Guide

Key rate duration explained

Harbor Capital’s fixed-income desk reported 5.2 years of modified duration on its $380M government-corporate blend. Risk dashboards showed a comfortable −5.2% price hit if all yields rose 100 basis points in parallel. Over one quarter, the 2-year Treasury yield jumped 85 bp on hawkish Fed guidance while the 30-year yield fell 40 bp on growth fears — a classic curve twist, not a parallel shift. The portfolio lost $18M anyway. Short-maturity paper repriced down; long bonds rallied; the net P&L bore no resemblance to the single duration number on the fact sheet. Modified duration assumed every point on the yield curve moved together. It did not.

Key rate duration (KRD) — also called partial duration or key rate sensitivity — decomposes rate risk by maturity bucket. Instead of one headline duration, you get a vector: how much portfolio value changes when the 2-year key rate moves 1 bp while other key rates are held fixed, then the same for 5-year, 10-year, and so on. KRD is how pension desks, insurers, and relative-value traders hedge shape risk on the curve, not just level risk. This guide defines key rates and partial durations, contrasts parallel vs twist scenarios, covers DV01 bucket hedging with futures and swaps, walks through Harbor Capital’s curve-risk refactor, and provides a technique decision table, pitfalls, and checklist. Pair with bond duration and rate risk for the baseline math and with liability-driven investing when liabilities have their own key-rate profile.

From one duration to many: why modified duration falls short

Modified duration answers one question: if the entire yield curve shifts up or down in parallel by Δy, approximately how much does price change? The first-order approximation is ΔP/P ≈ −D_mod × Δy. That is useful for quick stress tests and comparing two bonds of similar structure, but real markets rarely move in parallel.

Common non-parallel patterns include:

Bear flattening — short rates rise faster than long rates (Fed hiking into late-cycle slowdown).
Bull steepening — long rates fall more than short rates (recession cuts priced at the long end).
Twist — belly and wings move in opposite directions (mid-curve repricing on supply).
Butterfly — the 5–10 year sector moves differently from 2s and 30s.

A portfolio heavy in 3–7 year corporates has most of its cash-flow sensitivity tied to the intermediate key rates, not the 30-year point. A parallel +100 bp shock overstates loss if only the belly moves, and understates it if the front end spikes while the long end is pinned.

Key rates and partial duration defined

A key rate is a benchmark yield on the curve at a chosen maturity node — typically 2, 3, 5, 7, 10, 20, and 30 years for U.S. Treasuries. Key rate duration for bucket k measures the percentage price change of a bond or portfolio when only key rate k shifts by 1 bp and all other key rates are unchanged (with interpolation between nodes holding the curve shape elsewhere fixed).

For a single bond, KRDs sum approximately to modified duration when you shock all key rates in parallel — a sanity check risk systems run nightly. For a portfolio, KRDs are dollar-weighted sums of position-level partial durations. Report them in DV01 (dollar value of a basis point) per bucket for trading desks: “We are +$42k DV01 at the 5-year key rate” means a +1 bp move at that node costs $42,000.

Metric	What moves	Best for
Modified duration	Parallel shift, all tenors	Quick level-risk headline, fund factsheets
Key rate duration	One bucket at a time	Curve twist hedging, relative value
Effective duration	Parallel shift including options	Callable bonds, MBS, funds with prepay
Spread duration	Credit spread, not Treasury curve	Corporate and HY sleeves

Calculating and interpreting KRD vectors

Practitioners use one of three approaches:

Bump-and-reprice — shift each key rate +1 bp, reprice every holding with a fresh curve, divide ΔP by portfolio market value. Most accurate; requires a pricing engine.
Analytic partials — derive from cash-flow sensitivities under a multi-factor yield model. Faster for large books.
Regression on historical moves — infer effective KRDs from realized P&L vs PCA factors on curve changes. Useful for sanity-checking model risk.

Read the vector as a risk fingerprint. A barbell of T-notes and T-bonds shows positive KRD at 2y and 30y with a dip at 10y. A bullet of 7-year IG corporates concentrates risk at the 5–10y nodes. Mismatch between asset KRD and liability KRD is the hidden source of funding-ratio volatility in LDI programs.

Hedging curve shape: futures, swaps, and butterflies

Once you know the KRD vector, you hedge by adding instruments whose partial durations offset yours:

Treasury futures (2y, 5y, 10y, 30y) — liquid DV01 at specific nodes; basis risk vs cash bonds remains.
Interest rate swaps — pay-fixed/receive-float adds duration at the swap tenor; see interest rate swaps for fixed-leg sensitivity.
Butterfly structures — long belly / short wings (or reverse) when you need to neutralize 5y KRD without changing 2y or 30y much.
Options on rates — Swaptions for convexity and tail hedges beyond linear KRD.

Solve a small linear system: find notionals x in hedge instruments such that portfolio KRD + Σ x_i KRD_i ≈ 0 at each targeted node. In practice, desks hedge the largest three DV01 buckets and accept residual risk in minor nodes. Rebalance when KRD drift exceeds tolerance — coupon rolls, new issuance, and spread moves all shift the vector without any trade.

Harbor Capital refactor (worked example)

Harbor’s government-corporate blend was optimized on modified duration vs a 6.5-year liability benchmark. The risk team added a KRD report after the twist quarter loss.

Findings:

+$68k DV01 at 2y (T-bill roll and front-end corporates) — unhedged front-end spike hurt.
+$112k DV01 at 5y (core IG weight) — largest single bucket.
+$24k DV01 at 30y (small T-bond sleeve) — rallied, partially offsetting losses.
Net modified duration 5.2y masked offsetting long and short partials across the curve.

Changes shipped:

Replaced parallel-only stress with three scenarios: parallel +100 bp, bear flatten (+100 bp 2y / +25 bp 30y), bull steepen (−50 bp 2y / +25 bp 30y).
Added 5y and 10y Treasury futures overlays to flatten the 5y DV01 from +112k to +20k within tolerance.
Shifted $45M from 2–3y paper into FRNs to reduce front-end KRD without cutting overall yield target.
Matched liability KRD from actuaries at 5y and 10y nodes within 5% tolerance; reported funding ratio under twist scenarios in monthly board packs.

The next twist quarter (2y +60 bp, 30y −30 bp) produced a −$2.1M mark — within the pre-trade scenario band. Residual basis and spread noise dominated; curve shape was no longer the surprise.

Technique decision table

Approach	Use when	Avoid when
Modified duration only	Short horizon, parallel shocks, retail factsheets	LDI, barbells, twist-heavy macro regimes
Full KRD vector (7+ nodes)	Institutional books, active curve trades	Tiny portfolios where noise exceeds signal
PCA factor durations	Historical risk attribution, factor hedging	When you need tradable node-level DV01
Cash + futures overlay	Liquid government sleeves	Illiquid privates with stale marks
Swaption convexity pack	Large parallel moves expected	Linear twist hedging only (overpaying vol)

Common pitfalls

Assuming KRDs are static — as yields move, partial durations drift; monthly refresh minimum.
Ignoring spread duration — IG corporates have Treasury KRD and credit spread sensitivity; twist hedges do not fix widening.
Overfitting hedge ratios — perfect KRD neutrality on yesterday’s curve breaks on the next twist; leave tolerance bands.
Mixing key-rate conventions — vendor A’s 5y node may not match vendor B’s; reconcile before netting overlays.
Callable and MBS in bump-and-reprice — use effective partials; naive cash-flow bumps misstate prepay response.
Forgetting convexity at long nodes — large moves at 30y need second-order terms; KRD is first-order only.
Reporting duration without curve scenario — boards infer parallel risk from one number; label scenarios explicitly.

Production checklist

Compute KRD vector at standard nodes (2, 5, 10, 30y minimum).
Verify partial durations sum to modified duration under parallel bump.
Express risk in DV01 per bucket for trading and limits.
Run parallel and twist stress scenarios weekly.
Map liability KRD from actuarial or cash-flow models for LDI books.
Select hedge instruments with matching node liquidity (futures vs swaps).
Document basis risk between cash bonds and futures/swap hedges.
Separate Treasury KRD from spread duration on credit holdings.
Refresh KRD after major trades, rolls, and coupon dates.
Backtest P&L attribution: realized vs KRD-implied on past curve moves.
Set DV01 limits per bucket, not only aggregate duration.
Escalate when any single-node DV01 exceeds policy threshold.

Key takeaways

Modified duration is one scenario; KRD is the full picture.
Curve twists dominate many quarters more than parallel shifts.
DV01 by bucket is the language of hedge desks.
LDI match requires asset and liability KRD alignment, not just duration.
Hedge the largest partials; monitor drift and spread risk separately.