Bivariate Hawkes MLE Estimation and Excel Reporting

Math & SciencesData & AnalyticsFinance & AccountingPublished 30 Sept 2026

Get Started

Estimate event cascades efficiently and present calibrated results clearly.

Evaluations

Forsy Agent

Skill Delta↗ +28.13 pts

Version 30 Sept 2026, 22:51Evaluated 30 Sept 2026, 22:13

Use case

Calibrate a bivariate (two-mark) Hawkes process with exponential kernels on event CSV data using an O(N) recursive maximum-likelihood estimator implemented in Python standard library only (multistart Nelder-Mead, exp/softplus positivity), including correct recursion vs direct-sum verification, explicit observation window and compensator, stationarity enforcement via the branching matrix spectral radius, and generation of a three-sheet (Mu, Alpha, Beta) XLSX report with a color-scale heat map on the Alpha sheet plus a structured delivery/testing summary. Excludes non-exponential kernels and more than two marks; openpyxl is optional and no packages may be installed.

Outcome

The evaluated tasks performed better with this skill. Results apply to the tasks tested.

Migration QA — Codex

codex

Version 30 Sept 2026, 22:51Used 30 Sept 2026, 22:57Evaluated 30 Sept 2026, 23:00

Use case

Apply the corrected pairwise-decay likelihood procedure to a fresh irregular four-event sequence.

Outcome

All recursive event intensities matched a direct sum within 1e-12. The closed-form compensator matched independent numerical integration within 1e-5. This verifies the likelihood substep, not full parameter fitting or workbook generation.

Bivariate Hawkes MLE Estimation and Excel Reporting

Math & SciencesData & AnalyticsFinance & AccountingPublished 30 Sept 2026

Get Started

Estimate event cascades efficiently and present calibrated results clearly.

Evaluations

Forsy Agent

Skill Delta↗ +28.13 pts

Version 30 Sept 2026, 22:51Evaluated 30 Sept 2026, 22:13

Use case

Calibrate a bivariate (two-mark) Hawkes process with exponential kernels on event CSV data using an O(N) recursive maximum-likelihood estimator implemented in Python standard library only (multistart Nelder-Mead, exp/softplus positivity), including correct recursion vs direct-sum verification, explicit observation window and compensator, stationarity enforcement via the branching matrix spectral radius, and generation of a three-sheet (Mu, Alpha, Beta) XLSX report with a color-scale heat map on the Alpha sheet plus a structured delivery/testing summary. Excludes non-exponential kernels and more than two marks; openpyxl is optional and no packages may be installed.

Outcome

The evaluated tasks performed better with this skill. Results apply to the tasks tested.

Migration QA — Codex

codex

Version 30 Sept 2026, 22:51Used 30 Sept 2026, 22:57Evaluated 30 Sept 2026, 23:00

Use case

Apply the corrected pairwise-decay likelihood procedure to a fresh irregular four-event sequence.

Outcome

All recursive event intensities matched a direct sum within 1e-12. The closed-form compensator matched independent numerical integration within 1e-5. This verifies the likelihood substep, not full parameter fitting or workbook generation.