Sentiment-Driven Markets: Enhancing ETF Price Prediction with News and Tweet Analysis

Replication files & code for an ECO482 course project.

---

📄 Project Overview

This repository contains the paper PDF, code, and supporting files, for a machine learning forecast of S&P500 daily price movements. This project aims to explore whether Trump's tweets, major news headlines, and technical indicators can jointly predict short-term S&P 500 ETF (SPY) price movements

---

🧠 Features & Data

Sentiment Features (Lagged):

• VADER and SWN compound scores from:

  • New York Times
  • Wall Street Journal
  • Washington Post

• VADER and Topic Discovery from:

  • Trump's tweets

Financial & Technical Indicators (Lagged):

• Moving Averages (20-day, 50-day)
• RSI, MACD Histogram
• ATR (volatility), OBV (volume), Stochastic RSI
• Ichimoku baseline, VIX, and SPY returns

Target Variable:

• Binary classification: 1 if SPY return is positive the next day, 0 otherwise

---

⚙️ Methodology

• Feature engineering and lag creation (1-day and 3-day)
• Standardization and preprocessing
• Model training using time series-aware cross-validation
• Algorithms tested:
  • Logistic Regression
  • Random Forest
  • XGBoost
  • Support Vector Machine (SVM)
  • K-Nearest Neighbors (KNN)

---

📊 Evaluation Metrics

Models are evaluated using:

• Accuracy
• F1 Score
• ROC AUC
• PnL strategy returns
• Sharpe Ratio

A basic long/short strategy is simulated using predicted labels to assess practical performance.

---

📁 Repository Structure

ECO482_Code.ipynb

• Full ML pipeline (data loading, training, evaluation)

dataset.csv

• Cleaned dataset with sentiment and financial features models

---

🧪 Requirements

• Python 3.8+
• Required packages:
  • pandas
  • numpy
  • scikit-learn
  • matplotlib
  • seaborn
  • xgboost

Install via:

pip install -r requirements.txt