Periodic Stock Returns (Large cap Financials) v/s News Analytics

This paper looks at stock returns of large cap Financial Services companies listed on major stock exchanges around the globe and weighs its performance against the market news indicators as calculated by Infotrie.

For the purpose of our analysis, we have considered News Sentiment Data provided by Infotrie within the time period (1 Jan 2013 – 19, May 2014) and have assessed equities from NASDAQ, NYSE, Tokyo Stock Exchange and Paris Stock Exchange.

In this post we also attempt at establishing a deeper understanding of how news trends work, specifically we will look at how InfoTrie’s Buzz Index can be a significant indicator of stock returns trend over look back periods of 1,2 and 3 months.

Before we start looking at a detailed stock wise analysis we would like to mention certain key points and coin certain hypothesis that the reader can later observe in this paper.

  1. Buzz scores are nothing but normalized values of news volume over the past 7 days in the range 0 through 10, in simpler words the reader may consider it to be rate of change of news volume over past 7 days.
  2. Inference: Buzz scores may not good immediate indicators when considering wider look-back windows, they positively forecast a trend in returns, specifically a downward trend. Hypothesis: They represent the decay or rise in news volume thus reflecting the market’s interest or indifference. Interest helps a stock rise and indifference does the opposite.
  3. Sentiment: Ranges from 0 to 10. Values above 5 deemed as bullish while lower values are bearish indicators.
  4. News Volume: The higher, the more in-the-news a company is.

We correlated the Prices with Periodic Buzz, Volumes and Sentiment to pick up those with significant relationships (>40%). These companies will then be further analyzed to determine if our hypothesis were true.  In our research, we analyzed 16 companies but in the interest of being succinct, will only discuss 3 in this post. The full report will be available upon request.

1.     Berkshire Hathaway Inc. (BRK) [NYSE]

Sentiment News_Volume Buzz
Avg. 6.33 21.25 3.57
S.D 1.56 20.95 2.98

Beta: 0.39

The table above gives us an idea that BRK is “in the news” firm with a relatively bullish sentiment and a low buzz score implying a relatively stable news volume. Our correlation analysis suggests that we can observe that most significant indicator to be Buzz index over weekly and monthly intervals. Let’s see:

Fig 2: Weekly and biweekly returns v/s respective buzz momentum.

From Fig 2 we can observe value of returns having a healthy correlation with buzz scores as calculated and we can specifically point out that decay in buzz score are significantly related falling returns.

Fig 3: Monthly returns v/s SMA_60 and Buzz_60

From the above figure, we can say the buzz momentum follows a pattern similar but more precise than the pattern in Fig 2 thus strengthening our hypothesis.

2.     American Express Company (AXP) [NYSE]

Sentiment News_Volume Buzz
Avg. 6.41 16.88 5.45
S.D 1.69 16.19 3.52

Beta: 1.23

Essentially AXP has a bullish average sentiment with significant news volume. While at first glance it seems similar to Berkshire Hathaway (BRK), AXP is more volatile and has a positive correlation between simple moving averages (SMA) of Price and periodic Time values (unlike BRK). Thus, we find this worth observing.

Fig 6: Returns_90, Buzz_90 and SMA_90

Figure 6 reflects of a decay scenario where the decaying value of Buzz_90 relates to a steep decay in returns. We believe the steep decay has more to do with volatility rather than the magnitude of decay of buzz.

3. Capital One Finance (COF) [NYSE]

Sentiment News_Volume Buzz
Avg. 5.75 14.48 4.15
S.D 1.97 15.50 3.37

Beta: 1.60

COF offers a rather contradictory picture to our hypothesis. In this scenario, Buzz_90 is negatively correlated to 90-day returns. The most significant indicator was the 90-day volume, which also had a negative correlation to returns. From figure 7 below, we can conclude that this appears to be a non-causal relationship from the following figure.

Fig 7: Returns_90 and NewsVolume_90

Conclusion

If we consider company’s data samples to be a probabilistic hypothesis testing mechanism, our initial hypothesis regarding Buzz scores holds true 11 times out of 16.

Empirically we also saw that returns (60-90 days) for equities with a large news volume are more significantly correlated to the decaying Buzz values over 60-90 day look back windows while those with marginal or negligible news volume were related to upward as well as downward buzz trend.

One may have also noticed a marginal correlation between sentiment scores and returns over small periods (up to 2 weeks) which is consistent with SMA values reflecting generic correlations over small intervals of time.

However, is 11 out of 16 good enough?

In periods of gradual consistent growth, Buzz scores are more likely to behave as decay indicators and may not conform to our hypothesis.

Therefore, further research is needed to theoretically justify the pattern observed here and this has to be tested across all stock exchanges.

More blogs

Leave a Reply

Your email address will not be published. Required fields are marked *