Replacing a function by a polynomial
Near a point, a smooth function looks like a polynomial built from its derivatives there. This is the Taylor series: starting at x and moving a small step h, the function changes by its slope times h, plus a correction from its curvature, plus smaller corrections from higher derivatives.
f(x+h)≈f(x)+f′(x)h+21f′′(x)h2+61f′′′(x)h3+⋯(1) Four series are worth knowing near 0: ex=1+x+2x2+6x3+⋯, sinx=x−6x3+⋯, ln(1+x)=x−2x2+3x3−⋯ and 1−x1=1+x+x2+⋯. Truncating after a few terms gives a cheap approximation whose error is about the first missing term.
Finance uses this daily. A bond price P depends on the yield y; to second order ΔP≈−DPΔy+21CP(Δy)2, where D is the modified duration (the slope term) and C the convexity (the curvature term). The same shape appears as delta and gamma for an option: ΔV≈Δ⋅ΔS+21Γ(ΔS)2.
Two approximationsSine: at
x=0.5,
x−6x3=0.5−0.020833=0.479167, while the true
sin0.5=0.479426, an error of about
0.00026. Bond: with
P=100,
D=7,
C=60 and a yield rise of 1% (
Δy=0.01),
ΔP≈−7+21(60)(100)(0.0001)=−7+0.3=−6.7. The convexity term softens the loss.
Only good near the pointA truncated Taylor polynomial is an approximation near the expansion point, and its error grows quickly with the size of the step. A 1% move in yields is fine; a 5% move needs more terms or a full repricing of the bond.
Lab · what's graded
- The series of the exponential15 pts
- Approximating the sine15 pts
- The size of the error15 pts
- A bond price change20 pts
- Which term is the curvature10 pts
- A series by hand15 pts
- Functions and their series10 pts