UCLA

This study aims to predict West Texas Intermediate (WTI) crude oil spot price using ARMA models and sentiment analysis. Market sentiment is quantified using data from Twitter. Overall, four analyses are presented: 1) Baseline Model, 2) Lagged Regression, 3) AR(1), and 4) AR(1) + Sentiment. The baseline model simply uses the prior month’s average as the current month’s forecast. The lagged regression model uses Ordinary Least Squares (OLS) to regress the one-month lagged price on current price. The AR(1) model analyzes the monthly percent change and the AR(1) + Sentiment model adds market sentiment an additional predictor to the AR(1) model. Results indicate that an AR(1) + Sentiment is the best model and decreases the RMSE (as small RMSE is preferred) by 12.5% compared to the baseline model. However, the RMSE is quite large ($6.498/bbl) because the model fails to predict changes in trends and large jumps in price. Future work should mainly focus on improving these two items.