SSOE-Crude-Oil-Forecasting

Business Problem

1. SSOE has a derivatives trading desk which trades in Crude Oil and iron ore.
2. They are only allowed to make one trade per day.
3. So having a good idea about the market and its sentiment can potentially help increase the profits of the company.

Methods

1. Assess the effectiveness and accuracy of new data sources and data gathering techniques.
2. Understood the analytical tools and techniques and the ability to apply the relevant tool/technique to the given business problem.
3. Python based libraries for Forecasting, Natural Language Processing, Cloud analytics, GPU based computing etc.
4. One day ahead Forecasting based on different metrics for SSOE.
5. Exploratory Data Analysis (EDA) : Finding relationships between variables, outlier detection, plotting graphs of those relationships, correlations and preparing final data for modelling.
6. Time Series Univariate Models : ARIMA, SARIMA, Holt-Winters.
7. Multivariate Regression Models : Random Forest, XgBoost, Random Forest Quantile-Regression.
8. HyperParameter Optimization : Finding the best parameters using RandomSearch, HyperOpt.
9. Twitter Based News Analytics to find the market sentiment of crude oil for the next day.
10. Using libraries like GetOldTweets3, Python-Twitter to fetch historical Tweet data to train classifier models (LSTM, GRU).
11. Unsupervised techniques like Word Cloud and Topic Modelling for Twitter Data visualization.

Results

1. Achieved a Forecasting accuracy of 97% and a Directional accuracy of 55% for a test period of 5 months.
2. Baseline Twitter Classifier model able to acheive an accuracy of 60% in correctly classifying the price change in Crude oil for each day.