A demo of the dashboard screen
ShareFund: a democratized hedge fund that rewards good traders
We’ve been dissatisfied with the how much human error there is in conventional hedge funds. Humans are subject to emotions and factors that affect their ability to trade with a sound mind. Fear and fear management plays a huge role in hedge fund performance and we wanted to hedge that by creating a ShareFund.
What it does
Sharefund is a democratized hedge fund that allows you to gain more and more power over the control of the fund based on your trading history. It also takes advantage of algorithmic trading and machine learning in order to augment the voting decisions of the fund as a whole. So it values high performing users as well as high performing bots in a duality, giving us the best of qualitative and technical analysis when it comes to trading.
How we built it
Our fund is a weighted fund that allows all users who have contributed minimum amount of capital ($1000) voting power over where the fund invests. Right now it has seven Vangaurd ETFs as choices but we plan to add more exchange-traded stocks and funds in the future. A new user who has just joined has a very small amount of voting power over the direction of the fund and where the fund invests, but is able to gain more and more influence as they predict the stock market correctly.
The trading cycles happen at each day, and trade immediately at market open (9AM EST). After a enough days, our algorithm makes it such that users are able to gain influence and rewards users with the ability to acquire more and more control over the fund and have more “voting power”. It takes the votes the user made when they did not have any power and uses it to calculate a “voting power” number and rank the user among the many users who are in the fund.
Then our algorithm calculates what this means in the overall votes and weighs each user accordingly.
The voting power a user has is independent of the capital that they have in the hedge fund, which is why the barrier to entry is somewhat high, like a true hedge fund.
But that’s not the only thing. Users are also able to compete with bots, and they’re able to compete with bots that learn from historical data as well as our users themselves.
We’ve created a sample bot to show you how effective and how powerful the bot is at garnering a voting presence over the entire hedge fund. We currently only have one implementation of the bot but plan to open source the bot platform to multiple users so that they can create bots themselves and can garner a commission on the earnings that their bot manages to get.
We use a LSTM (Long Short-Term Memory) in our primary machine learning bot in order to predict the price of the stock market, and make buy or sell votes accordingly.
Challenges we ran into
It took us an extremely long time to figure out the algorithm and fine tune the coefficients to the perfect rate.
We struggled to find a perfect rate of climb for those users who consistently invested well, we also struggled with finding a way to punish users with high power who began to falter and struggle in the new stock market.
But we leveraged a sigmoid algorithm and used a small coefficient of learning and a high coefficient of punishment when it came to users with high voting power who made inadequate votes. In contrast, users who had low voting power who made consistently good votes would move up the rankings quickly after a reasonable history of valid voting patterns.
Accomplishments that we're proud of
We were able to get the machine learning model almost perfect, beating out many of our users in our simulation and completely taking the voting power of the fund overall. Since it is only a simulation, it is not representative of truly good human traders in the market as it leveraged a pseudorandom generator for trades. However, was nice to see our model completely outperforming its competition and beating random chance, in very little time.
What we learned
Overall, this project gave us a chance to learn more about the stock market in general. In the beginning, some of our team members didn't even know what a hedge fund was. We learned about ETFs and the dynamics of the stock market through our discussions to create a cost function and the logistics of a democratic hedge fund.
For the machine learning part of the project, we did some background research to better understand how to use an algorithm to predict the stock market trajectory. We discovered two types of artificial neurons - the perceptron and sigmoid neuron and read up on gradient descent.
Orange = reality, Green = predicted
Vangaurd Healthcare ETF:
Vanguard Industrials ETF:
What's next for ShareFund
Since it should represent what a hedge fund truly is, a high risk fund that hedges against the market as well as for it, we’re going to add options trading and allow the options to place bets against the market as well as for the market so that it allows for stability within our portfolio. In addition we’re planning on adding more than seven ETFs and opening it up to much more stocks. However, we need to develop a system to consolidate the stock options to a limited amount.