Stock Price Forecasting — StockSARIMA × Prophet
Stock forecasting system for IDX stocks using two time series models: SARIMA (Seasonal AutoRegressive Integrated Moving Average) and Prophet (Facebook). Both are trained in parallel per ticker and then combined through a weighted ensemble to produce the final forecast along with BUY/SELL/HOLD trading signals.
Detailed Insights
Data Pipeline
yfinance is used to download historical stock data from the IDX (2 years). The pipeline cleans the data, forward-fills gaps (for weekends/holidays), and uses get_close_series() to extract closing price features.
Training & Ensemble
SARIMA model (using auto_arima with BIC criterion) and Prophet (with multiplicative seasonality) are trained. The forecast results are combined using a 40% SARIMA and 60% Prophet weighting as a mitigation for the weaknesses of each model.
Trading Signal Generation
BUY (>0.5%), SELL (<-0.5%), and HOLD signals are generated with confidence scores based on the percentage change from the 7-day forecast compared to the current price.
Tech Stack
Key Results
- Weighted ensemble 40/60
- Auto ARIMA & Prophet changepoints
- Automated BUY/SELL/HOLD signals