Key Rate Duration: Locating Curve Exposure

Ask where the rate change occurs

Two portfolios can have similar overall duration while reacting differently when short rates rise and long rates fall. One may concentrate its present value around intermediate payments; another may combine early and distant receipts. Their shared headline duration describes a particular aggregate movement, not every possible reshaping. The useful next question is which parts of the discounting curve affect each collection of payments.

CFA Institute describes key rate durations as a way to identify sensitivity to changes in curve shape. That distinction matters even for a single coupon bond: the final principal payment is not its only exposure. Earlier coupons are discounted over shorter horizons. Assigning the entire bond to its maturity bucket can therefore hide information that a payment-level valuation would preserve.

Define the curve before defining its nodes

A key rate can mean a spot rate, a par rate or another specified curve input. These are not interchangeable coordinates. Bumping a par input and rebuilding the discount curve can alter several discount factors, while bumping a spot node follows a different mapping. A sensitivity report should name the coordinates before comparing its numbers with another report or adding them across holdings.

The ECB's methodology distinguishes estimated zero-coupon curves from derived forward and par curves. That provides a useful reporting discipline: record which curve object actually enters valuation. Also record the maturity nodes and behavior between them. A node at five years does not, by itself, tell a reader whether nearby four-year and six-year discount rates move with it, remain fixed, or change through recalibration.

Understand the local bump calculation

For node i, a common central-difference definition is KRD_i = (P_down - P_up) / (2 * P_base * h). Here h is the annual rate change in decimal units, and both shocked prices come from the same stated bump construction. A one-basis-point bump means h = 0.0001, not 0.01. The denominator uses the unshocked positive asset value, with consistent currency and price units.

This convention produces positive sensitivity when a rate increase lowers value. The corresponding first-order estimate is change in price divided by price approximately equal to minus the sum of KRD_i times each node's rate change. The minus sign belongs in that relationship even though reported durations are commonly positive. Record whether another system instead reports signed dollar derivatives before reconciling its output.

Trace an illustrative two-node scenario

Take a hypothetical USD 100,000 asset value and assumed key rate sensitivities of 1.2 years at the short node and 3.8 years at the long node. These are teaching inputs, not calibrated exposures. Let the short node rise by 25 basis points and the long node fall by 10 basis points. Hold cash flows and all other factors fixed for the approximation.

The short-node contribution is -1.2 * 0.0025 = -0.0030, or -0.30%. The long-node contribution is -3.8 * (-0.0010) = +0.0038, or +0.38%. Together they imply +0.0008, or +0.08%, equal to approximately +USD 80. Independently adding the dollar contributions, -USD 300 and +USD 380, gives the same result. This is a linear estimate, not an exact price forecast.

Make aggregation conditional rather than automatic

Key rate durations can approximately sum to a parallel curve duration when their bump shapes collectively reproduce the same parallel movement. That is a construction property, not a universal identity. Gaps between nodes, different endpoint rules or nonlinear rebuilding can break the comparison. A reported mismatch should first prompt a review of definitions before it is dismissed as harmless rounding or treated as a valuation error.

Portfolio aggregation also needs a common denominator. Market-value weighting combines proportional sensitivities when holdings use compatible curves, currencies and conventions. Dollar sensitivities can be added directly only within an appropriate common valuation basis. Netting a short exposure against a long exposure may reduce the total while leaving substantial gross exposure at separate nodes; the cancellation does not remove the underlying dependence on curve shape.

Challenge the approximation with targeted changes

A useful review compares the linear estimate with exact repricing under the identical specified curve shock. Small differences may reflect second-order effects; large differences can reveal a unit error, inconsistent rebuilding or a shock outside the local approximation's useful range. Testing only parallel moves leaves the distinctive purpose of key rate duration unexamined. Include a twist that moves separated nodes in opposite directions.

Bump-size checks should preserve the same curve definition while varying the numerical step. Very small steps can expose rounding noise; large steps can measure curvature as well as slope. Do not select the step that makes an inconvenient result disappear. Preserve both the sensitivity and the scenario residual so a reader can distinguish a useful local description from a complete explanation of the price change.

Respect the regular-bond calculator boundary

A calculator discounting every payment at one annual nominal yield cannot independently move curve nodes. Its modified duration concerns that one yield input, even if a neighboring article discusses broader curve risk. Constructing genuine key rate durations requires a term structure, an explicit bump map and a repricing method that consumes the rebuilt curve. Renaming single-yield duration does not supply those missing ingredients.

Use this guide to interpret an exposure report or design an educational calculation, not to infer a recommended hedge. The parent tools remain confined to regular noncallable fixed-coupon bullet bonds at a coupon date. Their supported annual, semiannual and quarterly payment frequencies do not create curve construction, optionality or default modeling. A clear boundary is part of the meaning of the result, not an incidental disclaimer.

Primary references

1. CFA Institute: Yield Curve Strategies
2. ECB Data Portal: Yield Curves Methodology