Swiss Multi-Asset Portfolio Optimizer

April 1, 2026 · View on GitHub

Python License Status Finance Crypto SMI swiss-finance-data

Advanced portfolio optimization on Swiss Market Index (SMI) equities with crypto integration, Black-Litterman modeling, and walk-forward validation.

Built with Python, yfinance, swiss-finance-data, scipy, matplotlib — professional-grade quantitative finance modeling with live market data.

Project Overview

This project demonstrates institutional-grade portfolio optimization techniques applied to the Swiss equity market (SMI) with alternative asset integration (BTC/ETH). Unlike typical student projects using FAANG stocks, this implementation:

  • Uses Swiss blue-chip equities (Nestlé, Novartis, UBS, Roche, ABB)
  • Addresses Markowitz instability with Black-Litterman extension
  • Integrates crypto as alternative assets with correlation analysis
  • Validates performance out-of-sample via walk-forward backtesting

Why this matters: Most finance portfolios stop at Markowitz. This project goes further by implementing the models actually used in asset management (Black-Litterman) and proving robustness with proper backtesting methodology.

Key Results

Part 1: Markowitz Mean-Variance Optimization

Performance (2018–2026):

  • Max-Sharpe: 13.4% return, 18.0% volatility, Sharpe 0.63
  • Equal-Weight: 9.6% return, 16.0% volatility, Sharpe 0.48

Key insight: Optimal portfolio concentrates on ABB (47%) and Novartis (53%), eliminating Nestlé, Roche, and UBS entirely. This demonstrates the corner solution problem of pure Markowitz optimization.

Part 2: Black-Litterman Extension

Why Black-Litterman? Markowitz produces extreme allocations (0% or 50%+) that are unstable and impractical. Black-Litterman incorporates market equilibrium priors and analyst views to generate more diversified, stable allocations.

Results:

  • Same Sharpe (0.63) but introduces Roche (4.5%) and UBS (8.4%)
  • Reduces concentration risk while maintaining performance
  • Lower volatility (17.6% vs 18.0%)

Views sourced from fundamentals (via swiss-finance-data + yfinance):

  • Macro calibration: SNB policy rate (risk-free rate) and Swiss CPI (inflation adjustment) fetched live from swiss-finance-data
  • Per-stock signals: trailing P/E, ROE, and earnings growth from yfinance — scored cross-sectionally and mapped to expected return views
  • Fallback to hardcoded views (ABB 18%, Novartis 15%, Nestlé 3%) if live data is unavailable

Part 3: Crypto Integration

Hypothesis: Low correlation between crypto and traditional equities provides diversification benefits.

Correlation matrix findings:

  • Bitcoin ↔ SMI equities: 0.04–0.22 (near zero)
  • BTC ↔ ETH: 0.82 (highly correlated)

Results:

  • Adding 6.3% BTC improves Sharpe from 0.63 → 0.65
  • Volatility unchanged (17.9% vs 18.0%)
  • Ethereum excluded by optimizer (lower Sharpe than BTC)

Why only 6.3%? Optimizer capped at 15% total crypto allocation. BTC offers better risk-adjusted returns than ETH on this period.

Part 4: Walk-Forward Backtest (Out-of-Sample Validation)

Methodology: Rolling 2-year training window, 6-month out-of-sample test. 12 windows from 2020–2026.

Out-of-Sample Results:

  • Max-Sharpe (OOS): 13.6% return, Sharpe 0.56
  • Equal-Weight (OOS): 10.6% return, Sharpe 0.51
  • Max drawdown: -28% (both strategies hit by 2022 Ukraine war)

Key insight: Performance holds out-of-sample. This is not overfitting. The model generalizes across 6 years and multiple market regimes (COVID, 2022 rate hikes, 2024+ recovery).

Per-window analysis shows:

  • W3 (2021): Sharpe 2.90 (bull market)
  • W5 (2022): Sharpe -1.61 (Ukraine war, rate hikes)
  • W8 (2023): Sharpe 3.18 (recovery)

Model adapts to changing market conditions by re-optimizing every 6 months.

🏗️ Technical Architecture

swiss-multiasset-portfolio-optimizer/
├── src/
│   ├── utils.py                    # Data fetching, metrics, config
│   ├── part1_markowitz_smi.py      # Mean-variance optimization
│   ├── part2_black_litterman.py    # BL model with analyst views
│   ├── part3_crypto_integration.py # Alternative asset analysis
│   └── part4_walk_forward.py       # OOS backtesting engine
├── reports/
│   └── figures/                    # 11 publication-ready charts
├── .gitignore                      # Git exclusions (venv, __pycache__, etc.)
├── LICENSE                         # MIT License
├── main.py                         # Pipeline orchestration
├── README.md                       # Project documentation
└── requirements.txt                # Python dependencies

Core dependencies:

  • yfinance — Live market data (Yahoo Finance API) + per-stock fundamentals (P/E, ROE, earnings growth)
  • swiss-finance-data — Swiss macro data (SNB policy rate, Swiss CPI) for Black-Litterman view calibration
  • scipy.optimize — Constrained portfolio optimization (SLSQP)
  • numpy/pandas — Numerical computing
  • matplotlib/seaborn — Professional visualizations

Note: No static data files — all market data fetched live via yfinance API.

Installation & Usage

Prerequisites

  • Python 3.10+
  • pip

Setup

# Clone repository
git clone https://github.com/EMen11/swiss-multiasset-portfolio-optimizer.git
cd swiss-multiasset-portfolio-optimizer

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Run Full Pipeline

python main.py

Output: 11 charts saved to reports/figures/ in ~10 seconds.

Run Individual Modules

python src/part1_markowitz_smi.py       # Markowitz baseline
python src/part2_black_litterman.py     # BL extension
python src/part3_crypto_integration.py  # Crypto analysis
python src/part4_walk_forward.py        # Walk-forward backtest

Swiss Data Integration

This project integrates swiss-finance-data — an open-source Python package providing clean access to official Swiss financial data (SNB, FSO).

Used in this project for:

  • SNB policy rate — live risk-free rate for Sharpe ratio calculations and Black-Litterman macro calibration
  • Swiss CPI — inflation adjustment for expected return view generation

swiss-finance-data is maintained by the same author and complements yfinance specifically for Swiss market data. The package fetches data live from official Swiss government sources — no scraping, no static files.

Data Sources

  • SMI Equities: Yahoo Finance via yfinance (tickers: NESN.SW, NOVN.SW, UBSG.SW, ROG.SW, ABBN.SW)
  • Stock fundamentals: Yahoo Finance via yfinance (P/E ratio, ROE, earnings growth — used for Black-Litterman view generation)
  • Swiss macro data: swiss-finance-data (SNB policy rate for live risk-free rate; Swiss CPI for inflation adjustment)
  • Crypto: Yahoo Finance (BTC-USD, ETH-USD)
  • Period: 2018-01-01 to present (live data)
  • Risk-free rate: SNB policy rate via swiss-finance-data (falls back to 2% if unavailable)

Why SMI over S&P500/FAANG?

  • Demonstrates local market expertise (relevant for Swiss asset managers like Pictet, UBS AM, Lombard Odier)
  • Less saturated than US tech stocks in student portfolios
  • Shows ability to work with non-USD markets

Methodology

1. Markowitz Mean-Variance Optimization

Maximize Sharpe ratio:

max⁡ww⊤μ−rfw⊤Σw\max_{w} \frac{w^\top \mu - r_f}{\sqrt{w^\top \Sigma w}}

Subject to:

  • ∑wi=1\sum w_i = 1 (fully invested)
  • wi≥0w_i \geq 0 (long-only)

Limitation: Produces extreme corner solutions (100% in 1-2 assets or 0% in others).

2. Black-Litterman Model

Combines market equilibrium (CAPM-implied returns) with analyst views:

μBL=[(τΣ)−1+P⊤Ω−1P]−1[(τΣ)−1π+P⊤Ω−1Q]\mu_{BL} = [(\tau \Sigma)^{-1} + P^\top \Omega^{-1} P]^{-1} [(\tau \Sigma)^{-1} \pi + P^\top \Omega^{-1} Q]

Where:

  • π\pi = implied equilibrium returns (reverse optimization)
  • PP = pick matrix (which assets have views)
  • QQ = view vector (expected returns)
  • Ω\Omega = uncertainty matrix (confidence)
  • τ\tau = scalar uncertainty (default 0.05)

Benefit: Forces diversification while incorporating forward-looking views.

3. Walk-Forward Backtesting

Avoids look-ahead bias:

  1. Train on 2 years of historical data
  2. Optimize portfolio weights
  3. Test on next 6 months (out-of-sample)
  4. Roll window forward and repeat

Result: 12 independent OOS tests from 2020–2026.

Learning Outcomes

This project demonstrates:

  1. Advanced optimization — Going beyond textbook Markowitz to real-world BL implementation
  2. Alternative assets — Crypto correlation analysis and integration
  3. Proper validation — Walk-forward backtesting (not just in-sample overfitting)
  4. Production code quality — Modular architecture, clean functions, professional visualizations
  5. Swiss market expertise — SMI vs generic FAANG portfolios

Use Cases

  • Asset management interview prep (Pictet, UBS AM, Credit Suisse AM)
  • Quantitative finance demonstrations (shows understanding of MPT limitations)
  • Portfolio research (extend with different models: HRP, CVaR, etc.)
  • Teaching material (clean code for finance students)

Future Extensions

  • Add Hierarchical Risk Parity (HRP) as alternative to Markowitz
  • Implement CVaR optimization for tail risk management
  • Extend to multi-asset classes (bonds, commodities, REITs)
  • Add transaction costs and rebalancing constraints
  • Build Power BI dashboard for interactive exploration

License

MIT License — Free to use, modify, and distribute.

Acknowledgments

  • Data: Yahoo Finance (yfinance library), Swiss National Bank / Federal Statistical Office via swiss-finance-data
  • Methodology: Markowitz (1952), Black & Litterman (1992)
  • Inspiration: Real-world asset management practices at Swiss private banks