Skip to content
IL
← Back home

WP-2026-002 · quant · ml · xai

Credit Risk & Options Pricing - Stochastic Models vs Deep Learning

Ibrahim Laklaa · 2026

5

classifiers

3

pricing methods

200k

Monte Carlo paths

8,000

borrowers

Abstract

Five classifiers, from logistic regression to an MLP, on 8,000 synthetic borrowers, explained by a consensus of feature importances. Alongside, a European call priced by Black-Scholes, by Monte Carlo on 200,000 antithetic GBM paths and by a neural network trained on 50,000 analytical prices, with Greeks and convergence analysis.

Two problems, one lens

Two classic questions of quantitative finance, credit default prediction and European option pricing, looked at side by side through accuracy and explainability.

Credit risk

Five classifiers on 8,000 synthetic borrowers with 15 features and a 15 % default rate: logistic regression, random forest (500 trees), gradient boosting (300 trees), an XGBoost-style configuration and an MLP (128, 64, 32). The logistic baseline reads

Models are compared on AUC-ROC, precision-recall, F1 and calibration curves with 5-fold cross-validation, and explained by a consensus of feature importances across the ensembles next to the logistic coefficients.

Option pricing

A European call is priced three ways across 10 scenarios. The analytical Black-Scholes price:

A Monte Carlo estimator on 200,000 GBM paths with antithetic variates:

And an MLP trained on 50,000 Black-Scholes prices to approximate the pricing map, with all five Greeks and a Monte Carlo convergence study.

Takeaway

Results, price surfaces and figures are in the report.