Quantum method accurately predicts long-term bond yields from short-rate models

Long-term yields of short-rate models as Feynman-Kac ground states: a variational quantum eigensolver with single-setting energy estimation on two NISQ platforms

Computational Engineering, Finance, and Science

Summary

Predicting long-term bond yields is important for finance but hard with some models. The authors show that these yields can be found by solving a quantum physics problem using a quantum computer technique called a variational quantum eigensolver (VQE). They test their method on popular financial models and real U.S. Treasury data, achieving very close estimates with fewer errors than traditional methods. This shows quantum computing could improve finance calculations even on current noisy devices.

What this means in practice

  • For quantitative finance teams: Compute long-term bond yield predictions more accurately and efficiently using quantum algorithms adapted for noisy quantum hardware.
  • For quantum hardware developers: Test and improve error mitigation methods and measurement strategies on near-term quantum devices using financial model benchmarks.

Authors

Doyeol "David" Ahn

Abstract

We show that the long-term yield of a short-rate model is the ground-state energy of a Schrodinger operator obtained from the Feynman-Kac bond-pricing equation, so that it can be computed by a variational quantum eigensolver (VQE). We treat the Vasicek, Cox-Ingersoll-Ross and Black-Karasinski models; the last has no closed form. Calibrated monthly to the U.S. Treasury curve from 1985 to 2026, the long-term yield follows the 30-year yield with an in-sample correlation of 0.98. The Hamiltonian is stoquastic, so its ground state is positive and the energy can be estimated from computational-basis samples with a single measurement setting. On ibm_cleveland, gate noise appears mainly as bit flips of the high-order qubits. A bit-flip deconvolution with a tail set fixed by a WKB criterion, combined with readout mitigation and zero-noise extrapolation, gives four-qubit estimates within 33-125 bp of the grid ground-state energies for all three models, and within 23-67 bp for Treasury-calibrated parameters. On IonQ Forte Enterprise the same analysis reduces the error in the same proportions. On the same circuits, a Pauli-measurement VQE deviates by 791-1,129 bp, so the single-setting estimator gives about ten times smaller errors with one setting instead of nine. An on-device VQE converges from 4,939 bp to 130 bp.