Module fit 

Fitting a parametric discount curve to a bond universe.

A smoother, not a repricer. Four parameters over fifty bonds cannot reproduce fifty prices and is not meant to: what it produces is a shape, and what makes the shape judgeable is the residual per bond. So the fit returns both, from one call — a residuals accessor nobody reaches for is how a smoother ships unjudged.

Use the calibration ladder (calibrate) where exactness matters; it fits one parameter per quote and reprices every one.

The Jacobian is seeded

The curve seam substitutes a mode-generic parameter vector, so the sensitivity of every bond’s model price to every parameter comes from one forward pass per bond — no bump size, no truncation error. Parameter counts wider than one seeding block loop over blocks, which is what par_quote_jacobian already does one layer up.

Structs

BondFitRow

One bond’s place in a fit.

FittedBondCurve

A fitted curve and the residuals that say how well it fits.

Enums

FitError

Why a fit produced no curve.

FitWeights

How price residuals are weighted.

FittedForm

Which functional shape a fit reaches for.

FittedFormKind

A parametric form named without its basis count.

Constants

LANES 🔒

The seeding width one pass covers. Matches the calibration Jacobian’s, for the same reason: it is the lane count JetK carries.

Functions

basis 🔒

A bond’s street basis — semiannual, the compounding a fitted government curve’s inputs are quoted on.

fit_bond_curve

Fit a parametric discount curve to quotes — (bond, clean price) pairs — as of reference.

model_clean 🔒

The clean price bond takes on curve, per 100 face, in mode R, where it settles from reference: the flows after settlement carried there.

residuals_and_jacobian 🔒

The weighted residual vector and its Jacobian at one parameter point.