Finance-Theory Β· Unit 7 Β· Video 6 Β· Interactive Practice

Convexity: The Bend That Duration Misses

IKey Formulas

FormulaNameWhat it says
Vm=1Pβˆ‚2Pβˆ‚y2=1(1+y)2βˆ‘k=1Tk(k+1)CkP (1+y)kV_m = \dfrac{1}{P}\dfrac{\partial^2 P}{\partial y^2} = \dfrac{1}{(1+y)^2}\displaystyle\sum_{k=1}^{T} k(k+1)\dfrac{C_k}{P\,(1+y)^k}ConvexityThe second derivative per unit of price, with yy and kk per payment period. For a semiannual bond quoted in years, discount at y/2y/2 per half-year and divide each k(k+1)k(k+1) by 4=224 = 2^2
P(yβ€²)β‰ˆP(y)[1βˆ’Dmβˆ—(yβ€²βˆ’y)+12Vm(yβ€²βˆ’y)2]P(y') \approx P(y)\left[1 - D_m^*(y'-y) + \tfrac12 V_m (y'-y)^2\right]Second-order Taylor approximationDuration supplies the term linear in the yield change, convexity the quadratic term
Dmβˆ—(P)=βˆ‘jPjPDm,jβˆ—,Vm(P)=βˆ‘jPjPVm,jD_m^*(P) = \displaystyle\sum_j \frac{P_j}{P} D_{m,j}^*, \quad V_m(P) = \sum_j \frac{P_j}{P} V_{m,j}PortfolioValue-weighted averages of the bonds' measures, assuming every yield moves by the same amount
E ⁣[Ξ”PP]β‰ˆ12Vm σ2E\!\left[\dfrac{\Delta P}{P}\right] \approx \tfrac12 V_m\,\sigma^2 when E[yβ€²βˆ’y]=0E[y'-y] = 0VolatilityThe linear term averages to zero; the quadratic term adds half the convexity times the variance of the yield change

Key Insight: the video's bond (4 years, 6% coupon paid twice a year, priced at par, every yield an annual rate compounded semiannually, so 3% per half-year) has Dmβˆ—=3.509846D_m^* = 3.509846 and Vm=14.805972V_m = 14.805972. For a rise from 6% to 8%, yβ€²βˆ’y=0.02y'-y = 0.02 and the two-term estimate is 100(1βˆ’0.0701969+0.0029612)β‰ˆ93.276427100(1 - 0.0701969 + 0.0029612) \approx 93.276427 against the exact 93.26725593.267255 (4% per half-year): a miss under 0.010.01. Duration alone gives 92.98030892.980308, almost 0.290.29 too low.

IITwo Terms Track the Curve

Duration's straight line drifts away from the exact price; the convexity term shrinks the miss to almost nothing.

IIIEqual Swings, Unequal Price Changes

For equal moves in yield, the gain from a fall beats the loss from a rise, so volatility alone raises the expected price.

IVPortfolio Convexity Is a Weighted Average

Mixing two zeros averages their durations and convexities, so the mix sits above the single zero of equal duration.

πŸ’‘ Both averages assume every bond's yield moves by the same amount. Investors know convexity is valuable, so a position with more convexity than another of the same duration trades at a lower yield, meaning a higher price: you pay for convexity when you buy it.

VQuiz Questions

Problem 1 Β· Apply the Two-Term Formula

Given: a bond priced at 100100 has Dmβˆ—=5D_m^* = 5 and Vm=40V_m = 40. Its yield rises by one percentage point, yβ€²βˆ’y=0.01y' - y = 0.01. Find the second-order estimate of the new price.

βœ… Correct! 100(1βˆ’0.05+0.002)=95.20100(1 - 0.05 + 0.002) = 95.20: duration takes off 5 points and convexity gives back 0.2.
❌ That is duration alone. 100(1βˆ’5Γ—0.01)=95100(1 - 5 \times 0.01) = 95 drops the convexity term 12Γ—40Γ—0.012=0.002\tfrac12 \times 40 \times 0.01^2 = 0.002.
❌ Check the sign of the quadratic term. 12Vm(yβ€²βˆ’y)2\tfrac12 V_m (y'-y)^2 is added: convexity is positive and the change is squared.
❌ Missing the one half. The Taylor term is 12Vm(yβ€²βˆ’y)2=0.002\tfrac12 V_m (y'-y)^2 = 0.002, not Vm(yβ€²βˆ’y)2=0.004V_m(y'-y)^2 = 0.004.
❌ Not quite. Use P(yβ€²)β‰ˆP(y)[1βˆ’Dmβˆ—(yβ€²βˆ’y)+12Vm(yβ€²βˆ’y)2]P(y') \approx P(y)\left[1 - D_m^*(y'-y) + \tfrac12 V_m (y'-y)^2\right].
Show solution

Linear term: Dmβˆ—(yβ€²βˆ’y)=5Γ—0.01=0.05D_m^*(y'-y) = 5 \times 0.01 = 0.05. Quadratic term: 12Vm(yβ€²βˆ’y)2=12Γ—40Γ—0.0001=0.002\tfrac12 V_m (y'-y)^2 = \tfrac12 \times 40 \times 0.0001 = 0.002.

P(yβ€²)β‰ˆ100 (1βˆ’0.05+0.002)=100Γ—0.952=95.20P(y') \approx 100\,(1 - 0.05 + 0.002) = 100 \times 0.952 = 95.20

Duration alone would predict 95.0095.00; the curve bends back above the tangent line, so the convexity term raises the estimate.

Problem 2 Β· When the Yield Falls

Given: the video's bond at 6% (price 100100, Dmβˆ—=3.509846D_m^* = 3.509846, Vm=14.805972V_m = 14.805972, yields compounded semiannually). The yield falls to 4%. Find the second-order estimate of the new price.

βœ… Correct! With yβ€²βˆ’y=βˆ’0.02y'-y = -0.02 both terms raise the price: 100(1+0.0701969+0.0029612)β‰ˆ107.3158100(1 + 0.0701969 + 0.0029612) \approx 107.3158, within 0.010.01 of the exact 107.3255107.3255.
❌ The quadratic term does not change sign. (yβ€²βˆ’y)2=(βˆ’0.02)2=0.0004(y'-y)^2 = (-0.02)^2 = 0.0004 is positive whichever way the yield moves, so convexity adds 0.00296120.0029612 here too.
❌ That is duration alone. The duration line understates the price on both sides; add 12Γ—14.805972Γ—0.0004=0.0029612\tfrac12 \times 14.805972 \times 0.0004 = 0.0029612.
❌ That is the estimate for a rise to 8%. Here yβ€²βˆ’y=βˆ’0.02y'-y = -0.02, so βˆ’Dmβˆ—(yβ€²βˆ’y)=+0.0701969-D_m^*(y'-y) = +0.0701969 and the price goes up.
❌ Not quite. Substitute yβ€²βˆ’y=βˆ’0.02y'-y = -0.02 into the two-term formula.
Show solution

Linear term: βˆ’Dmβˆ—(yβ€²βˆ’y)=βˆ’3.509846Γ—(βˆ’0.02)=+0.0701969-D_m^*(y'-y) = -3.509846 \times (-0.02) = +0.0701969. Quadratic term: 12Γ—14.805972Γ—(βˆ’0.02)2=+0.0029612\tfrac12 \times 14.805972 \times (-0.02)^2 = +0.0029612.

P(0.04)β‰ˆ100 (1+0.0701969+0.0029612)=107.31581P(0.04) \approx 100\,(1 + 0.0701969 + 0.0029612) = 107.31581

The exact price, discounting every cash flow at 2% per half-year, is 107.325481107.325481: the estimate is 0.00970.0097 low. Duration alone gives 107.0197107.0197, off by 0.3060.306. Against a rise to 8% the gain of 7.337.33 beats the loss of 6.736.73, while the duration line predicts 7.027.02 each way.

Problem 3 Β· A Two-Bond Portfolio

Given: a portfolio holds $60 million of bond A (Dmβˆ—=2.0D_m^* = 2.0, Vm=6.0V_m = 6.0) and $40 million of bond B (Dmβˆ—=9.0D_m^* = 9.0, Vm=95.0V_m = 95.0); both yields move by the same amount. Find the portfolio's modified duration, its convexity, and the estimated change in its value if yields rise by 50 basis points.

Portfolio modified duration Dmβˆ—(P)D_m^*(P)?

Portfolio convexity Vm(P)V_m(P)?

Estimated change in value for yβ€²βˆ’y=0.005y' - y = 0.005?

βœ… Correct! Weights 0.60.6 and 0.40.4 give Dmβˆ—(P)=4.8D_m^*(P) = 4.8 and Vm(P)=41.6V_m(P) = 41.6, so the value falls by about 2.348%2.348\% of $100 million.
❌ The weights are value shares, not one half each. Bond A is 60/100=0.660/100 = 0.6 of the portfolio's value and bond B is 0.40.4.
❌ The weights are swapped. The larger holding, bond A, carries weight 0.60.6.
❌ That is duration alone. βˆ’4.8Γ—0.005=βˆ’0.024-4.8 \times 0.005 = -0.024 leaves out the convexity term 12Γ—41.6Γ—0.0052=0.00052\tfrac12 \times 41.6 \times 0.005^2 = 0.00052.
❌ Check the sign of the quadratic term. The convexity term is added, which makes the loss smaller than duration predicts, not larger.
❌ Missing the one half. The quadratic term is 12Vm(yβ€²βˆ’y)2=0.00052\tfrac12 V_m (y'-y)^2 = 0.00052, not 0.001040.00104.
❌ Check the duration. Dmβˆ—(P)=βˆ‘jPjPDm,jβˆ—D_m^*(P) = \sum_j \frac{P_j}{P} D_{m,j}^* is a value-weighted average, never a sum.
❌ Check the convexity. Vm(P)=βˆ‘jPjPVm,jV_m(P) = \sum_j \frac{P_j}{P} V_{m,j} is a value-weighted average, never a sum.
❌ Check the change. Use Ξ”PPβ‰ˆβˆ’Dmβˆ—(P)(yβ€²βˆ’y)+12Vm(P)(yβ€²βˆ’y)2\frac{\Delta P}{P} \approx -D_m^*(P)(y'-y) + \tfrac12 V_m(P)(y'-y)^2 with the portfolio's own measures.
Show solution

The portfolio is worth $100 million, so the value weights are 60100=0.6\frac{60}{100} = 0.6 for A and 40100=0.4\frac{40}{100} = 0.4 for B.

Dmβˆ—(P)=0.6Γ—2.0+0.4Γ—9.0=1.2+3.6=4.8D_m^*(P) = 0.6 \times 2.0 + 0.4 \times 9.0 = 1.2 + 3.6 = 4.8 Vm(P)=0.6Γ—6.0+0.4Γ—95.0=3.6+38.0=41.6V_m(P) = 0.6 \times 6.0 + 0.4 \times 95.0 = 3.6 + 38.0 = 41.6 Ξ”PPβ‰ˆβˆ’4.8Γ—0.005+12Γ—41.6Γ—0.0052=βˆ’0.024+0.00052=βˆ’0.02348\frac{\Delta P}{P} \approx -4.8 \times 0.005 + \tfrac12 \times 41.6 \times 0.005^2 = -0.024 + 0.00052 = -0.02348

The value falls by about 0.02348Γ—100=2.3480.02348 \times 100 = 2.348 million dollars: βˆ’$2.348 million. Duration alone would say βˆ’$2.400 million.

Problem 4 Β· What Volatility Is Worth

Given: the video's bond at 6% (price 100100, Dmβˆ—=3.509846D_m^* = 3.509846, Vm=14.805972V_m = 14.805972). Rates grow more volatile: tomorrow the yield will be either 4% or 8%, each with probability 12\tfrac12. Find the expected price according to the second-order approximation.

βœ… Correct! E[yβ€²βˆ’y]=0E[y'-y] = 0 and E[(yβ€²βˆ’y)2]=0.0004E[(y'-y)^2] = 0.0004, so E[P]β‰ˆ100(1+12Γ—14.805972Γ—0.0004)β‰ˆ100.30E[P] \approx 100(1 + \tfrac12 \times 14.805972 \times 0.0004) \approx 100.30. The exact average, (107.325481+93.267255)/2=100.296368(107.325481 + 93.267255)/2 = 100.296368, agrees.
❌ Only the linear terms cancel. βˆ’Dmβˆ—(yβ€²βˆ’y)-D_m^*(y'-y) averages to zero, but 12Vm(yβ€²βˆ’y)2\tfrac12 V_m (y'-y)^2 is positive in both outcomes.
❌ Missing the one half. The expected gain is 12VmΟƒ2\tfrac12 V_m \sigma^2, not VmΟƒ2V_m \sigma^2.
❌ Volatility helps the bondholder here. The quadratic term is added, so the expected price rises above 100100.
❌ Not quite. Average the two-term formula over the two outcomes.
Show solution

The yield change is +0.02+0.02 or βˆ’0.02-0.02 with equal probability, so E[yβ€²βˆ’y]=0E[y'-y] = 0 and Οƒ2=E[(yβ€²βˆ’y)2]=0.022=0.0004\sigma^2 = E[(y'-y)^2] = 0.02^2 = 0.0004.

E ⁣[Ξ”PP]β‰ˆβˆ’Dmβˆ—β‹…0+12VmΟƒ2=12Γ—14.805972Γ—0.0004=0.0029612E\!\left[\frac{\Delta P}{P}\right] \approx -D_m^* \cdot 0 + \tfrac12 V_m \sigma^2 = \tfrac12 \times 14.805972 \times 0.0004 = 0.0029612

So E[P]β‰ˆ100Γ—1.0029612β‰ˆ100.30E[P] \approx 100 \times 1.0029612 \approx 100.30. Checking against exact prices: (107.325481+93.267255)/2=100.296368(107.325481 + 93.267255)/2 = 100.296368. Other things equal, the wider swing raises the bond's expected price; in the market, investors pay for that convexity up front.

Solved: 0 / 4