StockBot

Motivation

Financial illiteracy is an increasingly problem among college students. College students, mostly Amateurs, are afraid to trade and invest themselves because they have no experience, so discussing their thought process in a group of like-minded students is a good starting approach.

However, the problem with many group chats such as discord is that it doesn't offer resources to provide a clean way to discuss stocks. Example of some problems:

• Amateurs like to look at or follow Guru trades. However, in discord, it's hard for Guru's to specify when they will enter a trade and at what price (live trading). In this case, a discord bot command can specify the current price of a stock, and the Guru can use a bot command rather than using buy @here. This also allows amateurs to look at the trade afterwards and keep track of their paper trades.
• Amateurs want to know if their trade is successful, and one way is to use a bot stock forecast. However, note that markets are made up with emotions of greed and fear, which makes forecasting stock prices unreliable (at least of right now).
• Amateurs usually don't know what to pay attention to in premarket. Thus, the bot provides news and NASDAQ futures at 7:00 am for better preparation when market opens.

All of these functionalities will make it easier to discuss stocks on the discord group chat.

Technical Documentation

To use the bot locally

1. Install Python 3.6 or above. Install Python here.

2. If using git, run git clone https://github.com/zhuodannychen/StockBot

3. Run pip3 install -r requirements.txt to install dependencies.

4. To configure the DISCORD_TOKEN, there are two options: Create a .env in root and set DISCORD_TOKEN='YOUR TOKEN', or configure it in CONFIG.py and edit the comments in bot.py.

5. Edit the other TOKENS with your own tokens in CONFIG.py.

6. Run the bot with python3 bot.py.

To invite the bot to a server

7. To add the bot to your server, go to this link and follow instructions. https://discord.com/oauth2/authorize?client_id=759616214536028200&permissions=0&scope=bot

How the Bot Works

For getting real time data, like the !price command, !triple, and !profile, the data is scrapped from Yahoo Finance using BeautifulSoup.

For commands like !news and !snews, the newsapi is used to get headline news.

For sending news everyday at 7:00 am CDT, I used asyncio and datetime libraries to calculate the time until 7:00 am, then I called on commands of !news and !price.

For the forecast command, a LSTM model created with Tensorflow was used to forecast the stock prices. The data comes from alpha-vantage API. alpha-vantage provides all historical data in pandas format. However, I am only using the past 1200 days because the model trains quicker with a smaller data set. We also used an adam optimizer with mean squared error for loss.

User Documentation

! prefix

chart - Returns a tradingview link of a given stock symbol. Tradingview provides advanced charts and indicators for trading.

--- Example: !chart tsla

news - Returns (usually 10) headline news from a given source. Default is bbc-news. Here are a list of credible sources:

• business-insider
• cbc-news
• cnn
• fortune
• google-news
• hacker-news
• nbc-news
• reuters

--- Example: !news

--- Example: !news cnn

snews - Returns the headline news of a given stock symbol.

--- Example: !snews msft

price - Returns the current price and percent change of a given stock.

--- Example: !price aapl

profile - Returns the profile/about info of a given stock.

--- Example: !profile AMZN

triple - Returns the current price and percent change of the three major index (NASDAQ, DJI, and S&P 500).

--- Example: !triple

forecast - Returns the forecast prices of the next 5 days, from top to bottom, for a given stock symbol. Note: this command can take up to 60 seconds before returning results and is not very accurate.

--- Example: !forecast msft

If no stock symbol is inputted for any commands, MSFT will be used by default.

In addition, the bot sends news headlines and the premarket movement of NASDAQ 100 everyday at 7:00 am CDT.

Improvements

• The current stock forecasting model is not very accurate, so a better model can be developed.
• The current forecasting command takes a long time to execute because I am training it on the command. Saving the model after training and then predicting could lead to a better result and faster execution. The only problem is we don't want to overfitt the data.
• The current training data does not provide adjusted open, adjusted high, and adjusted low, so stocks with recent splits cannot be accurately predicted.
• Matplotlib plots cannot be sent over discord messages, so plotting the day or week trend of a stock cannot be created using a command. Implementation for directly displaying charts on discord messages is still needed.
• The time it sends news is 1 hour late (8:00 am) during daylight saving time, the bot can be improved to adjust accordingly.
• More commands can be implemented, such as earnings data, analyst recommendations, etc.