top of page
  • Writer's pictureAlpesh Patel

ChatGPT and The Future of Investing: Stocks It Likes Now

Updated: Nov 15, 2023

Last Tuesday on the BBC, I used ChatGPT to analyze a stock, the earnings of UBS. Let's look at the results, and actually, let's see how good it is at stock picking more generally.

I've put together a little presentation. That was me on the BBC with Sally Bundock.

I pointed to her how often she and I have seen earnings over the last 20 years and had to analyze them because we get a wad of documents from a company. In this case, it was only 15 pages from UBS. And give an instant reaction. I put it into ChatGPT to avoid cognitive biases.

But I want to go beyond that. I don't just want to talk about that. I want to talk about actually how you pick stocks more generally. And can ChatGPT help us to do that? Let's have a look at it.

First of all, just going back to the UBS, I think it was a British TV first. I cut and pasted and put it into ChatGPT. I'm currently using version four. And it's the paid version, and that was fine. And the outputs were rather good. It summarized 15 pages.

I could have put in more detailed prompts such as, "Pretend you are a forensic accountant, for instance. Pretend you are a TV presenter who is cynical about earnings results, and you've only got three minutes in which to convey the significant parts of these earnings." I could have put in all those prompts. And that's the great thing about this.

Now let's move on to stock selection. I put this data in.

This is where it gets fascinating. I put in not just what you see on screen but actually the data in segments. In segments of 10,000 stocks across a variety of perimeters, not just the ones you see on screen. Because there's a limit to how much data you can put into one go into ChatGPT. Okay. Now, that gets interesting. Then what happened?

By the way, this is how it looks inside ChatGPT once you insert it.

You and I could barely make sense of it. The great news is it can. It can work out which columns, where they belong, and where the following stock starts. At least that makes your life a hell of a lot easier.

What were the prompts and improvements? I asked it to help me select stocks. By the way, we did some backtesting as well. We gave it historical data and saw what happened. Wait till you see that. It'll blow your socks off.

I asked it, "Give me insights." I said, "If you have to guess which of these stocks would likely do the best in the next 12 months." Then I realized, you know what? I need to go into more detail than that.

And I guess it is not a good thing. I said, "Well, pretend you're Warren Buffet, and this is an academic exercise." Because, of course, it doesn't want to give financial advice, and rightly so. You've got to play around with it. And, of course, I could have improved those prompts even more. I'm not an expert. I'm sure many of you have got comments. Let me know because I'm reading quite a lot about this at the moment.

I also said, "Give me an even more detailed multi-factor analysis," because it gave me too little, too simple analysis. It could give me stuff I could read on the screen anyway. Or Excel could have just sorted. It told me, "Okay, this is the cheapest stock." I wanted a multi-factor analysis. Which is the cheapest, has the highest growth rates, best momentum, et cetera?

And then I asked it, of course, as I said earlier, "Pretend you're a journalist." And that was just more out of interest. What were the results? The results were alright, but there's more to it than that.

These were the ones that initially gave me from this year's data. And this is how it put it - Broadcom: Alpesh value, growth rating. That's my own proprietary rating based on the valuation of companies. And I weigh that importance based on academic literature, revenue growth, cash flow growth, etc. Strong CROCI cash return on capital invested. It's a formula invented by Deutsche Bank and used by Goldman Sachs Wealth Management for its wealthiest clients to forecast stocks: low PEG and recent performance.

It came up with these three names, which actually aren't that bad.

Having said that, I'd already led it down that route and pre-filtered quite a few. But it was interesting the names it came up with because, of course, I still had out of 10,000 stocks data, which was restricted only then to about 100. At least it gave me some names to narrow down on and further analyze.

But obviously, we want to go a bit deeper than that, don't we? I gave it the previous year's data. In fact, I picked a challenging time, December 2021. And asked it to forecast 12 months ahead, knowing full well that 2022 was awful for the markets, even for companies, the data of which the fundamentals look good. Well, what happened there?

This is where it continues to be interesting. It had a car crash.

If I told you, the top three stocks you selected fell the most into the subsequent three months because they gave me three names based on 2021 data. And I can tell you those three stank.

One was Aon. One, which interestingly now would hit my list. And you might say, "Well, you should have just extended your timeframe." There wasn't enough correlation. Java Securities and Cake Box. I said, "Well, what's going on?" And I told it. I fed back into the loop and said, "Actually, the ones you picked were pretty bad. What's going on?"

And once again, it improved because it learns. And this is where it's excellent as an educational tool for young analysts. And this is where we know there's still room for the greatest neural network of all, the human brain.

This is what it said. "When reassessing good and top stocks, consider the following factors—knowing full well that the ones it selected could have done better. Financial health, reevaluate the financial statements of the company. Reevaluate, don't just pick and then forget." This is why I do a weekly market update.

By the way, speaking of which, if you want that, go to I do a weekly market update on Telegram and YouTube because we've got to reevaluate the financial statements. We don't just pick and then close our eyes. Business fundamentals, are there any changes that might lead to stop price declines? In other words, we don't just buy, hold and forget. Because if we had with the stocks it picked, they look great at the time it picked it. The fundamentals look great, but they fell off a cliff.

And speaking in, what are we? In May 2023, the three it picked as its best also continued to stink, even though in performance, even though the fundamentals looked good at the time of selection. But market sentiment changes and macroeconomic factors change. But again, that wasn't enough. And this is great education for people who are new to investing. But again, that wasn't enough for me.

I asked it a few more questions.

I said, "Well, what was not in the data which could explain 2022 poor performance in the stocks you picked?" And again, some great educational information. The kind Warren Buffet might give you. Unexpected events, events unforeseen. Events like management scandals. One of the companies had a management scandal. All of these other factors are things that couldn't be in the data and couldn't have been foreseen. And I did ask it, "What could you have foreseen which was in the data? Now that I've told you that the stocks you picked looked great at the time but went on not to do too well."

The other things will seem obvious to you when you look back, but these are things that we so often forget. Again, why do I do a weekly update? Poor financial performance. The companies could have reported disappointing financial results. Now, there are two types. Academic research tells us about poor announcements—bad news, which is hard bad news, and bad news, which is soft bad news.

The soft bad news is the type in which the economic outlook looks challenging. The price falls but then recovers. The hard bad news is our figures are disappointing. Earnings have been missed, and it seems like they're not going to do well in the future. There you get a fall, and it doesn't recover. It just continues dropping. And that was the nature of these companies.

And there wasn't something in the data at the time which would've alluded to it, which is why of course, if there was something in the data, guess what? AI and I and everybody else would just be rich because we'd pick one stock. This is why we still need to have at least more than one, maybe not more than 40, or more than 30 misjudgments of growth prospects. The initial analysis might have been overly optimistic about the company's growth potential, which could have led to the stocks being overvalued and subsequently underperforming.

We allowed for that in the valuation, but it was at least still trying to give us some guidance. And this one was good. Number six is the change in investor sentiment. Investor sentiment might have shifted. And it had. It shifted, if you recall, from 2021 to 2022 against stocks generally. But it shifted away certainly from growth companies. It could have shifted, causing a decrease in stock demand and a subsequent decline in price.

And then, of course, geopolitical factors. Now if you map every single one of those things that ChatGPT said, there's a hell of a lot of stuff. The one thing that can map it the best is the human brain still. I could have created a bigger engine, which could have given all these data points every month, and fed it in through an API. And said to it every month, "Okay, these are your initial picks. At what point do you want to change this portfolio?" But I didn't have the resource or the manpower.

But also, that day will come when we can. That's the future of private investing. But also, there's more to come. I also would've then ended up trading the investments, which I didn't necessarily want to do. But there could be an algorithm on the hedge fund side where you are actively trading these rather than the fund management or TV presenting side.

Let's look at another slide here, which gets even more exciting.

"What clues were in the data," I asked it potentially. And it said, "Well, high valuation multiples if the stocks..." These were the ones which it picked with good quality but could have fallen. All of these should have stayed in its top picks because we allowed for this. We allowed for ensuring there were low valuation multiples or high debt levels or declining revenue, or declining earnings growth.

It suggests we need to be more sophisticated than looking at year-on-year earnings. We need to look at the delta, the rate of change of earnings growth—narrow profit margins. The direction of travel is important. Over-dependence on a single product. How are you going to incorporate that into the data? That's going to be qualitative data.

My point here is that there remains a role for the human neural network on qualitative data. Taking a vast amount of, not the actual physical data, the kind of data that a calculator can take on that ChatGPT can. But the qualitative data, is this overdependent on a single product? You could still algorithmize it. And put that into some form where you can tell it's only one product.

Negative industry trends or competitive pressures. These are things that, at the moment, humans don't do well. They rely on their qualitative nature too much and only look at the hard figures a little and, therefore, have a narrow restriction because humans can only do so much on data. There's a room for both parts. There is room for both parties to rely on each other. We haven't finished yet.

Anyway, the stocks it came up with for this year, you can have a look at my Telegram and YouTube channels. I'll share what some of those were. The point is I'll keep an eye on them, given how it had been done in the past. I don't necessarily want to go straight to them. Oh, I won't be a tease, then. Okay. I'll tell you which ones they were. I'm going to put it on screen now. There you go. It said, "Combining these factors, the following stocks stand out."

This is in May 2023. You can keep an eye out for these. I'm not saying they're necessarily my picks. This is what ChatGPT May 2023 has picked. And I told it, "Pretend you're Warren Buffett, and it's for the next 12 months." We'll see what happens. But imagine how much of that is to do less with the data and more with broader market conditions.

Don't forget Keep an eye on my Telegram channel on my YouTube, and I'll also give you more AI-based stock selections.

Alpesh Patel OBE


bottom of page