See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/273718224

Calculating Delta Risks and Hedges via Waves

Article in Wilmott · March 2015

DOI: 10.1002/wilm.10409

CITATION READS
1 3,645

1 author:

Patrick S Hagan
XBTO Inc.
106 PUBLICATIONS 2,651 CITATIONS

SEE PROFILE

All content following this page was uploaded by Patrick S Hagan on 16 October 2017.

The user has requested enhancement of the downloaded file.

 CALCULATING DELTA RISKS AND HEDGES VIA WAVES
PATRICK S. HAGAN
GORILLA SCIENCE
[email protected]

Abstract. A book’s delta risks are usually calculated by bumping the rate of each stripping instrument, re-stripping the
discount curve, and re-valuing the book using the new discount curve. The di¤erence between the new and old value of the
book is the book’s bucket delta risk with respect to that stripping instrument. After …nding the bucket delta risks with respect
to all the stripping instruments, the hedges are obtained by calculating the combination of stripping instruments needed to
o¤set this vector of bucket risks.
This methodology leads to several problems: risks bleeding between risk buckets, non-intuitive risk scenarios, over-
speci…cation of risks, and needless intertwining of the risk and stripping processes.
Here we present a simpler alternative method for calculating delta risks, the “wave” or “scenario” method, which is largely
free of these problems.

1. Conventional approach. The delta risks of a book are usually calculated by bumping the rate of
each stripping instrument, re-stripping the discount curve, and re-valuing the book using the new discount
curve. The di¤erence between the new and old values of the book is the book’s bucket delta risk with
respect to that stripping instrument. This methodology leads to several problems: bleeding, non-intuitive
risk pro…les, over-speci…cation of risks, and needless intertwining of the risk and stripping methodologies.
Here we present an alternative, more e¤ective, method of calculating delta risks and hedges, the “wave” or
“scenario” method, which has been developed to overcome these problems. Since the wave method directly
manipulates the forward curve itself, it can be used for any other curves (rate projection curves – like
Libor or Euribor projection curves, basis spread curves, survival curves for credit linked deals, . . . ) without
modi…cation.
1.1. Stripping. Let us brie‡y consider curve stripping. In depth accounts can be found in, e.g., [1],
[2]. Consider a discount curve
RT
f 0 (t)dt
(1.1) D(T ) = e 0 ;
where f 0 (t) is the corresponding instantaneous forward rate curve. Discount curves are obtained by stripping
a set of …xed income instruments –typically deposits, futures, and swaps. Let S1 ; S2 ; : : : ; Sm be the stripping
instruments used to obtain this discount curve. Parsing the market value of each instrument j yields a relation
of the form
nj
X
(1.2) VjS = (aj;i Rj + bj;i ) D(tj;i ) for j = 1; : : : ; m;
i=1

where VjS is the instrument’s current value (often 0), and Rj is the instrument’s rate (deposit rate, futures
rate, swap rate, . . . ) in the current market. The paydates tj;i and coe¢ cients aj;i and bj;i are determined
by the particulars of instrument j. Let us arrange these instruments in order of increasing maturity, so that
(1.3) 0 < T1 < T2 < : : : < Tm ;
where Tj = tinj is the …nal paydate of instrument j.
The art of stripping curves is to create a “good looking” instantaneous forward rate curve f 0 (t) which
satis…es all m …nancial constraints in 1.2. The simplest method is to use a piecewise constant forward curve,
(1.4a) f 0 (t) fj0 for Tj 1 < t < Tj ; j = 1; 2; :::; m 1;
(1.4b) f 0 (t) 0
fm for Tm 1 < t < 1;
1
where T0 0 is today. For the …rst inteval T0 < t < T1 , the constant forward rate f01 is chosen so that the
j = 1 constraint in 1.2 is satis…ed. For the interval T1 < t < T2 , the constant f02 is chosen to satisfy the
j = 2 constraint; this doesn’t upset the …rst constraint since the …rst constraint only involves dates t < T1 .
Continuing this bootstrap procedure through all m steps yields a piecewise constant forward curve f 0 (t)
which satis…es all m contraints.
Although the piecewise constant forward curve satis…es all …nancial constraints, the jumps in the forward
curve at the nodes Tj are clearly artifacts of our stripping methodology, and are unlikely to be present in
real discount curves. I.e., we could get picked o¤. To create more realistic continuous forward curves, more
sophisticated interpolation methods have been developed. Piecewise linear and cubic splines turn out to
be unstable, so to strip discount curves, one turns to using cubic splines under tension, smart quadratic or
smart quartic interpolations, or to monotone convex splines for the instantaneous forward curve. The most
common interpolation methods used for stripping are examined in depth in [1], [2], where the relative merits
and drawbacks of the schemes are discussed.
1.2. Coventional delta risks. After stripping to obtain the discount curve D(T ), one values the
book using this “base” curve to …nd the book’s current value, its mark-to-market. Under the conventional
approach, one then bumps the rate of the …rst stripping instrument,
(1.5) R1 ! R1 + ;
usually by 1 bp, and re-strips the discount curve, obtaining a new “bumped” curve D(T ^ ). The book is
re-valued using the new curve, and the di¤erence between the book’s new value and its base value is the
book’s “bucket delta risk” with respect to the …rst instrument, V1book . Putting R1 back to its original
value, one then bumps the second rate R2 , re-strips the curve, re-values the book, and subtracts the new
value from the base value to obtain the bucket delta risk with respect to the second instrument, V2book .
Repeating the procedure for all the stripping instruments yields the vector of bucket delta risks,
T
(1.6) Vbook V1book ; V2book ; : : : ; Vmbook :
The delta hedge is the linear combination of stripping instruments
m
X
(1.7) aj Sj
j=1

which cancels out the vector of bucket delta risks. It can be obtained by calculating the bucket delta risks of
the stripping instruments. So, after bumping each rate Rk , re-stripping the curve, and re-valuing the book
with the bumped curve D(T^ ), one also re-values each of the stripping instruments. Taking the di¤erence
between the new and old values yields the bucket delta risks of the stripping instruments with respect to the
k th instrument,
S
(1.8) Vjk ; for j = 1; 2; : : : ; m:
Consider the porfolio
m
X
(1.9) V book + aj Sj :
j=1

For this portfolio to be hedged, we need all the bucket delta risks of this porfolio to be zero. That is, we
need
m
X
(1.10) Vkbook + S
aj Vjk =0 for all k = 1; : : : ; m:
j=1
2
Solving yields
m
X
(1.11a) aj = Vkbook M 1
kj
k=1

1
where M is the inverse of the sensitivity matrix,
S
(1.11b) Mjk = Vjk :

1.3. Disadvantages of the conventional approach. An advantage of the conventional approach is

that the software for re-stripping the curve and re-valuing the books is the same software used in the base
case to obtain the book’s mark-to-market.
The …rst problem with the conventional approach is the bleeding of risks into di¤erent risk buckets.
Requiring forward curves f 0 (T ) to be continuous means that bumping the rate Rj a¤ects not only the
bucket Tj 1 < t < Tj , but must also a¤ect the rates in neighboring buckets so that continuity can be
preserved. In turn, these a¤ect other buckets, which a¤ect other buckets, until all the intervals are involved.
Consequently, a 2.5y swap may have signi…cant delta risks in the 5y, 10y, and even the 30y buckets, and
thus its hedges would include 5y, 10y, and 30y swaps. Clearly this is inappropriate since instruments cannot
depend on events after maturity.
A second problem is that bucket delta risks are highly non-intuitive. Suppose we were to use piecewise
constant forwards to strip the 1y, 2y, ..., 9y, 10y swaps. Consider what happens when we bump, say, the
4y swap rate. Since the …rst three swap rates have not changed, the instantaneous forward curve f 0 (t) does
not change for t 3y. For 3y < t < 4y, the instantaneous forward rate f 0 (t) must go up by roughly 4 bps
in order that the 4y swap rate goes up by 1bp. For 4y < t < 5y, the instantaneous forward rate f 0 (t) must
decrease by roughly 4 bps to leave the 5y swap rate unchanged. Beyond …ve years, the instantaneous rate
curve f 0 (t) should exhibit only minor changes arising from the swap coupons. Thus, the delta risk in the 4y
bucket measures payments received during year 3 minus the payments received during year 4. Worse, using
more sophisticated interpolation schemes, or stripping more complicated sets of instruments, leads to even
less intuitive risks.
A third issue is over-spec…cation of risks. Stripping a discount curve may involve 30 to 40 di¤erent
stripping instruments, which would lead to 30 to 40 di¤erent delta risk buckets and hedges. For some
purposes, such as hedging risks to a major currency swap book, this may be appropriate. But for risk
managers trying to understand an organization’s interest rate risks over twenty or more di¤erent currencies,
some method of systematically consolidating and simplifying the risk is needed.
A …nal problem is the coupling of risk calculations to the stripping process. Whenever the stripping
procedure changes, whether it’s the interpolation method or the choice of stripping instruments, this changes
the meaning of the bucket delta risks. So one cannot pick the best stripping method without considering its
e¤ect on the downstream risk processes.
2. The wave method. The “wave method,” is a scenario based method of calculating delta risks
and hedges which has been developed to overcome these problems. Abstractly, by bumping the rates and
re-stripping the curve, we have been generating m distinct, more-or-less reasonable scenarios, each of which
is a small perturbation from the base case. Since our hedged book is ‡at to these perturbations, our book
should be hedged to the extent that the real curve movements can be approximated by a linear combination
of these perturbations, at least until the changes become large enough for nonlinear e¤ects to set in. The
wave method follows the same approach, except that we directly specify the scenarios, instead of obtaining
them by bumping and re-stripping.
Suppose one strips a curve D(T ) using one’s best stripping methodology, and let f 0 (T ) be the resulting
instantaneous forward rate curve. One selects a set of hedging instruments, H1 ; H2 ; : : : ; HK . Usually these
3
hedging instruments are either the stripping instruments, S1 ; : : : ; Sm , or are a subset of these instruments.
Let us arrange these instruments in order of increasing maturity, and let
(2.1) 0 < T1 < T2 < < TK
be their …nal maturity dates. For each maturity date Tk (except the last one), we de…ne a new instantaneous
forward rate curve,
8
< 0 for t < Tk 1
(2.2a) fk (t) = f 0 (t) + 1 for Tk 1 < t < Tk
:
0 for Tk < t
for k = 1; 2; : : : ; K 1. Here f 0 (T ) is the instantaneous forward rate curve for the base case, and is
typically 1 bp. For the last curve, we de…ne
8
< 0 for t < TK 1
(2.2b) fK (t) = f 0 (t) + 1 for TK 1 < t < TK ;
:
1 for TK < t
so that its bump has ‡at extrapolation beyond the …nal date TK .Thus, for each scenario k, we have a new
discount curve
8
< 1 for t < Tk 1
(2.3a) ~ k (t) = D(t)
D e (t Tk 1 ) for Tk 1 < t < Tk k = 1; 2; : : : ; K 1;
:
e (Tk Tk 1 ) for Tk < t
and

~ K (t) = D(t) 1 for t < TK 1

(2.3b) D (t TK 1)
k = K:
e for TK 1 < t
~ k (t).
For each k, one values the book in scenario k; that is, one values it using the new discount curve D
The di¤erence between the book’s value in scenario k and the base scenario is the bucket delta risk with
repect to the k th scenario, Vkbook . Repeating for all k yields the vector of bucket deltas,
T
(2.4) Vbook V1book ; V2book ; : : : ; VKbook :
The delta hedge is the linear combination of hedging instruments which cancels out the vector of bucket
delta risks. Consider the hedged portfolio
K
X
(2.5) V book + aj Hj ;
j=1

and let
(2.6) Hjk ; for j; k = 1; 2; : : : ; K
be the k th bucket delta of hedging instrument j; that is, the di¤erence in the value of hedging instrument
j under scenario k and under the base case. Setting the bucket delta risks of the portfolio to zero for all
scenarios k requires that
K
X
(2.7) Vkbook + aj Hjk = 0 for k = 1; : : : ; K:
j=1
4
Solving these equations yields the hedges,
K
X
(2.8) aj = Vkbook H 1
kj
;
k=1

where H 1 is the inverse of the sensitivity matrix Hjk .

For the wave method, the sensitivity matrix Hjk is lower triangular. To see this, recall that the …nal
maturity date of hedging instrument Hj is Tj , and note that scenario k does not change the discount curve
for dates t < Tk 1 . So any scenario k with k > j cannot a¤ect instrument j. Thus,

(2.9a) Hjk = 0 for scenarios k > j:

The fact that Hjk is a lower triangular matrix makes inverting the matrix trivial, and also makes the
inverse of Hjk lower triangular,
1
(2.9b) H kj
=0 for k < j;

Thus
K
X
(2.10) aj = Vkbook H 1
kj
;
k=j

so the j th hedge is constructed only from the bucket delta risks from scenarios k j.
3. Conclusions. The wave method does not su¤er from bleeding. To see this, consider a deal whose
last paydate T < Tj for some j . Since scenario k only changes the discount curve for dates t > Tk 1 ,
clearly the deal’s bucket delta risk is zero for all scenarios k > j ,

(3.1) Vkdeal = 0 for k > j :

With 2.10, this means that

j
X
(3.2a) aj = Vkdeal H 1
kj
for j j ;
k=j
(3.2b) aj = 0 for j > j :

Thus, the hedges of a deal are exactly zero for all buckets which occur after the last pay date of the deal.
The scenarios used to calculate risks in the wave method also clear. Since these scenarios are completely
independent of the stripping process, one can develop and use the best stripping methods without considering
it’s e¤ect on downstream risk processes. Similarly, one can choose the best risk buckets to use without regards
to the stripping process. For some purposes, such as hedging a large swap book, it is appropriate to use all
the stripping instruments as the hedging instruments; one then gets very detailed delta risks and hedges.
For other processes, it may be more appropriate to choose, say, the 1y, 5y, and 10y swaps as the hedging
instruments. This then expresses the risks as short term (under 1y), medium term (from 1y to 5y), and long
term (over 5y) risks. Or one can choose a single instrument as the hedging instrument, so that the delta risk
is the risk to a parallel shift.
The wave method can be used to …nd the delta risks for any curves, not just discount curves. The risks
for projection curves, like Libor projection curves, the risks for basis spread curves, and the risks for survival
curves in credit problems can be handled in an identical manner.
5
 3.1. Compound IDs. An e¤ective way of implementing the wave method is to replace curve names
with compound curve names, and curve IDs with compound curve IDs. At the interface level, curves are
usually identi…ed by name, such as “USDSwap” or “EONIA.” The software identi…es these curves as, say,
curve 6 and curve 19, by comparing the names against the list of curves that have been created. Internally,
curves are usually speci…ed by these integers, the curve IDs.
To use compound IDs, one de…nes sets of scenarios by specifying a set of dates,

0 < T1 < T2 < < TK ;

and stores and names these scenarios, as in “USDRisks,” or “EMRiskSet.” Instead of using simple curve
names like “USDSwap” at the interface level, one uses compound names like

USDSwap j USDRisks j 7 j 1.5

This identi…es the curve as the base curve “USDSwap” with a shift of = 1:5bps in the seventh scenario
(k = 7) from the set of scenarios named “USDRisks.”See eqs. 2.2a, 2.3b. At the interface level, the compound
name is parsed into a compound ID (a curve ID, an ID specifying the set of scenarios, the scenario number,
and the shift amount). This compound ID threads through the system, with the scenario information being
ignored, until reaching the “discount factor” function. Only at this level does the scenario information get
used to modify the discount factors.
Compound names are also e¤ective for other market objects, such as vol cubes. Using compound names
for all market objects enables one to quickly create standard and customized risk scenarios, and then use
scripting to automate the downstream processes.

REFERENCES

[1] P.S. Hagan and G. West, Interpolation Methods for Curve Construction, Applied Math Finance, 13 No.2 (2006), pp. 89
– 129
[2] P.S. Hagan and G. West, Methods for Constructing a Yield Curve, Wilmott, (May, 2008), pp. 70 – 81

View publication stats