Equity Research & Quantitative Finance

Automated Equity Research
& Valuation Platform

An end-to-end machine-learning pipeline that ingests SEC EDGAR financials for 80 large-cap companies, engineers 22 financial features, and trains an XGBoost + LightGBM ensemble validated with walk-forward validation. Valuation outputs include company-specific WACC via CAPM, 5-year DCF projections, and 5,000-iteration Monte Carlo simulations, all compiled into per-stock analysis for AAPL, MSFT, GOOGL, META, NVDA, and AMZN.

●80-company training universe ●22 engineered features ●XGBoost + LightGBM ensemble ●Walk-forward validation ●5,000 Monte Carlo iterations ●SHAP explanations

PythonXGBoost LightGBMSHAP SEC EDGAR APIyfinance scikit-learnopenpyxl Monte CarloDCF / CAPM

Architecture

Pipeline & Methodology

Training companies (S&P 500 universe)

Engineered financial features

Monte Carlo iterations per stock

Large-cap stocks valued

Step 01: Data

SEC EDGAR + yfinance Ingestion

Annual income statements, balance sheets, and cash flow statements fetched via yfinance (primary) with automatic fallback to the SEC EDGAR XBRL API when yfinance returns empty data. Market data (beta, P/E, EV/EBITDA, sector) sourced from yfinance info. 80 companies ingested; each returning 4–6 years of annual data.

Step 02: Features

22-Feature Financial Engineering

Features span revenue momentum (YoY growth, 1-year lag, 3-year CAGR), profitability levels and trends (gross, EBITDA, net margins + trend direction), cash generation (FCF margin + lag), returns (ROE, ROA, ROIC), balance sheet (D/E, current ratio), investment intensity (CapEx, R&D), valuation multiples, and sector encoding. Lag and trend features capture momentum, the single largest driver of model improvement.

Step 03: Models

XGBoost + LightGBM Ensemble

Two regression targets: forward 1-year revenue growth and forward 1-year FCF margin. Both XGBoost (depth-3, high regularization) and LightGBM (15 leaves) are trained; predictions are blended 50/50. Validated with walk-forward validation: train on all years before year t, test on year t, ensuring zero future leakage. Baseline models (persistence + historical mean) provide comparison benchmarks.

Step 04: Valuation

WACC · DCF · Gordon Growth

WACC computed per-company via CAPM (K_e = R_f + β × ERP) with market-implied cost of debt. Revenue and FCF margin projections use model predictions with geometric decay toward long-run rates. Terminal value via Gordon Growth Model. DCF intrinsic value compared to current price to generate BUY / HOLD / SELL ratings.

Step 05: Risk

Monte Carlo Simulation

5,000 iterations bootstrapped from walk-forward OOF residuals (not in-sample residuals) to get honest noise estimates. Each iteration perturbs revenue growth, FCF margin, and WACC, re-runs the full DCF, and records the intrinsic value. Output is a full probability distribution over fair value, quantifying upside/downside risk beyond the point estimate.

Step 06: Explain

SHAP Waterfall Explanations

SHAP TreeExplainer decomposes each prediction into per-feature contributions, making the model fully interpretable. Beeswarm plots show global feature importance across the training set; waterfall plots explain each individual stock prediction: which specific financial characteristics are driving the revenue growth and FCF margin forecasts up or down.

Walk-Forward Validation

Unlike k-fold or group-k-fold, walk-forward is the only CV scheme that respects chronological ordering. The model is never allowed to train on future data when predicting the past, mirroring exactly how the model would be deployed in production.

Persistence Baseline

The persistence model predicts next year = this year, a surprisingly hard benchmark in finance. Beating persistence on revenue growth is the minimum bar. Beating it on FCF margin requires capturing genuine mean-reversion and company-specific dynamics the market hasn't priced.

Validation

Model Performance

Walk-Forward Validation: XGB+LGB Blend

SHAP Beeswarm: Global Feature Impact

Revenue Growth Model

FCF Margin Model

Mean |SHAP| Feature Importance

Cross-Stock

Peer Comparison

Side-by-side view of DCF upside/downside, WACC, predicted revenue growth, and predicted FCF margin across all six target stocks. Bar colors reflect the BUY / HOLD / SELL rating generated by the model.

Individual Coverage

Per-Stock Analysis

Each stock includes a 5-year historical + projection view, DCF bridge waterfall, Monte Carlo fair-value distribution, WACC sensitivity heatmap, and SHAP waterfall explanations for both the revenue growth and FCF margin predictions.

AAPL Apple Inc.

Historical Financials & 5-Year Projection