Summary
Generating correlation matrices is hard because they must follow strict mathematical rules and look the same no matter how you reorder their rows and columns. The authors created CorrFlow, which first transforms these matrices into a simpler form where all valid outputs are guaranteed. Then it trains a special neural network that respects the reordering symmetry to generate realistic examples. Testing on financial data shows CorrFlow produces valid matrices every time and fits the data better than competing methods. However, the authors note these results depend on their specific training setup and don’t prove it’s always better than other approaches.
What this means in practice
- •For financial data teams: Generate fully valid synthetic correlation matrices for portfolio risk modeling without manual correction steps.
- •For machine learning engineers: Train models that produce permutation-equivariant outputs for data with inherently reorderable structure, improving generative accuracy.
Tested on one dataset.
Abstract
Correlation matrices combine nonlinear positive-definiteness constraints with a simultaneous row-column permutation symmetry. Their generation therefore involves two distinct problems. First, every output must remain on the correlation elliptope, irrespective of how its variables are ordered. Second, the generator must learn the domain-specific distribution occupied by empirical data inside that valid space. Existing financial image generators mix these problems: they learn ordered matrix entries and repair invalid draws afterward. We introduce CorrFlow, which separates them. A generalized Fisher transform maps full-rank correlation matrices bijectively to unconstrained coordinates, guaranteeing valid outputs without projection. A graph flow-matching network learns the empirical distribution in those coordinates and is permutation equivariant: reordering the variables only reorders the output. We compare a conditional CorrGAN reimplementation, image diffusion, and CorrFlow on $80\times80$ Taiwan equity matrices using five seeds per neural method. CorrFlow produces 100% valid draws without repair and obtains feature sliced-Wasserstein distance $.712\pm.141$, versus $4.339\pm.252$ for CorrGAN and $4.610\pm.191$ for image diffusion after their invalid outputs are projected back to the correlation space. It also improves this distance over historical lookup, while lookup retains higher local coverage. These results support the combined construction under the evaluated training budget; they do not establish superiority over fully tuned image models or isolate the contribution of each component.