Interpolation Techniques Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Why interpolate? Straight line interpolation Cubic spline interpolation Basis spline interpolation
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 2
Why Interpolate
Structuring
Valuation
Project security cash flows Need forward rates on coupon dates Need spot rates on coupon dates
In either case coupon dates may not coincide with dates for which zerocoupon yields are known.
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 3
Interpolation Methods
Straight Line Polynomial
Splined polynomial
Single high order polynomial Unstable between points and at ends Low order polynomials linked together
Basis Splines
Represent discount function as weighted sum of other functions
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 4
Straight Line Interpolation – Pros and Cons
Simple to estimate intermediate points on curve Not accurate for undulating curves Gives different results on discount factors Produces discontinuous forward rate curve
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 5
Linear Interpolation
Intermediate values lie on a straight line between the nearest data points. T2 R2
Ti T1
Ri
R1
Ri = R1 + (R2 - R1 ) x (Ti - T1) / (T2 - T1)
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 6
Linear Interpolation: Rates or Discount Factors?
If interest rates lie on a straight line, discount factors do not Example:
Using Rates R1 = R2 = 5.00% T1 = 90 T2 = 180 Ti = 120 Ri = 5.00%
Copyright © 1996-2006 Investment Analytics
Using DF’s D1 = 0.9877 D2 = 0.9756 Di = 0.9836 Ri = 4.99%
Interpolation Techniques
Slide: 7
Linear and Exponential Interpolation
Linear interpolation on continuously compounded interest rates is equivalent to exponential interpolation on discount factors −R T −R T D1 = e
1 1
, D2 = e
2 2
Ri = (1 − α ) R1 + αR2 Ti − T1 α= T2 − T1 (1−α )
⇒ Di = D1 Copyright © 1996-2006 Investment Analytics
Ti T1
α
D2
Ti T2
Interpolation Techniques
Slide: 8
Cubic Spline Interpolation
A different cubic polynomial is fitted between each pair of data points The polynomials are twice differentiable Ensures that:
The slope of the curve is smooth The rate of change of the slope is smooth The curves “join” at the end points
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 9
Cubic Spline Curve Fitting Ri+1(t)
8.50%
8.00%
7.50%
Ri-1(t)
7.00%
Ri(t) = ai(t-ti)3 + bi (t-ti)2 + ci (t-ti) + di
6.50%
6.00%
5.50%
5.00% 0
200
400
600
800
Copyright © 1996-2006 Investment Analytics
1000
1200
1400
1600
1800
Interpolation Techniques
2000
Slide: 10
Natural Splines
End Curve Conditions
Conditions of the two ends of the yield curve must be specified for a solution.
Natural Spline
Second derivative (rate of change of the slope of the yield curve) equal to zero at both ends.
Slope of curve is constant at the ends
You typically only care about points in the belly of the curve
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 11
Cubic Splines – Pros & Cons
Smooth curve -twice differentiable at every data point Can be used on both rates and DF’s Works for undulating curves Produces continuous forward rate curve Not so easy to calculate Can suffer from oscillation
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 12
Lab: Building Yield Curves with Cubic Splines
Excel workbook; Yield Curve Modeling.xls Worksheet: Cubic Spline Curve Build 3m forward rate curve using:
Linearly interpolated DFs Linearly interpolated spot rates Cubic Spline interpolated spot rates
See Notes & Solution
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 13
Solution: Cubic Spline Forward Curves 8.50% 8.00% 7.50% 7.00% 6.50%
Linear Interp on DF
6.00%
Linear Interp on R 5.50%
Cspline Interp on DF
5.00%
0
500
Copyright © 1996-2006 Investment Analytics
1000
1500
Days
Interpolation Techniques
2000
Slide: 14
Basis Splines
Another widely used interpolation method Used for modeling discount function Typically combined with regression analysis
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 15
Regression: More Payment Dates than Bonds
This is the usual case, as bond coupon dates fall on different days in the year. Have to represent discount factors by a function
Insufficient bonds to estimate model parameters
Singular matrix
Use regression to determine parameters of the discount function
Then calculate discount factors on any chosen date
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 16
Representing the Discount Function by Basis Splines
Represent DF’s by function d(t): L
d (t ) = ∑ α l f l (t ) l =1
Bond prices can be expressed as the sum of discounted cash flows: n
L
j =1
l =1
Pi = ∑ C ij ∑ α l f l (t )
l = 1...L: the number of basis spline functions f. α: weights applied to each function
C: P:
Bond cash flows Bond price
Determine values of weights to fit bond prices to market data.
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 17
Estimating the Discount Function
Rearrange bond price equation:
Sum of discounted cash flows n
L
j =1
l =1
Pi = ∑Cij ∑αl fl (t)
Sum of weighted cashflow x spline function L
n
l =1
j =1
Pi = ∑ α l ∑ C ij f l (t )
Spline functions are defined over the whole period As long as we have more bonds than weighting factors α, regression can be used.
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 18
Basis Splines & Knot Points
Basis Splines
The discount function is a weighted average of a number of overlapping B-Splines. Cubic B-Spline functions usually selected. Individual spline functions are not linked.
Knot Points
Each spline function is non-zero over a welldefined interval. The start and end points of the splines are called “knot points”.
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 19
Basis Spline Curves
The discount function is the weighted sum of individual splines 2.00E-04 1.80E-04 1.60E-04 1.40E-04 1.20E-04 1.00E-04 8.00E-05 6.00E-05 4.00E-05 2.00E-05 0.00E+00 0
500
1000
1500
2000
2500
3000
Knot Points Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 20
Selection of Knot Points
Results can be sensitive to placing of knot points
Unless there is an even distribution of bonds.
Important to have an equal number of bonds with maturities between each knot point.
Reduces estimation error.
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 21
Building a Zero Coupon Curve from Treasury Bonds
Often have more payment dates than bonds
No unique set of discount factors that will price all bonds
Use Regression Analysis
Determine Least Squares Estimates of Discount Factors
Minimize the square of the difference between the observed bond prices and those based on estimated discount factors.
Discount factors must be linked by a functional form
Cubic splines have problems due to correlation. Basis splines are independent but watch “knot points”.
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 22
Lab: Building a Yield Curve with Basis Splines
Worksheet: Basis Splines Build yield curve using bond data Method:
Basis Splines & Regression
See Notes & Solution
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 23
Spot and Forward Rate Curves 7.5%
7.0%
6.5%
6.0%
5.5%
Spot Forw ard
5.0% 0
500
1000
Copyright © 1996-2006 Investment Analytics
1500
2000
Interpolation Techniques
2500
3000
Slide: 24
Confidence Intervals 8% 7% 7% 6% 6% 5% 5%
Spot Rate
4%
Upper 95%
4%
Low er 95%
3% 0
500
1000
Copyright © 1996-2006 Investment Analytics
1500
2000
Interpolation Techniques
2500
3000 Slide: 25
Interpolation Methods: Summary
Straight Line Interpolation
Cubic Splines
Inaccurate. Leads to discontinuous forward rates. Better than linear interpolation. Due to smoothness condition points on the yield curve are linked together. Linking causes multicollinearity. Accuracy of and one discount factor cannot be determined.
Basis Splines
Functions go to zero at defined points. Need to use a weighted combination of several B-Splines.
Copyright © 1996-2006 Investment Analytics
Interpolation Techniques
Slide: 26