Although the impact of a company is orders of magnitude larger than the impact of an individual, the onus is still on us to make any difference we can, and impact investing through a robo-advisor provides a low barrier of entry for people to make a difference in environmental, social, and corporate governance issues.
What it does
Impactnet enables users to select issues they are passionate about and builds a portfolio that is curated from companies with collinear interests. We developed a state-of-the-art recurrent neural network which employs natural language processing of corporate news articles as a means to classify companies on an ethics gradient. From this, we develop a balanced portfolio using a traditional portfolio theory (the Fama-French three-factor-model) paired with our ethics-based KPI.
How we built it
For the backend, we used the Django framework powered by a MongoDB database. We used Bing Search to find relevant news articles for hundreds of publicly traded companies, and used Diffbot to extract the text of each article. We fed this data on a per-company basis into two NLP algorithms, one of which we built using pytorch, scikit-learn, nltk, and numpy, and the other using Google Cloud's natural language content analysis API. To train the first algorithm, we used Mechanical Turk to crowdsource a dataset of human evaluations of articles pertaining to the ethics of companies. From here, we calculated a range of financial metrics on each company using Alpha Vantage and the Marquee API. We built the frontend using React, with data visualizations powered by the D3.js library.