Fill probability estimates in institutional bond trading with quantum computers (arxiv.org)

Researchers report that applying quantum-computed data transforms to production-scale intraday trade events can materially improve machine-learning estimates of fill probability for institutional corporate-bond orders. Using a quantum algorithm run on IBM Heron processors (and comparing to noiseless quantum simulators), the team integrated quantum-generated transforms as a decoupled offline component that models can query in low-latency trading workflows. Backtested on real trade event streams, models fed the hardware-transformed data achieved up to ~34% relative gains in out-of-sample test scores versus models using raw data or simulator-based transforms. The result is significant because it demonstrates a practical, near-term role for quantum computing in quantitative finance: not as a drop-in replacement for classical models, but as a complementary feature-engineering or data-transformation tool that can enrich inputs for statistical learning algorithms. Key technical takeaways include the decoupled-query architecture (minimizing latency impact), the use of production-scale intraday datasets, and an unexpected finding that current quantum hardware noise may be contributing to performance gains—suggesting noise-induced regularization or unique stochastic embeddings. The paper is empirical rather than theoretical proof of advantage; it calls for follow-up work on reproducibility, robustness across asset classes, comparisons with classical randomized transforms, and analyses of why noisy hardware helps.