AQR, a quantitative asset manager, is experimenting with ChatGPT, a generative artificial intelligence (AI) tool, in order to improve returns on investments. Promisingly, in the back tests, using any of the language models, including ChatGPT, would have helped investors outperform the market from 2004 until 2019.
Bryan Kelly, Head of Machine Learning at AQR, and Professor of Finance at Yale School of Management, explains why ChatGPT is not a very reliable predictor of future returns. This is because it was only trained once in 2021 and has limited ability to process stock market data in real-time. Markets, on the other hand, effectively incorporate information quickly.
However, Kelly described that some parts of ChatGPT’s language model can still be applied to investments in order to improve the portfolio construction process. This is done by translating publicly available financial text, such as analyst notes, into a numerical representation to anticipate returns (compression step) and transforming back into the original format (decompression step).
In addition, AQR has also considered other generative AI tools, such as Meta’s OPT and Google’s BERT, to help with optimizing portfolios.
AQR is a well-known asset management company that specializes in quantitative methods. Their strategies are based on a deep understanding of markets and investment management, containing a combination of economic and financial research. They believe that taking advantage of alternative perspectives and long-term investment horizons will eventually lead to superior returns.
Bryan Kelly is a highly experienced finance professor with extensive experience in the fields of economics, finance, and quantitative methods. He is the head of Machine Learning at AQR and is a professor at Yale School of Management. Kelly has written many articles related to quantitative approaches in finance, and has published multiple books in the field. He is actively involved in the global financial sector and is an important figure in the financial research domain.
