Capital Budgeting
Spreadsheets are perfectly suited for capital budgeting exercises. Those exercises involve a minimal amount of data (data = assumptions mostly) and only basic arithmetic, and we would like to view multiple tables (balance sheet, income statement, and cash flows) simultaneously, which we can do in a spreadsheet but not easily in python. How much value can AI add?
I think students should explore this question, with guidance, and answer it themselves. A good starting point is to provide a basic spreadsheet template, which students likely already have from prior classes (here’s an example) and then ask students to apply it to a specific situation. I like the Sneaker 2013 case. It has all of the normal features but no tricky issues like how much to charge for using existing equipment that is surplus now but might have other uses in the future.
We can ask students to chat with Julius, or whatever AI system they use, as they build the balance sheet, income statement, and cash flow forecasts. And, we can ask them to ask Julius to build a python model in parallel.
When building the models, it is useful to tell Julius what the desired table layouts are and to ask Julius to show the tables each time they are revised (and to ask it to transpose its tables when it shows them, because the natural inclination in python in assigning columns and rows is the opposite of what we normally do in spreadsheets). If there are discrepancies between the spreadsheet and python models, then students can ask Julius to analyze and reconcile them. The error may frequently lie with the LLM, because it has misunderstood something or made an incorrect guess about something.
Once the spreadsheet and python models are built, we can turn to sensitivity analysis. It is in this part of the process that AI may be of the most use. “AI + coding” can easily produce pivot tables and charts and help interpret the results.
This should be a multi-day exercise, I think, mostly done in class so students can get assistance, help each other, vent frustrations, etc. At the end, the discusson about the value of AI should be interesting. Students may conclude that AI is of no value or requires too much extra time for the value it adds. This is itself a useful lesson. And, they will have gained additional practice in capital budgeting, which cannot be a bad thing. I think it is likely that most students will have found the collaborative aspect of working with AI valuable (though possibly too time consuming), if only because it reinforced that they were doing the right things all along. It is generally true that, as they say, two heads are better than one, and AI is an always available and infinitely patient partner in whatever work we are doing.
First published on finance-with-ai.org
Also on substack at kerryback.substack.com