Research Notes

Notes from the quant lab.

Notes on factor judgment, risk design, and numerical methods: the discipline behind the dashboards.

Factor Modeling

From factor idea to verdict

Factor work earns trust through distribution, predictive power, performance, and risk concentration.

Risk Design

Top-bottom spreads can look clean until sector concentration turns a factor result into an accidental macro bet.

Derivatives

Lattices, LSMC, finite differences, Monte Carlo, Greeks: method choice is a product decision.

Factor Modeling

Distribution checks. Concentration checks. Rank tests. Portfolio tests.

Start with the cross-section. A factor with a broken or wildly skewed distribution needs different treatment before it becomes portfolio logic.
Inspect concentration. Sector, industry, and market-cap breakdowns reveal whether the factor actually measures the economic intuition it claims to.
Measure rank quality. IC, Spearman correlation, monotonicity, and the top-minus-bottom spread tell a richer story than a single return number.
Backtest buckets and the portfolio. Quintiles and deciles show whether the signal degrades smoothly or only works at the extremes.
End with a verdict. Users need to know whether the factor is promising, weak, noisy, sector-biased, or only conditionally useful.

Risk Design

Signal or sector bet. Separate them before trusting the spread.

Sector composition by quintile shows whether the top bucket is diversified or simply overexposed to one economic theme.
Median factor values by sector help separate stock-level signal from industry structure.
Industry-level breakdowns catch hidden clusters that broad sector labels can smooth over.

Derivatives

Method choice → visible risk surface.

Lattices work well when early exercise matters and you want the decision tree visible.
Least Squares Monte Carlo is strongest when the exercise policy has to be estimated from simulated paths.
Finite difference methods make value surfaces and Greeks easier to inspect across the grid.
Variance reduction is not cosmetic. It can determine whether a Monte Carlo estimate is usable in an interactive workflow.
For securitized products, rate scenarios and embedded optionality need to be reviewable side by side.