Empirical Results
Measured results for the neural methods. The theory behind these methods is in the theory section.
Architecture-specific results
Each page focuses on finding the optimal network size/shape for each problem within a single architecture. Numbers are single-seed and indicative, not a rigorous hyperparameter optimum.
| Architecture | Page | Network |
|---|---|---|
| Deep BSDE | deep_bsde.md | \(Z\) (and \(\Gamma\)) nets |
| Deep HJB (DGM) | dgm.md | value net |
| Neural operator | operator.md | branch/trunk (DeepONet) or spectral (FNO) |
Cross-architecture results
| Page | Contents |
|---|---|
| Results | Validation, activation comparison, reference agreement |
| Comparison | Training/execution time and error vs exact, split PDE/SOC |
| Coverage | Which architecture fits each exact-solution problem |
Problems considered
| Problem | Class | Exact solution | Math background |
|---|---|---|---|
| Black-Scholes European call | PDE | closed form | BSDE/DGM theory; PDE exact (Black-Scholes) |
| Merton log-utility portfolio | SOC | closed form | SOC exact (Merton portfolio) |
| Linear-quadratic regulator | SOC | closed form (Riccati) | SOC exact (LQ regulator) |
| Jump linear-quadratic regulator | SOC (jump) | closed form (Riccati/affine) | SOC exact (LQ with jumps); jump theory |
| Avellaneda-Stoikov market making | SOC | closed form (matrix exponential) | SOC exact (Avellaneda-Stoikov) |
| Merton jump portfolio | SOC (jump) | closed form | SOC exact (Merton with deterministic jumps) |
| Merton jump log-normal | SOC (jump) | numerical root | SOC exact (Merton with log-normal jumps) |
| Jump-to-default portfolio (log utility, L1) | SOC (jump) | closed form (default-chain matrix exponential) | Correlated price-default; operator theory |
| Jump-to-default portfolio (CARA L1) | SOC (jump) | no-default Merton; no-loss matrix exponential; L1 numerical | Correlated price-default |
| Jump-to-default portfolio (CARA L2/L3) | SOC (jump) | no closed form; DGM reference | Correlated price-default |
The Heston and Hawkes market-making extensions and unequal-\(k\) market making
are not yet in neural_solver; they are the target of the
neural-operator research plan
and are documented in the neural operator theory. The
log-utility L1 and CARA L1/L3 jump-to-default problems are implemented.
Selected CARA L3 result
The product-domain operator (Walsh on the default cube, MLP on prices) is
benchmarked against a per-instance DGM on the S-dependent CARA L3 family:
0.77% normalized held-out RMSE with the Walsh layer and 1.51% without it. The
operator trains in 0.51 s; the DGM reference is 1.37 s per instance. The
multi-source deep BSDE matches the DGM within 0.25-1.41% per instance. The
detailed table is in the source page operator.md.