Tanner Eastmond and Dr. Joseph Price, Economics Department

Introduction

Loss aversion is a well-documented behavioral phenomenon originally proposed by Kahneman and Tversky (2013). The idea is that people value losses more than they do commensurate gains. Many researchers have examined the effects of loss aversion on an individual level, but many economists think that these effects evaporate in highly competitive situations and when professionals are involved. This study seeks to examine whether individual loss aversion is reflected in aggregate stock prices.

We investigated whether individual perceptions of news about companies affect the stock price in the short run, namely if headlines perceived to be negative for a stock price lead people to sell, driving the stock price down. We expected that prices would drop twice as much in response to a negative headline as they would increase with a positive headline, which is a common prediction when considering loss aversion.

Methodology

To answer this question, we first needed to be able to collect information on headlines. We decided to focus on Fortune 500 companies, as they are often many of the companies in the forefront of public attention. We scraped around 20,000 headlines from the internet for these companies. Our focus was to use search results on Google, because we wanted to proxy what most people would see if they were trying to follow a stock on a day to day basis without having time to read many articles. For each headline, we gathered the text of the headline as well as the time when the headline was originally posted.

After collecting these headlines, we needed to be able to categorize what an individual would think about the headline. To do this, we used a Likert Scale for what the individual classifying the headline thought the effect would be on the given stock price, namely very positive, positive, neutral, negative, or very negative. Using this scale, we designed a task on Amazon Mechanical Turk, which is essentially a marketplace for human labor. A ‘requester’ puts up a ‘HIT’, or a human intelligence task, that ‘workers’ are paid to complete. In our case, we put up a survey where an individual classified 25 headlines based on our Likert Scale.

Then we used Google Finance to get daily stock price data for the focus stocks. Bringing this data all together, we then had a dataset of stock prices over time with categorized headlines.

We used linear regression to examine the effect of each type of headline (namely very positive, positive, etc.) on the closing stock price that same day and every day for the week following when the headline was published.

Results

With all of the headlines categorized, we found that individuals considered 32 percent of the headlines strongly positive, 32 percent positive, 25 percent neutral, 9 percent negative, and 2 percent strongly negative. This shows that for the most part, people thought that the headlines were actually good news for the stock price. Whether this is because people are generally optimistic about prices or because news outlets report more good news is beyond the scope of this study.

The main results did not reflect our expectations. Each individual headline had a small, statistically insignificant effect on the stock price both the same day and the following seven days.

Conclusion

Despite the inconclusive results, we are optimistic that we will better be able to answer this question in the future for a number of reasons. First, the majority of the headlines we scraped were search results from Google collected retroactively. This means that despite the fact that they came up on Google at a future date, it is possible that the headlines collected were not those that a person would see immediately on the day they were published. We are going to verify this further potentially by scraping headlines continuously in the future. This also solves another potential concern, which is that we did not catch a majority of the headlines. If we did not get most of the headlines, we may not find the correct affect of headlines on stock prices simply because we are not measuring enough headlines. The previous strategy will also help us to remedy this issue.

Another reason to be skeptical of the inconclusive results is that the stock data is collected only at the day level. It is possible that loss aversion is a short-term phenomenon on the aggregate, meaning that day level stock prices are likely not granular enough to observe the drop in price associated with the headline. We are thus going to obtain minute level stock price data in the future so that we can match the publishing of headlines with prices in the same minute.

One last extension we will pursue is having more than one individual classify the headline, then using the average rating for our analysis. This would avoid biasing our results by individual perceptions.

Despite these extensions we will pursue in the future, we found that the majority of stock headlines are not considered to be bad for the stock price by individuals reading the headlines. Two reasons for this may be that people are generally optimistic about stock prices or that news outlets largely report positive news. Further investigation is warranted here to assess the reasons behind this observation.

References

Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk. Econometrica, 47(2), 263-292.