Table 8 Performance of Heston–CaNN calibrating 5 parameters on a GPU over \(3125 \times 5\) (random seeds) test cases
From: A neural network-based framework for financial model calibration
Absolute deviation from \(\varTheta ^{*}\) | Error measure | Computational cost | |||
|---|---|---|---|---|---|
\(|\nu _{0}^{\dagger }-\nu _{0}^{*}|\) | 4.39 × 10−4 | J(Θ) | 2.52 × 10−6 | CPU time (seconds) | 0.85 |
\(|\bar{\nu }^{\dagger }-\bar{\nu }^{*}|\) | 4.54 × 10−3 | MJ | 7.18 × 10−8 | GPU time (seconds) | 0.48 |
\(|\gamma ^{\dagger }-\gamma ^{*}|\) | 3.28 × 10−2 |  |  | Function evaluations | 193249 |
\(|\rho ^{\dagger }-\rho ^{*}|\) | 4.84 × 10−2 |  |  | Data points | 35 |
\(|\kappa ^{\dagger }-\kappa ^{*}|\) | 4.88 × 10−2 |  |  |  |  |