Ridge-regularised FTSE 100 portfolio rebalancing with FCA compliance
| Metric | Value |
|---|---|
| Condition number improvement | k reduced from 2,862 to 62.7 (46x) |
| FCA compliance | PASS — 0 violations |
| Trades generated | 81 |
| Transaction cost | GBP 7,143 (0.01%) |
| Tests passing | 21 |
The charts illustrate sector allocation before and after optimisation. The initial portfolio shows a relatively balanced distribution, while the optimised portfolio reallocates weights to maximise exposure within regulatory constraints.
Post-optimisation, the portfolio maintains balance with slight alterations to sector allocations. Financials and Consumer see a slight increase while Energy and Industrials show a slight decrease — all sectors remain well within the FCA 30% limit.
FTSE 100 returns data is severely ill-conditioned — k(X'X) = 2,862 — due to within-sector correlation where stocks in the same industry exhibit near-linear dependence. Under these conditions, direct OLS via (X'X)^-1 X'y produces non-physical portfolio weights with magnitudes of +1000 or -1000, implying excessive leverage or unintended short positions.
This directly violates FCA regulations:
- No single equity position above 10%
- No sector allocation above 30%
- Minimum 5% in liquid assets
Direct OLS is therefore unsuitable for portfolio construction in correlated equity universes — regularisation is required to obtain stable, compliant weights.
Ridge regression replaces the OLS system with (X'X+lambda * I)w=X'y, adding lambda * I to lift the eigenvalues and reduce the condition number by approximately 46x. The regularisation parameter lambda is selected via 5-fold cross-validation.
FCA constraints are enforced using an iterative projection method. For version 1, this was a deliberate proof-of-concept decision, prioritising speed and reasonable results over optimality. A constrained quadratic programming (QP) formulation is identified as future work.
git clone https://github.com/LukeWardle/ftse-portfolio-rebalancer
cd ftse-portfolio-rebalancer
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
python main.py # runs full pipeline, prints FCA compliance and volatility
pytest tests/ -v # runs 21 tests
See DESIGN.md for full architecture and function signatures.
ftse100_portfolio_rebalancer/
|-- src/
| |-- data.py
| |-- ridge.py
| |-- constraints.py
| |-- rebalance.py
| |__ analysis.py
|-- data/
|-- tests/
|-- results/
|-- images/
|-- main.py
|-- verify_multicollinearity.py
|-- export_results.py
|-- visualise_portfolio.py
|-- DESIGN.md
|-- README.md
|-- .gitignore
|__ requirements.txt
- Upgrade data source to yfinance for real FTSE returns
- Upgrade solver to constrained QP (scipy SLSQP) for v2
- ILP upgrade for integer trade sizing
- Volatility comparison meaningful only with real returns data — synthetic data limitation
- y target is cross-sectional mean return — not a true optimisation target; upgrade with yfinance
MIT