GOALCWRU

Inspiration

We were inspired by the idea of taking a problem that is genuinely relevant to insurance and expressing it in a form that is natural for quantum optimization. We wanted to start from real Hartford data and ask a realistic portfolio-construction question: how should an insurer balance expected return, covariance risk, regulatory capital burden, and liquidity when selecting among asset-class sectors? That made the project feel much more meaningful than a generic finance demo.

Challenges

One of the biggest challenges was deciding how to reduce the original 50-asset Hartford workbook into a smaller problem that still preserved the important structure. Early on, we explored more aggressive simplifications. We considered clustering assets and then choosing one representative asset from each sector, which would have made the problem smaller and easier to encode. But that approach raised a major concern: it threw away too much information. A single representative asset could distort the return, risk, capital, or liquidity profile of an entire sector, especially if that sector was internally diverse.

That led us to rethink the modeling choice. Instead of choosing one asset per sector, we settled on an implementation that keeps all assets and aggregates them to the sector level using the scenario data directly. In practice, this meant building a sector-level scenario matrix, computing expected returns and covariance from those aggregated scenarios, and averaging the asset-level capital and liquidity characteristics within each sector. This was more work than using representatives, but it gave us a model that was much more faithful to the original dataset and much easier to justify. It also let us evaluate stress behavior using all 1,200 Hartford scenarios rather than relying only on covariance.

See the GitHub for more info!