AI-Assisted Financial Analysis
Spring 2025
Instructor
Kerry Back
(kerryback@gmail.com)
J. Howard Creekmore Professor of Finance and Professor of Economics
Meeting Schedule
TTh 12:30 – 2:00
3/18/2025 – 4/24/2025
Course Description
This is a project-based course. We will create workflows to visualize stock return data, do retirement planning, analyze financial statements, optimize portfolios, perform mutual fund performance evaluation, produce capital budgeting analyses, scrape and analyze short sell data, and model option values. We will use generative AI and python, and we will use generative AI to write the python code. The course builds on the knowledge gained in the core finance and applied finance courses and provides additional practice in applying the concepts learned there. The course also provides an introduction to python and to using generative AI. No prior experience with AI or python is needed.
The platform that we will use is Julius.ai. It is a wrapper around several LLMs (large language models) including, at this time, GPT 4o, Claude Sonnet 3.5, and Gemini. It includes an integrated python environment that is superior to OpenAI’s Data Analysis tool, because it allows for the installation of additional python libraries and it can access the internet to scrape data. Instructions entered through the chat window are wrapped in Julius’ prompt engineering enhancements and routed to one of the LLMs (you can choose which or use the default). If the instruction is best handled by coding, then the LLM will return code, and Julius’ servers will run the code to produce what was requested. If the code has an error, which happens to everyone, including LLMs, then the Julius servers will automatically send the error message to the LLM to parse the message and improve the code. The end result is usually successful execution of the instruction. So, we can run python to do things without having python on our computers and without writing the code ourselves. What python can do is just about anything. We can get data from the web or upload our own, perform data analysis, create fine-tuned visualizations, and save everything as jpegs, Excel files, Word docs, or even as web pages.
What we will do in the course is to create structured prompts, saved as Julius workflows, that can be shared with others. This is somewhat similar to creating a custom GPT. For example, in analyzing financial statements, we can specify which financial ratios we want to compute and what type of trend analysis and visualizations we want to create, etc., and create a workflow that can be used repeatedly for different tickers. Time in class will be alloted for students to build workflows. During class, we will share what we have built by sharing links using a Google Chat space. Grades will be based on workflows created as weekly assignments and on class participation.
A great feature of working with Julius is that it is interactive. After running a workflow, we may think of some additional analysis we want to perform. We can make the request to Julius in plain English and get the additional analysis we want.
In some cases, we may not need to work interactively. For example, I built a workflow to scrape the SEC EDGAR site that only requires the user to provide a ticker (SEC EDGAR 10-Q Scraper). In such cases, we could export the python code to run on our own computers. If generative AI is an integral part of the workflow (e.g., if we are using AI to summarize the MD&A section of a 10-K), then we would need to add code that accesses the LLM’s API to make requests. We can also use Julius to write the code to do that. If we wanted to take an additional step and get into software development, which of course we don’t in this course, then we could also ask Julius for the code to create a Windows or Mac executable to run the workflow.
Julius
Julius.ai provides a 50% academic discount. Sign up for a free account, then send an email using your Rice email account to team@julius.ai and ask for the academic discount. They will respond with a promo code to use. The Lite account ($8 per month after discount) allows 250 messages per month and may be ok. If the message limit becomes binding, you can always switch at that time to the Standard account ($18 per month after discount), which allows unlimited messages. Everything is run in the cloud from a web browser, so there is no software to download.
Honor Code
The Rice University honor code applies to all work in this course. Each student must do his or her own assignments, but students are allowed and in fact encouraged to seek advice from each other. Use of generative AI is of course permitted.
Disability Accommodations
Any student with a documented disability requiring accommodations in this course is encouraged to contact me outside of class. All discussions will remain confidential. Any adjustments or accommodations regarding assignments or the final exam must be made in advance. Students with disabilities should also contact Disability Support Services in the Allen Center.