How pooja2420/liquidity-scoring works
This repository builds corporate-bond liquidity scores and pre-trade cost estimates from bond reference, TRACE trade and macro data. Its principal workflow loads those inputs, engineers features and a composite score, trains temporal ML models, calibrates and shrinks impact parameters, then serves scores and cost estimates through an API and Streamlit dashboard. The README also documents model explainability, universe views and TCA/RFQ connectors. Arrows are limited to relationships visible in excerpts, the source index or README; unsampled details are not inferred.
Main entry points: run_pipeline.py Pipeline runner · feature_pipeline.py Feature builder · app.py Streamlit dashboard · main.py Scoring API
Components 24
Market data
- Bond universe
src/data/bond_reference.py - TRACE trades
src/data/trace_loader.py - Macro factors
src/data/macro_factors.py
Feature engineering
- Feature builder
src/features/feature_pipeline.py - Roll spread
src/features/roll_measure.py - Amihud ratio
src/features/amihud.py - Kyle lambda
src/features/price_impact.py - Trade frequency
src/features/trade_frequency.py - Inter-trade time
src/features/inter_trade_time.py - Feature store
src/features/feature_pipeline.py
Scoring models
- Model trainer
src/models/trainer.py - Walk-forward CV
src/models/cross_validator.py - SHAP explainer
src/models/shap_explainer.py
Impact and costs
- Impact calibrator
src/impact/impact_calibrator.py - Bayesian shrinkage
src/impact/bayesian_shrinkage.py - Pre-trade estimator
src/impact/cost_estimator.py
User products
- Scoring API
src/api/main.py - API schemas
src/api/models.py - Streamlit dashboard
app.py
Other components
- Analyst / trader
- Pipeline runner
run_pipeline.py - FRED
- TCA engine
- RFQ pricing engine
Connections 30
- Pipeline runner → Bond universe: loads
run_pipeline.py - Pipeline runner → TRACE trades: loads
run_pipeline.py - Pipeline runner → Macro factors: loads
run_pipeline.py - Macro factors → FRED: fetches data
src/data/macro_factors.py - Pipeline runner → Feature builder: builds features
run_pipeline.py - Feature builder → Roll spread: uses
src/features/feature_pipeline.py - Feature builder → Amihud ratio: uses
src/features/feature_pipeline.py - Feature builder → Kyle lambda: uses
src/features/feature_pipeline.py - Feature builder → Trade frequency: uses
src/features/feature_pipeline.py - Feature builder → Inter-trade time: uses
src/features/feature_pipeline.py - Feature builder → Feature store: persists features
src/features/feature_pipeline.py - Pipeline runner → Model trainer: trains models
run_pipeline.py - Pipeline runner → Walk-forward CV: runs validation
run_pipeline.py - Walk-forward CV → Model trainer: trains folds
src/models/cross_validator.py - Pipeline runner → SHAP explainer: computes explanations
run_pipeline.py - Pipeline runner → Impact calibrator: calibrates impact
run_pipeline.py - Pipeline runner → Bayesian shrinkage: shrinks parameters
run_pipeline.py - Impact calibrator → Bayesian shrinkage: provides fit parameters
README.md - Scoring API → Feature store: reads features
src/api/main.py - Scoring API → Pre-trade estimator: estimates costs
src/api/main.py - Scoring API → API schemas: uses schemas
src/api/main.py - Analyst / trader → Scoring API: requests scores
README.md - Analyst / trader → Streamlit dashboard: explores results
README.md - Streamlit dashboard → Feature store: loads feature data
app.py - Streamlit dashboard → Pre-trade estimator: estimates costs
app.py - Streamlit dashboard → SHAP explainer: shows importance
README.md - Streamlit dashboard → Impact calibrator: shows calibration
README.md - Streamlit dashboard → Feature store: shows universe map
README.md - Scoring API → TCA engine: integrates
README.md - Scoring API → RFQ pricing engine: integrates
README.md