Sentiment Analysis
Condense the company's investor communication into one number. Easily compare any company's sentiment over time and against peers
Peek Into the App
Take a look at the sentiment analysis of some of the top companies in the stock market, straight from the MarvinLabs app.
Sentiment scores can range from 0-100, with 0 being the most negative and 100 being the most positive score. The sentiment score is calculated weekly each Sunday.
Convert all text into a single number for easy comparison and trend spotting
Most analysts focus on numbers: we analyze the performance of companies by judging their revenue, margins, assets, or cash flow figures. We build models in spreadsheets, trying to understand the relationships between those figures. We value companies based on DCF analyses or multiples.
We’ve had good tools and systems helping us with our number obsession: from Bloomberg as a data vendor and Lotus 1-2-3 in the early days to Excel, databases, alternative data vendors, and so on nowadays.
We even took our obsession to company communication. We counted the numbers of “but”s in earnings calls. We had lists of “good” and “bad” words that we summed up to calculate sentiment. We even took to Twitter APIs and used models to derive sentiment from Tweets.
Today, we introduce modern sentiment analysis in Marvin Labs powered by large language models. We are not just counting words or using heuristics but using the technical advances of the last few months to understand the meaning of company communication.
Sentiment analysis is a great tool as a starting point for further, detailed analysis. It can help analysts discover investment opportunities that would have slipped through otherwise.
The value of sentiment analysis for outright portfolio construction has traditionally been limited. We understand that it is too crude and limited a signal to serve as the only factor in a quantitative portfolio construction process.
Return on a sentiment long-short portfolio. From 2018 to today a $100 notional would have yielded $60
Nevertheless, we want to show the returns of a hypothetical investment portfolio using only sentiment analysis1. We are building a weekly rebalancing portfolio with ten equal-weighted long positions and ten equal-weighted short positions. The long positions are the companies with the highest sentiment score in our universe, and the short positions are those with the lowest score. Such a portfolio with a long-notional of $100 would have yielded ca. $60 over the last five years without any meaningful drawdowns.
Footnotes
-
We wanted to clarify that this is not investment advice. Any such analysis from any software vendor - including Marvin Labs - should be taken with a grain of salt. ↩