kernel of the form (20), we have by (Qin and Linetsky,
2016, Theorem 3.3) that Pπ is the real-world probability measure: Pπ = P.

With the aim of deducing the dynamics of J under Pπ ≡ P, we compute

 Π(t , J ; T)  1
Z  
t
d = Π(t , Jt – + z ; T) – Π(t , Jt – ; T) N(dte , Jt – , dz )
Mt Mt
Π(t , Jt – ; T)  Π(t , Jt – + z ; T)
Z 
= – 1 N(dt
e , Jt – , dz )
Mt – Π(t , Jt – ; T)
Π(t , Jt – ; T)  η(Jt – ,z )
Z 
= e – 1 N(dte , Jt – , dz ),
Mt –
 Π(t , J + z ; T)   π(J + z ) 
t– t–
η(Jt – , z ) := log = log ,
Π(t , Jt – ; T) π(Jt – )
from which it follows that
dPπ Π(T, JT ; T)/MT  Z TZ   Z TZ 
η(Jt – ,z )
= = exp – e – 1 – η(Jt – , z ) ν(Jt – , dz )dt + η(Jt – , z )N(dt
e , Jt – , dz ) .
dQ Π(0, J0 ; T)/M0 0 0

Hence, by Girsanov’s Theorem for Lévy-Itô processes (see, e.g., (Øksendal and Sulem, 2019, Theorem 1.35)), we have
X π(j )
Eπt – N(dt , Jt – , dz ) = νπ (Jt – , dz )dt , νπ (i , dz ) := eη(i ,z ) ν(i , dz ) = g δ (z )dz . (21)
π(i ) i ,j j –i
j ̸=i

From (21), we see that the generator matrix of J under Pπ is given by (19).

6 Example: a two-state CTMC

Throughout this section, we suppose that the generator matrix G of J under Q and the matrix R of interest rates are
given by
! !
–λ λ 0 0
G= , R= ,
λ –λ 0 r

7
where λ, r > 0. The eigenvalues and corresponding (L2 normalized right) eigenvectors of G – R, denoted ρ± and π± ,
respectively, are given by
!
1 r ±γ
q
ρ± = (–2λ – r ± γ)/2, π± = p , γ := (2λ)2 + r 2 .
(r ± γ)2 + (2λ)2 2λ

Noting that (G – R) is symmetric, and thus its eigenvectors are orthogonal, we have from (11) that matrix of
Arrow-Debreu security prices is
! !
  e(T–t )ρ+ 0 π⊤
+
AT
t =e (T–t )(G–R)
= π+ π–
0 e(T–t )ρ– π⊤
–
!
1 – 1 (T–t )(γ+2λ+r ) (γ – r ) + (γ + r )eγ(T–t ) 2λ(eγ(T–t ) – 1)
= e 2 .
2γ 2λ(eγ(T–t ) – 1) (γ + r ) + (γ – r )eγ(T–t )

Next, we have from (9) that vector of zero-coupon bond prices is

! ! !
  e(T–t )ρ+ 0 π⊤
+ 1
BT
t =e (T–t )(G–R)
1= π+ π–
0 e(T–t )ρ– π⊤
– 1
!
1 – 1 (T–t )(γ+2λ+r ) (γ – 2λ – r ) + e γ(T–t ) (γ + 2λ + r )
= e 2 .
2γ (γ – 2λ + r ) + eγ(T–t ) (γ + 2λ – r )

From the above expression as well as the definition of the yield (10) we have

lim Y(t , i ; T) ≡ Y(t , i ; ∞) = (r + 2λ – γ)/2.

T→∞

In Figure 2, we plot the yield Y(t , i ; T) as a function of T for i ∈ {1, 2} with t = 0 fixed. For comparison, we also
plot limiting yield Y(t , i , ∞).

As the state space of J is S = {1, 2}, we need only one bond and the money market account to replicate an
Arrow-Debreu security. Thus, dynamics of the replicating portfolio X are of the form
 
(1) (1)
dXt = Δt – dB(t , Jt ; T1 ) + Xt – Δt B(t , Jt ; T1 ) r (Jt )dt .

(1)
Using (13), the number of bonds Δt one should hold in the portfolio in order to replicate the k th Arrow-Debreu
security with maturity T is
A(t , 2; T, k ) – A(t , 1; T, k )
 

 B(t , 2; T ) – B(t , 1; T )
 , if J t– = 1
 A(t , 2; T, k ) – A(t , 1; T, k )

(1) 1 1
Δt = = ,
A(t , 1; T, k ) – A(t , 2; T, k ) B(t , 2; T1 ) – B(t , 1; T1 )
, if Jt – = 2

 
B(t , 1; T1 ) – B(t , 2; T1 )
 

which, in this example, happens to be independent of the state of J.

In order to find the dynamics of J under the real-world probability measure Pπ we note that

(G – R)π+ = ρ+ π+ , π+ (i ) > 0, i ∈ {1, 2}.

It follows from Proposition 9 that the generator matrix of J under the real-world measure Pπ is
    2 
π+ (2) π+ (2)
– 2λ 2λ2
– (1)
λ π+ (1)
λ
Gπ =  ππ++(1) = γ+r   .
γ+r
π+ (1)
π (2)
λ – π (2)
λ γ+r
2 – γ+r
2
+ +

8
References
D.-H. Ahn, R. F. Dittmar, and A. R. Gallant. Quadratic term structure models: Theory and evidence. The Review
of Financial Studies, 15(1):243–288, 2002. 1

P. Carr and J. Yu. Risk, return, and Ross recovery. Journal of Derivatives, 20(1):38–59, 2012. 1

R. J. Elliott and R. S. Mamon. A complete yield curve description of a Markov interest rate model. International
Journal of Theoretical and Applied Finance, 06(04):317–326, 2003. doi: 10.1142/S0219024903001852. 1, 2, 3

D. Filipovic. Term-Structure Models. A Graduate Course. Springer, 2009. 1

B. Øksendal and A. Sulem. Stochastic Control of Jump Diffusions Stochastic control. Springer, 2019. 7

L. Qin and V. Linetsky. Positive eigenfunctions of markovian pricing operators: Hansen-Scheinkman factorization,
Ross recovery, and long-term pricing. Operations Research, 64(1):99–117, 2016. doi: 10.1287/opre.2015.1449. 1, 6,
7

S. Ross. The recovery theorem. The Journal of Finance, 70(2):615–648, 2015. doi: https://doi.org/10.1111/jofi.12092.
1, 6

9
Figure 1: Effective federal funds rate (EFFR) from January 1, 2022 to September 1, 2024. Source: https:
//www.newyorkfed.org/markets/reference-rates/effr.

0.10

0.08

0.06

0.04

0.02

5 10 15 20

Figure 2: For the model considered in Section 6, we plot the yield Y(t , i ; T) as a function of T with t = 0 fixed. The
solid curves below and above the dashed line correspond to i = 1 and i = 2, respectively. The dashed line is the yield
in the limit as maturity tends to infinity: limT→∞ Y(t , i ; T). Parameters used in this plot are λ = 1/2 and r = 0.1.

10