The Mechanics of Markov Chain Monte Carlo
R Metropolis sampler for 10-factor portfolio VaR/ES, benchmarked against the exact posterior to show how poor mixing understates parameter risk.
Projects are the working analyses behind the articles. Each one asks a specific risk question, builds a model to answer it, tests the result, and reports what it does and does not show. The code is open, mostly in R with some Python, and the data are synthetic or public and labeled as such, so you can see how a method behaves before it is applied to your own environment.
Risk Quantification. Bayesian Cybersecurity Loss Forecasting replaces a qualitative ransomware rating with an annual loss curve in dollars. Multi-Stage Attack Graph Simulation via Markov Chains quantifies dwell time and containment odds under baseline and hardened defenses.
Third-Party Risk. Quantifying the Blast Radius ranks structural risk in a 90-node fintech topology. Mapping the Invisible finds hidden privilege escalation paths in Active Directory. Operational Resilience: Quantifying Third-Party ICT Concentration Risk Under DORA shows eight ICT vendor entries collapsing to five failure domains.
Tail-Risk + Capital Modeling. Extreme Value Theory and Operational Loss Capital Estimation measures how much a log-normal model understates the tail. The Diversification Question quantifies the capital benefit of modeling dependence with copulas. Coding the Tail implements block bootstrap and EVT in R.
Fractional Risk Services. Auditing a Pre-Trained Hugging Face Financial Risk Model and Policy as Code show how model governance and compliance controls can be made reproducible.
If a project resembles a problem you are facing, get in touch. Any of these can serve as the starting point for an engagement on your own data.