07Collaborative Quantitative Research
Market Regime Modeling — Collaborative Research
Collaborative exploration of market regimes and regime-conditional classification and regression, with Jacob credited as a contributor rather than the repository owner.
Repository owned by supermogaboy; Jacob contributed 9 commits.
- Python
- Hidden Markov Models
- Classification
- Regression
- Feature Engineering
- Ownership
- Collaborative / Archive
- Timeline
- Collaborative contribution
- Status
- Collaborative / archived
- Repository owned by supermogaboy; Jacob contributed 9 commits.
Overview
Uses engineered market features, hidden-state regime analysis, and separate classification and regression experiments by regime. Retained as an honest record of collaborative work, not as a validated forecasting product.
Problem
Explore whether market behavior can be segmented into regimes, and whether regime-conditional models provide a useful framework for studying direction and returns.
My role
Contributed code and research iterations to a public repository owned by supermogaboy. Jacob is a contributor, not the founder, owner, lead, or sole developer.
What I built
- Market feature engineering with 17 features.
- Hidden Markov Model / hidden-state regime analysis.
- Regime-conditional probability and return experiments.
- Four model variants per regime in the visible implementation.
- Logistic regression for classification.
- GradientBoostingRegressor for return modeling.
Technical decisions
- Segment the series into hidden states before fitting per-regime models, rather than fitting one model across all conditions.
- Separate the direction question from the return question into distinct classification and regression experiments.
Testing and validation
- No CI, automated test suite, or reproducibility harness is present in the public repository.
- The audit found train/test overlap risk in the collaborative implementation, so no model-performance figure from this project is published here.
Measured evidence
9
Commits by Jacob
Counted from the public repository history at revision 1300b7d.
17
Engineered market features
Counted from the checked-in feature construction.
Contributed to a collaborative regime-modeling prototype spanning engineered market features, hidden-state regimes, and per-regime classification and regression. Reviewing the project later reinforced the importance of chronological validation, leakage controls, and explicit research limitations — principles applied in the newer Market Regime + Portfolio Risk Platform.
Limitations
- A collaborative, exploratory repository — not a product.
- No verified license.
- No verified live demo.
- Train/test boundaries are not demonstrably free of overlap, so no performance figure is published.
- Not the same project as the separately owned Market Regime + Portfolio Risk Platform.
- Repository-owner work must not be read as authored by Jacob.
Technology stack
- Python
- Hidden Markov Models
- Classification
- Regression
- Feature Engineering
- scikit-learn