1 of 15

FINANCIAL EFFECTS OF ALGORITHMIC NEWS CONSUMPTION

Seiji Sasaki

PPS361: Algorithms, Journalism, and The Public Interest

Prof. Napoli

April 19, 2019

2 of 15

OVERVIEW

Using stock data as a representation for fiscal health in various industries concerning Facebook and it’s news feed algorithm. Then seeing what time periods yields the highest volatility for stock returns. Furthermore, analyzing areas of high volatility using Facebook’s press releases.

​

​

​

3 of 15

SIGNIFICANT PRESS RELEASES

  • Read though hundreds of Facebook’s press releases and made a list of all relevant headlines (Facebook Newsroom)
    • Curation updates to the News Feed
    • Efforts to curb fake or low quality content
    • Acquisitions

4 of 15

DATA BEING USED FOR VOLATILITY FORECAST

  • Facebook Stock Adjusted Closing Price (6/19/13 – 4/17/19)
  • Google Stock Adjusted Closing Price (6/19/13 – 4/17/19)
  • S&P Stock Adjusted Closing Price (6/19/13 – 4/17/19)
  • NewsCorp Stock Adjusted Closing Price (6/19/13 – 4/17/19)
  • Disney Stock Adjusted Closing Price (6/19/13 – 4/17/19)
  • Twitter Stock Adjusted Closing Price (11/7/13 – 4/17/19)

​

​

5 of 15

WHY IS DISNEY INCLUDED?

  • NewsCorp Subsidiaries
    • Journalism
      • NY post
      • Wall Street Journal
      • Fox New
    • Entertainment
      • Sky network
      • Minor stake in Hulu

NewsCorp Discounted by Disney

​

Formula:

NewsCorpReturns –( DisneyReturns* NewsCorp%EntertainmentAsset)

6 of 15

PRELIMINARY DATA VISUALIZATION

7 of 15

FINDING TIMES OF HIGH VOLATILITY

  1. Moving Average
    1. Takes more than a day for changes in stocks to occur (can’t use daily returns)
    2. Stock shifts probably will not align with weeks (can’t use weekly returns)
  2. Calculate the Standard Deviation of the moving average returns
  3. Set the threshold to +-2 * standard deviation

​

​

8 of 15

JUSTIFYING THE THRESHOLD

  • Random Walk Theory
    • Traced Back to Jules Regnault (1863)
      • Popularized by Burton Malkiel (1973)
        • “Random Walk Down Wall Street”
    • Definition
      • Stock prices follow a random path and are unpredictable
    • Stock returns are normally distributed around their means
      • Thus 2 standard deviations would leave us with around 4% of the most extreme data

9 of 15

APPLYING THE THRESHOLD

10 of 15

11 of 15

SIGNIFICANT MATCH-UPS FOR FACEBOOK

​

  • “Facebook to acquire WhatsApp”
    • 2/19/14
  • “Using Qualitative Feedback to Show Relevant Stories”
    • 2/1/16

​

12 of 15

SIGNIFICANT MATCH-UPS FOR GOOGLE

  • “More Local News On FB”
    • 1/19/18

13 of 15

SIGNIFICANT MATCH-UPS FOR NEWSCORP

  • “Using Qualitative Feedback to show relevant stories”
    • 2/1/16
  • “Facebook to Remove Trending”
    • 6/1/18

​

​

​

14 of 15

CHALLENGES

  • Hard to find stock data on local news
    • Public news companies tend to be national news or huge conglomerates of local news
  • Correlation does not equal causation
    • Even in a RDD(Regression Discontinuity Design) other events could have caused volatility
  • Stock prices may not perfectly reflect a companies health
  • This type of analysis does not account for the long term effects
    • Algorithmic changes could and probably have a strong long-term benefit

​

15 of 15

CONCLUSION

  • Acquisitions and macroeconomic events have a far larger impact on Facebook stock than algorithm updates and other curational changes
  • NewsCorp seems to be pretty well insulated from acute fiscal effects stemming from Facebook’s curation algorithm
  • Improvements to algorithms lead to a lagged average positive return when the threshold is not implemented
  • NewsCorp is the only asset that has a negative average growth rate

​