Inspiration

When it comes to investing in the stock market, trading cryptocurrencies is seen as a high risk-high reward decision. Because cryptocurrencies like Bitcoin are relatively new and extremely volatile in market value, making accurate decisions on whether to buy or sell is difficult and complex. Currently, very few crypto-trading algorithms exist- none of which are available to the public. Only large trading companies have the tools to make informed decisions about Bitcoin investing. We want everyone to have access to a simple and low-risk system to trade Bitcoin without needing to fully understand the complexities of the crypto stock market.

What it does

In order to simplify and streamline crypto-trading, we created Midas, a machine-learning algorithm that uses quantitative historical data (e.g. market price, volume, etc.) put through a neural net to determine whether to buy or sell Bitcoin. We implemented Midas in a web interface that displays a graph of the current market value of bitcoin along with colored indicators showing whether to buy or sell bitcoin at a given time. Along with graph, we’ve implemented a profit calculator which determines what percentage profit you would make in a given time interval (buy at T1, sell at T2). The Midas interface allows for a streamlined and easily-accessible method for anybody to trade Bitcoin.

How we built it

We used the following scripts and languages: HTML, CSS, SASS, Javascript, and jQuery. HTML and CSS were used to create the website; SASS to make the site responsive; Javascript to create the Midas interface and jQuery to easier manipulate HTML and event handling. We are using tensorFlow and python to create the machine learning model.

Challenges we ran into

Turns out its harder to create a crypto-trading algorithm than we expected. We faced challenges with graphically displaying the bitcoin value along with the colored indicators of whether to buy or sell. We also had difficulties exporting bitcoin data into a .csv after running it through the machine learning algorithm.

Accomplishments that we're proud of

We were able to predict whether it was a good time to buy or sell bitcoin with around 60% accuracy, which is better than a simple guess or coin flip. We are also proud of the display interface we developed to show the bitcoin market price graph with the easily-readable color indicators on whether to buy or sell.

What we learned

We learned over the course of the weekend how to use Tensor Flow. We were able to pass in cleaned up bitcoin data, and have it return an index of whether it was a good time to buy or not. We were also able to learn how to use bootstrap to create a website layout that was aesthetically pleasing, as well as using javascript to create our own dynamic elements, like our bitcoin price and purchase index plot, which we made from scratch after we were unable to use canvasJS to plot multiple lines.

What's next for Midas

Currently, Midas is very similar to the other algorithms that the big trading companies are using. Midas only incorporates large amounts of quantitative historical data (market value, low/high price, Bitcoin volume, etc.) when determining whether to buy or sell Bitcoin. We will soon be adding even more data to the algorithm, like GPU prices for example.

The main differentiation factor, and what will make Midas more accurate than big trading firms, will be the implementation of Media Data and mass psychology to the algorithm. Using a web scraper and advanced natural language processing, Midas will be able to produce a media sentiment analysis that reveals the general public’s attitude towards Bitcoin, which we believe has a measurable impact on its market price. Using this Media Data, gathered from news organizations, social media, and other relevant sites, Midas will be able to more accurately determine the right time to buy and sell Bitcoin. We see massive potential for the future of Midas, and we hope to make the algorithm accessible to anyone who wants to start investing in the stock market.

Built With

Share this project:

Updates