Time series forecasting of the SPDR S&P 500 ETF (SPY) using a Seasonal AutoRegressive Integrated Moving Average (SARIMA) model.
This project demonstrates financial market forecasting with classical statistical modeling in Python.
The notebook builds and evaluates a SARIMA model to analyze and forecast SPY price movements. It covers:
- Historical data retrieval
- Time series preprocessing
- Stationarity testing
- Parameter selection
- Model training
- Forecast generation
- Visualization of predictions vs. actual values
The goal is to explore how seasonal ARIMA models perform on equity market data.
Market data for SPY is obtained via financial data APIs (e.g., Yahoo Finance through yfinance).
Typical fields used:
- Date
- Open
- High
- Low
- Close
- Adjusted Close
- Volume
- Python
- Jupyter Notebook
- pandas - data manipulation
- numpy - numerical computing
- matplotlib / seaborn - visualization
- statsmodels - SARIMA modeling
- yfinance - financial data retrieval
- Download historical SPY price data.
- Set date as index for time series structure.
- Handle missing values.
- Convert to appropriate frequency.
- Plot raw time series.
- Augmented Dickey–Fuller (ADF) test.
- Differencing applied if needed.
SARIMA parameters:
- p - autoregressive order
- d - differencing order
- q - moving average order
- P, D, Q, s - seasonal components
ACF/PACF plots guide parameter tuning.
- Fit SARIMA using
statsmodels. - Evaluate AIC/BIC for model quality.
- Generate out-of-sample predictions.
- Produce confidence intervals.
- Forecast vs. historical data.
- Residual diagnostics.
# Clone repo
git clone https://github.com/Jkovv/SARIMA_SPY.git
cd SARIMA_SPY
# Install dependencies
pip install pandas numpy matplotlib seaborn statsmodels yfinance notebook
# Launch notebook
jupyter notebook SARIMA_SPY.ipynb