Interest Rate Markets &

The Forward Rate - Discount Factor Relationship Explained

Saïd Business School, Oxford University

Nicholas Burgess
[Link]@[Link]
nburgessx@[Link]
October 2021

Abstract
In this paper we illustrate and explain the relationship between interest rates (or
forward rates) and discount factors. We present the forward-discount factor
relationship, which is popular and widely used in financial markets for yield curve
construction, and derive the exact formulae using a replication argument approach.
Furthermore we highlight the implicit assumption of simple compounded rates and
consider the disadvantages, side-effects and potential hazards of this assumption.
Finally to conclude we discuss how to relax this assumption and alternative
approaches.

Forward Rates
A forward rate is the interest rate applied for borrowing funds in the future for a
fixed period, three months say. It is a variable rate of interest that changes with
market conditions. Forward rates are typically determined two days before the start
of a loan period at which point they become known fixed rates and are no longer
variable. Forward rates are required to price mortgages, corporate loans and a wide
variety of financial instruments. In the London interbank markets LIBOR1 is a popular
forward rate used to reference over $350 trillion of transactions worldwide 2.

1
London Interbank Offered Rate
2
Burgess, Nicholas, Libor Benchmark Reform: An Overview of Libor Changes and Its Impact on
Yield Curves, Pricing and Risk (September 6, 2019). Available at
SSRN: [Link] or [Link]
Forward Rate Dynamics
In interest rate markets forward rates are instantaneous. At every point in the future
we have a distinct forward rate representing an interest rate for borrowing funds on
that fixing date; together the collection of forward rates forms a forward curve as
illustrated in (figure 1).

Consider the USD 3M LIBOR curve below, at every point on the forward curve we
have a distinct USD 3M LIBOR forward rate for borrowing USD funds for three
months.

Figure 1: USD 3M LIBOR Forward Curve Illustration

The forward rate f(S,T) from time S to T is in this case a distance of 3 months apart. The blue
arrows indicate other instantaneous USD 3M LIBOR forward rates all starting on the forward
curve and ending 3 months later. The area under the forward curve corresponds to the
discount factor, where for example P(S,T) is the discount factor from S to T and corresponds
to the area under the curve from time S to T.

Forward – Discount Factor Relationship

Forward rates of interest and discount factors are related via the following formula,
which is derived from a no arbitrage replication strategy and assumes we are
applying simple compound interest,

2
 𝑃 (𝑡, 𝑆)
𝑓(𝑆, 𝑇 ) = − 1 /(𝑇 − 𝑆) (1)
𝑃(𝑡, 𝑇)

The forward rate 𝑓(𝑆, 𝑇) which is the floating interest rate as at time t for borrowing
funds from time S until time T with t < S < T.

Replication Argument
To give some intuition as to where (equation 1) comes from, consider the following
replication argument. If we deposit funds today at time t making a long deposit until
time T then the interest received on this deposit should be the same as when making
a short deposit to time S and rolling this deposit forward to time T as shown in
(figure 2).

Figure 2: Forward Rate Replication Argument

The replication argument is equivalent to stating that the long discount factor is
equal to the short discount factor multiplied by the forward discount factor.
Mathematically we can represent this as follows,

𝑃(𝑡, 𝑆). 𝑃 (𝑆, 𝑇) = 𝑃(𝑡, 𝑇 ) (2)

3
Rearranging this for the forward discount factor we have,

𝑃(𝑡, 𝑇)
𝑃(𝑆, 𝑇) ≝ (3)
𝑃 (𝑡, 𝑆)

Simple Compounding
If we assume a constant interest rate r over the period [𝑆, 𝑇] with simple
compounding then forward discount factors can be represented as,

1
𝑃(𝑆, 𝑇) ≝ (4)
(1 + 𝑟(𝑇 − 𝑆))

Relabelling r as 𝑓(𝑆, 𝑇) to indicate it is a forward rate and substituting (equation 4)

into (equation 3) leads to,

1 𝑃(𝑡, 𝑇)
= (5)
(1 + 𝑓(𝑆, 𝑇)(𝑇 − 𝑆)) 𝑃 (𝑡, 𝑆)

Rearranging gives,

𝑃 (𝑡, 𝑆)
1 + 𝑓(𝑆, 𝑇)(𝑇 − 𝑆) = (6)
𝑃(𝑡, 𝑇)

Giving the forward rate result assuming interest rates with simple compounding,

𝑃(𝑡, 𝑆)
𝑓(𝑆, 𝑇) = − 1 /(𝑇 − 𝑆) (7)
𝑃 (𝑡, 𝑇)

Ghost Features
A disadvantage of this approach is that we require two discount factors for each
forward rate with the forward rate dependent on the discount factor on the forward
rate’s fixing start and end date e.g. for USD 3M LIBOR the fixing date and the fixing
date + 3M.

4
Modelling forward rates under this assumption, forward rates are dependent on the
discount factor on the fixing start date which is natural, as this is the market fixing
date and also on the fixing end date which is un-natural. Ideally interest rates should
not be dependent on discount factors or any market data after their fixing date.

When forward rates depend on pairs of discount factors forward rate volatility and
market jumps in value results in pairs of jumps in discount factors, namely a jump on
the forward fixing date and an artificial ghost jump on the forward end date. This can
cause a ripple effect on neighbouring forward rates in the forward curve that are
dependent on discount factors at the ghost point. Consequently this approach is
problematic for curve construction, risk-management and can give rise to unrealistic
/ unstable forward curves.

Furthermore one must note that this assumption acts as an approximation and
introduces errors when the underlying yield curve from which we imply forwards has
not been calibrated assuming simple compounding of forward rates.

Continuous Compounding
We can relax the assumption that interest rates compound with simple interest by
modifying our discount expression in (equation 4). If instead we assume a constant
interest rate r over the period [𝑆, 𝑇] that is continuously compounded we then
define the discount factor as,

𝑃 (𝑆, 𝑇 ) ≝ exp(−𝑟(𝑇 − 𝑆)) (8)

Once again relabelling r as 𝑓(𝑆, 𝑇) to indicate it is a forward rate and substituting

(equation 8) into (equation 3) leads to,

𝑃(𝑡, 𝑇)
exp −𝑓(𝑆, 𝑇)(𝑇 − 𝑆) = (9)
𝑃(𝑡, 𝑆)

Rearranging gives,

5
 𝑃(𝑡, 𝑇) 𝑃(𝑡, 𝑆)
𝑓 (𝑆, 𝑇)(𝑇 − 𝑆) = −𝑙𝑛 = 𝑙𝑛 (10)
𝑃 (𝑡, 𝑆) 𝑃(𝑡, 𝑇 )

Giving the forward rate result assuming interest rates with continuous compounding,

𝑃(𝑡, 𝑆)
𝑓(𝑆, 𝑇) = 𝑙𝑛 /(𝑇 − 𝑆) (11)
𝑃(𝑡, 𝑇)

Instantaneous Forward Rates

Whilst the interest rate for a given forward interval is indeed a single rate that is
reset or fixed on the given fixing date as shown in (figure 1); modelling interest rates
to be piecewise constant over forward rate intervals restricts the choice of yield
curve calibration instruments and can introduce curve inconsistencies where
intervals overlap. Ideally we want to discount factors to use instantaneous discount
factors and incorporate the full term-structure of forward rates.

To achieve this we can relax assumptions further to incorporate the term-structure

of forward rates by further modifying our discount expression in (equation 8) and
define the discount factor as follows,

𝑃(𝑆, 𝑇) ≝ exp − 𝑓 (𝑡, 𝑢)𝑑𝑢 (12)

Once again relabelling r as 𝑓(𝑆, 𝑇) to indicate it is a forward rate and substituting

(equation 12) into the forward rate replication formula from (equation 3) leads to,

𝑃(𝑡, 𝑇)
exp − 𝑓 (𝑡, 𝑢 )𝑑𝑢 = (13)
𝑃 (𝑡, 𝑆)

Taking the natural log of both the LHS and RHS and simple rearrangement gives,

6
 𝑃(𝑡, 𝑇) 𝑃(𝑡, 𝑆)
𝑓(𝑡, 𝑢 )𝑑𝑢 = −𝑙𝑛 = 𝑙𝑛 (14)
𝑃 (𝑡, 𝑆) 𝑃(𝑡, 𝑇)

Differentiating with respect to T,

𝜕 𝑃(𝑡, 𝑆)
𝑓(𝑡, 𝑇) = 𝑙𝑛 (15)
𝜕𝑇 𝑃(𝑡, 𝑇)

Noting that t is a free variable and setting 𝑡 = 𝑆 gives,

𝜕 𝑃(𝑆, 𝑆) 𝜕 1
𝑓 (𝑆, 𝑇) = 𝑙𝑛 = 𝑙𝑛 (16)
𝜕𝑇 𝑃(𝑆, 𝑇) 𝜕𝑇 𝑃(𝑆, 𝑇)

Using the definition from (equation 3) we have,

𝜕 𝑃(𝑡, 𝑆)
𝑓(𝑆, 𝑇) = 𝑙𝑛 (17)
𝜕𝑇 𝑃(𝑡, 𝑇)

This result from (equation 17) requires the forward curve interpolation function to
be continuous and sufficiently smooth. This is often not the case for forwards
derived from daily or overnight composite indices (OIS indices) such as US Fed Funds
and US SOFR, where forward market data is very spiky.

To avoid model complexity some market participants prefer to use the analytically
tractable solutions offered by (equation 7), (equation 11) however this might not be
the most suitable choice if one is to achieve the best forward curve fit or to model
the advanced features outlined in (Burgess 2021)3.

3
Burgess, N. (2021) - Advanced Yield Curve Calibration, Mixed Interpolation Schemes & How to
Incorporate Jumps and the Turn-of-Year Effect. Available at: [Link]