How OpenAI's GPT will affect Quantitative Finance

ChatGPT in Quant Finance

Since the advent of ChatGPT during the final inning of 2022, the world has been astonished by the power and potential for large language models to revolutionize society. This reaction, while slightly premature, is a direct response to how rapidly the field of artificial intelligence has evolved and how vivid the applications of this technology are. While some view this technology as an evil creature that will steal our jobs, others have already begun creating new products built off of GPT and are integrating it within their existing services. Many of these products have grown rapidly in popularity and are providing immense value to the world.

While a plethora of new products have been built with Open AI's GPT, very few are catered to the field of quantitative finance. Most software engineers today have focused on consumer-facing applications such as email writers, stock-image generators, etc. Nonetheless, large quant firms have recognized the potential for this technology and are already demonstrating interest in its applications. Citadel's CEO, Ken Griffin revealed in a recent interview, that he is seeking an enterprise-wide license for Chat-GPT at his company. Furthermore, Two Sigma, a quantitative hedge fund in New York, claimed that "Large language models (LLMs) are arguably the most important machine learning (ML) innovation of the past decade."

Given the excitement expressed by large institutional players in the quant industry, the following question arises: what are the potential applications of GPT in quantitative finance? In this article, we'll cover a handful of ways in which it may be incorporated and how it may impact the work quants do on a day-to-day basis. Rest assured, GPT will augment, but not automate, a quant's job. At least, for now, that is.

1. Accelerating the Research Process

Quantitative Researchers today spent an enormous amount of their time collecting data upon which they can perform analysis, test various hypotheses, and devise new trading strategies. Collecting this data can be quite cumbersome given that it requires collecting information from scattered sources and writing SQL queries to pull data from databases. Furthermore, this process is prone to error, as simple flaws in constructed SQL queries can have large impacts on the results of an analysis.

In this situation, LLMs can be leveraged to accelerate the collection of data. These models are capable of not only understanding human language but also programming languages. This means that LLMs can transfer human queries into efficient code in any desired language. Ultimately, this means that an application can be built, which takes as input a description of the data that is needed, and the model can automatically generate the corresponding accurate SQL queries and pull the data for the researcher.

The impact of this process is that researchers can spend less time working on "boring" data manipulation/collection tasks and more time creating alpha-generating strategies for the quant firm. Given the fast-paced nature of quantitative finance, any edge that one can get in the process of developing these strategies can have massive financial implications, further justifying the use of this technology.

2. Enabling Greater Documentation of Research

Most quantitative finance firms today are organized in a manner that fosters internal competition. Each team tries their hand at performing the best in order to have higher bonuses than their counterparts. While this internal competition can be motivating and drive more profits for the firm, this also leads to poor communication between teams, scattered documentation of tested research methodologies, and large amounts of repeated work.

With large language models, all of the research that quant teams conduct can be effortlessly summarized and documented for future reference. Over time, this documentation can grow into a large knowledge base, which many teams can read from and contribute to. Instead of repeating an analysis that was conducted a few weeks ago, a quant can pull the source code that was previously used and any accompanying notes as a starting point.

Ultimately, this enables greater efficiency in the process of conducting financial research. With teams able to build off one another, the R&D process will be greatly improved. Furthermore, the impact of this technology also extends to the hiring process. When bringing on new members to the firm, they will be able to easily get up to speed by referencing this growing knowledge base. Similarly, when quants leave the firm, their knowledge won't depart with them.

3. Improving the Developer Experience

One of the most prominent examples of the applications of LLMs in quantitative finance is their ability to improve the developer experience. This entails a number of different things including making it faster to churn out code, writing more efficient and error-free code, and easily transferring code between languages. This last application is particularly important for quantitative researchers as it would allow them to prototype new models in a more accessible programming language like Python and then transfer it over to a low-level programming language like C++.

For Quantitative Developers, this would allow more time to be allocated to thinking about high-level systems and interactions between these systems, rather than focusing on the actual details of the implementation. The value that developers bring is their ability to understand systems design and leverage LLMs to produce programs on their behalf.

Another potential outcome improving the developer experience is the consolidation of quant roles under one title. Today, there exists a trinity of quant roles: quantitative traders, developers, and researchers. However, if LLMs enable researchers to easily build production-quality code, there may be a union of these two positions.

Is this Technology Ready?

While the ideas covered in this article suggest possible ways that large language models may impact quantitative finance, this is by no means an all-encompassing list. Since these models are improving so drastically in a short time horizon, the potential applications of these models will evolve.

One of the current major bottlenecks for these types of models is the cost and speed of compute. Quant firms leveraging these models for prediction purposes will need the proper hardware to support quick predictions.

Closing Remarks

Overall, in this article we covered three ways that large language models, such as GPT, can push the frontier of quantitative finance. Namely, we identified that LLMs have the potential to accelerate the R&D efforts at a quant firm, improve the ability for researchers to document/distill their research, and improve the developer experience for all those who write code at the firm.

If you enjoyed this article, feel free to check out more interesting quant articles on our blog. Also, if you're currently looking for a job or internship in quantitative finance be sure to head over to our home page.