Predicting Stock Movements from Historical Data

Project Inspiration

Inspired by my courses I have taken at UC Davis, ECN 141: Economic & Financial Forecasting and STA 137: Applied Time Series, I wanted to harness ARIMA models for forecasting stock prices. These classes provided the foundational knowledge that fueled my desire to apply statistical techniques to real-world financial data.

Learning and Methodology

Time series analysis is inherently challenging, requiring a deep understanding of temporal data behaviors, such as trends, seasonality, and stationary. To tackle this complexity, I revisited and reinforced my learning from both courses, focusing on the intricacies of ARIMA models. The project involved segmenting the data into training and test sets—a technique often associated with machine learning—to evaluate the predictive power of our models reliably.

Challenges

The main challenge was the steep learning curve associated with time series analysis. Time series data can be unpredictable, and ensuring the model captures all underlying patterns without overfitting or underfitting was particularly challenging due to the inherent volatility of stock prices. Moreover, preparing the data, selecting the right model parameters, and interpreting the results necessitated a thorough re-education on topics covered in my previous coursework.

Conclusion

This project not only allowed me to apply theoretical knowledge in a practical setting but also deepened my understanding of financial forecasting. By comparing the predicted stock prices against actual historical data, I was able to gauge the accuracy of our models and gain valuable insights into the potential and limitations of economic forecasting using ARIMA models.

Built With

• r
• rmarkdown