EOF Lecture Notes - Bond Math - Sep 22

Download as pdf or txt
Download as pdf or txt
You are on page 1of 27

Economics of Finance

September 22, 2020


Richard Robb
Bond Market Math

We’ll work in as much economics as we can.

Compounding

Suppose you have $100. You deposit it in the bank. At the end of a year, the bank pays you $100
× (1+r). You deposit the original $100 plus the interest for a second year. At the end of two
years, you have $100 × (1+r)2. Continuing to save the interest and principal, at the end of n years
you’ll have $100 × (1+r)n.

If you needed $100 in n years and could invest according to the terms above, you’d need to start
with $100 × (1+r)-n. It would grow in n years to $100 × (1+r)-n × (1+r)n = $100. We could say
that $100 × (1+r)-n is the “present value” of $100 paid in n years.

Now let’s say the bank pays you r/2 every half-year. It pays semiannually. At the end of six
months, you’ll have $100 × (1+r/2). At the end of one year, you’ll have $100 × (1+r/2)2. At the
end of n years, you’ll have $100 × (1+r/2)2n. The interest is compounded semiannually. Using a
semiannually compounded interest rate, the present value of $100 paid in n years is
$100 × (1+r/2)-2n.

Let’s generalize this idea. Let t = time in years until a cash flow is paid. We’ll later take a more
precise look at the way market participants measure t, but for now, we’ll stay abstract.

Let t be any nonnegative number; it doesn’t have to be an integer number of years. If the
compounding frequency is F times per year:

1
Compounding frequency = F: PV = r Ft
(1+ )
F
These are probably the
Taking the limit as F→∞ only formulas you will be
expected to “memorize”
for the class.
Continuous compounding: PV  e  rt

where e is Euler’s constant, 2.7182… Real-world cash flows do not come in truly continuous
time, but are instead tied to dates. There’s no distinction between a payment in the morning of
Tuesday, September 18, 2020 and one in the afternoon of the same day. If you owe money due
September 22, you’ve satisfied your obligation no matter when on September 22 you transfer it.
When we use continuous time in our models, it is for analytical convenience, not realism.

Simple interest is standard for money markets—instruments that mature in one year or less.
Applying simple interest, $100 grows to $100 × (1+rt). This is convenient and compounding
doesn’t make much difference with normal interest rates (e.g., less than 10%) and t ≤ 1.

Simple interest would be misleading for long-term cash flows. Consider a single $100 cash flow
paid in 30 years. The cash flow is offered for sale at $17.411. The simple interest rate solves the
equation $17.411 × (1+ 30×r) = $100, so r = 15.8%. Looks like a great deal, right?

But is this the correct conclusion to draw? If you invested $17.411 at 6% annually compounded,
you’d get to $100 in thirty years: $17.411 ×1.0630 = $100.

The problem with the simple approach is that it ignores the potential to re-invest interest: you
don’t get the interest until the end, and this matters a great deal with long-term investments.

Interest rates expressed in one compounding frequency can be converted to another. To convert
6% annually compounded to semiannual,

 1  0.06  1 2  5.9126 %.


2

1  0.06  1  rs.a.  → rs.a. 


 2 

5.9126% paid semiannually is equivalent to 6% paid annually, assuming you can reinvest the
semiannual interest payments at 5.9126%.

To convert 6% annually compounded to monthly,

12
(1+0.06) = (1+ m ) → rm.  12  1  0.0612  1  5.841%.
r 1

12  

To convert to continuous compounding,

1  0.06   e r → r = ln(1.06) = 5.827%.

A 2% annual rate is equivalent to a 1.9901% semiannual rate. Compounding makes a bigger


difference with higher rates.

2
People who are selling things—loans, investments, projects—will often choose the compounding
frequency that flatters their offering. If you are a borrower, the prospective lender might try to
present the loan terms with monthly compounding. A 12% rate compounded monthly (borrower
pays 1% per month) equates to 1.0112–1 = 12.68% on an annual basis. Put another way, if the
borrower must also borrow the interest at a monthly cost of 1%, the total interest will reach
12.68% by year’s end. If you are an investor, then the seller of the asset might try for annual
compounding.

Here is the amount owed at the end of one year for a loan of $1 that pays 100%, quoted on
various compounding frequencies. Note that daily compounding is about the same as continuous
and the amount owed is converging to e:

Annual: 1 + 100% = 2.0000


2
 100% 
Semiannual: 1   = 2.2500
 2 
4
 100% 
Quarterly: 1   = 2.4414
 4 
12
 100% 
Monthly: 1   = 2.6130
 12 
365
 100% 
Daily: 1   = 2.7146
 365 

Compounding frequency doesn’t matter so much when rates are low. Here’s the same table
quoting a 2% rate:

Annual: 1 + 2% = 1.0200
2
 2% 
Semiannual: 1   = 1.0201
 2 
4
 2% 
Quarterly: 1   = 1.0202
 4 
12
 2% 
Monthly: 1   = 1.0202
 12 
365
 2% 
Daily: 1   = 1.0202
 365 
You will sometimes hear the word “biannual” applied to payments, but it is needlessly confusing
and I recommend against it. “Biannual” can mean semiannual, as in twice per year, but it can
also mean every two years. If you say “biannual coupon,” I guess the meaning has to be

3
“semiannual coupon” since there is no such thing as a coupon that pays every two years, but why
not make it simple and say “semiannual.”

Fixed-Rate Bond Overview

What’s the difference between a bond and a note? There’s no clear distinction between
these two terms, which are used interchangeably. “Note” can sound better when the
original maturity is ten years or less, but you don’t need to worry about this.

Let’s jump from the abstract to the very applied. We’ll replicate market-standard calculations
down to the last decimal place. Sure, you can rely on Bloomberg or other software, but doing this
once with our own hands should provide valuable insights.

The following example of a fixed-rate bond, John Deere Capital


(DE) 3.05%, is due January 6, 2028.

“John Deere is the brand name of Deere & Company, an


American corporation that manufactures agricultural,
construction, and forestry machinery, diesel engines, drivetrains
used in heavy equipment, and lawn care equipment. John Deere
Capital Corporation provides and administers financing for retail
purchases of new and used agriculture and turf equipment, and
construction and forestry equipment manufactured by John
Deere.”

4
DE issued this bond to investors on January 8, 2018 in exchange for $400 million. DE lends the
money to farmers and other customers to buy equipment sold by its parent. DE promised to pay a
coupon of 3.05% p.a. until the maturity date, January 6, 2028. These were originally 10-year
notes. The notes are semiannual, meaning DE generally pays ½ × 3.05% on the 6th day of every
January and July, starting on July 6, 2018. If January 6 or July 6 is a weekend or holiday, then
the coupon comes on the next good New York business day1; investors do not receive any
special compensation for waiting an extra one to three days. At the maturity date, it repays the
principal, i.e. investors get their money back plus the 20th and final coupon. (Principal is also
referred to as the bond’s “amount” or “face value.”) The first coupon, though, was “short” by
two days. Unlike the subsequent 19 coupons, the first coupon was (178/180) × ½ × 3.05%.

Citigroup, HSBC and Merrill Lynch acted as joint lead managers on the new issue. Together
with co-managers, this syndicate sold the bond to investors.

What specifically happened when DE issued the notes? This note is held in custody at a
clearinghouse called Depository Trust Corporation, or DTC. (The U.S. clearinghouses are DTC
for corporate notes and Fedwire for U.S. government and eligible agency bonds. Eurobonds clear
via Euroclear or Clearstream.) Investors must have an account with DTC or with a financial
institution that does. Once investors send money and instructions to DTC, DTC credits their
account with a certain amount of the DE notes. Every January 6 and July 6, DE sends 3.05% × ½
× $400 million = $6,100,000 interest to DTC. (The very first coupon on July 16, 2018 was only
$6,032,222.22). DTC checks who holds these notes on its books on that day and credits their
account with their share of the interest. This note can be held in minimum amounts of $1,000.

DE issued these notes at a yield of 62 basis points (bps) above comparable Treasury notes. We’ll
discuss yield calculation shortly.

Like most bonds, these DE notes are unsecured, meaning DE did not pledge collateral or security
that bondholders could seize if DE failed to pay. Were DE to default, a trustee would pursue
remedies in court on the bondholders’ behalf.

1
“New York business day” means days that banks are open in New York City for the purposes of making
payments in USD. Nonbusiness days are Saturdays, Sundays, New Year’s Day, Martin Luther King Jr.
Day, Presidents’ Day, Memorial Day, Independence Day, Labor Day, Columbus Day, Veterans Day,
Thanksgiving and Christmas. You can check a country’s business days on Bloomberg by entering CDR
<Go>.

5
Secondary Market Trade

John Deere, like most corporate bonds are freely negotiable. Bondholders can sell to anyone they
like at any price both parties accept. Secondary transactions are negotiated after the original
issuance. When the trade settles, the buyer must have USD in its DTC account, and the seller
must have the bonds. The buyer presents DTC with instructions to buy bonds for a particular
price from the seller, while the seller presents offsetting instructions. If both sides match, DTC
executes delivery vs. payment (DVP), switching the owner of the bond and the owner of the
money.

The following display shows a secondary trade:

Trade Date: The date on which an agreement is reached. Here, it’s Friday September 18, 2020.

Settlement Date: The date on which a trade is cleared by DVP. U.S. corporate bonds and most
European government bonds settle by convention T+2 (trade date plus two New York business
days); U.S. Treasuries settle T+1. The settlement date for this trade is September 22, 2020.

Price: In our example, the buyer and seller agreed to a price of 112.898 percent of the principal
amount of the bonds. Since this is a $1,000,000 transaction, the buyer must pay $1,128,980.00
plus accrued interest.

6
Accrued Interest: The buyer must also pay the interest that has accrued since the last payment
date, July 6, 2020. How much time has passed between July 6, 2020 and the settlement date,
September 22, 2020, has passed? By the market convention, we use 30/360 (pronounced “thirty
three sixty”): assuming there are 30 days in each month and 360 each year. So we have 30 – 6 =
24 for July (even though July has 31 days), 30 for August and 22 for September so it’s 24+30+22
= 76 days.
76
Accrued interest = × 3.05% × $1,000,000 = $6438.89.
360

This method of accrual leads to some obvious quirks. Buying a bond that settles February 28 in a
non-leap year and selling one day later for settlement March 1 results in three days’ accrued
interest. Buying for settlement March 30 and selling one day later for settlement March 31
results in no increase in accrued interest. Market participants are aware of this and adjust bid and
offer prices accordingly. You can compute days between dates via the 30/360 method using the
DAYS360 function in Excel.

Why is 112.898 the right price for a bond with a 3.05% coupon and remaining maturity of about
7.25 years? Let’s start by determining what return this price would deliver to the buyer, then
compare that return to alternative bonds.

Relationship between Price and Yield

If we discount all a bond’s cash flows at a common interest rate, the yield to maturity (or
“yield,” when the meaning is clear from context), then the present value of all cash flows must
by definition equal the price plus accrued interest. “Yield to maturity” is the same idea as
“internal rate of return” (IRR), but for bonds. By market convention, the yield’s compounding
frequency matches the bond’s payment frequency, and its discounting follows the bond’s day
counting method.

Estimating Approximate Yield Given Price

We can estimate yield with a simple approximation, which will be familiar to those of you who
took International Capital Markets. Unlike ICM, we’ll go on to calculate the exact yield.

First, find the current yield, the coupon divided by the price. In the case of the DE note:

Current yield = 3.05%/1.12898 = 2.70%.

But we must also adjust for loss of the 112.898 – 100 = 12.898 premium the buyer incurs when
the bond matures in about 7.3 years.

7
Amortization of discount ≈ 12.898%/7.3 ≈ 1.767% per year.

This loss is incurred on an investment of 112.898:

Amortization of premium per dollar invested ≈ 1.767%/1.12898 = 1.565% per year.

Current yield – amortization of premium per dollar ≈ 2.70% – 1.565% = 1.135% vs. the actual
yield to maturity, 1.197%. This approximation slightly understates the return by assuming the
investor loses the premium evenly throughout the life, while the yield to maturity assumes it
comes at the end.

We can use a similar method to amortize the discount of a bond trading below par. If the DE
price fell to 98, its yield would rise to approximately:

Current yield + accrual of discount ≈ 3.05%/0.98 + 2%/(7.3×0.98) ≈ 3.39%.

Here are some U.S. Treasuries September 18, 2020, around the same time as the DE price was
collected:

When the underwriters sold the bond in January 2018, the yield curve was upward sloping. Why
did they market it in terms of spread vs. the 2¼ of 11/27?

Treasuries trade in points and 32nds. Half of a 32nd is called a “plus” (written “+”). The
Treasury 2.75% due February 2028 is bid at 116-15 = 116.46875. This bond pays a 2.75%
coupon semiannually until February 2028. Although its yield is below the DE’s, its credit is
presumably better, and so offers investors an alternative. (Some Treasury notes mature on the
15th day of the month, some on the last day of the month. There is no way to know from this
table.)

Making this rough calculation for the DE bond’s return, investors get about a 1.135% - 0.478% =
0.657% annual pick-up in return for bearing DE credit risk.

8
Market Standard Calculation – Price given Yield

We will now calculate the rate of return on the DE bonds with greater precision. Our objective at
the moment is not to find the most rigorous method, but rather to match the industry standard.

In the DE example for September 22 settlement, 76 days have passed since the last coupon date
and 104 days remain until the next cash flow on January 6, 2021 using the 30/360 day method of
calculating time. The fraction of a year until the next cash flow is 104/360 = 0.28889. The time
until the following cash flow on July 6, 2021 is 104/360 + ½ = 0.78889 years. Subsequent cash
flows arrive every 0.5 years until the bond matures and repays principal. Each coupon payment
is exactly 3.05%/2 = 1.525% of par.
2 t
 y
With semiannual compounding at rate y, the discount factor for time t is 1   .
 2

DE 3.05% due 1/06/2028 – Price Given Yield


Yield = 1.1967135%
Cash Discount Present
Date Time
Flow Factor Value
Jan 2021 0.28889 1.525 0.99656 1.51975
July 2021 0.78889 1.525 0.99063 1.51071
Jan 2022 1.28889 1.525 0.98474 1.50173
July 2022 1.78889 1.525 0.97888 1.49280
Jan 2023 2.28889 1.525 0.97306 1.48392
July 2023 2.78889 1.525 0.96727 1.47509
Jan 2024 3.28889 1.525 0.96152 1.46632
July 2024 3.78889 1.525 0.95580 1.45759
Jan 2025 4.28889 1.525 0.95011 1.44892
July 2025 4.78889 1.525 0.94446 1.44031
Jan 2026 5.28889 1.525 0.93885 1.43174
July 2026 5.78889 1.525 0.93326 1.42322
Jan 2027 6.28889 1.525 0.92771 1.41476
July 2027 6.78889 1.525 0.92219 1.40634
Jan 2028 7.28889 101.525 0.91671 93.06870

Total 113.54190
Accrued 0.64389
Quoted Price 112.898

9
This matches the standard calculation according to market conventions. The price quoted is the
clean price, and the price including accrued interest is the dirty price. (Don’t worry if you
struggle to remember which is which—it rarely comes up outside the classroom.)

Written out algebraically,


 
t0 F  C C C 1 C 
P  y C  F  F  ...  F F 
 accrued  1   
100  F  F y  y
2
 y
n2
 y 
n 1

 1  1    1    1   
 F  F  F  F  
where
n = number of coupon payments remaining (n = 15)
C = annual coupon as decimal (C = 0.0305)
y = yield to maturity as decimal (y = 0.011967135)
P = price as percent (P = 112.898)
F = number of coupon payments per year (F = 2)
t0 = time until first coupon payment (104/360)

Applying the formula for the partial sum of a geometric series to simplify, we arrive at

P  y
 t0 F
 C  y
 ( n 1)
C C
 accrued  1   1  1     .
100  F  y  F  y F 

10
In the DE example,

P 76
+0.0305× =
100 360
104
0.011967135 -2×360 0.0305 0.011967135 -(15–1) 0.0305 0.0305
(1+ ) [(1– ) (1 + ) + + ].
2 0.011967135 2 0.011967135 2

Solving for P, P = 112.898.

Yield depends on price; a higher price raises the cash flow discount rate and so corresponds to a
lower yield. There’s no equation to calculate the exact yield given a bond’s price. You (i.e., the
computer) must rely instead on iterative methods. That is, guess a yield, and if it generates too
high a price, guess higher in the next round. Repeat until you match the given price.

Yield to maturity summarizes a bond’s rate of return. Since market conventions for day counting
are arbitrary and yield ignores the yield curve’s shape, yield to maturity is not necessarily the
most meaningful measure of bond value. Additionally, it assumes that cash flows are received on
the scheduled payment date, while in practice, cash flows scheduled for non-business days are
not paid until the following good date.

Applying the yield-to-maturity calculation to the U.S. Treasury 2.75% due March 2028 results in
a yield of 0.478%. DE, then, provides a pickup of 1.197% – 0.478% ≈ 71.9 bps vs. the
comparable Treasury note. An investor considering the switch must decide whether the
additional 0.719% p.a. is worth DE’s increased credit risk. This is a little more accurate than the
rough calculation suggesting a yield differential of 65.7 bp.

DE is slightly wider vs. Treasuries than when it was issued two years ago, mostly due to
uncertainty surrounding the pandemic and global trade. The world experiencing a glut of corn
and soybeans, while trade wars threaten U.S. exports of farm equipment. DE’s operating income
for the fiscal year ended 10/31/19 was a solid $4.06 bn, the same as fiscal 2018 and up from $2.7
bn in fiscal 2017. Operating income in the quarter ending 7/31/20, was a solid $1.53 bn.

Duration

How much does the price of a fixed-rate bond change as the yield changes? We could just as
easily pose the question the opposite way: “how much does the yield change as the price
changes?” Both questions address the relationship between change in yield and change in price.

Suppose the DE bond yield rises by 10 bps. This might follow a 10 bps increase in eight-year
U.S. Treasury yields; day to day, the spread between the yields on the DE bonds and the

11
Treasuries maturing at the same time should stay relatively constant. This increase might arise
from an increase in the real rate or expected inflation.

How much would the price fall? Let’s begin with a simple approximation to develop intuition. It
should drop by about 0.1 × 7.3 years = 0.73%—if the bondholder pays 0.73% less for the bond,
she’ll enjoy (approximately) 10 bps extra return per annum. If the bond is trading at a premium
to par, she’ll suffer 0.73% smaller loss premium. If it’s trading at a discount, she’ll realize an
accrual of 0.73% bigger discount over 7.3 years.

Now consider the same 10 bps yield increase for a bond maturing in only one year. The price of
this bond would fall by about 0.1%. That would boost the return over one year by about 0.1%.

Of course, this is only approximate. For greater precision, apply our formula for price as a
function of yield:

A bond consists of a sequence of cash flows:

 t1F t2 F
 y  y
P  accrued  c1 1    c2  1    ...
 F  F
Taking the derivative,
 t1F 1  t 2 F 1
P  y  y
 c1t1 1    c 2 t 2 1    ...
y  F  F
Rearranging terms,

t F t F
P  y  y 1  y 2
1   c1t1 1    c 2 t 2 1    ...
y  F   F  F
 t F t F
 Duration .
P  AI  y 1  y 2
c1 1    c 2 1    ...
 F  F

Duration (or more precisely, Macaulay duration), as defined above, is the weighted average time
in years until cash flows are paid. The weights are the PVs of the associated cash flows. For a
single cash flow, duration is simply the time until payment is complete.

The above equation isn’t yet ideal for measuring the interest rate sensitivity of a bond or, more
 y
generally, a set of cash flows. To rid the left-hand side of the nuisance term 1   , divide both
 F
 y
sides by 1   and define
 F

12
Duration
Modified Duration = .
y
1
F

This provides a more useful diagnostic: modified duration (MD), the approximate percent
change in a bond or portfolio’s value with respect to a 100 bps interest rate fluctuation (i.e., with
respect to yield). If MD = 5, then a decline in interest rates from 4% to 3% increases the
portfolio’s value by approximately 5%.

To determine Macaulay duration for the DE example, take the sumproduct of the “Time” and
“Present Value” columns on the table “Price Given Yield,” then divide by the sum of the PV.
The Macaulay duration is 6.6039 years.

Modified duration = 6.6039 / (1+ 0.011967 / 2) = 6.565.

Basis point value (BPV) measures the change in price per one million face value under a one
basis point change in yield. It’s also called “DV01” (pronounced “dee vee oh one”), delta of the
value for one basis point.

BPV = DV01 = MD × (Price + Accrued Interest)

BPV = 6.656 × 113.542 = $745.37.

If the yield rose by one basis point, the bond’s price would fall by about $750.21 per $1 million
face value. If the yield declined by one basis point, the price would rise by about the same
amount.

Because the relationship between price and yield is not linear, duration can only be an
approximate measure of interest rate sensitivity. Although MD is more useful in practice than
Macaulay duration, we still express MD in years because it closely resembles Macaulay.
Intuitively, why is duration important for fixed-rate bonds? If you have a fixed-income security,
you are stuck with the income it pays you. The longer the time till maturity, the longer you are
stuck, and the more sensitive the price is to a change in yield.

Since Macaulay duration is mainly an academic concept and “modified duration” takes longer to
say, people use “duration” and “modified duration” interchangeably when meaning is clear from
context.

13
Convexity

Convexity measures the change in duration as interest rates change. Remember, yield and price
do not have a linear relationship.

Let’s consider convexity in terms of a paradox. Assuming the yield curve is flat and continuously
compounded interest rates are 5%, compare the following two portfolios, each worth $100:

 Bullet Portfolio: $100 invested in a single cash flow to be paid in five years. Principal
amount = $100 × e5%×5 = $128.40. (That is, $128.40 in five years is worth exactly $100
today.)

 Barbell Portfolio: $50 in cash and $50 invested in a single cash flow to be paid in 10
years. Principal amount = $50 × e5%×10 = $82.43.

The duration of both portfolios is five years. If continuously compounded interest rates rise by
one basis point, the value of both portfolios declines by 0.05 to 99.95.

But the barbell portfolio has more convexity: if rates rise, its duration falls faster and it loses less
money. If rates fall, its duration rises faster and it makes more money. Formally, the absolute
value of the barbell’s second derivative of price with respect to yield is greater. To see why the
barbell is better in terms of yield shocks, consider extreme cases. If interest rates approach
infinity, the bullet portfolio becomes worthless, as does the ten-year cash flow in the barbell
portfolio. But the cash maintains its value, so the barbell portfolio approaches $50. If interest
rates drop to zero, the bullet portfolio converges to its face value, $128.40, while the barbell
portfolio converges to $50 + $82.43 = $132.43. The following chart shows profit or loss as a
function of yield.

14
40
Two Portfolios of Equal Duration with Parallel Yield Curve Shifts
30

Barbell
20

10 Bullet

0
0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%
-10

-20

-30

While each portfolio’s duration moves in the right direction, no matter how interest rates change,
the barbell always does better. A trader who buys the barbell and shorts the bullet appears to win
in every outcome. How can that be?

It turns out, we’re pushing too hard on the assumption that the yield curve is flat and can only
shift up or down. If that were the way the world worked, everyone would buy barbell portfolios,
driving down their yields relative to portfolios with equal duration and less convexity.

Let’s think about the economics. Suppose short-term interest rates fall sharply due to a downturn
in the business cycle that lowers expected inflation. A five-year note substitutes for lending short
term over a five-year horizon, so the yield on five-year notes declines. But short-term rates are
unlikely to stay low forever, and the central bank will eventually get inflation back on target.
Thus the yield on the ten-year note will decline by less than that on the five-year note, and the
yield curve will steepen, disappointing traders who bought the barbell and sold the bullet.

Long-term yields are generally less volatile than short-term yields because the short rate tends to
revert to its mean of around 3%. In terms of our model of consumption and exchange:

Real Rate: Shocks to the real rate tend to wear off over time, so it returns to the rate of time
preference. For example, if the capital stock declines due to an earthquake but the labor supply
remains constant, the productivity of capital will rise. This raises the real interest rate,
incentivizing people to save until the capital stock is rebuilt, the marginal product of capital falls
and real interest rates return to normal.

15
Inflation: Over time, inflation returns to the target rate, around 2% for developed countries. If
inflation rises, say due to unanticipated shocks in money demand, the central bank will tighten
and reduce inflation to the target level.

In my view, it’s a mistake to push the mathematics of convexity too far. Any insights it provides
hinge on a particular and unrealistic assumption about how the yield curve moves around. I’ve
heard “convexity” described as something bond salesmen talk about to sell bonds. Bonds with
more convexity (holding duration constant) might entice the unsuspecting, but the underlying
theory is unsound.

Do not confuse yield to maturity “convexity” with the negative convexity of callable bonds,
fixed-rate residential mortgages, or securities backed by such mortgages. With the latter,
investors receive principal back prematurely when interest rates decline and borrowers refinance.
It’s a real and unambiguous phenomenon that hurts investors, other things equal. We’ll discuss
callable bonds later in the semester.

Floating-Rate Notes

The following display shows a USD floating-rate note (FRN) issued by Citigroup, Inc.,
Citibank’s parent company, on June 12, 2018. Every three months, it pays USD three-month
Libor + 1.25%, calculated on an actual/360 basis. Citigroup will pay par plus the last interest
payment. The note matures on July 1, 2026.

16
Unlike the DE bond, the Citigroup notes are callable by the issuer. They mature on July 1, 2026.
If Citigroup likes, it has a one-time call option on July 1, 2025. Odds are pretty high that
Citigroup will call the bonds unless it’s in dire trouble at the time. Not calling means that it
borrowers for a year at L+1.25%.

The note’s interest rate resets effective the 1st day of every January, April, July and October. (It
was actually issued on June 12, 2018 rather than July 1, so the first coupon was set two days
before June 12. The first coupon accrued for the full period from June 12 to October 1). Payment
scheduled for non-business days roll to the next good business day. (Rolling is subject to one
quirk: it doesn’t roll into a new month. If rolling forward would take it into a new month, it rolls
backwards to a good business day. Of course, rolling into a new month is not an issue for the
notes in this example where payments are scheduled on the first day of the month.) Effective
June 12, three-month USD Libor was 2.32631%. Thus for the 111-day period from June 12,
2018 to October 1, 2018, the notes paid 2.32631% + 1.25% = 3.57631%. Bondholders received
3.57631% × (111/360) = 0.011026956% on October 1, 2018.

Libor peaked at 2.797% as the Fed tightened the overnight rate in September 2018 and again in
the following December. Libor started drifting down as the Fed reversed course eased starting in
July 2019 (keep in mind that the three-month rate not only reflects the current overnight rate
anticipated future Fed actions over the next three months). The rate on April 1, 2020 was high
relative to Fed funds at the height of pandemic uncertainty. We can’t yet determine the interest
rate for October 1, 2020 to Monday January 4, 2021, since it will depend on three-month Libor
effective October 1, 2020.

The prices of floating-rate notes are not sensitive to interest rates like those of fixed-rate bonds.
Why not? Consider a note that pays interest at today’s three-month Libor, L3M. Suppose a buyer
purchases the note for settlement on three months before it matures, say December 31. For

17
simplicity, assume (i) the note does not pay a margin over Libor, (ii) investors consider it a
perfect substitute for bank deposits, and (iii) investors can borrow and lend at Libor. The value of
the note on September 30, three months before maturity must be par:

90
100  100  L3M 
360  100.
90
1  L3M 
360
The numerator is the cash flow an investor receives in three months, while the denominator is the
discount factor. Put another way, were the price under 100, an investor could borrow money at
the rate L3M, buy the note, wait three months and earn a profit. If the price were higher than 100,
no one would buy the note, since they could instead invest at Libor and earn a better return.

Now consider the same note six months before it matures, on June 30. It pays three-month Libor
for the three-month period from June 30 to September 30 and we already know it will be worth
par on September—we can apply the argument again from the perspective of June 30 (as if the
note matured on September) and conclude that its value as of June 30 is equal to par. By
induction, a floating-rate note of any maturity is theoretically worth par, at least on a reset date.
Intuitively, if interest rates rise, a floating-rate note pays a higher interest rate, but so does
everything else, so the note does not become more valuable.

Calculating the Discount Margin on a Floating-Rate Note

In May 2007—before the financial crisis began—Citigroup issued ten-year floating rate notes
with a margin of 27 bps over three-month Libor. During the crisis, when Citigroup was teetering
on the brink of insolvency and was partly nationalized by the U.S. Treasury, it did not issue
floating rate notes at all—the spreads would have been too high to stomach.

By 2016, Citigroup was back on its feet. In March 2016, it issued for five years at L+1.38%. In
June 2018, it issued in ten years at L+1.25%. But by the standards of today, even this spread
appears generous; the notes are trading at a premium price of around 101.32. An investor who
buys $1,000,000 of these notes in the secondary market will earn L+1.25% applied to a principal
amount of $1,000,000 plus the $1,000,000 principal in July 2025 (probably), assuming Citigroup
does not default. The investor pays 101.32% to buy the bonds, a premium that will cut into her
return over 4.75 years. Loss of the premium deducts approximately 1.32%/(4.75×1.0132) = 27.5
bps p.a. since about 4.75 years remain until the note comes to an end. By our rough
approximation, the return equates to L+1.25% – 27.5% ≈ L+ 0.975 %. That is, buying a bond
that pays L+1.25% and matures in 4.75 years at a price of 101.32 delivers approximately the
same return as buying one that pays L+0.975% at par. (We’re being particularly crude here by

18
ignoring the way the premium cuts into the current return. That is, the buyer has to pay a little
more than par to get the L+1.25% running yield.)

As with fixed-rate bonds, buyers of floating-rate bonds pay sellers the interest that has accrued
since the last payment date. Consider an investor who agrees on September 18, 2020 to buy
$1,000,000 of the Citigroup FRNs for a price of 101.32. As before, the notes settle in two New
York business days, on September 22, 2020. Whenever we say a certain amount of bonds, we
mean “principal” or “face” value, so the face value is $1,000,000. The investor pays the interest
that has accrued since the last payment date, Wednesday July 1, 2020. Following the money
market convention, interest accrues on an actual/360 basis: the floating rate × actual number of
days on the calendar / 360.

Eighty-three actual days elapsed between the last payment date, July 1, 2020, and the settlement
date, September 22, 2020. Since three-month Libor reset at 0.29613% effective July 1, 2020, the
rate for the current period is 0.29613% + 1.25% = 1.54613%. The accrued interest for a buyer of
$1,000,000 is 1.54613% × (83/360) × $1,000,000 = $3,564.69. A buyer of $1,000,000 at a price
of 101.32 pays $1,013,200.00 plus $3,564.69 accrued interest for a total of $1,016,764.69 to
acquire the bonds on September 22, 2020.

Two pages ago, we informally estimated a discount margin (DM) of 97.5 bps. DM is the
equivalent to yield to maturity for floating-rate notes. (The term is inelegant, because it applies
equally to bonds trading at a discount or a premium.) As the screen above reveals, the

19
approximation was close but slightly overestimated the market standard calculation by 0.5 basis
points. (That’s mostly because we ignored the fact that you spend more than par to acquire the
L+1.25%.) To calculate the precise discount margin on a floating-rate note, find the bond’s
expected cash flows, assuming that each future Libor rate will match the most recent one, i.e., on
the trade date. Then discount each payment (except the first) at a fixed spread over that same
Libor rate according to actual/360 and the bond’s reset frequency. (Assume the length of each
period is (365.25/360) × 1/f where f = reset frequency.) For the first payment, simply use the
actual amount. Solve for the fixed spread that matches the bond’s price plus accrued interest to
arrive at the discount margin. This is the floating-rate equivalent to yield to maturity. (You don’t
have to memorize this calculation.)

It may seem crude to assume three-month Libor will remain constant and discount at a spread
over that single value. Why not take factors like the shape of the yield curve into account? It
turns out not to matter much, since the effects of raising future Libor and raising the discount
factors nearly offset. Redoing the exercise with Libor 100 bps higher would decrease the DM by
less than 1 bp. Assuming Libor 100 bps lower would raise the DM by less than 1 bp.

This calculation, like yield to maturity, is not my personal idea about how to compute discount
margin. It’s the market convention, as reflected on Bloomberg. That doesn’t mean you have to
use it when assessing whether to buy or sell. You are free to do as you like, but need to use the
industry-standard calculation to communicate with the rest of the world.

20
Duration of a Floating-Rate Note, Revisited

Let’s think more deeply about the duration of floating-rate notes. Earlier, we argued that if
analysis is restricted to quarterly dates just before the reset, a floating-rate note with no margin
and no credit risk must be priced at par. But this fails to describe the Citigroup FRNs, as well as
other real-world situations.

The margin, s, of the Citigroup FRNs is 1.19%. Assume n periods remain until the note matures.
Let f1,f2,…,fn be the Libor rates that apply to each period expressed as a percent. Assume perfect
certainty. (We could instead consider the case with uncertainty and let f1,...,fn be expected Libor
rates.) The length of each period is p1,p2, …, pn.

For simplicity, begin with n=2. The present value of the two interest payments and principal paid
at t=2 is:

( f1  s) p1 1  ( f 2  s) p 2  1 1 
PV =   1  s   (*)2
1  f1 p1 (1  f1 p1 )(1  f 2 p2 )  1  f1 p1 1  f1 p1 1  f 2 p2  

Not surprisingly, PV=1 if s=0 for any value of f1 and f2. This confirms that the notes have no
duration apart from the spread; even as f1 and f2 go up and down, PV=1 so long as s=0.
Otherwise, PV=1 plus the present value of the spread payments.

If we mindlessly consider the interest rate sensitivity of the PV in (*), we will find positive
duration; the spread payments have duration like a bond.

But does that seem right to you? Should a note that pays L+s, priced at par, have interest rate
sensitivity?

n
1
2
In the general case, PV = 1  s  1  f p     1  f p  .
i 1 1 1 i i

21
Consider an FRN priced at par. We can see an immediate defect in (*) since the formula always
gives a price that is greater than par. In this case, (*) makes a false statement of fact: it asserts
that the FRN is valued at a premium.

In order to determine a more plausible duration, let’s take a step back. Formula (*) is not doing
its job. Once we have a better model, one that matches the market data (i.e., the price of the
note), we can change f1, … fn and see what happens to the note price.

Clearly, this model must incorporate the possibility of default—otherwise we have no chance of
explaining why the note is only worth par. Keep in mind that our purpose is to calculate duration;
a complicated discussion of risk preferences would only distract us. Assume instead:

Risk Neutrality: price= present value of expected cash flows without compensation for
uncertainty. A risk neutral investor would pay $500,000 for the prospect of $1,000,000 if a fair
coin came up heads.

For a note priced at par, we can model the Libor margin as compensation for default risk. This
margin should be the approximate annual probability of default, assuming zero recoveries. That
is, investors get s extra every year, but also have an s percent probability of losing everything.
Assume the default probability each period is s×pi conditional on survival up to the beginning of
the period. The probability of surviving the ith period is 1 – s×pi ≈ 1/(1+s × pi).

1
Probability of surviving up to the end of period i = .
1  s  p1 1  s  p2 1  s  pi 

The expected present value is each cash flow’s present value multiplied by the probability of
survival. Consider a three-year quarterly floating-rate note, with fi = 2% for all i and s=3%.
Weighting by the survival probability brings us very close to reproducing a price of par, although
the result is inexact due to approximations from discrete time. (Intuitively, at the end of each
quarter, survival is rewarded by an amount s×pi, which the investor gets to keep with probability
1/(1+ s×pi), but the loss is 1+fi×pi, which occurs with probability 1 – 1/(1+ s×pi). The expected
benefit does not precisely offset the cost, but they converge as the time period, pi, gets smaller.)3

3
This is easy to demonstrate in continuous time. Consider a bond that matures at T. r = coupon rate, s =
instantaneous default probability.

T
Price = 
0
e  st e  rt ( s  r )dt  1  e  rT e  sT  1 .

 st  rt
The term e is the probability of surviving up to t; e is the discount factor and (s+r) is the bond’s rate of
 rT  sT
payment. The term 1  e e is the present value of the principal multiplied by the probability of survival up to
T. Since the price = 1 regardless of the interest rate, the bond must have zero duration.

22
Rather than assuming a default rate of s and zero recoveries, we could assume some recovery
rate, λ, and a correspondingly higher default rate = s/(1-λ). The assumption that λ = 0 is just for
convenience but not essential.

Libor = f1,..,f12 = 2%; s = 3%. Default probability = 3% p.a.


Time (years) Cash DF @ Libor Survival
0.25 1.278 0.9951 0.99264
0.50 1.278 0.9900 0.98509
0.75 1.278 0.9850 0.97759
1.00 1.278 0.9800 0.97015
1.25 1.236 0.9752 0.96301
1.50 1.278 0.9702 0.95568
1.75 1.278 0.9653 0.94841
2.00 1.278 0.9604 0.94120
2.25 1.250 0.9556 0.93419
2.50 1.278 0.9507 0.92708
2.75 1.278 0.9459 0.92003
3.00 101.278 0.9411 0.91303

Present Value 108.83910 99.95727


PV if L=2.01% 108.83766 99.95707
Duration 0.13 years 0.02 years

“Cash” is cash flow assuming no default with Libor set at 2%. The next column is the discount
factor, and the last is the chance of survival up to the end of the period assuming a default
probability of 1 – 1/(1+s × pi). The present value of the unweighted cash flows is 108.83910.
Weighting each cash flow by the survival probability brings the expected present value down to
99.9572. Raising f1,..,f12 to 2.01%, we can compute the basis point value, then multiply by 100
and divide by the price to find the modified duration. It turns out to be almost exactly zero. If we
failed to multiply by the survival probability, we would pick up duration from the 3% fixed cash
flow stream.

23
Let’s now shift our focus from a bond priced at par to one trading at a discount. Assume f1,..,f12 =
2% and s=3%, but the discount margin is 18%. Applying the model and assuming a survival
probability of 1/(1+ 18%×pi) each period, we match the price almost perfectly:

Libor = f1,..,f12 = 2%; s = 3%. DM = 18%


Time (years) Cash DF @ L + 18% Survival
0.25 1.278 0.9529 0.95740
0.50 1.278 0.9065 0.91529
0.75 1.278 0.8624 0.87504
1.00 1.278 0.8205 0.83656
1.25 1.236 0.7818 0.80092
1.50 1.278 0.7437 0.76570
1.75 1.278 0.7076 0.73202
2.00 1.278 0.6731 0.69983
2.25 1.250 0.6411 0.66969
2.50 1.278 0.6099 0.64024
2.75 1.278 0.5802 0.61209
3.00 101.278 0.5520 0.58517

Present Value 66.41097 66.25049


PV if L=2.01% 66.25450
Duration -0.606 years

When interest rates rise, price increases and negative duration is not negligible. Why? If we can’t
develop some intuition, it is hard to trust the result.

When interest rates rise, floating-rate payments go up, but so do discount factors. In addition, we
shift the value in the bond closer to the present, when it’s least likely to default. If, as an extreme
example, interest rates jump to 100%, the FRN’s price would go to 84.67, holding the annual
default probability at 18%. With 100% short-term interest rates and high floating-rate sets, the
first-year cash flows represent most of the value. Since these cash flows are scheduled to arrive
when the bond is most likely to survive, the result in the table above is plausible.

Up till now, we have considered FRNs at the moment before Libor is set. But the current set can
lead to duration, since it’s a fixed cash flow. Everything else held equal, the duration of a
quarterly floating-rate note the day after it resets is 0.25 years.

The point of this analysis is not to obsess over computing floating-rate note duration,
but to illustrate how a simple economic model can provide useful insights. We pursued
the following steps: (i) Make assumptions suited to the problem at hand (risk
neutrality, equal annual default probability, zero recoveries in the event of default). (ii)
Fit the model, which gives bond price as a function of survival probability and interest
rates, to our data. (iii) Change the interest rate to measure sensitivity, as was our
objective. (iv) Finally, think about the result to be sure it makes sense.

24
Yield – What does time have to do with it?

C
At the beginning of today’s class, we wrote PV   C  (1  r ) t without saying why time
(1  r ) t

enters into the equation. Good things (including interest) are supposed to come to those who
wait, but there ought to be more to it than that.

We can arrive at several logically distinct answers. Any one of them can explain the connection
between time and the return or they can work in combination:

 Real risk-free rate: We established last week that the real interest rate should equal the
marginal product of capital. Production operates in time: it is the marginal product of
capital per year. We also noted that in equilibrium, the real rate equals the rate at which
consumers are willing to trade present for future consumption—the cost of delay is
naturally experienced per unit of time.
 Inflation: This, too, arises from the rate of increase in money supply relative to demand
per unit of time. We saw last week that we could add the real rate and the rate of inflation
(approximately) to arrive at the nominal rate.
 Default risk: The more time, the greater the chance of default. In the case of DE bonds,
for example, yield spread was 0.72% vs. Treasuries. This is the market-clearing price if
the expected loss from default of DE is 0.72% per year and investors are approximately
risk neutral. (If we think that U.S. Treasuries might default, we can say that “0.72% is the
difference between DE and Treasury default risk per year.”)
 Opportunity cost: Consider an investment manager who believes she can unearth market-
beating opportunities. Does this contradict rational choice and efficient markets? I don’t
think so, but it’s a longer discussion for later on in the course. For now, let’s assume this
is the case. Assume further that the economy has deflation equal to the real interest rate,
so the nominal interest rate is zero. Suppose the manager finds a deal that she believes
will turn $100 million into $110 million. There is no default risk but some other source of
uncertainty, such as the outcome of a court case. With nominal rates equal to zero, what’s
the hurry? If it takes one year or five years, she’ll outperform the market by a present
value of $10 million. On the surface, earning a 2% return for five years is as good as 10%
for one year. But for all sorts of reasons, she has a finite lifetime and her investors have
limited patience. All else equal, better for the manager to find five one-year deals that
turn $100 million into $110 million, one after the other, then one five-year deal with a
10% return.

25
YTM, Duration Revisited

Here’s a more rigorous way to view yield to maturity and duration for fixed-rate bonds formally
considering default risk.

Assume risk-neutrality. Define d = annual default probability. Assume zero recoveries. The
probability of surviving for the first period is 1– d and probability of surviving up through period
j
j is (1-d) . Assuming for simplicity that payment frequency F=1 and t=0 is a coupon payment
date (i.e., the next coupon is paid at t=1), with n remaining coupons, the expected present value
is

2 n-1 n
P 1-d 1-d 1-d 1-d
= C [ ] +C [ ] +…+ C [ ] +(1+C) [ ] (**).
100 1+i 1+i 1+i 1+i

The term in square brackets can be approximated:

1-d 1+d 1-d2 1


× = ≈
1+i 1+d 1+i+d+i×d 1+i+d
If we set y = i + d, we arrive at the conventional expression:

P C C C 1+C
= + +…+ + .
100 (1+y) (1+y) 2
(1+y)n-1 (1+y)n

Thus, the yield to maturity is approximately the risk free rate plus the annual default probability.
To put it another way, the difference between the yield to maturity and the risk-free rate is
approximately the default probability assuming zero recoveries.

We are interested in knowing how much the bond price changes when the nominal risk-free rate
changes—that will define interest rate risk. Taking the derivative of (**) with respect to i, we
arrive at:
∆P 1
| | (1+i)= Macaulay Duration = Modified Duration × (1+ y)
∆i P

If we wanted to be pedantic, we would define Modified Duration as Macaulay Duration / (1+i)


rather than Macaulay Duration / (1+y). But that would only change the outcome by a factor of
(1+i)/(1+y) and in the semiannual case by even less: (1+i/2)/(1+y/2). This would add complexity
to the duration calculation in exchange for a phony precision; flat yield curve, constant
probability of default and so on. As long as we’re approximating, we can interpret the standard

26
modified duration as the percentage change in a bond (or a portfolio) with respect to the risk-free
interest rate, holding the spread constant.

Thus, we can conceive of duration as the sensitivity of the bond’s market value to the risk-free
interest rate, i.e., the market risk that is common to all bonds. So duration asks, holding the credit
spread, d, constant, how much would the price change when r or π changes?

We didn’t need the assumption that recoveries are zero. We could assume d = expected loss =
default probability × loss given default and the proof goes through as above.

27

You might also like