Companies have succesfully created app only consumer banks to help users manage their money, and offer innovative features like budget monitoring. With being app only these companies have high operational efficiency and as a result can offer lower fees to their customers. We aim to bring this disruption to consumer investing allowing regular people to manage their investments and grow their money.

What it does

Molecule is an app only hedgefund, powered by deep learning. Molecule is designed to change the way people invest. The app is designed to make financial services accessible to the many, and not hidden behind large fees, or large minimum investments. We have 3 funds the adventurous, which carries more risk, safe which has less rsk but lower returns and the ethical fund which doesnt invest in fossil fules or wepons manufacturers and favours companies that work to improve the environment. Each day users can place money from their wallet into a fund, and watch it (hopefully) grow.

The money in each fund is invested in various financial markets using the neural network we trained to predict the price of certain assets at the end of the day users can withdraw money from the fund into their wallet or leave it in the fund for it to continue being invested. Molecule has a price prediction model which allows the money in funds to be invested in various financial markets.

The model uses various technical indicators and prices of correlated assets to predict whay the price of an asset will be tomorrow.

E.g. For Goldman Sachs, molecule uses data such as: *Similar companies such as JPMorgan Chase *Composite indices (NYSE, FTSE100, Nikkei225 …) *Fourier transforms of the price of GS with varying somponents to analyse small and large-scale trends.

How we built it

  • Trained an LTSM on various technical indicators about a stock price and the prices of correlated asset, to try to predict the price tomorrow.
  • Built a react-native app with a nodeJS reat API and mongoDB to allow users to deposit money ad invest it in funds.
  • Created a react-native ap with a nodeJS rest api to allow users to deposit money and invest in funds.

Challenges we ran into

  • Getting a recurrant neural network to train properly.
  • Accomplishments that we're proud of
  • We are proud we managed to deliver a price prediction model and a UI.
  • Our Expo mobile application had difficulties connecting to our node.js API

What we learned

We learned how LSTMs work.

What's next for Molecule

  • Bigger dataset
  • Try to include an adversarial architecture (we looked at Metropolis-Hastings GAN)
  • Include fundamental analysis into our predictive model (sentiment analysis of news articles relating to the asset as a given time as well as yearly shareholder reports)
Share this project: