📈 McDonald's Stock Direction Prediction with LASSO in R

This project predicts whether McDonald's (MCD) stock price will go up or down in the following quarter using quarterly financial statements, macroeconomic indicators, and adjusted stock prices. The model is based on LASSO logistic regression, estimated with multiple lambda values. The entire workflow is built in R, integrating data via the SimFin API and a local macro dataset.

---

🎯 Goal

• Forecast stock direction (Up/Down) one quarter ahead.
• Use only public and explainable financial and macroeconomic inputs.
• Build a robust predictive pipeline with visualization and accuracy metrics.

---

📦 Data Sources

• 📉 McDonald's quarterly P&L and balance sheet data from SimFin API.
• 💹 Historical adjusted closing prices (quarterly average).
• 🌍 Quarterly macroeconomic indicators (inflation, unemployment, GDP growth).

---

🔍 Modeling Approach

• Merge and preprocess financial, macro, and price data (2018–2025 Q1).
• Engineer lagged price features and previous direction.
• Create a binary target (1 = Down, 0 = Up).
• Estimate LASSO logistic regression for 10 values of lambda (0.01–0.10).
• Evaluate accuracy, precision, recall, and F1 on training data.
• Predict for 2025 Q2 and visualize key outputs.

---

📊 Results

• Best λ = 0.02

• Prediction for 2025 Q2:
  📈 UP (growth) with 66.33% implied probability
  📉 Probability of decline: 33.67%

• Confusion Matrix Metrics (Train Set):

  • Accuracy: 92.6%
  • Precision: 1.000
  • Recall: 0.800
  • F1-score: 0.889

• Direction Accuracy (based on price change):
  57.7% over historical quarters

---

📂 Project Structure

├── Plots/                         # Output visualizations
├── cache/                         # SimFin cached data
├── integrated_data.csv            # Final dataset used for modeling
├── macro_data.csv                 # Macroeconomic input (CSV)
├── lasso_lambda_coefficients.xlsx # Variable importance for 10 lambdas
├── demodata.RData                 # SimFin raw data (optional)
├── mcd_lasso_model.R              # Main R analysis script
└── README.md

---

▶️ How to Run

1. Install R packages:

install.packages(c("simfinapi", "glmnet", "caret", "ggplot2", "xgboost", "randomForest",
                   "e1071", "pROC", "PRROC", "openxlsx", "imputeTS", "readr", "patchwork"))

2. Set your SimFin API key:

sfa_set_api_key("your-api-key-here")

3. Run the script:

source("mcd_lasso_model.R")

---

📁 Key Outputs (Plots/)

• 01_adjusted_closing_price.png – Historical adjusted price
• 02_absolute_weights_lambda_0.01_and_0.03.png – LASSO variable importance
• 03.1_..._estimated_probabilities_lambda_0.01_and_0.03.png – Probabilities + errors
• 04_final_model_coefficients.png – Final model coefficients (λ = 0.02)
• 05_roc_curve.png – ROC curve (AUC = 1)

---

🧪 Other Models

Random Forest, XGBoost, and SVM were tested for comparison only. Their performance is not central to the analysis.

---

📌 Notes

• Last available quarter (2025_Q1) is excluded from training.
• Input features are quarterly and fully explainable.
• Cache and intermediate files are ignored via .gitignore.

---

📄 License

This project is open for educational and research purposes. Attribution is appreciated.