Machine-Learning-Stock-Trading-Bot

Objective: Day trading bots are ubiquitous projects. Swing trading bots are not as common, especially one that performs both fundamental and technical trading methodology with machine learning implementation. Staying true to its swing trading objective, the program is only designed to execute selling and buying once per day.

Mechanics: --> Implements sentiment analysis on equity related "tweets" to identify optimistically trending stocks --> Performs data collecting and scrubbing of 3 months worth of price history, press release, sentiment polarity, and subjectivity data --> Predicts and calculates target price through machine learning random forest model for the next 20 business days

a.) config.py -- Storage of all api keys: ALPHA VANTAGE, NEWSAPI, ALPACA

b.) forecast_library.py -- Data scrubbing and machine learning function

1. def importing_twitdata(id1, id2) --> Imports stocktwit data from 2 different automated accounts that posts daily trending stocks --> Data is scrubbed by deleting twits that does not literally contain the ticker symbol of a stock

2. def twit_sentiment_analyzer(raw_data) --> Uses TextBlob to determine the twits sentiment on particular stocks --> Disregards any twit with polarity lower than 0.4 (Polarity ranges from (negative) -1 to 1 (positive))

3. def pulling_price_history(ticker) --> Pulls the latest 3 months data of stock price history from alphavantage and puts into a DataFrame --> Supplements DataFrame by adding press release dates and news sentimentality and subjectivity to the DataFrame --> NewsAPI data was stored in clustered dictionary and list; thus it was hectic handpicking appropriate sections of data --> Format and scrub DataFrame so that it is ready for machine learning

4. def random_forest_forecast(data) --> Chose the random forest regressor model for its powerful prediction capability on short term forecasts and limited data --> Lacking data (merely 3 months worth of stock price history) discouraged implementing other model (free open data was used) --> Random forest regressor had the highest score --> Train 80% of the data and test 20% --> Predicts the stock's price for the next 20 business days and return the maximum price multiplied by it's confidence level

c.) main.py -- Executing Trade on Alpaca (https://alpaca.markets/)

1. def buy() --> Executes created functions in forecast_library.py in the following order: *** def importing_twitdata(id1, id2) *** def twit_sentiment_analyzer(raw_data) *** For each candidate stock, the bot avoids buying stocks already in position, calculates number of shares to buy *** qty_desired = round(((target_price - current_price) / current_price) * buying_power) *** Checks if the portfolio can afford to make an order *** Store buy history and target prices in a text file.

2. def sell() --> If position has more than -7.5% loss or Unrealized Gain > Target Price, the bot orders an immediate sell order on market price --> Else, the bot pushes stop/limit order: limit order = newly updated target price
   stop limit = -7.5% of purchased price

3. if __name__ == '__main__':
# Account Status Review
alpaca_api = tradeapi.REST(base_url=ALPACA_BASE_URL,
                    key_id=ALPACA_API_KEY,
                    secret_key=ALPACA_SECRET_KEY
                    )
# Liquidity
buying_power = float(alpaca_api.get_account().buying_power)

# Current Holding Positions
open_positions = alpaca_api.list_positions()
my_positions = set(map(lambda index: open_positions[index].symbol, range(len(open_positions))))

# Open Orders
open_orders = alpaca_api.list_orders(status='open')
my_orders = set(map(lambda index: open_orders[index].symbol, range(len(open_orders))))

# Executing Sell & Buy Orders
sell()
buy()
sys.exit()