A Practical Guide to

Interest Rate Curve Building Validations

(w/ Excel Replica of Bloomberg Libor @ GitHub)

Xianwen Zhou
Vilen Abramov

February 5, 2019

Contents
1 Which Curves? 2

2 Validation Process 5

3 Instrument Selection 7
3.1 Interest Rate Instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Fixed Income Instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

4 Bootstrapping 8
4.1 Exact Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.2 Smoothing Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.3 Excel Replica of Bloomberg Libor . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.4 Prioritizing Market Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

5 Market Developments 11
5.1 OIS Discounting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.2 Basis Spreads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
5.3 CME and LCH spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

6 Interpolation and Extrapolation 13

6.1 Linear Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
6.2 Non-Linear Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
6.3 Extrapolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
6.4 Reconciliation of Interpolation Terminology Among Vendors . . . . . . . . . . . 14

7 Desired Curve Characteristics 16

7.1 Positivity and Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
7.2 Curve Smoothness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.3 Curve Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.4 Localness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

Electronic copy available at: https://ssrn.com/abstract=3329563

8 Common Validation Tests 19
8.1 Round-trip Consistency Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
8.2 Benchmarking Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
8.3 Sensitivity Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

9 Conventions and Maths 22

10 Outlook 23

11 Appendix A: Swap Manager 25

12 Appendix B: Bloomberg Tickers for Bootstrapping Instruments 26

13 Appendix C: Bloomberg Tickers for Pre-Bootstrapped Curves 29

Abstract
Curve building is an information extraction process defined by a set of bootstrapping in-
struments B, bootstrapping procedures P, interpolation technique I, and extrapolation
technique E. The bootstrapping procedures allow one to deduce/bootstrap information
about risk factors (such as future Libor rates) that can be used to derive a fair value of a
financial instrument with payoff function dependent on these risk factors (such as interest
rate swap). The bootstrapping procedures deduce information about either expected val-
ues of the risk factors (such as forward rate curve) or their distributions (such as volatility
surface). The curve building process involves a number of steps that include instrument
selection, market data prioritization, bootstrapping, interpolation, and extrapolation. The
paper provides information about market conventions, common curve building issues, val-
idation techniques, standard tests, and validation criteria. We will focus on the bootstrap-
ping of the expected values of interest rates (such as future forward rates). Some of the
validation techniques described in this paper (such as round-trip testing) can be extended
onto other markets and other types of the risk factors (including volatility surfaces). The
paper has a reference to the published Excel spreadsheet that demonstrates how to repli-
cate Bloomberg Libor curve. The spreadsheet can be downloaded from GitHub repository.

1 Which Curves?
Let’s start our discussion with interest rate curves. Interest rate curves reflect time value of
money concept. Time value of money needs to be recorded in a way so that different market
participants can have a consensus. Interest rate curves are built for this purpose. Market par-
ticipants express their views of rates (such as forward rates, discount factors, etc.) via market
quotes of different financial instruments (such as forward rate agreements, Eurodollar futures,
etc.). Hence, curve building can be viewed as a process of extracting market information.

Different market segments view the time value of money differently. Hence, this information
can be recorded using different types of curves. The curve families are usually classified by
the underlying index (e.g., LIBOR, Prime, SIFMA, etc.) and rate type (e.g., zero, forward,
discount, etc.). Since different forms of curves are conveying the same information (time value
of money), they can be converted into each other under certain conventions.

Interest rates curves are not directly observable in the markets, they are usually built (boot-
strapped) by utilizing a discrete set of market quotes of rate locking instruments. Curve

Electronic copy available at: https://ssrn.com/abstract=3329563

 1.8

3mL Forward Rate

1.7
3mL Zero Rate

1.6
1.5440 %
Rate, %

1.5

1.4
1.4062 %

1.3

3 Month 3 Month 3 Month 3 Month 3 Month

1.2
T/N 1m 2m 3m 4m 5m 6m 7m 8m 9m 10m 11m 1y 1y1m 1y2m 1y3m 1y4m 1y5m 1y6m 1y7m 1y8m 1y9m 1y10m 1y11m

Term

Figure 1: Comparison of the 3-month LIBOR Zero and Forward Rates

building process often needs to supplement the missing information via some sort of interpo-
lation assumptions. This topic will be discussed in more details later. Let’s take a step back
and review different curve types:

• Discount Factor Curve: representing current price of zero coupon bond for that curve
index with notional of $1

• Zero Rate Curve: the one rate to summarize the holding period return by assuming
money is invested in the curve index and rolled based on index term until the end of
holding period

• Forward Rate Curve: current expectation of the future index rate starting from certain
period from now

• Swap Rate Curve: the fixed rate to equate the series of floating rate payments indexed
to the floating rate for the life of the contract

It should be noted that rate indexes are always associated with time, i.e. tenors. The in-
dex rate tenor and the curve tenors (or nodes) are different. The index rate tenor (such as
3-month LIBOR, 1-month LIBOR, etc.) reflects the liquidity and credit risk embedded in the
underlying index, it is fixed for any given index. The curve tenors (or nodes) can vary, one
can build a more granular or less granular curves by adding or removing the bootstrapping
instruments. These nodes should be interpreted as either the total investment period or the
time between current and future effective dates. Let’s consider 3-month LIBOR (3mL) curve
as an example: based on Figure 1, current market implied 1-year 3-month LIBOR forward
rate is 1.544% and 1-year 3-month LIBOR spot (zero) rate is 1.4062%.

The market implied information can be interpreted in the following manner.

• The expected annualized rate to be offered on a 3-month term deposit starting 1 year
from today is 1.544%. We will denote this rate as f3mLibor (0; 1, 1.25).

Electronic copy available at: https://ssrn.com/abstract=3329563

 • Average annualized funding cost/investment yield by rolling short term 3 month time
deposit contract for 1 year starting from today is 1.4062%. We will denote this rate as
r3mLibor (0; 0, 1).
Generally speaking, a curve at time t can be viewed as an array of a form {r(t; t i , Ti }ni=1
under market convention C (market conventions include zero rate, forward rate, etc.) with
[ti , Ti ] representing investment periods. Spot or zero rates have starting points t i s fixed at the
current time t = 0 (ri = r(0; 0, Ti )), while forward rates have fixed investment period length
δ = Ti − ti (f (0; ti , ti+1 ) = r(0; ti , ti + δ)).

As stated above, different curves can be converted into each other. Let’s demonstrate how
zero rates can be converted to discount factors. Given zero rate r t , the discount zero-coupon
bond price Z(0, t) (i.e. discount factors Z(0, t)) can be calculated as follows under continuous
compounding:

Z(0, t) = e−rt t
The zero rate rt and forward rate (f (0; ti , ti+1 )) are connected in the following manner.
rti+1 ti+1 − rti ti
f (0; ti , ti+1 ) =
ti+1 − ti
Market conventions begin to play a role as the determination of (ti+1 − ti ) (measured in years)
depends on the way one deals with holidays and day counts. The day counting rules vary
by jurisdiction and instrument type. Normally, 30/360 or Actual methods are used for day
counting. This calculation should take into consideration holiday calendars. In Bloomberg
terminal, the swap manager function (SWPM <Go>) provides market convention details for
different curve indexes (please, refer to Figure 7 in Appendix A).

So far, we have talked about interest rates market. The same logic can be applied to other
markets where forward rates play an important role. This includes: commodity, foreign ex-
change, and equity markets. On the contrary, credit markets deal with credit spreads or par
spreads that reflect expected default rate. One can use different instruments to imply mar-
ket participants’ expectations of the forward rates. Some instruments (such as interest rate
swaps, commodity swaps, foreign exchange swaps) can be priced off of these market-implied
forward rates. However, option-embedded instruments require information about distribution
of forward rates rather than just their expected values. The distributional information is
typically compressed into the implied volatility parameter that implicitly assumes log-normal
distribution. As stated earlier, in this paper we will focus on the curve building associated
with expected rates rather than distributions.

Summarizing the information outlined above, in order to specify a curve one needs to specify
the following items.
• the reference rate (LIBOR, SIFMA, etc,)
• the term associated with the reference rate (LIBOR 1M, LIBOR 3M, etc.)
• the “as of time” t
• investment periods [ti , Ti ]
• type of rates and/or market conventions C

Electronic copy available at: https://ssrn.com/abstract=3329563

 Identify All Curves Pre-built Curves (e.g., BBG)

Curves Boot- Check the bootstrapped

strapped Internally curves for market consistency

Replicate interpolation and

extrapolation methodology

Identify bootstrap-
Identify “parent”
ping instruments and
and “child” curves
prioritization rules

Validate consistency of
the market conventions
for the spread curves

Figure 2: Curve Building Validation Process

2 Validation Process
Let’s describe a curve building validation process outlined in Figure 2. The first step is to
identify all the curves. As described in the first section, there is a number of curve attributes
that need to be identified. There will be a number of curves corresponding to different markets,
products, underlying entities, etc. Once the curves are identified, they can be grouped into
two buckets - curves that are built/bootstrapped internally and curves that are pulled from a
third party (such as Bloomberg, Reuters, etc.). It is very common for market risk management
teams to pull pre-bootstrapped curves from a third party. In Appendix C we give examples
of Bloomberg tickers corresponding to different pre-bootstrapped curves. Now we will focus
on the validation procedures associated with internally built curves. It is the best practice,
to apply the same validation standards to the pre-bootstrapped curves. However, there will
be certain limitations that may limit validator’s ability to fully replicate third party’s curve
building process.

One needs to start with the identification of the bootstrapping set B and bootstrapping pro-
cedures P. What are they? The bootstrapping set is a set of market-traded instruments.
The payoff function of these instruments should depend on the index rate tenor (such as 3-
month LIBOR). In order to price these instruments one needs to estimate future values of
the corresponding index rate tenor. A bootstrapping procedure allows one to invert the pric-
ing process to derive assumptions about the future values of index rate tenor from market
prices of the instruments. This inversion problem is typically under-determined and requires
additional assumptions (in the form of interpolation method) about implied rates that are
not dictated/observed by/in the market. This interpolation method should be viewed as part
of the bootstrapping procedures P. It is important to check the bootstrapping instruments

Electronic copy available at: https://ssrn.com/abstract=3329563

 ×105
2.5

1.5
Volume

0.5

0
FEB 19 SEP 20 MAR 23 SEP 25 DEC 28
Futures Month

Figure 3: EDF Trading Volume on 1/29/2019

Source: CME Group (reporting available here)

and index rate tenor for consistency. For example, one can’t use Eurodollar futures (EDF)
to build 1-month LIBOR curve since EDFs are linked to the 3-month LIBOR rates. How-
ever, one could still utilize them by incorporating 3s1s (3-month vs 1-month) basis spread.
Furthermore, different instruments may imply contradictory rates. This can be caused by
the market illiquidity. In this case, one need to define prioritization rules that would allow
a bootstrapping mechanism to overcome market anomaly. These prioritization rules are part
of the bootstrapping procedures as well. It is important to keep in mind that liquidity plays
an important role when defining a bootstrapping basket. It is well known that EDF marks
tend to be less liquid for longer tenors. The trading volume corresponding to different EDFs
is displayed in Figure 3. The trading volume beyond 4 year tenors appear to be very minimal.
It is interesting to note that there are some liquidity gaps even on the short end.

Now, when all the inputs and outputs are clearly defined, it is time to discuss the validation
testing activities. In the testing phase, it is important to replicate the bootstrapping process.
Different bootstrapping procedures have different pros and cons that we will discuss later. It
is important to confirm the curve’s consistency with the market. In other words, a theoretical
price of any instrument from the bootstrapping set should match the market price when priced
off of the bootstrapped curve. This is a so-called round trip test.

The bootstrapped curve gives rates corresponding to a discrete set of time points. When pric-
ing instruments that do not belong to the bootstrapping set (such as over-the-counter trades),
one needs to use some sort of interpolation and extrapolation techniques. A model validator
needs to replicate and assess these techniques.

In addition to the standard “parent” curves (LIBOR, EURIBOR, Treasury, etc.), model de-
velopers build “child” curves. The “child” curves are built as a spread over “parent” curve.
Here are some examples of the “child” curves. The tenor basis curves (1-month, 6-month,
and 12-month tenors curves), product basis curves (prime, commercial paper, bma), etc. One

Electronic copy available at: https://ssrn.com/abstract=3329563

needs to make sure that basis spreads that are used to build these “child” curves are being
incorporated in accordance with the market conventions. All these topics will be discussed in
more details in the following sections.

3 Instrument Selection
3.1 Interest Rate Instruments
Curve bootstrapping starts with market inputs. The bootstrapping instruments are a discrete
set of rate locking instruments traded in the market. For validation purpose, it’s critical to
make sure that the curve bootstrapping inputs are pulled and interpreted correctly.

A validator needs to understand the market conventions of the bootstrapping instruments.

For example, for LIBOR curves, Eurodollar futures have moving term points while FRAs can
provide fixed term points. Another example is, for basis curves, some basis spreads are quoted
as an add-on spread curve (e.g., LIBOR 1M, LIBOR 6M, LIBOR 12M) while other basis
spreads are quoted as a ratio of the base curve(e.g., BMA).

Another important aspect is the liquidity of the market instruments. As different rate locking
instruments have different liquidity profiles, bootstrapping the curve beyond the liquid term
point requires some data manipulation. A good reference to the liquidity profile of common
bootstrapping instruments can be found in the following research paper [11].

A more recent example on OIS curve bootstrapping can be found in the Bloomberg white
paper [12]. The paper points out some of the challenges concerning instruments availability
for the long tenors of the OIS curve. “Currently for USD, OIS rates are not widely available
in the marketplace beyond the 10-year maturity. In order to support OIS/DC stripping, the
OIS curve is extended beyond the 10-year maturity by harnessing USD Fed Funds (FF) basis
swap quotes, that are available to 30-year maturity, since both OIS and FF basis are stripped
to project forwards of the FF Effective Rate, FEDL01, and sensible inferences can be made
from each other in an arbitrage-free environment. Two methods have been developed to imply
OIS rates from FF basis swaps.” Hence, when validating the OIS curve, a validator should
verify how the system extends the OIS raw input curve beyond 10 year-maturity and replicate
the process by taking into consideration conventional difference between the OIS and LIBOR
curves (such as day count, compounding, averaging method, etc.).

Common bootstrapping instruments and their corresponding Bloomberg tickers have been
provided in Appendix A. In order to have a fixed term points curves, FRA contracts rather
than Eurodollar futures are selected. The four curves captured in Appendix A are standalone
discounting curve (OIS), base index curve (LIBOR 3M), add-on basis curve (LIBOR 1M), and
ratio basis curve (BMA). The curves can be located in Bloomberg terminal using the following
commands.

• OIS curve: ICVS 42 <Go>

• LIROB 3M Curve: ICVS 23 <Go>

• LIBOR 1M Curve: ICVS 50 <Go>

Electronic copy available at: https://ssrn.com/abstract=3329563

 • BMA curve: ICVS 357 <Go>

3.2 Fixed Income Instruments

The most straightforward bootstrapping process is the Treasury curve bootstrapping. Trea-
sury curve serves as the base curve in the fixed income world. The other fixed income in-
struments will trade based on the spread over treasury (e.g. muni curves, agency curves,
etc.). Taking on-the-run treasury instruments, we can derive the daily par yield published by
the Treasury department (Table 1). Since treasury bonds are coupon bearing instruments,
directly using the yield for discounting can’t generate arbitrage free results. And in order to
obtain the zero rate, one needs to perform the par yield to zero rate conversion.

Table 1: Daily Constant Maturity Treasury Quote

Source: The Treasury’s Website (available here)
Date 1M 2M 3M 6M 1Y 2Y 3Y 5Y 7Y 10Y 20Y 30Y
1/2/2019 2.4 2.4 2.42 2.51 2.6 2.5 2.47 2.49 2.56 2.66 2.83 2.97
1/3/2019 2.42 2.42 2.41 2.47 2.5 2.39 2.35 2.37 2.44 2.56 2.75 2.92
1/4/2019 2.4 2.42 2.42 2.51 2.57 2.5 2.47 2.49 2.56 2.67 2.83 2.98
1/7/2019 2.42 2.42 2.45 2.54 2.58 2.53 2.51 2.53 2.6 2.7 2.86 2.99
1/8/2019 2.4 2.42 2.46 2.54 2.6 2.58 2.57 2.58 2.63 2.73 2.88 3

The bootstrapping process gets more complicated when one deals with corporate bonds. There
may not be enough name specific bonds in the market in order to build a name specific yield
curve. Hence, modelers use bond information from similar (in terms of credit rating and in-
dustry sector) companies. This may lead to conflicting information since bonds from different
companies may imply different information. In this case, the non-exact bootstrapping meth-
ods are used. We will discuss these methods in the Bootstrapping section.

4 Bootstrapping
Curve building is both an art and science because one needs to make educated guesses about
rates not directly observed in the market. Hence, there is no one correct curve building
method. One can talk about the quality of a bootstrapping method with respect to a number
of desired characteristics (e.g., smoothness). In the following subsections, we will discuss a
number of bootstrapping topics. Please, refer to [5] for a thorough review of the bootstrapping
methods. We have also published an Excel spreadsheet [6] that demonstrates how to replicate
Bloomberg LIBOR curve. The spreadsheet can be downloaded from GitHub repository here.

From a very high level, one can break bootstrapping methods into two major categories -
exact methods and non-exact methods. Exact methods allow modeler to replicate market ex-
actly, but require additional assumptions regarding rates that would need to be interpolated
using some sort of parametric approach (e.g., cubic spline interpolation). On the other hand,
non-exact methods start with a functional form that represent a curve. Consequently, this
functional form can be tied to the market data via calibration. In this case, a modeler is not
guaranteed to replicate the market exactly. Figure 4 outlines dependencies between exact (E)
and non-exact (NE) methods with parametric (P) and non-parametric (NP) representations

Electronic copy available at: https://ssrn.com/abstract=3329563

 E P

NE NP

Figure 4: Bootstrapping Approaches

of the curve.

4.1 Exact Method

In mathematical terms, exact bootstrapping can be interpreted as solving a system of equations
[5].

Cd = p,
where C is an instruments’ cash flow matrix, d is a discount factor vector, and p is a price
vector. This system of equations is usually under-determined and, therefore, allows for in-
finitely many solutions. In other words, the number of rates that we need to imply is larger
than the number of liquid instruments that are quoted in the market. A typical way to deal
with this problem, is to complement a system of equations with artificial/interpolated ones.
These artificial equations represent our belief (i.e. interpolation assumption) about rates not
observable in the markets.

In our Excel spreadsheet [6], there are three instrument types in the calibration basket - cash
deposits, Eurodollar futures, and interest rate swaps. The cells corresponding to different
instrument types are marked in different colors. The data is pulled from Bloomberg as of
June 4th of 2018. The EDF dates were pulled from CBOE page here.

4.2 Smoothing Methods

In mathematical terms, non-exact bootstrapping can be interpreted as optimization.
N
X
min (yi − f (Ti ))2
α1 ,...,αn
i=1

Notice, that there can be variations of this general approach based on the object of modeling
and object of optimization. In the formula above, both the object of modeling and the object of
optimization is yield. In the famous Nelson-Siegel approach, the object of modeling is forward
rate rather than yield. In other words, function f (u) represents instantenouos forward rate.
The yield in this case can be calculated as follows.

Electronic copy available at: https://ssrn.com/abstract=3329563

 Ti
1
Z
yi = f (u)du
Ti 0
The optimization problem can be rewritten as follows.
N 
X Z Ti 2
min T i yi − f (u)du
α1 ,...,αn 0
i=1

In the optimization problem above, the functional form of the curve is pre-defined. But how
good that assumption is? What functional form is the “best?” The answer to this question
is given by the Lorimier’s theorem [7]. In this theorem, the author solved the following
optimization problem.
Z T N 
X Z Ti 2
min (f 0 (u))2 du + α T i yi − f (u)du ,
f ∈H 0 0
i=1

where H is a Hilbert space of absolutely continuous function on interval [0, T ]. The functional
above consists of two terms - the first term represents the smoothness while the second one
represents yield accuracy. The parameter α reflects the importance of the accuracy term. The
solution to this problem is given by quadratic splines.

4.3 Excel Replica of Bloomberg Libor

In this section we will describe our Excel replication of Bloomberg Libor curve [6]. The
spreadsheet can be downloaded from GitHub. There are six tabs.

• BBG Replica

• BBG Data Table

• BBG Input

• BBG Input Tickers

• BBG Output

• BBG Output Tickers

On the “BBG Input” tab, one can see Bloomberg screen shots of the input data. Input tab
screen shot has three instrument groups - Cash Rates, Contiguous Futures, and Swap Rates.
Instruments that are used for bootstrapping are highlighted in orange. The Zero Rates Chart
at the bottom is the output curve. This curve is broken into three segments - short end,
middle, and long end. Notice, that Contiguous Futures quotes contain prices and convexity
adjustments that can be converted into forward rates. The day counting conventions, maturity
dates, and terms are also captured on this tab. We have recorded input data on the “BBG
Data Table” tab. The data from this tab will be referenced on the main calculation tab, the
“BBG Replica” tab.

On the “BBG Output” tab, one can see Bloomberg output - zero rates and discount factors.
In our calculations, we will focus on the discount factors. Our calculated discount factors are

10

Electronic copy available at: https://ssrn.com/abstract=3329563

compared to these Bloomberg output rates on the “BBG Replica” tab. Bloomberg tickers
corresponding to the input and output rates are captured on the “BBT Input Tickers” and
“BBG Output Tickers” tabs.

Finally, the discount factors are calculated on the “BBG Replica” tab. Rows corresponding
to different instrument types are filled with different colors - red for cash, blue for EDF, and
white for swap. Instrument specific discount factor formulas are captured at the bottom of
this tab. The calculated discount factors are compared to the Bloomberg output in the last
two columns. There is also a check on the swap value in rows 22-23, columns E-G. The value
of the interest rate swap with fixed rate set to the spot swap rate should be 0.

4.4 Prioritizing Market Data

Due to the market imperfections, it is natural to see cases when the same or similar rates
implied from different market instruments do not agree with each other. For example, the
front end LIBOR forward rate can be implied from either Euro-dollar Futures(EDF) con-
tract or Forward Rate Agreement (FRA). Due to market supply and demand and the trading
mechanism difference (exchange versus OTC), these two rate locking instruments may indicate
different rates for the same underlying time window. Basically, the liquidity profile difference
between different instruments leads to the transactional cost differences.

The Eurodollar contract is much more liquid for the front end LIBOR curves, thus the hedging
of interest rate exposure by Eurodollar is more cost efficient. However, market risk system
needs to generate scenarios which is ideally free from interpolation noise. Generating scenarios
using FRA contracts will be preferable as the historical market movements are consistent due
to the fixed term points.

The underlying position will also play a role in the market data selection consideration. The
exposure concentration of the underlying position will lead to the model user to prefer certain
term points to be smooth or at least not subject to interpolation noise. This consideration
will justify the addition of instruments pointing to a certain tenor in the bootstrapping set.

5 Market Developments
5.1 OIS Discounting
OIS discounting is a new market trend after the 2008 financial crisis. Prior to the crisis, one
LIBOR curve(3M) has been used for projecting future floating rate reset and discounting the
future value into present, implying market agrees that the LIBOR curve is a good proxy for
risk free rate.

During the financial crisis, however, the LIBOR OIS spread spiked to historical high level,
forcing the market to rethink about a better proxy for risk free rate, at least a better rate
index so that it’s credit risk free under market stress. OIS becomes a market standard for
derivatives pricing as it can better capture the risk free rate change without worrying the

11

Electronic copy available at: https://ssrn.com/abstract=3329563

credit component.

A very good example can be found in the Bloomberg white paper section-Extension of USD
OIS Curves [8].

5.2 Basis Spreads

During the financial crisis, previous convergence of market rates that refer to the same time
interval does not hold any more. LIBOR rates quotes for different terms cannot neglect the
liquidity and credit risk premium demanded by market participants. Two options for the mar-
ket to adapt to the new regime: develop model which can incorporate the credit and liquidity
premium under the original framework or begin the market segmentation so that different
terms of the same curve (LIBOR) will be treated as different risk factors.

Obviously, after crisis market embraces for the second option, basis spreads become a new
critical market risk factor in bank’s market risk system.

From model validation perspective, the market convention for how the basis spreads are ac-
tually used needs to be examined. For add-on basis curves, the quoted basis spreads are the
difference added to the reference floating index so that the two floating legs can be priced the
same. And the spread is usually added to the shorter term leg:

For example, for the LIBOR 1M v.s 3M basis swap, the two legs are:
• 3M leg: reset and pay quarterly on 3M LIBOR rate

• 1M leg: reset monthly on 1-month LIBOR compounded quarterly + basis spread

While for LIBOR 3M v.s. 6M basis spread, the two legs are:
• 3M leg: reset and pay quarterly on 3M LIBOR rate compounded to 6M + basis spread

• 6M leg: reset and pay semi-annually on 6M LIBOR rate

5.3 CME and LCH spread

Around mid-May 2015, bank swap books suffered as CME-LCH basis explodes: the relative
cost of clearing interest rate swaps at CME versus LCH experienced a 13-fold increase. The
basis has been quoted around 0.15 basis point, but the basis for 10-year US dollar swap surged
to 2 bps peak. So for a swap trade with notional of 1 billion, a difference over the life of the
trade will amount to 2 million.CME has released some official comments as:

“Recently, a pricing differential, or basis, has developed in the interest rate swap market. The
most common explanation given for the cause of this pricing differential is a structural imbal-
ance between the supply and demand for pay-fixed and receive-fixed swaps in the marketplace
as a result of the 2013 clearing mandate for interest rate swaps. The issuers of corporate debt
are exempt end-users who hold their receive-fixed swaps bilaterally outside of clearing, but the
buy side parties who purchase this debt must clear their pay-fixed swaps. With swap dealers
acting as the counterparty for both sides of these trades, it is believed that this has caused a

12

Electronic copy available at: https://ssrn.com/abstract=3329563

large accumulation of client pay-fixed swaps in clearing, where the dealer receives fixed. And
CME is trying to build the CME specific swap curve for dealers to better estimate the clearing
costs.”

The basis between exchanges has some implication to the market participants. A business
entity should manage this basis risk by either dealing with the same exchange for all the
derivatives trades or hedging the basis risk as in the case for Libor 1M vs 3M.

6 Interpolation and Extrapolation

6.1 Linear Interpolation
Linear interpolations assume a linear relationship between the unknown rate(discount factor)
and the known rates(discount factor). As the same rate can be represented in different forms
(e.g., spot, forward, discount factor, etc.), the linear relationship can also be built on each
different form for any t in (ti−1 , ti ).

Table 2: Linear Interpolation Methods

Linear on the zero rate r(t) = tt−t i−1
i −ti−1
r(ti ) + ti −t
ti −ti−1 r(ti−1 )
Linear on the log zero rate ln(r(t)) = tt−t i−1
i −ti−1
ln(r(ti )) + ti −t ti −t
i−1
ln(r(ti−1 ))
t−ti−1 ti −t
Linear on the discount factor Z(t) = ti −ti−1 Z(ti ) + ti −ti−1 Z(ti−1 )
Linear on the log of discount factor r(t)t = tt−t i−1
i −ti−1
ti −t
r(ti )ti + ti −t i−1
r(ti−1 )ti−1
t−ti−1 ti −t
Linear on the swap rate s(t) = ti −ti−1 s(ti ) + ti −ti−1 s(ti−1 )
Linear on the forward rate f (t) = tt−t i−1
i −ti−1
ti −t
f (ti ) + ti −t i−1
f (ti−1 )

Special attention should be paid to the method- linear on the log of discount factor, which is
a benchmark method for yield curve bootstrapping process. The method is easy to implement
and very stable. The log of discount factor equals to −tr(t). The interpolation method is
essentially interpolating in the tr(t) space. A noticeable feature of this interpolation method
is that all instantaneous forward rates are positive.

A common problem with the linear interpolation is the continuity of the forward rates. Even
the benchmark method-linear on log of discount factor has jumps in the transition points. Lin-
ear on swap rates is not common, since the method makes assumption on observable market
inputs. Linear on forward rate is known to be undesirable either, on one hand, the continuity
of forward rates will be addressed; on the other hand, the method will introduce zig-zag be-
havioral [3].

6.2 Non-Linear Interpolation

In this section we will consider one of the most common interpolation techniques - cubic
splines. Cubic splines can be defined as follows.

r(t) = ai + bi (t − ti ) + ci (t − ti )2 + di (t − ti )3

13

Electronic copy available at: https://ssrn.com/abstract=3329563

Spline is supposed to alleviate the problem of the zig-zag behavior of the single polynomial
fitting. Cubic splines are a set of cubic functions trying to connect the discrete nodes cor-
responding to traded market instruments. Several desired properties are pursued by the
practitioners.

• The splines pass all the nodes derived from the traded market instruments

• The first order derivative is continuous at the transition points

• The second order derivative is continuous at the transition points

Continuity can be defined via right and left limits that should give the same value. For cubic
splines, the right and left derivatives can be expressed in the following manner.

∂+ r(ti ) = bi + 2ci (ti − ti ) + 3di (ti − ti )2 = bi

∂− r(ti ) = bi−1 + 2ci−1 (ti − ti−1 ) + 3di−1 (ti − ti−1 )2

In order to ensure continuity/smoothness, one needs to make both right and left derivatives
equal. These ”continuity/smothness” equations can be constructed for all transition points.
In order to ensure continuity in the first place, one needs to match the left and right limits at
each transition point. The left and right limits can be expressed as follows.

lim r(t) = ai + bi (ti − ti ) + ci (ti − ti )2 + di (ti − ti )3 = ai

t→t+
i

lim r(t) = ai−1 + bi−1 (ti − ti−1 ) + ci−1 (ti − ti−1 )2 + di−1 (ti − ti−1 )3
t→t−
i

One can consider additional properties when interpolating. Following [3], monotone convexity
is a highly desirable property in order to build a “good” curve. Hagan demonstrates that
this method is the best choice among various yield curve bootstrapping methods in terms of
forward rates positivity, smoothness, stability, localness, and hedge’s localness. Please, refer
to Hagan’s paper [3] for complete details.

6.3 Extrapolation
The rates need to be extrapolated (rather than interpolated) for instruments with maturity
dates going beyond the last available point on the discount curve. For example, FinCAD
function “aaDFCurve Extend” allows one to obtain extended discount rates. The curve is
extended by assuming a constant par swap rates for the extended dates.

6.4 Reconciliation of Interpolation Terminology Among Vendors

Let’s take a look at how different vendors define the bootstrapping and interpolation methods.

FinCAD defines bootstrapping as follows [4]:

The process of building a discount factor curve from market quotes is known as bootstrapping.

14

Electronic copy available at: https://ssrn.com/abstract=3329563

Bloomberg provides the following definition [8] of bootstrapping:

The process of building of the IR curve a.k.a. curve stripping is a process of creating a curve
object which would correctly price a set of N given instruments, e.g. produces correct discount
factors and forward rates used in these instruments.

These definitions appear to be very similar, however, there are some differences in the im-
plementation. In FinCAD, the curve building process can be broken down into two steps -
bootstrapping and interpolation.

As discussed in the Bootstrapping section, the bootstrapping problem is underdetermined and

requires additional assumptions about rates. Hagan stated in [3], “the interpolation method
is intimately connected to the bootstrap, as the bootstrap proceeds with incomplete infor-
mation. This information is “completed” using the interpolation scheme.” FinCAD offers
multiple methods to deal with this step - the linear swap rate (LSR), linear forward rate (LFR
and LFR2), and quadratic forward rate (QFR). These methods are, in essence, dealing with
interpolation required for converting prices of the bootstrapping instruments into rates. Once
this step is completed, one can start using these rates to price other instruments. However,
the other instruments may require information corresponding to the time points that were not
captured in the bootstrapping instruments. Hence, additional interpolation is required. Fin-
CAD offers a number of interpolation methods for the final rate output (i.e. discount factors)
- the linear discount factors, linear spot rates, cubic spline discount factor, and exponential
discount factors. The two interpolations discussed above can be named as bootstrapping in-
terpolation and pricing interpolation.

It is important to note that interpolation objects may vary as well. In FinCAD, the bootstrap-
ping interpolation is performed on the forward rates and pricing interpolation is performed
on the discount factors. While Bloomberg allows the user to choose 4 bootstrapping interpo-
lation methods - piecewise linear performed on simple-compounded zero rate and piecewise
quadratic, step-function, and piecewise linear performed on continuously-compounded forward
rate.

It should be noted that the so-called “raw method”, which is the constant forward rate method
in FinCAD and step-function forward method in Bloomberg actually mean the same thing.
Hence, different vendors can name the same method differently, a validator can easily form
nature benchmarks based on the understanding of this fact. Secondly, interpolation methods
allow for alternative definitions in terms of different interpolation objects. For example, linear
interpolation performed on the product of zero rate and time is equivalent to the exponential
interpolation performed on the discount factor. In order to facilitate the benchmarking of
different vendors’ bootstrapping methodologies, Table 3 groups different bootstrapping and
pricing interpolation methods based on the vendor, interpolation object, and linearity of the
interpolation method.
It is worth mentioning that a model validator should assess whether bootstrapping interpola-
tion and pricing interpolation “agree” with each other. Combination of certain methods may
lead to the curve anomalies. This topic has been discussed in [4]. The results and recommen-
dations of the FinCAD analysis can be found in Figure 4. In addition to the interpolation
assumptions, one may use some sort of smoothing techniques. In FinCAD, this can be achieved
by applying post-smoothing. In Bloomberg, this can be achieved by using a global method

15

Electronic copy available at: https://ssrn.com/abstract=3329563

 Table 3: Reconciliation of the Interpolation Terminology
Linear Non-linear
Zero Rate FinCAD (linear spot rate)
Bloomberg (piecewise linear simple)
Bloomberg (piecewise linear continuous)
Forward FinCAD (linear forward) Bloomberg (quadratic forward)
FinCAD (enhanced linear forward) FinCAD (quadratic forward)
Bloomberg (step function forward)
Discount Factor FinCAD (Linear discount factor) FinCAD (exponential discount factor)
FinCAD (cubic spline discount factor)
Swap FinCAD (linear swap)

rather than a bootstrap method.

Table 4: FinCAD Recommendations Concerning Bootstrapping and Pricing Interpolations

Bootstrapping Interpolation Post-smoothing Required? Recommended Pricing Interpolation
Linear Swap Rates (LSF) Yes Any
Constant Forward Rates (CFR) No Exponential
Original Linear Forward Rates (LFR) Yes Any
Enhanced Linear Forward Rates (LFR2) Yes Any
Quadratic Forward Rates (QFR) No Cubic Spline

Summarizing the topics discussed in this section, a model validator needs to identify a num-
ber of curve building parameters in order to compare different interpolation methods. These
parameters include bootstrapping instruments set, bootstrapping interpolation method along
with interpolation object rates type, pricing interpolation method along with interpolation
object type, and equilibrium assumptions.

7 Desired Curve Characteristics

The desired curve characteristics actually depend on the purpose or use of the bootstrapped
curve.

For example, on the front desk, it is quite likely that the best curve bootstrapping method is
subject to many different conditions. As noted by Amir Sadr,

“When selecting a method, one has to decide on the right tradeoff between smoothness and
the sensibility of the prescribed hedges. Because each trader may have a different opinion
on where the right tradeoff is, there has yet not emerged a universally accepted curve build
method.”

While for risk management purpose, the liquidity of the instruments may not be the most
critical consideration, the curve which offers less interpolation noise and easy scenario gener-
ation will be more favored.

Though different users have different preferences to the curve characteristics, some common
desirable properties can be identified, such as:

16

Electronic copy available at: https://ssrn.com/abstract=3329563

 • Forward rates are positive

• Forward rates are continuous

• Forward rates are smooth

A validator is concerned about how the curve looks like now based on current market inputs
and how the curve will look like given some shock to the inputs. The current curve is desired
to be positive, continuous, and smooth. The shocked curve is desired to have stability and
localness. No method possesses all the merits at once, one has to sacrifice some for the others.
For validation purposes, there is good reference [3] for the pros and cons associated with
various bootstrapping options. For the sake of completeness, we present Hagan’s summary
table here (Table 5). Depending on the users, the best curve is the one which can maximize
his/her utility function.

Table 5: Hagan’s Summary [3]

Yield curve type Fwd > 0? Smoothness Localness? Stablity? Hedge Localness?
Linear on discount No Not Continuous Excellent Excellent Very Good
Linear on rates No Not Continuous Excellent Excellent Very Good
Raw(linear on log of discount) Yes Not Continuous Excellent Excellent Very Good
Linear on the log of rates No Not Continuous Excellent Excellent Very Good
Piecewise linear forward No Continuous Poor Very Poor Very Poor
Quadratic No Continuous Poor Very Poor Very Poor
Natural Cubic No Smooth Poor Good Poor
Hermite/Bessel No Smooth Very Good Good Poor
Financial No Smooth Poor Good Poor
Quadratic Nature No Smooth Poor Good Poor
Hermite/Bessel on rt function No Smooth Very Good Good Poor
Monotone Piecewise Cubic No Continuous Very Good Good Good
Quartic No Smooth Poor Very Poor Very Poor
Monotone Convex(unamerliorated) Yes Continuous Very Good Good Good
Monotone Convex(amerliorated) Yes Continuous Good Good Good
Minimal No Continuous Poor Good Very Poor

7.1 Positivity and Continuity

The condition for positive forward rates is another way of saying discount curve should be
monotonically decreasing. Forward curve positivity can be guaranteed through the design
of interpolation method. For example, in a normal market(where yield curve are upwardly
sloping), the raw method can guarantee that instantaneous forward rate is always positive.
The positivity of forward rates is unique to financial problem as it is consistent with the time
value of money concept.

The positivity of forward curve is also a check point for scenario designs. For example, during
the stress testing process, the market curve will be shocked by some artificial market scenarios
to identify the portfolio’s vulnerability. One should keep in mind that in this scenario design
process, the shocked market should also preserve certain properties.

The condition for continuous forward rates helps to guarantee that no jumps happen at the
connecting points where bootstrapping instruments switch.

17

Electronic copy available at: https://ssrn.com/abstract=3329563

7.2 Curve Smoothness
Curve smoothness is a desired feature for curve building, however, a measure for the smooth-
ness is not a straightforward concept. One reference can be found in the research work by Ken
Adams, in which, he referred to the strain energy concept (g(x) represents the curve shape
function):
Z b 2
∂ 2 g(x)
dx
a ∂x2
The measure is trying to quantify the “bending” of the elastic spline, also the smaller this
measure of smoothness, the smoother the interpolated curve.

Curve smoothness is a step further from the curve continuity. Some conclusions drawn by
Adam:

For forward rate: “The interpolating function that guarantees the smoothest continuously
compounded instantaneous forward-rate curve is a quartic spline.”

For zero rate: “A financial cubic spline denotes a cubic spline that is constrained so that its
derivative at its right-hand end is zero, and its second derivative at the left-hand end is also
zero. No other interpolation function that is subject to the same constraints as the financial
cubic spline, and which fits the given data, is smoother than the financial cubic spline that
interpolates that data.”

7.3 Curve Stability

Per Hagan’s definition, the stability of the interpolation method is measured by “given a
change in one of the inputs, how much can the output interpolated curve be changed.” Hagan
developed a norm for both the yield curve(r) and forward curve(f).

 
 ∂r(t) 
kM (r)k = sup max  
t i ∂ri 
 
 ∂f (t) 
kM (f )k = sup max  
t i ∂fid 
However, Hagan also notes that the norms usually cannot be determined analytically but
estimated empirically. The practical approach for estimating the norm is to measure the max-
imum difference between original bootstrapped curves and any of the curves which arise when
any of the original bootstrapping nodes are changed by 1 basis point.

While in FinCAD, the stability is the way how curve looks. For example, the exponential in-
terpolation (constant forward rate) approach is deemed as a good example for curve stability.

7.4 Localness
By definition, some methods are global method and some methods are local method. The lo-
calness measures the sensitivity of the curve output to the input change. If we view the curve

18

Electronic copy available at: https://ssrn.com/abstract=3329563

stability as the magnitude of the output change, localness can be viewed as the direction of the
output change. So a curve which is stable and have small spill-over when input rate changes
should be preferred. The interval for yield curve values change is denoted as (t i−l , ti+u ).

The localness concept has also been expanded to hedging side. Per Hagan’s definition, if
assuming the admissible hedging instruments are exactly those used to bootstrap the yield
curve, it’s important to note whether most of the delta risk get assigned to the hedging
instruments that have maturities close to the given tenors. For the sake of completeness, we
present Hagan’s localness indices for various curve interpolation methods in Table 6.

Table 6: Hagan’s Localness Indices [3]

Interpolation Method l u
Linear on discount 1 1
Linear on rates 1 1
Raw(linear on log of discount) 1 1
Linear on the log of rates 1 1
Natural Cubic i−1 n−i
Hermite/Bessel 2 2
Financial i−1 n−i
Quadratic Nature i−1 n−i
Bessel on cap function 2 2
Monotone Piecewise Cubic 2 1
Quartic i−1 n−i
Monotone Convex(unamerliorated) 2 2
Monotone Convex(amerliorated) 3 3
Minimal i−1 n−i

8 Common Validation Tests

Validation tests for curve bootstrapping process provide the assurance that the curves coming
out of the bootstrapping process is reasonable in terms matching current market and predict-
ing future market. One needs to pay special attention to the order of the validation activities.
As one wants to make sure the building blocks are validated first.

For interest rates curves, it’s important to keep in mind the validation sequence should be
consistent with the curve building sequence. For example, under dual curve framework, the
OIS curve is usually built first as a standalone discount curve. Then, LIBOR 3M curve can be
built based on the LIBOR indexed instruments and the OIS discount curve. After OIS curve
and LIBOR 3M curve are ready, other spread curves such as: LIBOR 1M, BMA curve can be
built based on the previously bootstrapped OIS and LIBOR 3M curves.

This validation sequence will help identify any issue more effectively. The following sections
will demonstrate various kinds of tests applicable to the bootstrapped curves.

19

Electronic copy available at: https://ssrn.com/abstract=3329563

8.1 Round-trip Consistency Test
Round-trip consistency test refers to the process of using the bootstrapped curve to reprice
the exact bootstrapping instruments. Round-trip test confirms that the curve output is con-
sistent with the market bootstrapping instruments. As we know that bootstrapping input is
a set of discrete market instruments, when we mention market consistency achieved based on
round-trip results, we only refer to matching those instruments.

Round-trip result is neither a necessary or sufficient condition for a “good” curve. As if one
opts for exact method, round-trip results will guarantee the match but it may run into the
problems of over-fitting, or if one opts for non-exact method, round-trip results will not be
guaranteed.

Round-trip testing should be conducted in the same system where the curve is bootstrapped.
Taking Bloomberg as an example, SWPM function can be used to perform the round-trip
testing for the curve built under ICVS function. And one has to be aware what exactly the
round-trip test tries to tie back to the market.

For example, the following table summarizes sample instruments for several curve indices.

Table 7: Market Instruments for Round-trip Testing

Curve Index OIS LIBOR 3M LIBOR 1M BMA
Front Cash Point Cash Point Cash Points Cash Point
Mid and Long OIS Swaps EDF/FRA/Swaps Basis Swap Spreads Basis Swap Spreads

Taking LIBOR curve round-trip test as an example, the cash point, EDF, or swaps should be
priced the same as market quotes by the bootstrapped curve. Cash point or spot LIBOR should
match the forward rate 2 business day from today. Referring back to the concepts mentioned
in the first section, the spot LIBOR quote should match f3mLibor (0; 0 + 2D, 0 + 2D + 3M ). For
the middle portion, the Eurodollar contract adjusted by convexity should match the forward
rate terms which are the effective dates of the ED contracts. For the long end, the fixed rate of
the swap indexed on 3M LIBOR curve should match the market quotes for the swap contract.

If the curve building is under dual curve framework, one needs to keep in mind that the front
end of of the curves (the portion prior to the swap segment) should be exactly the same as
that under single curve environment since no discounting is needed (up to the first swap point).

It should be noted that round-trip test does not guarantee that the curve output from the
bootstrapping process is reasonable. As the main purpose of the curve bootstrapping process
is to be able to price instruments which are not directly observable in the market (instruments
not captured in the bootstrapping set).

Another reminder is that under dual curve framework, round-trip test only confirms that the
combined OIS curve and OIS adjusted LIBOR forward curve can reprice market instruments
from the bootstrapping set, but we cannot infer whether we have fully achieved the desirable
results for standalone LIBOR curve or OIS curve.

Thus, round-trip consistency test should be accompanied by other tests to further assess the

20

Electronic copy available at: https://ssrn.com/abstract=3329563

 2.5 10
Raw Method
Fitted Zero Method
Difference in bps
8
OIS Forward Rates, %

Difference, bps
6

4
1.5

1 0
1w 1y 5y 10y 20y 30y 50y
Tenors

Figure 5: OIS Forward Rate Curve Benchmarking

quality of the bootstrapped curves.

8.2 Benchmarking Tests

It is always important to benchmark the curves bootstrapped from one method to another
method or from one system to another system. The curve benchmarking is an effective way
to confirm the pros and cons embedded in each bootstrapping method. For example, curves
built from advanced interpolation methods are usually benchmarked to the raw method as a
starting point.

An example can be seen in Figure 5 for a benchmarking analysis between “raw” method and
a “fitted zero” method for OIS forward rate curve.

The benchmarking can also be used for the dual curve discounting. As noted in the Bloomberg
white paper, “while effects on the forward rates from using dual-curve stripping are convoluted,
the formula below can be used to provide a helpful estimate. Let

δr (T ) = rsw (T ) − rOIS (T )

be the spread in bp between the swap rate and the OIS rate for maturity T, d sw (T ) be
the annual rate of change in bp of the swap rate curve for maturity T (a measure of curve
steepness). Then the forward rate change in bp, δf (T ) can be approximated in the following
manner.”

0.5 ∗ δr (T ) ∗ dsw (T ) ∗ T 2
δf (T ) ≈ −
10000
In this test, the forward curve difference due to discounting can be approximated and quan-
tified. If one uses the BBG function and tests different smoothing methods, the benchmark
result can be drawn as in Figure 6.

21

Electronic copy available at: https://ssrn.com/abstract=3329563

 6
Piecewise Linear
5
Piecewise Linear (Continuous)
4 Smooth Forward
Step Forward
3
Difference, bps

-1

-2

-3
0 5 10 15 20 25 30 35 40 45 50
# of Tenor points

Figure 6: Forward Rate Difference Benchmarking due to OIS and LIBOR Discounting

A quick take away from the formula above is: the OIS adjusted LIBOR forward rate will not
necessarily be smaller than the original LIBOR forward curve. One may have an impression
that since the OIS rate is generally lower than LIBOR, to value the OIS based swaps the same
as under LIBOR based discounting, the OIS adjusted LIBOR forward rates need to be lower
to compensate for the smaller discount factor. But the reality is the actual sign of the forward
rates difference depends on the shape of the forward curve. That’s why the difference can be
either positive or negative.

8.3 Sensitivity Tests

Sensitivity tests are connected with curve stability and curve localness. The test can help the
validator gain confidence how the model reacts to shocks to the inputs.

For localness test, validator could shock the input instruments by small amount and see the
range of the output curve gets impacted. Though, based on Hagan’s table, the localness of
different bootstrapping methods have been quantified, however, still, the validator could verify
whether the model output is consistent with Hagan’s table(Table 6). Also, the fact that sim-
ple methods have a better localness performance over complex methods can be checked as well.

For the stability test, the validator could rely on the similar shock performed for the localness
test, but focusing on the maximum change along the output curve. Since it’s hard to have
a closed-form result for this test, the validator should measure the relative performance from
different options.

9 Conventions and Maths

In the BBG function SWPM, the curve conventions have been established as the default set-
ting for different curve indexes. The screen can also be used for the round-trip benchmarking

22

Electronic copy available at: https://ssrn.com/abstract=3329563

for the swap trades.

The attention should be paid to the fields such as: reset frequency, day count, pay frequency,
etc for both legs. As any discrepancy with market convention could result in inaccurate round-
trip result.

Sometimes, the validator should also look at the conversion transformation process, as dif-
ferent indexes having different conventions may need to be combined. The workaround is to
make sure the market quotes get transformed in a way different instruments can be combined.
A good example will be for the OIS swap rates conversions, since the market only quotes
LIBOR swap and OIS-LIBOR basis swap. The Bloomberg white paper “OIS Discounting and
Dual-Curve Stripping Methodology at Bloomberg” has good details about this process.

10 Outlook
Alternative Reference Rate Committee1

The Financial Stability Oversight Council (FSOC) recommended in its 2014 Annual Report
that U.S. regulators cooperate with foreign regulators, international bodies, and market par-
ticipants to promptly identify alternative interest rate benchmarks anchored in observable
transactions and supported by appropriate governance structures, and to develop a plan to
accomplish a transition to new benchmarks while such alternative benchmarks were being
identified. Greater reliance on alternative reference interest rates will make financial markets
more robust and thus enhance the safety and soundness of individual institutions, make finan-
cial markets more resilient, and support financial stability in the United States. The Financial
Stability Board (FSB) has also called for the development of alternative, nearly risk-free ref-
erence rates.

In response to the FSOC’s recommendations and the objectives of the FSB, the Federal Re-
serve convened the Alternative Reference Rates Committee (ARRC) on November 17, 2014
in a meeting with representatives of major over-the-counter (OTC) derivatives market partic-
ipants and their domestic and international supervisors and central banks. The ARRC was
convened to identify a set of alternative reference interest rates that are more firmly based
on transactions from a robust underlying market and that comply with emerging standards
such as the IOSCO Principles for Financial Benchmarks and to identify an adoption plan with
means to facilitate the acceptance and use of these alternative reference rates. The ARRC
was also asked to consider the best practices related to robust contract design that ensure
that contracts are resilient to the possible cessation or material alteration of an existing or
new benchmarks.

New benchmark rate selection process has considered the following aspects [16].

• Benchmark Quality The degree to which the benchmark design ensures the integrity
and continuity of the rate. The underlying market was evaluated according to its liq-
uidity, transaction volume, and resilience.
1
https://www.newyorkfed.org/arrc/index.html

23

Electronic copy available at: https://ssrn.com/abstract=3329563

 • Methodology Quality The degree to which the benchmark construction could sat-
isfy the IOSCO Principles for soundness and robustness, including standardized terms,
transparency of data, and availability of historic data.

• Accountability Evidence of a process that ensures compliance with the IOSCO Prin-
ciples.

• Governance Evidence of governance structures that promote the integrity of the bench-
mark.

• Ease of Implementation Anticipated demand for and relevance to hedging/trading

and Existence of, or potential for a term market in the underlying rate.

Based on the criteria above, the two favored candidates are: Overnight Unsecured Lending
Rates (OBFR) and Secured Lending Rates (Treasury Repo).

SOFR- the LIBOR Replacement

In June 2017, the ARRC identified SOFR(Secured Overnight Financing Rate) as its preferred
alternative to USD LIBOR index. The index represents a volume weighted median rate of
repurchase agreements(Repo) that are secured by treasuries for overnight term. Unlike the
LIBOR index which is an arithmetic average of the participating banks’ offer rates, the SOFR
is a transactions based volume weighted index. SOFR index has been published since April
2nd 2018. 2

Soon after the daily fixing of SOFR is published, the one-month and three-month SOFR fu-
tures came to CME listing in May 2018. CME SOFR futures are the leading source for price
discovery for the forward SOFR fixing. 3

Cash instruments indexed to SOFR index has entered the market since July 2018. As of Jan
2019, over $40 billion SOFR indexed securities have been issued in the market place since
Fannie Mae’s first issuance in July 2018.

The major challenge facing the SOFR index is the lack of term rate setting. Market is wit-
nessing higher and higher futures volumes for SOFR index but still the open interests beyond
one year are quite small. With more and more longer term cash instruments are traded in the
market, the open interests for longer term SOFR futures are expected to pick up.

On July 18th 2018, LCH announces its first clearance of SOFR swaps. Credit Suisse, Goldman
Sachs, and JP. Morgan are among the first participants. On Oct 9th 2018, the CME group
announced its clearance of the first OTC SOFR swaps: five market participants with trades
notional over $200 million. The cleared trades include SOFR OIS and basis swaps against
LIBOR and Fed Funds. Though the market is still unclear how the term structure of SOFR
will be determined, the proliferation of SOFR linked derivatives undoubtedly helps to pave a
smoother market transition from LIBOR regime to the new era.
2
https://apps.newyorkfed.org/markets/autorates/sofr
3
https://www.cmegroup.com/trading/interest-rates/secured-overnight-financing-rate-futures.html

24

Electronic copy available at: https://ssrn.com/abstract=3329563

11 Appendix A: Swap Manager

Figure 7: BMA Swap Conventions (SWPM <Go>)

25

Electronic copy available at: https://ssrn.com/abstract=3329563

12 Appendix B: Bloomberg Tickers for Bootstrapping Instru-
ments

Table 8: Sample BBG tickers for Bootstrapping OIS curve

OIS Rate Term BBG Tickers
O/N FDFD Index Cash
1w USSO1Z Curncy Swap
2w USSO2Z Curncy Swap
3w USSO3Z Curncy Swap
1m USSOA Curncy Swap
2m USSOB Curncy Swap
3m USSOC Curncy Swap
4m USSOD Curncy Swap
5m USSOE Curncy Swap
6m USSOF Curncy Swap
9m USSOI Curncy Swap
1y USSO1 Curncy Swap
1y6m USSO1F Curncy Swap
2y USSO2 Curncy Swap
3y USSO3 Curncy Swap
4y USSO4 Curncy Swap
5y USSO5 Curncy Swap
6y USBG6 Index Swap
7y USBG7 Index Swap
8y USBG8 Index Swap
9y USBG9 Index Swap
10y USBG10 Index Swap
12y USBG12 Index Swap
15y USBG15 Index Swap
20y USBG20 Index Swap
25y USBG25 Index Swap
30y USBG30 Index Swap

26

Electronic copy available at: https://ssrn.com/abstract=3329563

Table 9: Sample BBG tickers for bootstrapping LIBOR 3M curve
LBIRO 3M Rate Term BBG Tickers
O/N US00O/N Index Cash
3m USFR00C PREB Curncy FRA
4m USFR0AD PREB Curncy FRA
5m USFR0BE PREB Curncy FRA
6m USFR0CF PREB Curncy FRA
7m USFR0DG PREB Curncy FRA
8m USFR0EH PREB Curncy FRA
9m USFR0FI PREB Curncy FRA
10m USFR0GJ PREB Curncy FRA
11m USFR0HK PREB Curncy FRA
1y USFR0I1 PREB Curncy FRA
13m USFR0J1A PREB Curncy FRA
14m USFR0K1B PREB Curncy FRA
15m USFR011C PREB Curncy FRA
16m USFR1A1D PREB Curncy FRA
17m USFR1B1E PREB Curncy FRA
18m USFR1C1F PREB Curncy FRA
19m USFR1D1G PREB Curncy FRA
20m USFR1E1H PREB Curncy FRA
21m USFR1F1I PREB Curncy FRA
22m USFR1G1J PREB Curncy FRA
23m USFC1H1K PREB Curncy FRA
2y USFR1I2 PREB Curncy Swap
3y USSWAP3 BGN Curncy Swap
4y USSWAP4 BGN Curncy Swap
5y USSWAP5 BGN Curncy Swap
6y USSWAP6 BGN Curncy Swap
7y USSWAP7 BGN Curncy Swap
8y USSWAP8 BGN Curncy Swap
9y USSWAP9 BGN Curncy Swap
10y USSWAP10 BGN Curncy Swap
11y USSWAP11 BGN Curncy Swap
12y USSWAP12 BGN Curncy Swap
15y USSWAP15 BGN Curncy Swap
20y USSWAP20 BGN Curncy Swap
25y USSWAP25 BGN Curncy Swap
30y USSWAP30 BGN Curncy Swap
40y USSWAP40 BGN Curncy Swap

27

Electronic copy available at: https://ssrn.com/abstract=3329563

Table 10: Sample BBG tickers for bootstrapping LIBOR 1M curve
LBIRO 1M Rate Term BBG Tickers
1m LIBT1M Index Cash
3m USBAAC CMPN Curncy Basis Swap
6m USBAAF CMPN Curncy Basis Swap
9m USBAAI CMPN Curncy Basis Swap
1y USBA1 CMPN Curncy Basis Swap
1y6m USBA1F CMPN Curncy Basis Swap
2y USBA2 CMPN Curncy Basis Swap
3y USBA3 CMPN Curncy Basis Swap
4y USBA4 CMPN Curncy Basis Swap
5y USBA5 CMPN Curncy Basis Swap
7y USBA7 CMPN Curncy Basis Swap
10y USBA10 CMPN Curncy Basis Swap
12y USBA12 CMPN Curncy Basis Swap
15y USBA15 CMPN Curncy Basis Swap
20y USBA20 CMPN Curncy Basis Swap
30y USBA30 CMPN Curncy Basis Swap

Table 11: Sample BBG tickers for bootstrapping BMASFIMA curve

BMA Rate Term BBG Tickers
1w MUNIPSA Index Cash
1y USSML1 Curncy Basis Swap
2y USSML2 Curncy Basis Swap
3y USSML3 Curncy Basis Swap
4y USSML4 Curncy Basis Swap
5y USSML5 Curncy Basis Swap
7y USSML7 Curncy Basis Swap
10y USSML10 Curncy Basis Swap
12y USSML12 Curncy Basis Swap
15y USSML15 Curncy Basis Swap
20y USSML20 Curncy Basis Swap
30y USSML30 Curncy Basis Swap

28

Electronic copy available at: https://ssrn.com/abstract=3329563

13 Appendix C: Bloomberg Tickers for Pre-Bootstrapped Curves

Table 12: Sample BBG tickers for pre-bootstrapped LIBOR 3M zero curve
Tenor BBG Tickers
1D S0023Z 1D BLC2 Curncy
1W S0023Z 1W BLC2 Curncy
1M S0023Z 1M BLC2 Curncy
2M S0023Z 2M BLC2 Curncy
3M S0023Z 3M BLC2 Curncy
6M S0023Z 6M BLC2 Curncy
9M S0023Z 9M BLC2 Curncy
1Y S0023Z 1Y BLC2 Curncy
18M S0023Z 18M BLC2 Curncy
2Y S0023Z 2Y BLC2 Curncy
3Y S0023Z 3Y BLC2 Curncy
4Y S0023Z 4Y BLC2 Curncy
5Y S0023Z 5Y BLC2 Curncy
7Y S0023Z 7Y BLC2 Curncy
10Y S0023Z 10Y BLC2 Curncy
15Y S0023Z 15Y BLC2 Curncy
20Y S0023Z 20Y BLC2 Curncy
25Y S0023Z 25Y BLC2 Curncy
30Y S0023Z 30Y BLC2 Curncy
40Y S0023Z 40Y BLC2 Curncy
50Y S0023Z 50Y BLC2 Curncy

29

Electronic copy available at: https://ssrn.com/abstract=3329563

References
[1] Sadr, A., Interest Rate Swaps and Their Derivatives. A Practitioner’s Guide, Wiley
Finance, 2009.

[2] Adams, K., Smooth Interpolation of Zero Curves, Algo Research Quarterly, Vol. 4, NOs.
1/2, June 2011.

[3] Hagan, P., & West, G., Interpolation Methods for Yield Curve Construction, Applied
Mathematical Finance, Vol. 13, No. 2, 89129, June 2006.

[4] FinCAD, Revisiting “the Art and Science of Curve Building”, 2011. Available: here.

[5] Filipović, D., Term-Structure Models: A Graduate Course, Springer, 2009.

[6] Abramov, V., Zhou, X., Bloomberg Curve Building Replication.xlsx, 2019. Available: here.

[7] Lorimier, S., Interest Rate Term Structure Estimation Based on the Optimal Degree of
Smoothness of the Forward Rate Curve, Ph.D. Thesis, University of Antwerp, 1995.

[8] Bloomberg, Building the Bloomberg Interest Rate Curve Definitions and Methodology,
March 2016.

[9] OpenGamma, Interest Rate Instruments and Market Conventions, 2013.

[10] FitchSolutions, Charting a Course Through the CDS Big Bang, 2009.

[11] Credit Suisse, A Guide to the Front-End and Basis Swap Markets, Fixed Income
Research, Credit Suisse, 2010.

[12] Bloomberg, OIS Discounting and Dual-Curve Stripping Methodology at Bloomberg, 2012.

[13] FRB, Capital Planning at Large Bank Holding Companies: Supervisory Expectations
and Range of Current Practice, August 2013. Available: here.

[14] The Department of The Treasury, Daily Treasury Yield Curve Rates, Daily Report.
Available: here.

30

Electronic copy available at: https://ssrn.com/abstract=3329563

[15] FHFA, Generating Historically-Based Stress Scenarios Using Parsimonious Factoriza-
tion, Working Pater 13-2, August 2014. Available: here.

[16] ARRC, Rate Evaluation Process, ARRC Presentation Materials, June 16, 2017. Avail-
able: here.

31

Electronic copy available at: https://ssrn.com/abstract=3329563