Skip to content
QuantCalc Tools
Directory

Bond Duration Convexity Math

Par Swap DV01 vs Zero-Rate PV01: Jacobian Curve Transformation Math

Risk engines compute sensitivities to bootstrapped zero-coupon spot rates, whereas trading desks execute hedges in par SOFR OIS swaps; inverting the lower-triangular Jacobian matrix bridges the two.

By Elena Rostova, FRM, CQFReview not recorded
Par Swap DV01 vs Zero-Rate PV01: Jacobian Curve Transformation Math — QuantCalc Tools 2026 quantitative & supervisory data card
QuantCalc Tools (Modified Duration, Convexity, DV01 & Key-Rate Hedging): Lower-triangular bootstrap Jacobian J = dz/ds, coupon tail cancellation, and spline traps.Source: QuantCalc Tools Institutional Research DeskLicense: CC-BY-4.0

Internal pricing engines value exotic swaptions, corporate bond books, and structured notes off a bootstrapped Zero-Coupon Spot Curve $z = (z1, z2, \dots, zn)^\top$, outputting a vector of Zero-Rate PV01s ($\nablaz V = \partial V / \partial zi$). Yet you cannot call an interdealer broker to trade a "10-year zero-coupon swap"—live liquidity sits in Par NY Fed SOFR Index Overnight Index Swaps (OIS) $s = (s1, s2, \dots, sn)^\top$! Converting zero-rate risk into exact par swap hedge notionals requires the Multivariate Chain Rule Jacobian Matrix $J = \partial z / \partial s$, underpinning both Key Rate Duration (KRD) Bucketing and ISDA SIMM v2.6 sensitivity grids. Anchor scalar duration definitions in our Modified Duration, Convexity & DV01 master guide.

Deriving the Lower-Triangular Bootstrap Jacobian Matrix $J = \partial z / \partial s$#

Annual-pay par SOFR OIS swaps satisfy $si \sum{j=1}^{i} \alphaj P(0, Tj) + P(0, Ti) = 1$, where $P(0, Tj) = e^{-zj Tj}$ (or $(1 + zj)^{-Tj}$). Because bootstrapping solves $z1$ from $s1$, then $z2$ from $(s1, s2)$, and generally $zi$ from $(s1, \dots, si)$, changing a long par swap rate $sk$ never alters shorter zero rates $zj$ ($j < k$)—making the Jacobian $J{ij} = \partial zi / \partial sj$ strictly lower-triangular! By the multivariate chain rule: $$\nablas V = \left(\frac{\partial z}{\partial s}\right)^\top \nablaz V = J^\top \nablaz V, \qquad H{\text{notional}} = -\text{diag}(PV01{\text{swap}})^{-1}\,\nabla_s V$$ Look at $J^\top$: because $J$ is lower-triangular, $J^\top$ is upper-triangular, meaning we solve exact par-swap hedge notionals cleanly by backward substitution from the 30-year tenor inward to 1 year!

Worked $3 \times 3$ Jacobian Inversion: Hedging a $100M 3-Year Zero Exposure with 1Y, 2Y, 3Y Par Swaps#

Formula or model from the articleNot independently verified
Worked 3-Node (1Y, 2Y, 3Y Annual OIS) Zero-PV01 to Par-Swap Hedge Transformation
  • Market Par OIS Curve: s1 = 4.00%, s2 = 4.20%, s_3 = 4.50% (Annual ACT/360 = 1.0 accrual for clarity)
  • Bootstrapped Discount Factors & Zero Rates (Annual Compounding):
  • T=1Y: DF1 = 1 / 1.0400 = 0.961538 => z1 = 4.0000%
  • T=2Y: 0.0420*(0.961538 + DF2) + DF2 = 1 => DF2 = 0.920999 => z2 = 4.2042%
  • T=3Y: 0.0450*(0.961538 + 0.920999 + DF3) + DF3 = 1 => DF3 = 0.875977 => z3 = 4.5138%
  • Exact Lower-Triangular Bootstrap Jacobian J = dzi / dsj:
  • [ ds1 ds2 ds_3 ]
  • dz_1 [ 1.0000 0.0000 0.0000 ]
  • dz2 [ -0.0206 1.0210 0.0000 ] <-- Notice dz2/ds_1 = -0.0206 < 0!
  • dz3 [ -0.0155 -0.0315 1.0475 ] <-- Notice dz3/ds2 = -0.0315 < 0 & dz3/ds_3 = 1.0475 > 1!
  • Target Portfolio Risk: Pure $100,000,000 3-Year Liability Zero => Zero-PV01 Vector nabla_z V = [ $0, $0, -$25,144 ]^T / bp
  • Apply Upper-Triangular J^T to get Par-Swap PV01 Vector nablas V = J^T * nablaz V:
  • ParPV011Y = (-0.0155) * (-$25,144) = +$389.73 / bp! (POSITIVE sensitivity to 1Y par swap!)
  • ParPV012Y = (-0.0315) * (-$25,144) = +$792.04 / bp! (POSITIVE sensitivity to 2Y par swap!)
  • ParPV013Y = (+1.0475) * (-$25,144) = -$26,338.34 / bp! (4.75% LARGER than zero PV01!)
  • Sum of Par PV01s = +$389.73 + $792.04 - $26,338.34 = -$25,156.57 / bp (Preserves parallel shift PV01!)

Examine those off-diagonal signs in $J$:

  • Why did a pure 3-year zero liability generate positive $+$389.73/\text{bp}$ and $+$792.04/\text{bp}$ Par-Swap PV01 at 1Y and 2Y? Because when you hedge the 3Y principal by receiving fixed on a 3Y Par Swap, that 3Y Par Swap introduces unwanted 1Y and 2Y fixed coupons! You must pay fixed on small 1Y and 2Y swaps ($-$390/\text{bp}$ and $-$792/\text{bp}$) to strip out the 3Y swap's intermediate coupons!
Lower-triangular zero-to-par Jacobian curve transformation and backward-substitution OIS swap hedge sizing workflow
Lower-triangular zero-to-par Jacobian curve transformation and backward-substitution OIS swap hedge sizing workflowVerified Exhibit

Crucial Production Warning: Cubic Spline Curve Builders Break Lower-Triangularity!#

Watch out for a classic quantitative developer bug when building $J$ numerically:

  • Under piecewise-flat forward or linear-on-log-discount bootstrapping, $J$ is strictly lower-triangular ($J_{ij} = 0$ for $j > i$).
  • Under natural cubic spline or monotone convex (Hagan-West) global curve fitting, bumping the 30-year par swap rate $s{30}$ causes tiny "spline waves" (ringing) back into the 10-year and 5-year zero segments ($J{5,30} \neq 0$)! Institutional derivatives engines therefore compute direct instrument bump-and-rebootstrap Par PV01s ($V(\text{rebootstrap}(s + e_k \cdot 1\text{bp})) - V(s)$), which implicitly captures the exact full-rank Jacobian regardless of spline choice! Test overnight leg mechanics in our SOFR overnight compounding & carry calculator.

Comparing Curve Sensitivity Representations Across Desk & Regulatory Systems#

Sensitivity Coordinate SystemMathematical DefinitionMatrix Sparsity vs Tradable SwapsOff-Diagonal Coupon Feedback?Primary Institutional Consumer
Zero-Coupon Spot PV01 ($\nabla_z V$)$\partial V / \partial z(T_i)$Upper-triangular $J^{-\top}$ neededZero (Pure single-date cashflow)Quant pricing & XVA engines
Instantaneous Forward PV01 ($\nabla_f V$)$\partial V / \partial f(T{i-1}, Ti)$Bidiagonal step mappingStep-forward telescopic sumHJM / Libor Market Model calibration
Par OIS Swap DV01 ($\nabla_s V$)$(\partial z / \partial s)^\top \nabla_z V$Identity ($I_n$) — Direct hedge!Yes (Sign-flipped short tails)Linear rates desks & ISDA SIMM
PCA Factor DV01 ($\nabla_{PC} V$)$E3^\top \nablas V$ (Top 3 eigenvectors)$3 \times n$ orthogonal compressionCompressed into Level/Slope/CurvMacro risk limits & VaR engines
Comparison of diagonal Jacobian multiplier J_ii = dz_i/ds_i across 2Y, 5Y, 10Y, 20Y, and 30Y par swap maturities at 4.5% curve
Comparison of diagonal Jacobian multiplier Jii = dzi/ds_i across 2Y, 5Y, 10Y, 20Y, and 30Y par swap maturities at 4.5% curveVerified Exhibit

Connecting Jacobian Algebra to Key-Rate Tents & PCA Hedging#

That $1.46\times$ diagonal amplification at 30 years explains the exact mathematical root of the negative intermediate Par-KRDs highlighted in Key-Rate Duration tent functions and partial DV01! Compress 12-tenor Par Swap DV01 vectors into 3 tradable macro factors using PCA level, slope, and curvature DV01 hedging, inspect institutional clearing venues in our proprietary trading software directory, and explore the Modified Duration & Key-Rate Hedging Topic Hub.

Frequently asked questions

Why is the diagonal entry dz_30 / ds_30 of the swap Jacobian substantially greater than 1.0 (around 1.46)?

When you raise the 30-year par swap coupon by +1 bp while holding 1Y through 20Y par rates fixed, all first 20 annual coupons still discount at unchanged rates—forcing the final 20Y–30Y zero discount factor to drop extra hard (+1.46 bps) to keep the swap at par!

What happens if a trader uses raw Zero-Rate PV01s directly as Par Swap hedge notionals without Jacobian inversion?

At the 30-year tenor, skipping Jacobian inversion over-hedges 30Y par swap notional and leaves unhedged intermediate coupon exposure across 2Y–20Y, creating up to a 25% curve-steepener hedge error!

Does ISDA SIMM v2.6 require Zero-Rate PV01 or Market Par-Swap PV01 across its 12 interest-rate vertices?

ISDA SIMM specifies market par-rate sensitivities (bumping par OIS andibor/term swap curves by +1 bp at each of the 12 vertices from 2w to 30y), matching live interdealer hedge instruments directly.

Why do modern XVA engines use Adjoint Algorithmic Differentiation (AAD) to compute the swap Jacobian?

Bumping and re-bootstrapping 12 vertices across 10 currencies requires 120 full Monte Carlo re-valuations. AAD propagates sensitivities backward through the bootstrap graph in ~3x the cost of a single valuation!

Publication record

Published
14 November 2026
Updated
14 November 2026
Author
Elena Rostova, FRM, CQF
Review
Review not recorded
Automated fact check
Passed in the stored pipeline record
Source URLs
3 recorded

AI assistance recorded: anthropic/claude-opus-4.6. This field does not establish human review or source verification.

Academic & Institutional Attribution

Cite This Research & Embed Live Data Card

Open Attribution License (CC BY 4.0)

Standard academic & institutional research paper citation format.

Elena Rostova, FRM, CQF. (2026, November 14). Par Swap DV01 vs Zero-Rate PV01: Jacobian Curve Transformation Math. QuantCalc Tools. https://quantcalctools.com/guides/par-swap-dv01-vs-zero-rate-pv01-jacobian-curve-transformation
Explore Bond Duration Convexity Math Hub (20) →

QuantCalc Tools Data, Benchmarks & Contributor Records

Par Swap DV01 vs Zero-Rate PV01 Jacobian Matrix Guide | QuantCalc Tools