Goldman Sachs, Golden Stacks

Inspiration

The thought-provoking question Goldman Sachs provided made us think about analyzing trends. Doing that visually is easy enough but could we do better?

What it does

The model tries and predicts an upward or downward trend for a sector given stock history information, and environmental data. We created 3 models:

1. one with no environmental factors
2. one with relevant factors
3. one with irrelevant factors. We hoped to show that 1 & 3 performed about the same, with an improvement in model 2.

How we built it

Our own assumption was the stocks in one sector are correlated to each other performance: for ex, BP and Chevron can be competitors to each other, but still under the same influence of environmental factors. We used the Support Vector Machine architecture to create a quick and linear machine model given our own time constraint. There were three models: one with only stock history data as the baseline model, one similar to the first one but add Co2 emission and rainfall level, and the last one is with might be irrelevant mortality rate, population growth, and accessibility rate to electricity.

Challenges we ran into

Preprocessing the data was the most challenging part, we had to pick datasets to augment the stock data that had information more frequent than annual.

Accomplishments that we're proud of

Preprocessing and combining datasets.

What we learned

Data science skills such as preprocessing and training a model.

What's next for Golden Stacks

More time would allow for the use of more refined data and more sophisticated model architecture. We believe the transformer architecture would excel on this type of time-series data. This is because the transformer would allow the model to learn how each data point relates to other data points, basically learning the relationship they have with one another.

Built With

• python
• scikit-learn
• snowflake
• sql