MarketReader is designed to help users understand what is driving asset price movement in real time.
But for non-professional users, it can be tricky to understand our system’s technical, data-based output in its raw form.
So, we have begun leveraging OpenAI’s ChatGPT to summarize the key takeaways for market moves in a way that is easy for “humans” to digest. The text summaries may also be a quick point of access for professional users.
ChatGPT has received a lot of accolades, but one of its primary criticisms is a lack of useful output for real-world situations. In order to ensure that ChatGPT output contains the most relevant and important information pertaining to each market move, we had to optimize our queries to properly weight the immediate contextual information that is most crucial to know in the moment. With these tweaks, we have found that ChatGPT is very good at quantifying unstructured data (e.g. a tweet) to evaluate its relevance to other information.
One of the keys to MarketReader’s success in using ChatGPT is utilizing the highest quality news and market data content as inputs to ensure that the output is valuable. We’ve found that this combination of cutting-edge AI technology with structured, filtered data underneath is a winning formula for providing clear, concise explanations of what is driving the market.
Every 10 minutes, MarketReader identifies significant price moves and feeds our raw data to ChatGPT, which then outputs concise summaries of what is happening and why.
On February 8th, Alphabet Inc. stock (GOOGL) dropped 9% after its Bard AI chatbot gave a wrong answer at a launch event. MarketReader automatically provided an explanation for the price move at 8:50am ET, hours before any news outlets, and before GOOGL dropped an additional 5%:
The news was not reported by Bloomberg at all until 10:05am, and no full coverage was released until 11:27am. CNBC did not cover the story until 11:19am.
The MarketReader system quickly identifies whether stock prices are moving because of changes in analyst ratings:
Or earnings results, as seen with $LYFT after market close on February 9:
Or more nuanced investor dynamics:
In addition to slicing data into 10-minute bars, the MarketReader system also synthesizes data collected over the course of the trading day to provide insights into why a stock price is moving on a given day. We refer to these insights as “Day-to-Moment.”
ChatGPT excels at summarizing disparate data sources from MarketReader’s Day-to-Moment output into coherent text that provides a full picture of what is driving a stock’s price movement.
In this case, our system provided ChatGPT with data on several 10-minute bars for $DIS that were identified as unusual during after-hours trading between February 8th and 9th, and it returned a summary containing all of the most relevant information neatly packaged in a readable paragraph format:
This combination of MarketReader technology, which can identify market drivers along multiple dimensions using industry leading finance models, and OpenAI technology, which can concisely craft prose explaining MarketReader’s sophisticated analysis in an easy to digest form, will revolutionize financial news coverage.