Proxy bonds cut stress-test errors by 72–95 percent
Proxy bonds reduce valuation errors by 72–95 percent versus traditional methods in Bank of Spain research using U.S. Treasury yield curves from 1985–2025. Aggregate portfolio characteristics suffice, reducing disclosure requirements.
Two cash flows bound losses
Alexandra Matyunina, Andrey Pankratov and Federico Severino construct proxy bonds matching market value and selected risk characteristics.
Their bullet method uses duration and produces a lower price bound.
Upper and lower barbells each use two cash flows to match duration and convexity; the lower barbell also requires maturity information.
Under synchronous yield changes, the barbells deliver the tightest possible price bounds given value, duration, convexity and a maturity cap.
Their bounding roles reverse when yields fall.
A 50-50 average barbell supplies a midpoint estimate, not a guaranteed conservative valuation.
In a 30-year bond example, a 300 basis point increase produces 98.7 percent less valuation error than the duration method.
Less reporting, smaller errors
The authors test a simulated 50-bond portfolio against 180-day Treasury yield-curve changes sampled every 90 days over 1985–2025.
With whole-portfolio data, average-barbell root mean squared error is 0.287 percent, versus 1.257 percent for duration and 1.205 percent for duration–convexity.
Three maturity brackets reduce it to 0.217 percent; finer reporting adds little accuracy.
In parallel-shock tests up to 500 basis points in magnitude, four proxy cash flows replace 4,261, with errors no greater than 0.38 percent.
Bounds have their boundaries
The portfolio bounds require synchronous rate changes and non-negative weights.
For non-parallel scenarios, the reported accuracy is empirical rather than a universal bounding guarantee.
The authors identify large slope changes, curvature changes and extremely steep initial curves as sources of larger errors.