Golden Stacks

By Anh Nguyen, Anh Hoang, Kaleb Dickerson, and Phu Nguyen

Problem Statement:

• Discover and document how the environment and the markets are tied together
• Extract further insights into how impact the environment and markets are on one another.

• Briefly visualize the data to select the appropriate environmental metric for our model
• Pre-process the data
• Create and train the machine learning prediction model
• Stress-test the model with stocks from all sectors
• Draw insights and conclusions

​

Approaches:

Visualize Metrics and Stocks Trends

Possible Correlation?

Total CO2 emission vs US energy sector stocks

Challenges

​

• Manually identified a database related to the United States
• Perform SQL queries to organize and extract relevant data
• Result? Relatively clean dataset with stock information and the specified environmental metric
• Metrics that were collected monthly and annually, the metric is assigned uniformly across the daily stock data to study the effects.

​

Data Usage & Integration

Pre-process the Data

​

• Mismatched time metrics (Data collected daily vs monthly/annually)
• Environmental datasets have vast amounts of information to filter

​

Pre-process the Data (continued)

​

• Managing complexity
• Maintaining a consistent metric to organize the data (order by date and enforcing strict date equality)
• Maintaining clean data
• Filtering out irrelevant information by removing outliers and null values
• Coherent transformation
• Stock data is time sensitive and use the date/month to maintain consistency
• Pair-programmed table join operations to ensure correct logic implementation

​

Data Normalization & Serialization

Training the models

• Due to time constraint, we decided to identify the challenge as a binary classification problem.
• Classify in given year, will there be an uptrend or a downtrend for a given sector
• Support vector machine
• Training
• Trained the model with energy sector due to due to our personal assumption
• Three models created
• Base model: Trained without environmental features
• Irrelevant model: Trained with irrelevant environmental metrics
• U.S birth rates, mortality rates, population growth
• Relevant model: Trained with our identified environmental metric

​

Result

Insights and Conclusion

Tentative Conclusion: Accounting the environment factor, we can more accurately predict stock trends

• Assumption: Stocks in a specific sector is correlated to each other
• There might be some correlation between the environment and the markets.
• Interestingly, predictions for other sectors such as tech and healthcare improved as well
• How to Improve:
• A transformer architecture would likely perform better with this time-sensitive data
• Perform a feature importance test to choose the appropriate feature to add to our model.
• Graph the data distribution to identify outliers such as XOM, PNRL, etc.

​