Investing Insights from Big Data

Big Data and AI in Investment Strategy

• Goldman Sachs defines "big data" as the proliferation of digital information, noting that 90% of global data was created in the last two years, yet less than 2% is currently analyzed.
• The firm leverages machine learning and natural language processing (NLP) to consume unstructured text data—such as news articles, patent filings, and regulatory disclosures—at a speed and scale surpassing traditional manual analysis.
• Investment strategies supplement traditional financial metrics with alternative data sets, including credit card transactions, geolocation (GPS) data from mobile devices, and web traffic analytics.

Economic Impact Data from the Pandemic

• Foot Traffic Declines (Year-over-Year):
  • Casualty rates in physical retail and leisure sectors were severe, including:
    • Movies and entertainment: Down 85%
    • Casinos and gaming: Down 84%
    • Department stores: Down 79%
    • Amusement parks (leisure facilities): Down 80%
    • Apparel retail: Down 75%
    • Hotel resorts: Down 63%
• Growth Sectors:
  • Food retail increased 25%.
  • Home improvement retail increased 26%.
  • Hypermarkets and supermarkets increased 15%.
• Web Traffic Trends:
  • Educational services saw the highest growth with web traffic increasing over 80%.
  • Interactive home entertainment (video games) rose 40%.
  • Home improvement content rose 50%.
  • Streaming services increased 35%.
• Web Traffic Declines:
  • Airlines and hotel/cruise line sectors both saw traffic drop 70%.
  • Leisure facilities dropped 54%.
  • Casinos and gaming dropped 40%.

Quantitative Risk Management Framework

• Portfolio managers aim to quantify and neutralize "COVID sensitivity" to avoid outsized exposure, treating the pandemic as a specific investment factor similar to geopolitical or natural catastrophe risks.
• The firm designed a "COVID basket" or factor based on four specific dimensions to diagnose portfolio sensitivity:
  1. Revenue exposure from China.
  2. Presence of unique customers or suppliers based in China.
  3. Existence of subsidiaries located in China.
  4. Classification within industries expected to be most acutely impacted by a pandemic.
• This framework allows managers to calculate a "beta" specific to COVID-related names, aiming for neutral exposure rather than betting on the event's magnitude.
• The approach mirrors strategies used for other market-moving events, such as Brexit, the 2016 U.S. presidential election, and the 2011 Japan earthquake/tsunami, where investor uncertainty drove market volatility more than immediate fundamental changes.
• Alternative data is favored over traditional fundamental data for its real-time frequency and enhanced predictive forecasting power, as traditional data is often reported with a significant lag.