The Impact of Donald Trump’s Tweets on Financial Markets

The Impact of Donald Trump's Tweets on Financial Markets

Student Name: Krishan Rayarel

Module: L13500 Economics Dissertation 2018 Supervisor: Spiros Bougheas Word Count: 7,345

This Dissertation is presented in part fulfilment of the requirement for the completion of an Undergraduate degree in the School of Economics, University of Nottingham. The work is the

sole responsibility of the candidate.

I am happy for this dissertation to be made available to students in future years if chosen as an example of good practice.

Abstract: This study tests the efficient market hypothesis (EMH) by analysing the effect of Donald Trump's company-specific tweets on financial markets for the period between 8 November, 2016 (the U.S. presidential election date) to 24 January, 2018 (a year after inauguration). Using a sample of 24 company-specific tweets, the results show that a tweet by Trump leads to statistically significant abnormal returns that last for 2 to 3 trading days. This is inconsistent with the semi-strong form of EMH. This is the first paper to test the attention-based investing hypothesis by Barber and Odean (2005) using Trump's tweets. Attention-based investing is a possible reason for market inefficiency as Trump's tweets lead to an abnormal trade volume of 43.54% on the day of the tweet and an increase in Google search activity on the week of the tweet.

1

1. Introduction ..................................................................................................................... 3 2. Background...................................................................................................................... 4 3. Literature Review............................................................................................................. 5 4. Data.................................................................................................................................. 8

4.1. Tweet Collection 4.2. Financial Data Collection 4.3. Sentiment Classification 5. Methodology .................................................................................................................... 9 5.1. Event Studies

5.1.1. Sample Selection 5.1.2. Normal and Abnormal Returns 5.1.3. Significance Tests for AAR & CAAR 5.2. Average Abnormal Trading Volume (AAV) 5.2.1. Significance Tests for Average Abnormal Trading Volume 5.3. Google Search Activity 5.4. Hypotheses 6. Results: Testing the Efficient Market Hypothesis (EMH) ........................................... 17 6.1. Market Model Coefficients 6.2. Average Abnormal Returns (AAR) 6.3. Cumulative Average Abnormal Returns (CAAR) 7. Results: Testing for Attention-Based Investing .............................................................22 7.1. Average Abnormal Trading Volume (AAV) 7.2. Google Search Activity 8. Discussion.......................................................................................................................23 8.1. General Analysis 8.2. Justifications 8.3. Limitations 9. Conclusion ......................................................................................................................24 10. Bibliography ..................................................................................................................26 11. Appendices ....................................................................................................................28

2

1. Introduction The social networking platform Twitter was established in 2006. It has since grown in popularity with over 330 million active users (Statista, 2018). An avid user of Twitter is Donald Trump. Recently elected as the 45th President of the United States, Trump, unlike his 44 predecessors, actively uses Twitter to express his views on global affairs. One would think that Trump uses Twitter to build support for his campaign however, Trump is renowned for using Twitter as a strategic tool to prevent US companies from moving operations overseas and publicly berating political leaders. With over 50 million followers on Twitter, Trump has become the most followed world leader and investors are now closely monitoring his tweets as indicators of future policy. This gives Trump exclusive power to influence financial markets with just 140 characters, as shown below:

Figure 1: 1-minute chart of Lockheed Martin's (LMT) stock price. Trump tweeted about Lockheed Martin on 22 December, 2016 after the market had closed. The next trading day, Lockheed Martin's stock price dropped by 2% and this decreased its market value by $1.2 billion. Trump may have been able to move the market this much because investors feel Lockheed Martin will be targeted by Trump in future policy and therefore this new information is reflected today through a decrease in its stock price. Figure 1 also shows that the stock recovered on the same day. This may imply markets rapidly incorporate new information and thus the efficient market hypothesis (EMH) holds.

3

On the other hand, some of Trump's tweets are non-informative. For instance, Trump responded to Nordstrom's announcement of removing Ivanka Trump's clothing line:

@realDonaldTrump: "My daughter Ivanka has been treated so unfairly by @Nordstrom. She is a great person ? always pushing me to do the right thing! Terrible!"

This tweet merely draws attention to Nordstrom's actions and is non-informative. Immediately after the tweet, Nordstrom's stock plummeted by -0.3%. However, after just 4 minutes (Marketwatch, 2017) the stock price recovered and at market close, the stock price had actually increased by 4%. The fact that the market reacts to a non-informative Trump tweet, suggests that the EMH does not hold. A possible explanation of these movements is Trump's tweets catch the attention of retail investors1 who cannot process all available information. This paper tests this hypothesis.

These examples highlight that analysing the impact of Trump's tweets on financial markets is a unique way of testing the EMH. Specifically, this paper will first test the hypothesis that a positive (negative) Trump tweet leads to a higher (lower) stock return. The study will then test whether stock prices rapidly incorporate the informational content of Trump's tweets, thus testing the semi-strong form of the EMH. Finally, this paper tests the attention-based investing hypothesis by Barber and Odean (2005) through analysis of trading volume and Google search activity.

2. Background Eugene Fama (1965) pioneered the EMH, which states that all available information at a certain time is fully incorporated into security prices. The intuition behind this is that, in the absence of frictions, this information disperses so quickly that security prices adjust before an investor has time to trade. There are three forms of the EMH; the semi-strong form is analysed in this paper. This form states that all public announcements are priced into the market. Public announcements may include earnings announcements and dividend changes. By classing Trump's tweets as public announcements, the semi-strong form of EMH is tested.

The literature review shows that there is no consensus on whether the EMH holds in practice. One possible reason for this is that the EMH assumes that investors consider all available

1 A retail investor is a non-professional investor that trades for personal reasons rather than for an institution. They tend to trade in much smaller amounts than institutional investors do. They have fewer resources to work with.

4

information when investing. When choosing what to invest in, investors have the option of many different stocks. Barber and Odean (2005) suggest retail investors have high search costs and therefore are likely to only consider information that grabs their attention rather than all available information. On the other hand, institutional investors are less prone to indulge in attention-based purchases as they have relatively more time and resources. This means that the EMH does not hold, as some investors do not consider all available information. This study will test this hypothesis by analysing abnormal trading volume to assess whether a tweet by Trump does actually generate investor attention. Furthermore, this study will contribute to Barber and Odean (2005) as it will monitor the Google search activity of these stocks during Trump's tweets. Google search activity determines whether small retail investors act upon Trump's tweets and drive market inefficiency, assuming small retail investors are likely to use Google to search for stocks, unlike institutional investors who will use Bloomberg Terminals.

3. Literature Review The EMH appears frequently in economic literature. A common methodology used to test the EMH is event studies2. This dissertation also uses event studies and therefore, the review starts by comparing key event study papers. These papers tend to analyse the effect of financial announcements, such as stock splits, on markets. Recently, Economists have focused on the impact of non-financial announcements, such as social media posts, on stock markets to test the EMH. Trump's tweets are non-financial announcements and therefore key papers that analyse the effect of these specific announcements on financial markets will be the focus of the latter parts of the review.

Fama, Fisher, Jensen and Roll (1969) were the first to test the semi-strong form of EMH using event studies. They analysed the impact of stock splits3 on stock prices using monthly data. They pioneered the market model4 in order to control for general market movements and focus purely on the effect of the stock split announcement. In the pre-announcement period, the returns on stocks are very high. This is because these companies have experienced dramatic increases in expected earnings and dividends, leading to an increase in stock price (hence the need for the stock split). The returns on these securities are even higher a few months after the split. This implies that the market is inefficient as it takes time to incorporate this new information. However, the authors note that co-existing events could be driving stock price changes. More

2 Event studies are a statistical method used to measure the impact of an event on a firm's value or stock price 3 A stock split is a way of reducing the stock price of the company in order to make the stock more affordable to investors. 4 A market model is a statistical method of finding expected returns of stocks. It is devised through a regression, which will be shown in the methodology section.

5

specifically, dividend payments usually change after a stock split and therefore this may affect the stock price. Once they control for dividend announcements, abnormal returns5 are insignificant for periods after the stock split announcement. This means that the market is efficient as stock prices adjust rapidly to new information. Similarly, this dissertation uses event studies and the market model to test stock market efficiency. In contrast, instead of using monthly data, this study uses daily data as Trump is likely to have a short-term effect on stock prices. This paper by Fama et al. (1969) initiated an array of research on the effect of different types of public announcements on financial markets such as dividend policy changes and quarterly earnings announcements.

Charest (1978) analysed the impact of dividend changes on the stock prices of New York Stock Exchange (NYSE) companies over the period 1947-1967. He finds that an increase in cash dividend leads to abnormal returns of 1%. A decrease in cash dividend leads to abnormal returns of -3.18%. He suggests there is a stronger negative effect as a decrease in cash dividend announcement is usually in the wake of other bad news. Moreover, he finds significant abnormal returns in the months following dividend changes. This implies the market is inconsistent with the semi-strong form of EMH. Similarly, Ball and Brown (1968) found evidence inconsistent with the semi-strong form of EMH. However, they analysed the impact of annual corporate earnings announcements on security prices. They found that stock prices do not fully incorporate new information instantly and abnormal returns are present many days after the announcement. They call this the post earnings announcement drift (PEAD) and it represents the delay in stock price adjustment to equilibrium levels. Bernard and Thomas (1989) test the EMH through analysis of quarterly earnings announcements on stock prices. Similarly, to Ball and Brown (1968), they find a PEAD of 60 days. Therefore, these findings are inconsistent with the EMH. Many studies analyse the impact of financial announcements on stock markets. This dissertation contributes to the existing literature by focusing on a non-financial announcement. Specifically, it focuses on how social media posts can affect financial markets.

The growing influence of media and social media has led to a flurry of research into the impact of media on financial markets. Tetlock (2007) analyses the interactions between a popular Wall Street Journal column and the stock market. In this journal column, brokerage houses, Analysts and other professionals give their views on stocks. By categorising the content of these columns into pessimistic, negative and weakly negative, he finds that high values of media pessimism lead to temporary downward pressure on corresponding stock prices and temporarily high values of

5 Abnormal returns are the actual returns of a stock minus the expected returns of a stock (which is calculated through the market model).

6

trading volume. This downward pressure lasts for two trading days. After making a fair assumption that these newspaper columns are non-informative about the fundamentals of the company, he argues that the market is inconsistent with the semi-strong form of EMH. Ranco et al. (2015) instead analyse the impact of Twitter on markets. During U.S. earnings season, a positive tweet leads to a cumulative average abnormal return (CAAR)6 of 4.22% on the day of the tweet whereas a negative tweet leads to a CAAR of -5.64%. These effects last for up to 10 days after the tweet, implying inconsistency with the semi-strong form of the EMH. Similarly, positive tweets posted during non-earnings periods lead to a CAAR of 3.65% whereas negative tweets lead to a CAAR of -3%. These effects last for up to 8 days after the tweet. This confirms that the EMH does not hold. Similarly, this dissertation analyses the impact of social media on financial markets, however it focuses on one Twitter user, namely Donald Trump. Ranco et al. (2015) show that including earnings announcement periods lead to an inflated figure for abnormal returns. Therefore, in this study, Trump's tweets that occur during earnings season are excluded in order to find the exact impact of a Trump tweet.

Malaver and Vojvodic (2017) focus on the effect of Trump's tweets on the Mexico Peso against the U.S. Dollar. They analyse the impact of 7,429 Trump tweets between June 16, 2015 and February 21, 2017. The results show that a negative tweet by Trump leads to an increase in the daily volatility of the Mexican Peso against the U.S. Dollar by 21.6%. This shows that Trump can influence foreign exchange markets. This dissertation will focus on stock markets rather than the foreign exchange market as methods for testing the EMH in foreign exchange markets are not as widely documented as methods for testing the EMH in stock markets.

Born, Myers and Clark (2017) test the semi-strong form of EMH by analysing the impact of Trump's tweets on stock markets. They analyse stock returns during the president-elect period7. The results show that CAAR are statistically significant for five trading days after the tweet, implying inefficient markets. They also compute abnormal trading volume and Google search activity in order to monitor whether noise trading is present. They find that the pattern of trading volume and Google search activity implies small noise traders are reacting to Donald Trump's tweets and therefore driving market inefficiencies. A caveat of this paper is a small sample size of only 15 tweets. Ge, Kurov and Wolfe (2017) build upon the limitation of Born et al. (2017) by considering a larger sample size of 48 tweets. They consider tweets from the pre-

6 Cumulative average abnormal returns are simply the sum of average abnormal returns over the event window. This is used to test how long it takes markets to incorporate new information. 7 The president-elect period is a period between the election date and inauguration whereby a candidate has won the election but has not entered office yet. For Trump, this period was between November 8, 2016 and January 20, 2017.

7

inauguration and post-inauguration period8 rather than just the president-elect period. Using event studies, they find that positive and negative Trump tweets, on average, move stock price by 0.80% and there is no asymmetry in the impact of positive and negative tweets. They find abnormal returns are statistically significant for 2 days after the tweet. This is inconsistent with the semi-strong form of EMH. A limitation of event study methodology is that co-existing events could move stock prices and therefore have an impact on the sample of stocks. For instance, preceding company or earnings announcements may cause misinterpretation of the impact of Trump's tweets. These papers do not consider the effect of earnings and company announcements that occur in proximity to Trump's tweets. Ge et al. (2017) note that only 18 out of the 48 presidential tweets do not have preceding related news events. To build upon these limitations, this dissertation categorises Trump's tweets into informative and non-informative tweets. Informative tweets are used to test the EMH and non-informative tweets will be discarded from the sample.

To summarise, this dissertation is building upon the limitations of previous literature. The sample size of Trump's tweets is increased by increasing the sample range. 24 tweets are obtained, which is an enlargement of the sample size of previous literature9. Furthermore, a subsample of Trump's tweets that are not responses to company announcements and do not occur during earnings announcements is created. This subsample is used to test the EMH. The key contribution of this paper to existing literature is the explanation for why markets may be inefficient. More specifically, analysis of abnormal trading volume and Google search activity provides insight on whether Trump's tweets lead to attention-based investing, which in turn drives market inefficiency.

4. Data 4.1. Tweet Collection An algorithm devised by is used to collect a full sample of Trump's companyspecific tweets. Tweets posted by Donald Trump's personal account (@realDonaldTrump) are used rather than Donald Trump's presidential account (@POTUS) as Trump simply retweets his personal account tweets on his presidential account and hence activity on this account provides no new information.

8 Inauguration, in this case, is the ceremony to mark the start of Donald Trump's presidency. The ceremony was held on January 20, 2017. 9 Considering Ge et al. (2017) have only 18 informative tweets.

8

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download