Skip to content

Using natural language processing, sentiment analysis, and machine learning algorithm in order to determine whether Elon Musk’s tweets have an impact on Tesla’s stock price and its volatility.

Notifications You must be signed in to change notification settings

tamobee/Tweeting-Tesla

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

54 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Tweeting Tesla

Project 2: Tamari Bachulashvili, Aidan Dowdle, Lisa Esberger, Minyeong Han, Carlos Tacchi

Introduction

Using NLP, sentiment analysis, and other features we used machine learning to determine whether
Elon Musk’s tweets have an impact on Tesla’s stock price controlling for the market. We then used
these findings to develop algorithmic trading programs to trade in Tesla Stock.

Some of the key questions that we wanted to answer:

  1. Is there a correlation between Elon Musk’s Tweets and returns?
  2. Can we predict the change in daily and weekly returns using machine learning?
  3. Could investors and day traders benefit from an algorithm that tracks Tweets?
  4. Which (buy, sell, hold) strategy could be the most profitable?

Data Collection/Cleaning

The data collection and cleaning process was lengthy. We pulled the Twitter data from a scrapper
called Twint that pulled Elon Musk's tweet as well as engagement metrics that we thought would be
valuable to use as Machine Leaning features.

elon_tweets

After uploading the tweets, we encountered some practical problems during the cleaning process.
We had to write lengthy code to push weekend and off hour tweets and engagement into the next
trading day. Once this was accomplished, we pulled the stock data from Yahoo Finance as it
incorporated stock splits in the data. We also pulled dates on earnings announcements from
Beautiful Soup as this could also be an important feature for movement in stock price.

From the twitter data itself we used NLP to understand the sentiment of the tweets. From
the compound score we also created positive, negative and neutral labels. After running
frequency analysis on most common words, we manually intervened to remove non-relevant stop
words for the most accurate results.

frequency_analysis

Data Exploration

As we looked through our data to determine our features in our model we focused on a few key ideas

twitter_engagement

We found there to be a correlation between engagement from other twitter users and Tesla stock movement
We also found there to be a high degree of correlation between sentiment and Tesla stock movement.

It is important to note that when determining our y variables for Tesla Stock, instead of using Tesla
stock price change we used a daily and weekly basis. We also considered for macro-climate economic
bias by controlling with a tech-stock based index fund (QQQ). Our output was the difference between Tesla
and QQQ on daily/weekly basis.

Model Analysis

We used random forest regression machine learning as a way to reduce over fitting in our model so we could
look at mean prediction of the individual trees as output. Our x variables included Sentiment analysis (compound, positive, neutral, negative),
engagement (retweets, replies, likes, etc), keywords
(both individual keywords and keyword strength) and quarterly earning report dates were used. After training
our model we achieved a success with a MSE score on daily return for test data at .001 (train data MSE .0005).
The weekly returned similar results with our test data at .002 (train data MSE .006). Overall feature importances
were skewed towards twitter engagement and compound sentiment score.

features

Algo Results/ Project Conclusions

We took our model results and looked at buy/sell flags for Tesla. When running an algo that traded these flags
we found that our results were not as good as buy and holding strategies for Tesla given its quick rise.
We had tested the 5-day and 10-day (profit of $1,021,691.88), 10-day and 20-day ($1,840,235.96), 20-day and 50-day ($1,386,095.99), and 50-day and 100-day ($1,914,389.90). With Tesla’s relatively continuous upward trajectory from 2016-2020, buying and holding this position throughout all four years would have most likely yielded the greatest reward.

We looked at four different buy and hold strategies:
features

Overall, we believe we answered our questions that we set out initially.

We discovered a correlation between Elon Musk’s Tweets and changes in Tesla’s returns using sentiment analysis.
Our model appeared to predict daily and weekly returns well as it achieved a mean squared error close to zero.
Our algo trading exploration provided mixed results and although we felt it could not be used for day trading,
it could be used to indicate timely entry points to the market.

About

Using natural language processing, sentiment analysis, and machine learning algorithm in order to determine whether Elon Musk’s tweets have an impact on Tesla’s stock price and its volatility.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published