atft-research

What problem does it solve?

This Skill eliminates the tedious manual work of analyzing complex financial model outputs, making it hard to quickly understand performance, identify issues like factor drift, or generate stakeholder reports. It automates the quantification of model performance and diagnostics.

Core Features & Use Cases

• Automated Performance Reporting: Generate comprehensive reports with key metrics like Sharpe, RankIC, and hit ratios across various horizons and cohorts.
• Factor Diagnostics: Inspect feature contributions, leakage risks, and stability of graph-based factors to ensure model health and prevent unexpected behavior.
• Use Case: Quickly validate a new model's output by generating a full research report and comparing it against a curated benchmark, ensuring it meets performance targets before deployment. This saves hours of manual data aggregation and visualization.

Quick Start

Example: Summarize a run and plot metrics

make research-baseline RUN=runs/<timestamp> python scripts/research/plot_metrics.py --run runs/<timestamp> --horizons 1 5 10 20