Dashboard for Geopolitical Event Probabilities

Visualizing a research project on predicting the GBPUSD rate, supervised by Eugene Neduv

• Features are sentiment scores and volumes in news articles, and option-implied volatility.
  • sentiment scores is defined as a normalized number of news sentiment.
  • We gathered all news with Brexit tags from Reuter news archive from 2015 to 2018.
• We take those three features and express them as a combined probability function

Used Bokeh and ran a Bokeh server on my laptop

My responsibilities include

• writing this dashboard from the scratch.
• consolidating and verifying existing works.
• writing a prediction model using XGBoost

Screenshots

Two Plots on Sentiment Analyses

1) Sub-topic: F/X Market

• Aggregation frequency: Weekly
• Aggregation method: Proportion of positive news
• This sub-topic was manually identified after applying topic modeling

2) Sub-topic: Global Financial Market

• Aggregation frequency: Weekly
• Aggregation method: Proportion of positive news

A Plot on Combined Probability

• Top: combined probability function with sliding bars for various weights applied to this function
• Bottom: three features are plotted
  • Option-implied volatility, sentiment scores (polarity), news volumes in order (left to right)

Predicting a direction of GBPUSD rates

• Accuracy: 0.58
• Feature importances: 3 came from non-traditional features, 1 came from traditonal feature, option-implied volatility
  • News counts(Most contributed to prediction), Subjectivity (3rd), Positive polarity(4th)
  • Option-implied volatility (2nd)

Photos

• Date: April 3, 2019
• Presented on Data Science day at Columbia University