Holding a long Treasury bond pays in three ways when the yield curve is steep: the coupon, the roll-down as the bond ages into a lower-yielding part of the curve, and the term premium that investors demand for carrying duration. When the curve is flat or inverted, those payments shrink or turn negative, and short duration is the safer place to be.
The bet is that the slope of the curve, 10-year minus 3-month yield, tells when to hold duration and when to step away from it. The idea comes from the academic carry literature (Koijen, Moskowitz, Pedersen and Vrugt) and from work on return predictability (Cochrane). Nobody is forced to trade; the claim is a risk premium that the slope makes visible.
How it was tested
Pre-registered on 2026-05-25. The slope was evaluated at the month-end close and traded at the next open, across three Treasury ETFs: TLT (20+ years), IEF (7-10 years) and SHY (1-3 years). Five variants: a TLT/SHY switch at two slope thresholds rebalanced monthly, the same switch quarterly, a three-way rank rotation, and TLT or cash. No shorts. In-sample 2014-2020, out-of-sample 2021-2026, with five variants requiring t > 2.57.
Where it died
Rejected the same day it was registered. The average out-of-sample return per holding period was negative in every variant, so 0 of 5 passed the raw gate. Against SPY, with beta assumed equal to 1 as the instrument of the time did, the result was worse than insignificant: the best variant had an alpha t of −2.09 and the worst −3.68. The rule did what it was built to do; the out-of-sample window simply opened just before a rate cycle.
The record
Thesis
The slope of the Treasury curve (10-year minus 3-month yield) predicts excess returns on long-duration bonds: when the curve is steep, coupon plus roll-down plus term premium pay for holding duration; when it is flat or inverted, step down to short duration.
Koijen, R., Moskowitz, T., Pedersen, L. & Vrugt, E. (2018). Carry. Journal of Financial Economics.
Cochrane, J. (2007). The Dog That Did Not Bark: A Defense of Return Predictability. Review of Financial Studies.
Specification
signal
slope = 10-year yield − 3-month yield, evaluated at month-end close
Measured with beta assumed = 1 (the instrument of the time). Every variant is significantly negative, not merely insignificant.
Why it died
The signal said "steep curve, hold long duration" in 2021 — just before the Fed took its funds rate from 0% to 5.5% in about 18 months and TLT fell 47% peak to trough. By the time the curve inverted and the signal rotated to short duration, the damage was done. Every variant lost money out of sample, and the worst reached alpha t −3.68.
What it taught
A slope signal is slow; a rate cycle is fast. The rule reacted to the level and shape of the curve, with no defence against the speed of rate rises. The pre-registered out-of-sample window happened to contain the worst regime in 40 years for duration — extending the backtest to earlier decades to soften that result was explicitly declined as post-hoc.