Download presentation
Presentation is loading. Please wait.
Published byLilian Hodge Modified over 9 years ago
1
1 Estimating the Term Structure of Interest Rates for Thai Government Bonds: A B-Spline Approach Kant Thamchamrassri February 5, 2006 Nonparametric Econometrics Seminar
2
2 Introduction Interest rate in modern financial theories Fixed income market (bonds and derivative securities) Other market securities (for time discounting) Corporate investment decisions (alternative opportunities and cost of capital) The term structure of interest rates Representing relationship between bond yields and maturities Useful in pricing coupon bonds Introduction
3
3 Bond Pricing Spot rate: Forward rate: P(t) is the price at time t of a zero coupon bond of par value = 1 (also called discount factor) r(t) is the instantaneous spot rate at time t f(t) is the instantaneous forward rate at time t Theoretical Framework
4
4 Bond Price, Spot Rate and Forward Rate Relationship Discount function = price of zero-coupon bond P(t) Forward rate f(t) Spot rate = zero-coupon yield r(t) Theoretical Framework
5
5 Methods for Extracting the Term Structure Simple linear regression Polynomial splines Exponential splines Basis splines (B-splines) Nelson and Siegel (1985) and its variants Bootstrapping and cubic splines Theoretical Framework
6
6 Splines Spline is a statistical technique and a form of a linear non-parametric interpolation method. A k th -order spline is a piecewise polynomial approximation with k-degree polynomials. A yield curve can be estimated using many polynomial splines connected at arbitrary selected points called knot points. Some conditions are applied: continuity and differentiability Theoretical Framework
7
7 B-Splines of Degree Zero Theoretical Framework Recurrence relation
8
8 B-Splines of Degree One Theoretical Framework Simplified to
9
9 B-Splines of Degree Two Theoretical Framework
10
10 B-Splines of higher degrees is the p th spline of k th degree. and are the pre-specified knot values. Theoretical Framework B-Splines of Higher Degrees
11
11 Degree of polynomials (k) Interval of approximation (n) Number of basis functions (p) = n+k Number of knots (n+1+2k) Theoretical Framework B-Splines of Degree Three (k=3)
12
12 B-Splines of Degree Three (k=3) Knot specification [-3, -2, -1, 0, 5, 10, 15, 20, 25, 30] In-sample knots: 0, 5, 10, 15 Out-of-sample knots: -3, -2, -1, 20, 25, 30 Approximation horizon: [0, 15] Approximation intervals (n): 3 Number of knots (n+1+2k) = 10 Number of basis functions (p) = n+k = 6 Theoretical Framework
13
13 B-Spline Basis Functions (k=3) B1B1 B2B2 B3B3 B5B5 B4B4 B6B6 Theoretical Framework
14
14 The Term Structure Fitting Using B-Splines Approximation by curve S λ p are coefficients corresponding to the p th -spline that determines S(t) Bond pricing Q represents bond price C is the cashflow matrix Theoretical Framework
15
15 The Term Structure Fitting Using B-Splines Bond pricing regression Q represents bond price C is the cashflow matrix Theoretical Framework
16
16 The Term Structure Fitting Methodology Bond pricing the price of the coupon bond u is a linear combination of a series of pure discount bond prices t m is the time when the m th coupon or principal payment is made. h u is the number of coupon and principal payments before the maturity date of bond u. y(t m ) is the cashflow paid by bond u at time t m. P(t m ) is the pure discount bond price with a face value of 1 Methodology
17
17 The Term Structure Fitting Methodology Model formulation P(t) is the price at time t of a zero-coupon bond (par value = 1) Spot rate: Forward rate: Methodology
18
18 Discount Fitting Model Bond price Discount function Discount fitting function Restriction Methodology
19
19 Spot Fitting Model Bond price Pure discount bond price Spot function Spot fitting function Methodology
20
20 Forward Fitting Model Bond price Pure discount bond price Forward function Forward fitting function Methodology
21
21 Data & Estimation Setup Trading data on January 13, 2006 from the ThaiBMA 12 treasury-bills and 28 government bonds (LB series) Input: time to maturity, coupon rate, weighted average yield, weighted average price B-Splines of degree k = 1, 2, 3, 4 Approximation intervals n = 1, 2, 3, 4, 5 Knot specification Estimation horizon = 0 – 15 years Within-sample knots are integers (1 to 14) Out-of-sample interval length = horizon/n Methodology
22
22 Indices for Evaluation of Regression Equations Generalized cross validation (GCV) RSS is residual sum of squares k is the degree of B-spline polynomials n is the number of approximation intervals m is sample size Methodology
23
23 Mean integrated squared error (MISE) is the yield curve derived from the B-spline approximation is the ThaiBMA interpolated zero-coupon yield curve Methodology Indices for Evaluation of Regression Equations
24
24 Estimated Results Generalized cross validation (GCV) Mean integrated squared error (MISE) Comparison with the ThaiBMA Empirical Results
25
25 Minimum Values of Generalized Cross Validation (GCV) Empirical Results
26
26 Model Estimation, GCV (k = 3, n = 2) Empirical Results (%)
27
27 Fitted Term Structures of Interest Rates Using Different Fitting Models (k = 3, n = 2) Empirical Results
28
28 Confidence Intervals for Estimated Coefficients (Spot Fitting, k = 3, n = 2) Note. * denotes statistical significance at 1% level. Empirical Results
29
29 Confidence Intervals of Spot Fitting Model (k = 3, n = 2) Empirical Results
30
30 Minimum Values of Mean Integrated Squared Error (MISE) Empirical Results
31
31 Model Estimation, MISE (k = 3, n = 3) Empirical Results (%)
32
32 Fitted Term Structures of Interest Rates Using Different Fitting Models (k = 3, n = 3) Empirical Results
33
33 Confidence Intervals for Estimated Coefficients (Restricted Discount Fitting, k = 3, n = 2) Note. * denotes statistical significance at 1% level. Empirical Results
34
34 Confidence Intervals of Restricted Discount Fitting Model (k = 3, n = 2) Empirical Results
35
35 Fitted term structures: GCV, MISE in Comparison to the ThaiBMA Yield Curve Empirical Results
36
36 Confidence Intervals of Restricted Discount Fitting/ Spot Fitting with ThaiBMA Empirical Results Spot Fitting (GCV) Restricted Discount Fitting (MISE)
37
37 Conclusions Discount fitting can give unbounded term structures at very low maturities. Spot fitting is generally has lower GCV values than forward fitting (at k = 3). Suggested model: spot fitting Suggested B-splines degree = 3 interval = 2 knot position [-22.5 -15 -7.5 0 3 15 22.5 30 37.5] Conclusion
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.