Fixed-Income Return Attribution: Carry, Curve, Spread¶
One-line definition¶
Splitting a bond portfolio's realized return into carry (income earned by holding), curve/duration effect (P&L from rate moves), and spread effect (P&L from credit spread moves) — Brinson's sector/security split doesn't fit fixed income well, since the same bond can be over/underweight and have a rate view and a credit view simultaneously.
Formula¶
No new math — this combines pieces already built elsewhere:
- carry ≈ yield earned over the holding period
- curve effect: key_rate_return_approximation (src/pm/fixed_income/curve.py)
- spread effect: spread_pnl (src/pm/fixed_income/credit.py)
fixed_income_return_decomposition(carry, curve_effect, spread_effect) in
src/pm/attribution.py just labels and totals them for reporting.
Why PMs care¶
A fixed-income portfolio can show a flat total return while carry, curve, and spread are all large and offsetting — knowing which is true changes what you'd do next (e.g. a curve loss offset by carry is very different from a spread loss offset by curve).
Common mistakes¶
- Expecting the three effects to sum exactly to realized return. Carry, curve, and spread are each approximations built on duration measures; the residual is the part the first-order model does not explain, and a large residual is information rather than an error to hide.
- Double-counting carry and rolldown. Rolldown is a price effect from moving along an unchanged curve and belongs in the curve term, not in carry — see Carry and rolldown.
- Attributing a spread move to credit when the benchmark curve itself moved. Spread is measured against a curve; if the curve shifted under it, part of the apparent spread effect is a curve effect.
- Reaching for Brinson instead. Sector/security attribution assumes allocation and selection are separable, which does not hold when one bond carries a rate view and a credit view at once.