Open to research & analytical roles

Raj
Thakur

Mathematical sciences student and actuarial analyst building ML pipelines and quantitative risk models, with a standing research interest in tail risk, random matrix theory, and applied econometrics on emerging markets.

BSc Mathematical Sciences · GPA 3.64 SOA Exams P · FM · FAM: Cleared Research-Oriented
3
SOA Exams Cleared
P · Probability  |  FM · Financial Math  |  FAM · Actuarial Math
3
Working Papers In Progress
Labor migration · Inflation forecasting · AI & labor markets
Open Theoretical Questions
Tail risk · Random matrices · Optimal transport · Mean field games
About

Quantitative mind,
rigorous by nature.

I'm Raj Thakur, a final-year Bachelors in Mathematical Sciences student at Tribhuvan University, focused on actuarial science, quantitative modelling, and applied research. I'm a candidate with the Society of Actuaries (SOA), having cleared Exams P, FM, and FAM.

My work sits at the intersection of mathematics, risk, and applied machine learning: from classical probability and extreme value theory to LightGBM pipelines and econometric forecasting. I'm as interested in the open theoretical questions (tail dependence, random matrices, optimal transport) as I am in shipping something that runs in production. Most of what I build starts from the same question: what does a thin, noisy, emerging-market dataset actually let us claim?

Alongside the technical work, I place real value on clear writing, cross-departmental collaboration, and translating dense quantitative results into decisions non-specialists can act on.

P
Probability
Foundational probability theory, distributions, and stochastic processes underlying actuarial and risk models.
✓ Cleared
FM
Financial Mathematics
Time value of money, interest theory, annuities, and derivatives fundamentals.
✓ Cleared
FAM
Fundamentals of Actuarial Mathematics
Life contingencies, survival models, and long-term actuarial and financial mathematics.
✓ Cleared
Experience

Applied practice,
real institutions.

Actuarial Intern
United Ajod Insurance Ltd.
Jul 2026 – Aug 2026
  • Designed an ORSA calculation engine and a dynamic risk-monitoring dashboard in Excel to evaluate enterprise capital adequacy and solvency requirements
  • Built Key Risk Indicators mapped across a 161-row cross-departmental risk register, establishing quantitative thresholds for operational, market, and underwriting risk
  • Ran a programmatic data-quality audit of the risk register, catching threshold typos, status mismatches, and blank rows that manual review had missed
  • Collaborated with stakeholders across departments to align individual risk inputs with overall enterprise risk management objectives
Research

Beyond coursework,
into open questions.

I treat mathematics as a toolkit for open empirical questions, particularly where developing-economy data is thin, noisy, and under-studied. Below are papers currently in progress, and the theoretical areas I'm working toward next.

Working Papers
Machine Learning Approaches to Nepal CPI Inflation Forecasting
Working Paper

Benchmarking LASSO, Elastic Net, Random Forest, and Gradient Boosting against AR/VAR baselines for Nepal CPI, with SHAP-based explainability to keep the forecasts interpretable.

Healthcare & Insurance Behavioral Analytics among Young Adults
Dissertation · Complete

A logistic regression model in R examining how financial literacy shapes future healthcare and insurance decisions among young adults in Kathmandu, built on original survey data (n = 135).

Theoretical Interests
📉
Extreme Value Theory & Tail Risk
Peaks-over-threshold and GPD-based estimation of severe downside risk in thin, illiquid emerging markets.
🎲
Random Matrix Theory
Marchenko–Pastur filtering to denoise correlation structure in NEPSE covariance matrices for more stable portfolio estimation.
🔀
Optimal Transport
Wasserstein distributionally robust optimization as a way to build portfolios that stay robust under model misspecification.
🌊
Mean Field Games & Rough Paths
Longer-horizon reading interest: how large-population strategic interaction and rough-path methods might extend classical stochastic finance.
Skills

Expertise &
Capabilities.

Actuarial & Quantitative
Actuarial Analysis Financial Mathematics Quantitative Modelling Risk Analysis Extreme Value Theory Time Series Analysis
Mathematics & Statistics
Probability Theory Statistical Inference Regression Analysis Survival Analysis Linear Algebra Calculus
Machine Learning & AI
Machine Learning Large Language Models Agentic AI PyTorch Scikit-learn LightGBM Hugging Face
Data Engineering & Tools
Python R SQL PostgreSQL FastAPI MLflow Docker GitHub Actions MS Excel
Professional
Technical Writing Cross-cultural Communication Stakeholder Collaboration
Projects

Work &
Projects.

NEPSE Bank Volatility Forecasting
● Live
End-to-end MLOps pipeline forecasting next-day volatility for Nepal Stock Exchange bank stocks

Built a fully automated forecasting system for 19 NEPSE-listed commercial bank equities, from data acquisition through a deployed, publicly-queryable API, designed and shipped entirely on free-tier infrastructure.

  • Reverse-engineered NEPSE's obfuscated authentication scheme to build a resilient daily scraper against an undocumented exchange API, with automated market-status detection and failure handling via scheduled GitHub Actions
  • Engineered a leak-free volatility feature set (Parkinson high-low estimator, HAR-style realized-volatility lags) and validated a pooled LightGBM model using purged, expanding-window walk-forward cross-validation against naive baselines
  • Tracked experiments and model versions with MLflow, including automated promotion of the best-performing model to production on each weekly retrain
  • Deployed a Dockerized FastAPI service on Render serving live forecasts from a PostgreSQL backend, with a fully decoupled serving layer that never runs live model inference
Python PostgreSQL LightGBM MLflow FastAPI Docker GitHub Actions Render
Modelling Tail Risk in NEPSE with Extreme Value Theory
● Complete
Peaks-over-threshold EVT applied to NEPSE index returns to quantify severe downside risk

Standard risk models assuming normally-distributed returns systematically underestimate the frequency and severity of extreme market crashes. This project applies Extreme Value Theory to the NEPSE index to produce tail-risk estimates that don't rely on that flawed normality assumption.

  • Applied the Peaks-Over-Threshold approach, fitting a Generalized Pareto Distribution to extreme daily losses to estimate Value at Risk (VaR) and Expected Shortfall (ES)
  • Used Mean Excess Plots to identify the optimal GPD threshold, and fit shape (ξ) and scale (σ) parameters via SciPy's genpareto
  • Complemented tail estimates with market microstructure and liquidity proxies, the Amihud Illiquidity Ratio and the Corwin-Schultz high-low spread estimator, to contextualise extreme moves against trading conditions
  • Found that large NEPSE losses coincide with high turnover rather than low liquidity, the opposite of the standard positive liquidity-risk correlation assumption, and adjusted the risk framework toward holding-period extension scaled to NEPSE's T+3 settlement cycle
Python Pandas / NumPy SciPy Matplotlib / Seaborn Extreme Value Theory
Contact

Let's
connect.

I am open to research collaborations, internships, and opportunities in actuarial science, data science, risk analysis, and AI. Please feel free to get in touch.