SillyCon - Algorithmic Trading Assistant

Introduction

SillyCon is an algorithmic trading bot that uses Elon Musk’s tweets and machine learning to predict cryptocurrencies’ price movement, then trades accordingly and backtests its performance.

How to run the app

Option 1: sillycon.herokuapp.com

Option 2: python3 main.py

Background

Near the inception of Fintech Project Two, Bitcoin and Dogecoin prices skyrocketed after Elon Musk's tweeted about each coin. Similarly, the price of Gamestop was causing trouble on Wallstreet after Reddit users "manipulated" the price of Gamestop (GME) shares. As such, the team theorised that there is a strong correlation between social media and the price movement of Bitcoin and Dogecoin. The team decided to build an algorithmic trading assistant to analyse this in further detail for crypto traders.

Hypothesis

Silicon Valley tech entrepreneurs tweets (e.g. from Elon Musk) have a high correlation to crypto price movements, and that this movement can be successfully predicted with machine learning models.

User Input

• A drop-down list consisting two source selections:

  • Elon Musk Tweets

  • Google Trends Data

• A drop-down list for choosing the desired crypto coin:

  • Bitcoin

  • Dogecoin

App Output

The app has an interface which provides the following output plots across multiple tabs:

• General insights plots

  • tweeting_price_curve_btc
  • tweeting_price_curve_doge
  • cumulative_return_curve_btc
  • cumulative_return_curve_doge
  • price_curve_btc
  • price_curve_doge

• Plots that show the results of a Random Forest model (Price only)

  • rf_ema_closing_prices_btc
  • rf_ema_daily_return_volatility_btc
  • rf_bollinger_closing_prices_btc
  • rf_predicted_vs_actual_btc
  • rf_predicted_vs_actual_last_ten_btc
  • rf_cumulative_return_btc
  • rf_ema_closing_prices_doge
  • rf_ema_daily_return_volatility_doge
  • rf_bollinger_closing_prices_doge
  • rf_predicted_vs_actual_doge
  • rf_predicted_vs_actual_last_ten_doge
  • rf_cumulative_return_doge

• Plots that show the results of a fixed trading strategy (Buy upon relevant Tweet, Sell 24hrs later)

  • entry_exit_price_plot_btc
  • entry_exit_portfolio_plot_btc
  • portfolio_evaluation_table_btc
  • entry_exit_price_plot_doge
  • entry_exit_portfolio_plot_doge
  • portfolio_evaluation_table_doge

• Plots that show the results of an algorithmic trading based on RNN LSTM (Price + Tweets)

  • rnn_predicted_positive_return_curve_btc
  • rnn_cumulative_return_plot_btc
  • rnn_predicted_positive_return_curve_doge
  • rnn_cumulative_return_plot_doge

• Plots that show the results of an algorithmic trading based on Random Forest (Price + Tweets)

  • rf_predicted_positive_return_curve_btc
  • rf_cumulative_return_plot_btc
  • rf_predicted_positive_return_curve_doge
  • rf_cumulative_return_plot_doge

• Plots that show the results of an algorithmic trading based on RNN LSTM (Price + Google Trends)

  • google_predicted_positive_return_curve_btc
  • google_cumulative_return_plot_btc

Libraries Used

pandas, pathlib, hvplot, tensorflow, sklearn, dotenv, numpy, random , os, json, pickle, re, time, bs4 , urllib, requests, datetime, sys, collections

APIs Used

Twitter API , Cryptocompare API

Explanation of Each File and Folder

• main.py = Main file which co-ordinates the entire app and calls functions in the other .py files (used for launching the app)

• retrieve folder (various files) = Fetches raw data from Twitter and Cryptocompare, and creates Pickle files for processing

• clean_data.py = Processes raw data (Pickle files) into suitable clean dataframes for consumption by machine learning models

• algo_trading_rf.py = Prepares data and trains model for Random Forest (Price + Tweets)

• algo_trading_rnn.py = Prepares data and trains model for Recurrent Neural Network (Price + Tweets)

• algo_trading_rnn_google.py = Prepares data and trains model for Recurrent Neural Network (Price + Google Trends)

• model folder (various files) = Stores Machine Learning models

• data folder (various files) = Stores ready-to-use datasets (e.g. clean DataFrames for analysis, algo trading results, etc.)

• process_data.py = Prepares data and trains model for Random Forest (Price only). Runs all models and creates plots for Trading Strategies and General Insights

• make_word_cloud.py = Generates WordCloud images for plotting

• build_dashboard.py = Creates dashboard layout for Interact function outputs

• chatbot folder = Contains AWS Chatbot concept model (json)

• image folder (various files) = Contains WordCloud and ReadMe images