I hold an MSc in Financial Engineering & Management from the University of Twente (2026) and work in M&A at Moore MKW Corporate Finance. My work sits where applied ML meets finance: interpretable models, honest baselines, and reporting null results when that's what the data says.
I'm moving toward empirical AI economics: how AI adoption is actually changing professional work, and why adoption patterns diverge from theoretical capability.
| Repo | What it shows |
|---|---|
| post-merger-operating-performance-ml | My MSc thesis: can pre-deal information predict post-merger operating performance? XGBoost under chronological validation on 4,229 public-firm deals, judged across five outcome measures. Predictability is target-dependent (asset-turnover R² 0.117 vs Healy CFROA 0.022), and SHAP concentrates the signal in operational and financial structure rather than revenue-synergy stories. Comes with the interactive defense deck and a 46-second explainer video. |
| ma-premium-bunching | Do acquisition premia bunch at round numbers like offer prices do? On 320 US cash takeovers, offer prices cluster hard (37% end in .00) but premia do not (21% on a 5% multiple vs 20% by chance); a spine test shows the faint premium roundness is mechanical inheritance from round prices. |
| ai-credit-risk | Is AI displacement risk priced in corporate bonds? A firm-level exposure measure (industry automatability times labor intensity) on monthly Z-spreads for 344 issuers, with finance subsidiaries mapped to parents through the LSEG ultimate-parent id. A clean, well-identified null: point estimates are right-signed but only the rating-controlled cross-section is marginally significant, and the adoption channel does not separate. |
| ai-exposure-firm-outcomes | Does a firm's exposure to AI show up in its stock returns or fundamentals? A firm-level AI measure built from 10-K text, tested with Fama-French event studies and fixed-effects panels. No robust effect, and the paper shows why industry-level exposure measures are too collinear and why exposure event studies carry pre-trends. |
| rl-trading-agent | DQN comparing four reward designs under transaction costs. The agent loses to buy-and-hold out of sample, and the result is how much reward design alone changes behaviour. |
| credit-risk-pd | Macro-conditional default probabilities via the Belkin one-factor model, with parameter-recovery tests on synthetic data. |
| capital-structure-rf | Random forest vs linear baselines for corporate capital structure, served via a Shiny app (R). |
| jump-diffusion-pricing | Merton jump-diffusion simulation, calibration, and tail-risk pricing vs Black-Scholes. |
| click-fund-derivatives | Closed-form and Monte Carlo pricing of a capital-protected structured product. |