Fixed income lab · limits

Where the model is wrong

A model is a set of assumptions. Knowing where they fail matters as much as the formulas, so here they are, roughly in order of how much they matter for the numbers on screen.

The curve is a fit, not trades

Large effect

Par yields read off the Treasury's own curve

The published yields are not prices of real bonds. They are points on a curve the Treasury fits to recently auctioned securities, quoted at fixed maturities no bond actually has. The lab treats each one as a par bond issued today, so it prices a hypothetical bond exactly and a real one only approximately. A real Treasury with an odd coupon or maturity trades away from this curve by a few basis points, more in stressed markets.

The fix: build the curve from traded prices of individual Treasuries (CUSIP-level data).

Moderate effect

No on-the-run premium, no liquidity effects

The newest issue at each maturity usually trades richer than older ones, because it is the easiest to trade and to borrow in repo. The fitted curve smooths that over, and the lab does not model it.

Small effect

No extrapolation past 30 years

A bond maturing after the curve's last published maturity is refused instead of priced off an extrapolated rate. On days without a 30-year yield (2002 to early 2006), that limit is 20 years.

Corporates have no credit curve

Large effect

Treasury curve plus one flat spread

A corporate bond is priced as the Treasury curve plus one spread added to every maturity. Real credit spreads have a term structure of their own, depend on the issuer's default risk and expected recovery, and widen sharply in a crisis. The Z-spread the lab reports is a summary number, not a credit model: it does not tell you the chance of default.

The fix: a credit curve per issuer or rating, from CDS or a set of the issuer's bonds.

Risk numbers are local

Moderate effect

Durations and key-rate DV01s hold for small moves

DV01, duration and key-rate durations are first derivatives: they describe a 1 bp move. For moves of 50 or 100 bp, convexity matters, and the duration estimate misses. The Basics area's tangent model shows by how much: −32.6% for a 400 bp move on a 50-year bond.

Key-rate durations also depend on the interpolation scheme. With a non-local scheme such as a cubic spline, moving one pillar moves the curve elsewhere too, so the same bond can show a different key-rate profile under a different scheme.

Idealised hedges

Moderate effect

Par bonds issued today, no costs

The hedge instruments are Treasuries issued at par on the curve date, maturing exactly at each pillar. Real hedges use the bonds that exist, often futures, with bid-ask spreads, repo financing and margin. The sizes are exact fractions of a dollar, while trades come in round lots.

A hedge holds only for the curve it was computed on. As rates move and time passes the exposures drift, and the hedge has to be rebalanced.

The rate tree is simple

Large effect

One volatility, annual steps, recombining rates

The Kalotay-Williams-Fabozzi tree uses a single volatility for every year and every rate level, with one step per year. Real rate volatility differs by maturity and rate level, and calls are exercised on specific dates with notice periods, not once a year. The callable bond's value is sensitive to the volatility you choose, and the lab takes it as an input, not something fitted to swaption prices.

The fix: calibrate the volatility to market option prices, use finer steps, or use a model with mean reversion such as Hull-White.

The convertible has no credit risk

Large effect

Everything discounted at the risk-free rate

The coursework model discounts the whole convertible at the risk-free rate, as if the issuer could not default. Convertible issuers are usually riskier companies. Desk models split the value into an equity part discounted at the risk-free rate and a debt part discounted at a credit-adjusted rate (Tsiveriotis and Fernandes, 1998). The model also has no issuer call, no put, no dilution, and a constant stock volatility.

The coursework models are textbook

Small effect

Whole periods and 30/360

The Basics models price on coupon dates only (whole periods, no accrued interest) with 30/360 day counts, as the coursework did. They teach the shape of each effect. For a real bond on a real settlement date, use Price a bond.

Data timing

Small effect

One curve a day, at the close

The Treasury publishes one curve per business day, late in the afternoon New York time. Nothing in the lab moves during the day. A price from it is an end-of-day estimate, not a quote you could trade on.