Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

YIELD CURVE ANALYSIS USING PRINCIPAL COMPONENTS

*Swamynathan V.

There has been a phenomenal slide in the same is then extended to delve on the
interest rates which has been of interest to methodology to derive the sensitivity of a
the corporates and academicians alike. To portfolio of Fixed Income Securities to
put the facts in perspective, the following change in interest rates.
graph depicts the level of the yield curve on
What are the traits associated with a brand
six sample dates:

Yield Curve Movement

12
11
S pot R a te (%)

10
03-Apr-01
9 01-Oct-01
8 01-Apr-02
01-Oct-02
7
01-Apr-03
6
01-Aug-03
5

4
1 Year 5 Year 7 Year 10 Year 15 Year 20 Year 25 Year 30 Year
Figure. 1

While the movement in the interest rates can like ‘Wheel’ of HLL?
be loosely explained in terms of a fall of
What are the features one desires when
around 400 bps. in the 10-year segment and
buying a C-Class car? What are the
the like, one is tempted to enquire about a
psychological traits that cause a person to
statistical framework to validate such
contribute to a charity institution?
developments.
Market research firms develop
The following note attempts to explore the
questionnaires seeking responses on
yield curve movement by introducing one of
personal background, social status, tastes,
the statistical methods used extensively. The
financial strength, et. al. to identify various

*Swamynathan V. is Assistant Manager, Risk Management,


The Clearing Corporation of India Limited.
factors that result in a person perceiving a security 7.37% GS 2014. This implies that:
brand, preferring a particular feature of a car while there may be 30 Factors (for the sake of
or contributing selflessly to a social cause. convenience) causing the daily movement in
the curve, there seems to be actually fewer
Factor Analysis – a stream in Statistical
variables (Components) that drive the yield
research – is used widely to collate and
curve. Hence, an effort has been made by us
analyse responses and arrive at certain
to identify the underlying factors through
common Factors which cause a particular
Principal Components Analysis (PCA).
result. It’s a multi-variate statistical
technique for combining of seemingly The Principal Components Decomposition
diverse material responses into few unique, is a particular kind of Factor Model. Its role
unrelated (and perhaps abstract) underlying is to reduce the dimensionality of the subject
Factors. (in our case, the yield curve factors) i.e. to
reduce the number of underlying sources of
The work on application of a technique of
uncertainty or market factors. These factors
Factor Analysis to yield curve movements is
are not specified in advance; rather they are
credited to Robert, Litterman and
derived out of the data analysed.
Sheinkman. The trio published an article
‘Common Factors Affecting Bond returns’ The number of Principal Components
in the Journal of Finance (1991) employing derived from the yield curve analysis will be
the technique of Principal Component equal to the number of factors which drive
Analysis to the yield curve movements. the yield curve (i.e. 30 nos.). At first read, it
suggests that there may not be much
The Indian Sovereign yield curve starts from
advantage using the PCA decomposition.
the overnight rate and extends upto 30 years.
However, in the area of Finance, the data are
Every tenure may be considered as a Factor
so highly correlated with each other that the
which contributes towards the movement in
first few principal components explain most
the yield curve; for the sake of convenience,
of the variability of the yield curve.
we may state that there are only 30 factors
Thenceforth, one may work with the
(corresponding to the 30 tenures) which
Principal Components which explain most
effect the yield curve.
of the changes; than working with all the
But we know that the movement in the causal factors.
7.95% GS 2032 is related to the change in the
Our analysis covers the change in yield curve
YTM of 6.01% GS 2028. The change in the
since Jan-2000 till Feb-2004, a period
prices of the relatively illiquid securities in
spanning just over 3 years. The data yield
the maturity bucket 7 to 12 years are driven
curve has been developed by the Nelson-
by the change in the YTM of the benchmark
Siegel model.
We initiated a process to identify the Effect of Principal Components
0.6000
Principal Components in the Matlab
environment by feeding data of the 0.4000
yields of past 3 years. Instead of 30
yield points as factors as stated above, 0.2000

for the sake of our analysis, data

V a lue s
0.0000
relating to the yields of the tenures 0.5, 0.5 1 5 7 10 15 20 25 30

1, 5, 10, 15, 20, 25 and 30 years which is -0.2000

9 in number were entered in the


-0.4000
software. The software has derived 9
components (as expected, it’s equal to -0.6000
Tenures (Yrs.)
the number of tenure points of
Figure 2 PC1 PC2 PC3
analysis which is 9 in no.) which have
caused the movement in the yield curve. PCA also gives the contribution of
The effect of each of the components on each variability by the principal components to
tenure point of the yield curve is tabulated the yield movements.
below:
Table 1
PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9

0.5 Year -0.3102 -0.5499 -0.3851 0.4135 -0.4438 -0.2705 0.1037 0.0341 0.0094

1 Year -0.3080 -0.5151 -0.2529 -0.2214 0.5785 0.4026 -0.1632 -0.0570 -0.0164

5 Year -0.3260 -0.2159 0.3800 -0.5558 0.0082 -0.4078 0.3825 0.2626 0.1107

7 Year -0.3356 -0.0894 0.4473 -0.1055 -0.3108 0.0352 -0.4286 -0.5479 -0.2941

10 Year -0.3437 0.0548 0.3926 0.3133 -0.1470 0.4673 -0.1271 0.4614 0.3969

15 Year -0.3469 0.2034 0.1587 0.4152 0.3081 0.0616 0.5272 -0.0821 -0.5073

20 Year -0.3451 0.2866 -0.0798 0.1934 0.3357 -0.4006 -0.1192 -0.3696 0.5754

25 Year -0.3423 0.3373 -0.2748 -0.0989 0.0471 -0.2668 -0.4807 0.4851 -0.3795

30 Year -0.3396 0.3707 -0.4268 -0.3772 -0.3745 0.3788 0.3054 -0.1866 0.1048

It’s comforting to note that PC1 has PC1 contributes around 97.5% to the change
consistent negative values for all tenures. in the overall yield curve; followed by PC2 of
PC2 has negative values for the shorter around 2.23%. The third component
tenures; it increases for the medium to changes the curve to the extent of 0.1%;
longer terms. PC3 has a negative impact on remaining components have a negligible
the yield curve for the shorter and longer effect on the curve dynamics.
tenures while its effect is positive for the
medium tenures.
Table 2 PC3 is the Curvature effect. It has an effect
similar to changing the curvature of the
Variance % Contribution Cum. Contribution
PC1 0.0037189000 0.9759 0.9759 curve of a day. This may be interpreted as the
PC2 0.0000849730 0.0223 0.9982 curve taking a longer (or shorter) time to
PC3 0.0000064787 0.0017 0.9999 reach the longest tenure rate.
PC4 0.0000002734 0.0001 1.0000
The values of the first three Components as
PC5 0.0000000364 0.0000 1.0000 rd
on 23 Feb’04 are as follows:
PC6 0.0000000028 0.0000 1.0000
Table 3
PC7 0.0000000001 0.0000 1.0000
Zero
PC8 0.0000000000 0.0000 1.0000 Tenures PC1 PC2 PC3
Rates
0.5 4.29% -0.3102 -0.5499 -0.3851
PC9 0.0000000000 0.0000 1.0000
1 4.34% -0.3080 -0.5151 -0.2529

We may, hence, conclude that the 5 4.91% -0.3260 -0.2159 0.3800


7 5.20% -0.3356 -0.0894 0.4473
dimensionality of the yield curve has even
10 5.56% -0.3437 0.0548 0.3926
got reduced from 9 to 2 (or 3). While we 15 6.01% -0.3469 0.2034 0.1587
haven’t assigned any name to the 20 6.29% -0.3451 0.2866 -0.0798

components, it is interesting to realize that 25 6.48% -0.3423 0.3373 -0.2748


30 6.61% -0.3396 0.3707 -0.4268
PC1 effects a change in the level of the curve.
-0.1664 0.0184 -0.0053
It has had a negative impact in the yield
curve in the past 3 years i.e. the fall in yield of The Principal Components derived are
1 year (of 400 bps.) through 30 years (of 500 linear combinations of the variables (30
bps.) is explained by PC1 (parallel yield curve tenures) under study. Each
movement) to the extent of 97.59%. From component is, thus, arrived at by the
Figure 2, the effect of PC1 is uniform across summation of the products of the Zero Rates
the tenures indicating a parallel movement and the contribution of the component to
in the curves. This is also diagrammatically the respective tenures.
depicted in Figure 1.
To get an intuitive feel of the Principal
PC2 can be considered as one that effects the Components, we have plotted the effects of a
Slope of the curve. PC2 seems to cause a fall 100 bps. change in the Components on the
in the short term (due to its negative values yield curve of 23rd Feb’04 in Figure 3.
in the shorter tenures) and a rise in the
longer term (due to the positive values in As is evident, PC1 results in a change in the
Table 1). This is evident from the Figure 2. LEVEL of the curve; PC2 effects the SLOPE
where a dip in rates is observed in the shorter of the curve; PC3 results in a change in the
term and rise in the longer term. CURVATURE of the curve.
Figure 3
Component Effects
0.08
0.07
0.07
0.06
Ra te s (% )

0.06
0.05
0.05
0.04
0.04
0.03
1 2 3 4 5 6 7 8 9
Tenures
YC 23/02/04 YC shift due to PC1
YC shift due to PC2 YC shift due to PC3

A Risk Manager would be interested in using for a 1 bp. change in each of the components
the concept of PCA to arrive at the volatility is computed. This is i.e. the change in the
of a portfolio of fixed income securities. price of the kth bond for a basis point
One finds it convenient to deal with a th
change in the i Principal Component. This
volatility matrix of 3 factors as compared to is similar to the Duration of a bond (which is
a volatility matrix of 30 x 30. the change in a price of a security for a small
We have tried to implement the PCA model change in its yield). The values of b ki are
to a portfolio comprising of Zero Coupon tabulated below for a 1 bp change in each of
bonds - 25% of face value in a 1 year, 35% in 5 the principal components.
year and 40% in 20 year Zero coupon bonds It is notable that the effect of the change in
rd
based on the yields of 23 Feb, 2004. The PC1 is all negative (LEVEL), change in slope
present value of the portfolio is as follows: due to PC2 (SLOPE) and change in curvature
Table 4 caused by PC3 (Curvature).
Zero Rates 4.34% 4.91% 6.29%
Present Value (Rs.) 95.84 95.32 94.08 The Variance of the portfolio is derived
Portfolio weight 0.25 0.35 0.4 from the table above using the formula:
2 2 2
The sensitivity of the yield curve to changes Portfolio Variance = b p1l1 + b p 2 l2 + b p 3 l3
in the components is given by the respective Where l is the Variance of the respective
contributions to volatility of the Principal Principal Component.
Components as listed in Table 2.
The above equation can be regarded as a
Firstly, the change in the price of the bonds product of the Duration and Volatility of a
bond to arrive at the risk exposure. be a classical case for implementing PCA.
Our results have identified three factors
This results in the Variance of the portfolio
which can explain the change in the
which can then be extended to arrive at the
complete yield curve. These factors can be
Value-at-Risk figures. Assuming a Normal
used to compute the Variance of a portfolio.
Distribution, the VaR can, then, be
computed by adopting the parametric One may extend the implementation of PCA
method. to Stress a portfolio by changing the
Principal Components with respect to the
Conclusion
anticipation of yield changes in future.
The Indian yield curve movement seems to

You might also like