What the finance class was teaching

Fragment — recovered from a conversation of 2026-06-15. June 2026.

Thirty years ago, in a college finance class, they drilled one number into me harder than any other: bond duration. I half-remembered it as the point where interest-rate risk and maturity risk get balanced against each other — which, it turns out, is the spirit right and the mechanics a little off. Duration is the present-value-weighted average time to a bond’s cash flows, and the closely related thing, its price sensitivity to a move in yield. For a zero-coupon bond it equals maturity exactly.

But the reason it was the number is the part that matters, and it’s the part that came back to me this week. Duration collapses three separate characteristics of a bond — maturity, coupon, yield — into a single statistic that captures most of what you actually care about. A complicated instrument’s risk has a low effective dimensionality. A couple of numbers carry nearly all the signal. That instinct is one of the most durable truths in quantitative finance, and I sat there all week watching the harness rediscover it from scratch, in a corner of the market nobody had cleanly measured: out of every signal we tested, DSCR and the maturity wall were the only two that survived.

And the mapping is cleaner than a coincidence. Any debt instrument’s risk decomposes onto two axes. Will it pay? — credit. When, and at what rate, does it roll? — duration, refinancing. My two survivors land exactly on those axes: DSCR is the credit axis (can the property’s income carry the debt), and the maturity wall is the duration axis (when does the balloon come due, and into what rate environment). The harness didn’t find “two factors because duration is one factor.” It rediscovered, empirically, that CRE distress lives on the same two axes fixed income has always said govern any debt.

There’s a twist that makes it sharper. Classic duration is a Treasury-world idea — it assumes the bond pays and asks only what it’s worth if rates move; it deliberately excludes default. So my two factors aren’t duration. They’re duration’s world plus the one thing duration is built to leave out. Which is exactly the right pair for an instrument that can actually default. And one of them is closed-form math you compute from a schedule; the other is a measurement I discovered by testing. A theorem and a regression pointing at the same two axes — that convergence is its own kind of validation.

The field keeps making this discovery in different clothes. Macaulay’s duration in 1938, one factor. Litterman–Scheinkman: the entire yield curve is level, slope, curvature — three numbers. Fama–French: market, size, value. My harness: DSCR and maturity. Every one of them the same finding — the effective dimension is far below the feature count.

And the part that actually stopped me: duration is computed entirely from the indenture, the bond’s legal disclosure document. DSCR and maturity come entirely from the ABS-EE tape, the loan’s legal disclosure document. In both cases the sufficient statistic lives in what the instrument is required to say about itself. My college finance class was teaching me the data thesis thirty years before I wrote it down. I just didn’t have the standpoint yet to hear it as anything but an exam question.