Covfefe & NLP — Do Trump’s tweets transfer the Inventory Market?

Illustration by Robert Connolly, Barron’s

Twitter, one of many world’s largest social media companies, has now change into a platform for politicians, organizations, and corporations to present updates to their followers. Customers representing an organization or a political celebration use Twitter to state views on present information, push their political campaigns and even affirm official coverage selections. Necessary figures which have used the location to broadcast their ideas to thousands and thousands of followers embrace Tesla founder Elon Musk, European Fee President Donald Tusk, and present UK Prime Minister Boris Johnson. The usage of social networks can have massive monetary/enterprise implications — for instance, Elon Musk caught quite a lot of unfavorable consideration when he tweeted that ‘funding was secured’ to take his electrical car firm Tesla personal. Tesla’s share value rose as a lot as eight.5% (1) and this resulted in a punitive investigation into Musk by the Securities and Alternate Fee. This led to Musk being ousted as Chairman for the corporate that he based, in addition to a $20 million effective — all due to one tweet. The President of america, Donald Trump, at present has 66.8m followers on Twitter, underneath the deal with “@realDonaldTrump” (2). Trump is very energetic on his Twitter account, averaging roughly 10 tweets a day since 2016 (three). Trump is candid along with his tweets, usually making ‘advert hominem’ remarks towards his critics, particularly towards the Democratic Get together. Twitter performed an enormous half within the earlier common election (held in 2016), with Trump’s feedback ultimately prompting an investigation into former rival presidential candidate Hillary Clinton.

The Environment friendly Market Speculation

In a broad sense, merchants and portfolio managers make selections on whether or not to purchase, promote or maintain particular equities (shares of an organization) primarily based on the basics of the agency, its administration, and its anticipated worth sooner or later. It’s due to this fact important to maintain up with firm information, which can contain earnings calls, new product bulletins, and extra lately, information delivered by tweets. This ties in with an concept known as the ‘environment friendly market speculation’. The EMH maintains that shares commerce at their intrinsic worth on inventory exchanges, i.e. all details about an organization, its administration, and its valuation are constantly expressed within the present share value. This suggests that when Trump tweets, this data might be mirrored in monetary markets even earlier than it hits headlines. So now, let’s acquire the info.

Information Assortment

To research Trump’s tweets, we’ve collected roughly 3000 tweets from 28 July 2019 to 16 November 2019 (retweets included), utilizing the Twitter API by way of the Python module Tweepy. To do that, we created a developer account on Twitter after which used the cursor methodology to acquire the tweets, placing them right into a dataframe:

dtweets = tweepy.Cursor(api.user_timeline, id="realDonaldTrump", include_rts=True, depend=200, tweet_mode="prolonged")df = pd.DataFrame(columns=["Date Created","RT Count", "Tweet"])

The tweets have then been pre-processed to ‘clear’ them, eradicating hyperlinks and emojis.

def clean_tweet(tweet):
return ' '.be a part of(re.sub('(@[A-Za-z0-9]+)|([^0-9A-Za-z t])|(w+://S+)', ' ', tweet).cut up())#The lengthy string of what looks like nonsense primarily takes away any characters that aren’t regular english phrases/letters.
df["Clean Tweet"] = df['Tweet'].apply(lambda x: clean_tweet(x))

Subsequent, we’ve utilized a method known as pure language processing, or NLP. At a excessive degree, NLP makes use of computational methods to research sentences to realize a particular final result. Though it might sound advanced, you possible use NLP every single day. Widespread examples embrace predictive textual content on cell phone keyboards, language translation on webpages, and search outcomes. The NLP software we’ve used is known as ‘sentiment evaluation’, which primarily determines whether or not every Tweet is optimistic, impartial, or unfavorable. The Python module ‘Textblob’ goals to find out the sentiment of a tweet by parsing by way of every sentence and giving it a rating between -1 and 1. As an illustration, a Tweet together with the phrase “fallacious” would often be given a unfavorable rating. Furthermore, a Tweet containing the phrase “thanks” would often be given a optimistic rating. In fact, there are methods wherein a Tweet might not at all times be given the right sentiment, particularly for financial variables. For instance, “excessive unemployment” could also be given a optimistic sentiment when low unemployment is definitely higher for the economic system. Nevertheless, for this process, we’ve used the default Textblob performance because it has already undergone testing on massive datasets by way of machine studying. Right here’s a snippet of the code used to use sentiment evaluation to the tweets:

def analyze_sentiment(tweet):
evaluation = TextBlob(tweet)
if evaluation.sentiment.polarity > zero:
return 'Optimistic'
elif evaluation.sentiment.polarity ==zero:
return 'Impartial'
return 'Detrimental'

df["Sentiment"] = df['Clean Tweet'].apply(lambda x: analyze_sentiment(x))

for tweet in dtweets.objects(3000):

if i == 3000:
#code stops at 3000 tweets topic to Tweepy's limits

Primarily, this code defines a perform that classifies a tweet primarily based on its sentiment and offers it rating utilizing Textblob, by way of the usage of ‘if’ ‘else’ statements. We then create a brand new column in our pandas dataframe to the ‘clear tweet’, iterating by way of every tweet. As soon as we run this code, the dataframe ought to look one thing like this (any empty/nonsensical clear tweets have been faraway from our evaluation):

Information Visualisation

Utilizing the info we’ve collected, we will clarify its distribution visually. Right here Under we present the distribution of sentiment for every of Donald Trump’s tweets. Of the roughly 3000 tweets analyzed, 662 tweets got a unfavorable sentiment, 798 impartial and 1415 optimistic.

Supply: Twitter, Python. Tweets collected utilizing Twitter API & Tweepy.

The chart beneath exhibits that the frequency of Trump’s tweets has dramatically elevated from the averages of 2016 (contains retweets).

Supply: Twitter, Python. Tweets collected utilizing Twitter API & Tweepy.

To measure the impact these tweets have on inventory market returns, we’ve determined to make use of the S&P 500 Index, a weighted common of the 500 largest firms buying and selling on the US inventory trade.

Filtering down the info

To be able to use essentially the most consultant knowledge attainable, we’ve filtered tweets which are associated to the inventory market in accordance with particular phrases, similar to ‘commerce’, ‘China’, ‘Powell’ and extra, as they’re extra prone to have the potential to maneuver markets. The whole listing of phrases and phrases is beneath:

“commerce”, “charges”, “inflation”, “economic system”, “progress”, “manipulation”, “foreign money”, “greenback”, “china”, “Fed”, “Powell”, “Xi”, “tariffs”, “impeach”.

This trimmed down our complete tweet depend to round 500. The tweets have been then given a rating relying on their sentiment. A optimistic tweet was given a rating of zero.eight, a unfavorable tweet -1.1 and a impartial tweet was given zero.four. These scores have been later remodeled right into a sequence, rebased on the worth of the S&P 500 at our begin date. When giving the scores, we thought of that buyers are loss averse — which means that the disappointment that they get from a loss is larger than the happiness they acquire of the identical magnitude. This is the reason a unfavorable tweet has a better magnitude than a optimistic one. Additionally, one might count on a impartial tweet to be given a rating of zero — right here, we assumed that the steady-state of the inventory market was above zero, as development world progress is above zero.

Supply: Yahoo Finance (four)

After we mix the 2 sequence collectively, we get the chart beneath:

Supply: Yahoo Finance, Twitter


Visually, we will see that the development of the inventory market and Trump’s tweets appear to maneuver collectively. When operating a correlation between the 2 sequence, the optimistic correlation of zero.63 implies that Trump’s tweets do in actual fact transfer with the inventory market. Nevertheless, importantly, correlation doesn’t suggest causation. In different phrases, we’ve no indication of the route of the connection. There are additionally many particular person elements that will have an effect on inventory returns similar to rates of interest, inflation expectations, and different idiosyncratic elements. The interpretation of the connection we’ve discovered is extra possible that the tweets coincide with market-moving occasions similar to a shocking financial knowledge launch or a financial coverage choice after markets have reacted to the info. Subsequently, we can not say that with 100% confidence that Trump’s tweets pre-emptively have an effect on the inventory market. Nevertheless, it’s greater than possible that the sentiment of the tweets correlates positively with market returns.

Thanks for studying! Please be happy to depart any feedback for any insights you might have.

Leave a Reply

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