Know these first
- M-B01 Derivatives
Math Atoms
By the end you will be able to
- 1Approximate a curve near a point with its tangent line: the first-order approximation.
- 2Add a second-order term with the second derivative to bend the line into a parabola.
- 3Estimate f(a + Δx) to first and second order, and compare the errors.
- 4Explain why duration and convexity are the first- and second-order terms for a bond price.
Why you need it
A risk manager wants to know what a 1 percentage point jump in rates would do to a portfolio of thousands of bonds, quickly. Two numbers per bond, and one line of algebra, give a very good answer. This atom is that line.
The idea
A line, then a bend
You know a curve well at one point: its value, its slope and how sharply it bends. Can you guess its value a little way off? Think of driving: knowing where you are and your speed, you can guess where you will be in a minute. Knowing whether you are speeding up makes the guess better. Step through three guesses for a bond price curve.
Three guesses, each better than the last
Step 1 of 3- 1
Order 0: nothing changes
A flat line at today's value. Right at the dot, wrong everywhere else.
That is the whole idea of a Taylor expansion: build a simple curve that matches the real one at a point, first in level, then in slope, then in bend. Each extra term is a higher order.
The rules
To measure the bend we need one new tool, the second derivative.
Now build the approximation one order at a time.
For a bond. Put the bond price into the same formula, and two famous names drop out.
The formulas, for reference
The error of a first-order estimate shrinks roughly with , and of a second-order estimate with : halve the step, and they shrink about 4 and 8 times.
Try it
Push the step Δx out in both directions and compare the two guesses with the true value (the open circle). The tangent always sits below a curve that bends upwards, so it overstates the fall and understates the rise. Then switch to the 30-year bond: more bend, so the second-order term matters more.
Try itTaylor overlay
Curve
True value at a + Δx
50.8349
Halve the step and watch the errors: order 1 shrinks about 4 times, order 2 about 8 times.
| Δx | order 1 error | order 2 error |
|---|---|---|
| 2 pt | −1.13719 | 0.087853 |
| 1 pt | −0.294945 | 0.011315 |
| 0.5 pt | −0.075129 | 0.001436 |
Worked examples
1. √102 without a calculator
Step 1 of 4- 1
What we know at a = 100
2. A bond price after a 1 point move
Step 1 of 4- 1
What we know at 5%
100 paid in 10 years, annual compounding: . At 5%: , , .
3. The same, in duration and convexity
Step 1 of 3- 1
The two numbers
Common mistakes
Each line below is a tempting slip. Spot what is wrong, then check.
Reaction
Check what stuck
Question 1 · objective 1
f(a) = 50 and f′(a) = −4. Use a first-order approximation to estimate f(a + 0.5).
Question 2 · objective 2
Same point, and now f″(a) = 6. What is the second-order estimate of f(a + 0.5)?
Question 3 · objective 3
Estimate √26 with a first-order approximation from √25 = 5. (The derivative of √x is 1 ÷ (2√x).)
Question 4 · objective 2
Which term of the expansion captures how the curve bends?
Question 5 · objective 4
In ΔP/P ≈ −DΔy + ½C(Δy)², duration D belongs to which part of the Taylor expansion?
Glossary
- Second derivative
- The derivative of the derivative, written f″(x). It measures how fast the slope changes: how sharply the curve bends.
- Taylor expansion
- An approximation of f near a point a built from f and its derivatives at a: f(a) + f′(a)Δx + ½f″(a)(Δx)² + …
- Order
- The highest power of the step Δx used in an approximation. First order uses Δx, second order adds (Δx)².
- Curvature
- How much a curve bends away from a straight line.
- Parabola
- The U-shaped (or upside-down U) curve of a quadratic, such as y = x².
- Modified duration
- D = −P′/P: the first-order sensitivity of a price to its yield, in percent per unit of yield.
- Convexity
- C = P″/P: the second-order term, measuring how the price-yield curve bends.