Begin with a question
A model is a means to answer a decision-relevant question. Each investigation states what it measures, why it matters, and which baseline it must beat.
METHODS & RESEARCH STANDARDS
Show the work.
State the limits.
Useful risk research makes its assumptions inspectable. These principles connect every investigation and interactive experiment in the lab.
Keep evaluation separate
Training, calibration, and evaluation have distinct roles. Time-dependent observations require chronological separation; simulated experiments require independent evaluation paths.
Compare the same problem
Strategies share paths, costs, horizons, and information sets. Forecasts target the same quantity. More complexity must improve an outcome that matters.
Make uncertainty visible
Report variability, coverage, sensitivity, and failure cases alongside point estimates. Conformal guarantees depend on assumptions; a risk penalty is not a hard constraint.
Leave a reproducible record
Results carry parameters, seeds, model versions, and data provenance. Saved snapshots support inspection without a running calculation service.
Name the boundary
Synthetic data establishes controlled behavior, not external validity. Scenario outcomes are conditional. Educational implementations are not validated production or regulatory systems.
DATA & EXTERNAL CONTEXT
Sources inform the question.
The market lab uses Kenneth French industry portfolio returns and simulated stress. Forecast and hedging labs use seeded synthetic data. Official publications provide context for stress and climate research; the lab does not present its constructed scenarios as official forecasts.
Reading a result
Check the source label, timestamp, units, seed, and parameters first. Saved snapshots and live calculations are explicitly distinguished. Use the data tables and configuration download to inspect the numerical output behind a chart.