INDIAN INSTITUTE OF MANAGEMENT CALCUTTA

Studying the Behaviour of Indian Sovereign Yield Curve Using Principal Component Analysis
Fixed Income Markets
Group 3 9/10/2013

This paper analyses the factors responsible for affecting term structure changes in the context of Indian sovereign yield curve. We apply Principal Component Analysis on our data to determine these factors.

Contents
Introduction ........................................................................................................ 2 Previous Literature .............................................................................................. 4 Data Source ........................................................................................................ 6 Methodology ........................................................................................................ 7 Results and Analysis ........................................................................................... 9 Conclusion ........................................................................................................ 13 Bibliography...................................................................................................... 13

Introduction
The yield curve is sometimes called the term structure of interest rates, and is often presented as a simple chart showing the annualized yield investors receive for loaning out their money for different time periods. These can range from overnight to as long as 30 years in the case of some government bonds. Most often, analysts look at the yield curve in terms of sovereign securities, since there should be (in theory) no problems with credit quality that could distort the picture. Golaka Nath (2012) identifies the following advantages of estimating the yield curve reasonably: The yield curve serves as a benchmark in the economy as private corporate entities raise funds by paying a credit spread for the risk inherent in them; Investors use the sovereign yield curves to demand an appropriate price for their investment risk; Banks and other financial institutions use the yield curves to not only price the illiquid securities in their books but also match the duration of their assets and liabilities; Central banks use the information from secondary market yield curves to monitor the policy interest rate synchronization with the economic effective rate in the inter-bank market; At macroeconomic level, the yield curve has a predictive power for the state of economy. The Indian sovereign bond market is underdeveloped in comparison with the developed markets of the world. A reasonably good measure of the level of development of the bond market is the outstanding debt to GDP ratio. The following graph illustrates the outstanding debt as a percentage of GDP for the top 10 economies of the world (March 2013).

Figure C1: Outstanding Debt as a percentage of GDP 230% 150%

100%

84% 65%

92% 60% 35% 22%

52%

50%

38%
11%

Government Debt/GDP

0%

Source: BIS Quarterly Review, IMF World Economic Outlook, CBonds.info

As shown above, India ranks 7 th (out of 10) as far as the sovereign debt to GDP ratio is concerned. The Bank for International Settlements (BIS) holds the view that sovereign debt can be a deterrent to growth only if this ratio exceeds the 85% level. India, with a ratio close to 38%, has a long way to go in this regard. Another measure of the level of development of the bond market is liquidity, measured in terms of number of bonds traded versus the total bonds outstanding. Barring the benchmark securities and a few other bonds, sovereign bonds in India are characterized by market illiquidity, especially the bonds with higher tenors which are often held-till-maturity by insurance companies and pension funds. These shortcomings pose significant challenges towards estimating the yield curve for India. One popular estimation model is the Nelson-Siegel (NS) functional form which has been used by NSE since 1999 to calculate the spot interest rates. The main purpose of this paper is to study the term structure dynamics and to figure out the common factors of the Indian term structure and its volatility as it helps to understand the pricing mechanism of various OTC and other underlying and derivative products. Previous research studies have indicated that the three biggest determinants of term structure are level, slope and curvature of the yield curve. We will use Principal Component Analysis (PCA) to analyze the effect of these factors on the term structure. 3

Previous Literature
Influence of monetary policy on the term structure of interest rates: 1. A very important work done in this field was Ang, A., J. Boivin, S. Dong, and R. Loo-Kung (2010): Monetary Policy Shifts and the Term Structure . The existence of historical shifts in monetary policy, or different monetary policy experiments, provided an opportunity to statistically estimate the effects of changes in the policy rule on the term structure. According to this model, short rates move due to movements in the i) ii) output gap component, and the inflation component

The study found that the endogenous response of inflation to past changes in inflation loadings is an important component of how bond prices reflect monetary policy risk under the risk-neutral measure. In contrast, policy shifts in output gap loadings exhibit little time series variation, so almost all changes in monetary policy stances have been done with respect to inflation. If investors assigned no value to monetary policy shifts, then the slope of the yield curve would have been, on average, up to 50 basis points higher than the data. The term spread would have also been significantly more volatile without activist monetary policy that is priced by investors. This valuable contribution of monetary policy discretion is due to the risk discount assigned by investors for monetary policy shifts. 2. According to the research paper done by "Ying He and Carlos Medeiros" on "An Assessment of Estimates of Term Structure Models for the United States", 2011 This paper assesses estimates of term structure models for the United States. In this context, it first describes the mathematics underlying both the Nelson-Siegel and Cox, Ingersoll and Ross family of models and estimation methodologies. It then presents estimations of some of these models within these families of modelsa three-factor, yield-only Nelson-Siegel model, a four-factor Svensson model, and a preference-free, two-factor CIR modelfor the United States from 1972 to mid 2011.. These estimations encapsulate the changes in expectations of short-term future interest rates, while confirming that the yield-factors of the term structure of

interest rateslevel, slope and curvaturesprovide a good representation of the term structure.

Effects of Treasury Yields, Ination, Ination Forecasts, and Ination Swap Rates on the term structure of interest rates: 1. The flagship journal Estimating Real and Nominal Term Structures using Treasury Yields, Inflation, Inflation Forecasts, and Inflation Swap Rate s, was published in 2011 by Haubrich, J., G. Pennacchi, and P. Ritchken. Under this theory, the term structures are driven by state variables that include the short term real interest rate, expected ination, a factor that mod els the changing level to which ination is expected to revert, as well as four volatility factors that follow GARCH processes. The study found that allowing for GARCH eects is particularly important for real interest rate and expected ination processes, but that long-horizon real and ination risk premia are relatively stable . 2. According to the research paper done by "Adrian, T., and H. Wu" on The Term Structure of Inflation Expectations, Working Paper, Federal Re serve Bank of New York", 2009 "They presented estimates of the term structure of inflation expectations which was derived from an affine model of real and nominal yield curves. They found out that model-implied inflation expectations can differ substantially from breakeven inflation rates when market volatility is high. These differences are highly correlated with market volatility measures such as the VIX equity implied volatility index. Intuitively, as implied volatility increases, risk premia increase, and breakevens tend to overpredict inflation expectations." Some other useful studies worth mentioning are: 1. According to Buraschi, A., A. Cieslak, and F. Trojani (2010): Correlation Risk and the Term Structure of Interest Rates, 1) Within the economy, the predictability of excess bond returns is supported by the single forecasting factor of Cochrane and Piazzesi (2005). 2) The dynamic correlations of yields and the co-movement in their volatilities lead to realistic properties of conditional hedge ratios between bonds 5

3) Some state variableswhile loading weakly on yieldshave an economically significant impact on the prices of interest rate caps, thus evoking the notion of un-spanned factors.

2. According to Fontaine, J.S., and R. Garcia (2011): Bond Liquidity Premia, The pattern across interest rate markets and credit ratings is consistent with accounts of flight-to-liquidity events. However, the effect is pervasive even in normal times. The evidence points toward the importance of aggregate liquidity and aggregate liquidity risk compensation in asset pricing. They find that measures of changes in the stock of money and measures of the availability of funds in the banking system are important determinants of our measure of aggregate liquidity. To a lesser extent, the liquidity factor varies positively with transaction costs and aggregate uncertainty.

Data Source
We performed our analysis on data collected from Bloomberg. PCA was applied to monthly yield changes data from Jan08 to Sep13 (69 data points) for the set of maturities given in the table below:

Table T1: Descriptive Statistics of Historical Term Structure of Interest Rates (in %) 3M Mean StDev Max Min Median 6.91 1.85 10.42 3.53 7.33 6M 7.12 1.81 10.53 3.78 7.58 1Y 7.01 1.59 9.87 4.06 7.62 2Y 7.29 1.18 9.33 4.81 7.70 3Y 7.52 0.94 9.32 5.30 7.76 4Y 7.71 0.79 9.31 5.75 7.90 5Y 7.83 0.74 9.29 5.73 7.95 7Y 7.96 0.61 9.33 6.19 8.02 10Y 7.92 0.71 9.13 5.79 8.09 15Y 8.49 0.48 10.07 7.07 8.53 20Y 8.58 0.49 10.07 7.29 8.61 30Y 8.91 0.56 10.71 7.62 9.00

Source: Bloomberg

Also, the following figure illustrates the historical G-Sec rates in India for certain key maturities using the data collected. Implications of T1 and C2 are discussed in a later section. 6

Figure C2: Historical G-Sec Rates in India 10.00% 9.00% 8.00% 7.00% 6.00% 5.00% 4.00% Dec-08 Mar-09 Dec-09 Mar-10 Dec-10 Mar-11 Dec-11 Mar-12 Dec-12 Mar-13 1-Y 3-Y 5-Y 10-Y 20-Y

Jun-11

Jun-09

Jun-10

Jun-12

Sep-08

Sep-09

Sep-10

Sep-11

Sep-12

Jun-13

Source: Bloomberg

Methodology
Principal Component Analysis: An Overview
PCA is a useful statistical technique that can be used for finding patterns in data of high dimension. These findings can then be used to highlight similarities and differences between different data points. One chief advantage of PCA is that data can be compressed after identification of patterns without much loss of information. Since the PCA model explicitly selects the factors based upon their contributions to the total variance of interest rate changes, it may help in hedging efficiency when using only a small number of risk measures. In general, data reduction and summarization is popularly done using a technique called factor analysis. Factor analysis is a statistical method used to describe variability among observed, correlated variables in terms of a potentially lower number of unobserved variables called factors. Factor analysis is done in the following circumstances:

Sep-13

To identify underlying dimensions, or factors, that explain the correlations among a set of variables To identify a new, smaller, set of uncorrelated variables to replace the original set of correlated variables in subsequent multivariate analysis (regression or discriminant analysis)

To identify a smaller set of salient variables from a larger set for use in subsequent multivariate analysis

Mathematically, a factor model with n factors can be represented as: where, Xi = i th standardized variable Ri1 = standardized multiple regression coefficient of variable i on common factor j Fi = common factor i Ci = standardized regression coefficient of variable i on unique factor j Ui = the unique factor for variable i The common factors themselves can be expressed as linear combinations of the observed variables: where, Fi = estimate of the i th factor Wi = Weight or factor score coefficient k = number of variables In PCA, factor weights are computed in order to extract the maximum possible variance, with successive factoring continuing until there is no meaningful variance left.

PCA and the Indian Sovereign Term Structure

PCA can be successfully applied to determine the chief factors affecting movements in the term structure. Thus, the yield curve shifts can be assumed to be a function of different realizations of principal components. As mentioned before, height, curvature and slope are considered to be the three major principal components 8

affecting these shifts. According to PCA, not all components have equal significance in explaining changes. The first component explains the maximum percentage of the total variance of interest rate changes. The second component is linearly independent (i.e., orthogonal) of the first component and explains the maximum percentage of the remaining variance, the third component is linearly independent (i.e., orthogonal) of the first two components and explains the maximum percentage of the remaining variance, and so on. If yield curve shifts result from a few systematic factors, then only a few principal components can capture yield curve movements. Moreover, since these components are constructed to be independent, they also help in simplifying the task of managing interest rate risk. The principal components with low eigenvalues make little contribution in explaining the interest rate changes, and hence these components can be removed without losing significant information. This not only helps in obtaining a low-dimensional parsimonious model, but also reduces the noise in the data due to unsystematic factors (Nawalkha, Soto and Beliaeva).

Results and Analysis

Table T1 on descriptive statistics shows that the difference between maximum and minimum yields is far higher in the short-term than the long-term. This is because short-term rates are guided by monetary policy rates and liquidity factors. For instance, RBI recently increased short-term borrowing rates to curb the downfall of rupee. In the aftermath of financial crisis in 2007-08, RBI supported the market by infusing huge liquidity along with bringing down policy Repo rate and reserve ratios for the Banks. This helped in lower interest rates at the shorter end but the longer end remained more stable. Similarly, volatility (measured by standard deviation) is far higher in the short-term than in the long-term. Again, this is because shortterm rates are the most affected by monetary policy changes. Hence, they experience greater volatility than long-term rates. These observations can also be observed through Figure C2. Both range and variability of interest rates fall sharply with an increase in maturities. Principal Component Analysis was applied on the different maturities specified earlier from Jan08 to Sep13. The following table illustrates key results obtained.

Table T2: Eigenvalues of the Covariance Matrix Factors F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 Eigenvalue 9.3366 2.1607 0.2662 0.1180 0.0490 0.0427 0.0126 0.0051 0.0042 0.0028 0.0018 0.0002 Difference 7.1760 1.8945 0.1482 0.0690 0.0063 0.0301 0.0075 0.0009 0.0015 0.0010 0.0015 Proportion 0.7781 0.1801 0.0222 0.0098 0.0041 0.0036 0.0011 0.0004 0.0004 0.0002 0.0001 0.0000 Cumulative 0.7781 0.9581 0.9803 0.9901 0.9942 0.9978 0.9988 0.9993 0.9996 0.9999 1.0000 1.0000

Source: Bloomberg, Own Calculations

Table T-2 shows that the first three components account for 98.03% of the total variance of interest rate changes. Component F1 accounts for 77.81% of the variance, while F2 and F3 account for 18.01% and 2.22% respectively. Thus, F1, F2, and F3 combined can be used to estimate the changes in term structure in India with reasonable degree of confidence. Figure C3 shows the impact of these three components on the yield curve.
Figure C3: Impact of F1, F2 and F3 on yield curve 1.200 1.000 0.800 0.600 0.400 0.200 0.000 -0.200 -0.400 -0.600
Source: Bloomberg, Own Calculations

Factor 1 Factor 2 Factor 3 3m 6m 1y 2y 3y 4y 5y 7y 10y 15y 20y 30y

F1 represents a parallel shift in the yield curve. Thus, it is also called the height factor. F2 represents a change in the steepness, and is called the slope factor. F3 10

affects the curvature of the yield curve by inducing a butterfly shift. It is called the curvature factor. Thus, as discussed before, height, slope and curvature explain about 98% of the shifts in yield curve. The height factor dominates the rest and its coefficients are always positive. The slope factor dominates the remaining portion and its coefficients turn from negative to positive with maturity. The curvature factor accounts for the least of the 3 biggest factors. It carries a negative value initially, becomes positive towards the middle, and again becomes negative at the end part of the yield curve.
Table T3: Eigenvectors of 3 principal components F1 3m 6m 1y 2y 3y 4y 5y 7y 10y 15y 20y 30y 0.875 0.891 0.908 0.954 0.983 0.993 0.988 0.981 0.944 0.655 0.641 0.650 F2 -0.451 -0.431 -0.401 -0.278 -0.157 -0.056 0.043 0.117 0.101 0.670 0.741 0.695 F3 0.118 0.098 0.082 0.041 0.008 -0.047 -0.122 -0.059 -0.274 0.322 0.099 -0.159

Source: Bloomberg, Own Calculations

Table T3 shows that for a 10-year bond, a unit increase in factor 1 causes the 10year rate to increase by 0.944%. A unit increase in all 3 factors causes 10-year rate to increase by 0.774%. Figure C4 plots 2 factors against each other based on correlations with yield changes in different maturities. On the left, F1 vs F2 plot shows high correlation of F2 with yields of higher maturity securities. F1 has high correlation with yields of all maturities. On the right, F5 shows hardly any correlation with any security. Similar plots are observed for factors F6 to F12. Hence, this plot further proves the dominance of F1, F2 and F3 as major principal components in the analysis of the sovereign yield curve.

11

Figure C4: Plots of Proportion explained by factors for Different Maturities

Source: Bloomberg, Own Calculations

A scree plot is a plot of the Eigenvalues against the number of factors in order of extraction. Experimental evidence indicates that the point at which the scree begins denotes the true number of factors. In figure C5, one may choose to consider 3 or more factors using the scree plot given. This is consistent with our findings using the eigenvalue criterion.
Figure C5: Scree Plot of Eigenvalues and Cumulative Variability

Scree plot
10 9 100

8
7

80

Eigenvalue

6 5 4 3 2 1 0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12

60

40

20

axis

Source: Bloomberg, Own Calculations

12

Cumulative variability (%)

Conclusion
A wide variety of research has been conducted to determine the factors contributing to changes in yield curve. In this study, we used Principal Component Analysis to identify these factors in the context of the sovereign yield curve of India. Our results are in line with factors suggested by various research studies. Height, slope and curvature are the three biggest factors responsible for changes in the yield curve. Height accounts for 77.81% of the variability, while slope and curvature account for 18.01% and 2.22% respectively. Together these three account for 98.04% of the total variability. Hence, one can estimate yield curve changes with great certainty using information on these 3 changes.

Bibliography
Adrian, T., and H. Wu (2009): The Term Structure of Inflation Expectations, Working Paper, Federal Reserve Bank of New York. Golaka Nath (2012): Estimating term structure changes using principal component analysis in Indian sovereign bond market A Tutorial on Principal Components Analysis (2002): Lindsay I Smith Fontaine, J.-S., and R. Garcia (2011): Bond Liquidity Premia, Review of Financial Studies, forthcoming Buraschi, A., A. Cieslak, and F. Trojani (2010): Correlation Risk and the Term Structure of Interest Rates, Working paper, University of Lugano. Haubrich, J., G. Pennacchi, and P. Ritchken (2011): Estimating Real and Nominal Term Structures using Treasury Yields, Inflation, Inflation Forecasts, and Inflation Swap Rates, Working Paper, Federal Reserve Bank of Cleveland. Ang, A., J. Boivin, S. Dong, and R. Loo-Kung (2010): Monetary Policy Shifts and the Term Structure, Review of Economic Studies, forthcoming.

13