With the growth and development of ChatGPT, it’s worth considering how our clients might utilize this tool to streamline their daily tasks, reducing dependence on formulas or analytics tools. I’ve taken the initiative to explore this possibility firsthand.

In this experiment, I’ve constructed two portfolios, each containing thirty securities, rebalanced semi-annually from 2020 to 2023. These portfolios follow two distinct weighting schemes. Traditionally, I would have used Portfolio Analytics (PA) to investigate and compare their performance.

However, this time, I’ve taken a different approach. I’ve created a SQL snippet to fetch the daily returns from our ‘fp_total_returns_daily’ data feed. These returns, alongside the daily weights, are then fed into the GPT-4 code interpreter to plot the portfolios’ performance. (note that I’ve omitted details regarding the actual securities in these portfolios to ensure data privacry).

Despite requiring a bit of guidance in instructions, the process is strikingly simple and effective, yielding the following beautifully rendered charts:

The output produced is noteworthy both for its visual appeal and its quality. Moreover, by simply adding a column for country data, the tool instantly generates a pie chart displaying country exposure. This efficiency is impressive:

When I compared the results from GPT with our Portfolio Analytics (PA), I found a small difference in the returns. This shows that PA is still very important when it comes to getting precise results.

However, this test shows that GPT can be really handy for two main things. First, it can save time because users don’t have to learn our complicated formulas; they can just pull data straight from the database. Second, it can do some of the work that Portfolio Analytics (PA) does, like working out important numbers like the Sharpe ratio and drawdown, and even making charts. So, overall, GPT can make these processes quicker and easier.