Quantitative portfolio optimization and factor analysis engine built with Python and Streamlit. Implements mean-variance optimization (Markowitz, 1952), Fama-French factor regression, and market regime detection to decompose portfolio returns into risk premia and analyze performance across bull, bear, and high-volatility regimes.
- Efficient Frontier + Capital Market Line — traces optimal risk-return tradeoff via constrained optimization (SLSQP)
- Three Optimization Strategies — Maximum Sharpe Ratio, Global Minimum Variance, Risk Parity
- Monte Carlo Simulation — up to 50,000 random portfolio allocations to visualize the feasible region
- Portfolio Backtesting — equity curves comparing all strategies vs equal-weight benchmark
- VaR / CVaR — Value at Risk and Conditional VaR at configurable confidence levels
- Rolling Analysis — rolling volatility and pairwise correlation over configurable windows
- Regresses portfolio excess returns on academic risk factors from Ken French's Data Library
- Four model choices: FF3, FF3+MOM, FF5, FF5+MOM
- Reports alpha (annualized), factor betas, t-statistics, R², and adjusted R² per strategy
- Significance-coded coefficients (*** 99%, ** 95%, * 90% confidence)
- Side-by-side factor-loading comparison across all three strategies
- Classifies each trading day into Bull / Bear / High-Volatility regimes using SPY as the market proxy
- Bear trigger: > 10% drawdown from rolling peak
- High-vol trigger: 20-day realized volatility in the top 15% of the sample
- Regime-shaded equity curves and regime-conditional Sharpe ratios
- Identifies which strategies are genuinely diversified vs. which only perform in calm uptrends
- Real historical prices from Yahoo Finance via
yfinance - Daily factor data from Ken French's Data Library (Tuck School of Business, Dartmouth)
- CSV export of optimized weights
| Layer | Technology |
|---|---|
| Frontend | Streamlit |
| Visualization | Plotly |
| Optimization | SciPy (SLSQP) |
| Numerical | NumPy, Pandas |
| Market Data | yfinance |
| Factor Data | Ken French Data Library (CSV/ZIP) |
| Deployment | Streamlit Cloud |
Daily simple returns are used throughout for mathematical consistency with portfolio aggregation:
R_i,t = (P_i,t / P_i,t−1) − 1
This ensures the portfolio return is exactly the weighted sum of asset returns, R_p = w · R — a property that does not hold for log returns. Annualized statistics assume 252 trading days per year.
Given n assets with expected return vector μ and covariance matrix Σ, the portfolio return and variance are:
R_p = wᵀ · μ
σ²_p = wᵀ · Σ · w
Subject to:
- Sum of weights = 1 (fully invested)
- 0 ≤ w_i ≤ 1 (long-only, no leverage)
Solved numerically with Sequential Least Squares Programming (scipy.optimize.minimize, method="SLSQP").
Maximum Sharpe Ratio — maximizes the risk-adjusted excess return:
maximize (wᵀμ − r_f) / √(wᵀΣw)
Global Minimum Variance — minimizes total portfolio variance:
minimize wᵀΣw
Risk Parity — equalizes each asset's contribution to total portfolio risk:
minimize Σᵢ (RC_i − σ_p/n)²
where RC_i = w_i · (Σw)_i / √(wᵀΣw)
Traced by minimizing variance subject to a target return constraint, producing the set of Pareto-optimal portfolios.
The line from the risk-free rate through the tangency (Max Sharpe) portfolio:
E[R_p] = r_f + [(E[R_T] − r_f) / σ_T] · σ_p
At confidence level α:
VaR_α = percentile of portfolio returns at (1 − α) × 100%
CVaR_α = E[R_p | R_p ≤ VaR_α]
CVaR (Expected Shortfall) is more sensitive to tail risk than VaR alone.
Portfolio excess returns are regressed on factor returns via OLS:
R_p,t − R_f,t = α + β₁·MKT_t + β₂·SMB_t + β₃·HML_t + … + ε_t
Where factors are drawn from Ken French's Data Library:
- MKT-RF — market excess return (market beta)
- SMB — small minus big (size tilt)
- HML — high minus low book-to-market (value tilt)
- RMW — robust minus weak profitability (quality tilt, FF5 only)
- CMA — conservative minus aggressive investment (quality tilt, FF5 only)
- MOM — winners minus losers (momentum tilt, optional add-on)
The intercept α is the portion of excess return unexplained by known factor exposures — the "skill" or "mispricing" component. A positive, statistically significant α (|t| ≥ 1.96) indicates outperformance beyond the factor model's predictions. T-statistics are computed from OLS standard errors:
β = (XᵀX)⁻¹ Xᵀ y
SE(β) = √(σ² · diag((XᵀX)⁻¹))
t = β / SE(β)
Each trading day is classified using SPY as the market proxy:
BEAR if drawdown from rolling peak > 10%
HIGH_VOL else if 20-day realized vol > 85th percentile of sample vols
BULL otherwise
Realized volatility is annualized: σ_realized = std(r_20) · √252.
For each strategy, regime-conditional statistics are computed:
Ann. Return_regime = mean(r_daily | regime) · 252
Ann. Vol_regime = std(r_daily | regime) · √252
Sharpe_regime = (Ann. Return_regime − r_f) / Ann. Vol_regime
| Metric | Formula |
|---|---|
| Annualized Return | (final / initial)^(252/N) − 1 |
| Annualized Volatility | σ_daily · √252 |
| Sharpe Ratio | (R_ann − r_f) / σ_ann |
| Sortino Ratio | (R_ann − r_f) / σ_downside,ann |
| Max Drawdown | min((P_t − max(P_{≤t})) / max(P_{≤t})) |
| Calmar Ratio | R_ann / |MDD| |
- Long-only portfolios — no short selling or leverage
- Historical returns and covariances used as forward-looking estimates (standard assumption; subject to estimation error)
- No transaction costs or taxes modeled in backtest
- Mean-variance framework assumes either normally distributed returns or quadratic utility; real returns exhibit skewness and fat tails (the Asset Stats tab surfaces this)
- Weights are optimized once in-sample — no walk-forward or rolling re-optimization
- Factor regression uses OLS with homoskedastic standard errors (no Newey-West adjustment for autocorrelation)
- Regime classification uses SPY only; not suitable for non-US or non-equity benchmarks without substitution
- Mispricing of alpha: a statistically significant alpha over a backtest window does not imply persistence out-of-sample
git clone https://github.com/Marin-X/QuantFolio.git
cd QuantFolio
pip install -r requirements.txt
streamlit run app.pyThe Fama-French factor data is fetched on-demand from Ken French's Data Library and cached for 24 hours.
- Markowitz, H. (1952). "Portfolio Selection." The Journal of Finance, 7(1), 77–91.
- Sharpe, W. F. (1966). "Mutual Fund Performance." The Journal of Business, 39(1), 119–138.
- Fama, E. F. and French, K. R. (1993). "Common Risk Factors in the Returns on Stocks and Bonds." Journal of Financial Economics, 33(1), 3–56.
- Fama, E. F. and French, K. R. (2015). "A Five-Factor Asset Pricing Model." Journal of Financial Economics, 116(1), 1–22.
- Carhart, M. M. (1997). "On Persistence in Mutual Fund Performance." The Journal of Finance, 52(1), 57–82.
- Maillard, S., Roncalli, T., and Teïletche, J. (2010). "The Properties of Equally Weighted Risk Contribution Portfolios." Journal of Portfolio Management, 36(4), 60–70.
- Rockafellar, R. T. and Uryasev, S. (2000). "Optimization of Conditional Value-at-Risk." Journal of Risk, 2, 21–42.
Factor data: Ken French's Data Library, Tuck School of Business, Dartmouth College.
Built by Marin Xhemollari · Portfolio · LinkedIn