Investing With Claude AI: Screening the S&P 500 to Find Undervalued Stocks

0:00You can have a great thesis. You can actually be 100 % correct in terms of your thesis, but be completely wrong in terms of the actual investment result.

0:09Debs: Have you ever had something where Claude or GPT said, this is just a crappy idea, don't do this? I am not a coder. I am not someone who has Python coding skills or anything like that. I've done a little bit of Visual Basic in my time. But what Claude did was basically provide a screen that does pull in data from an API. How do we determine if something is good value or is just a crappy company? Retail investors can and do now build their own screens. They even now have the ability to connect the outputs of their screens to their trading platforms. you can in theory trade automatically off the back of your investment screening.

0:52Debs: You got to make sure you've got a lot of conviction here before you actually use one of these tools to pull the trigger. Thanks to all our listeners for joining us for this week's episode of What's the Big Deal? Graham, please do tell our listeners, what is it that we're going to be talking about this week? What is the big deal? I'm really excited for this week. We are going to switch gears a little bit. We're still going to use AI to help us with some of this finance work. But we're going to talk public market investing. Now, this is something I really don't know too much about. I mean, of course, I do at a high level.

1:26Debs: But in terms of some of the detail that goes into how you pick a company to invest in, how you, in essence, make the decision to pull the trigger on buying a public stock, it's not something I've done a whole lot of. All my experience is private market. So I'm really excited to talk to you about this today. I know, I know. And we're going to be using AI to enhance our discussions. Of course, where would we be without Claude? And Claude's done a bit of running some screens. That's going to be the focus of our discussions, isn't it? How we can screen to identify investment opportunities in the public markets.

2:05So let's start off with a little bit about, you know, what screens are. Maybe that's probably a good place to start the discussions, isn't it, Graham? Yeah.

2:12Debs: And I mean, at a high level, you can think about this, whether we're talking public market, private market, we're going to talk about a certain number of factors that we're going to apply that will let us filter down a potential universe of investments and really create a short list that's going to help us decide whether or not we want to invest. Now, in the private market investing that I used to do, we had some overall themes in terms of, say, sectors that we liked more than others. We'd look at most stuff across sectors. Now, this is going to vary massively depending on the individual firm or the house.

2:51Debs: Some only play in certain sectors. We looked at pretty much everything. but ultimately we didn't we weren't in the position because you're you're dependent on transaction volume weren't really the position to just say okay i like i like this company i like this sector so we're going to buy at this point you got to wait for something to to actually come up public market investing you can just decide at any point to buy or sell so it's it's some in some ways it's a bit more it's a bit more real time it's a bit more fun in that sense because you can look at these factors at any time and make a really quick decision.

3:23Debs: So what are the factors that you are filtering for today? August 24th to 26th, we're running our investment banking AI bootcamp. Three days in New York. This isn't theory. It's a direct adaptation of the AI training that we deliver inside the world's top investment banks, taught by the same instructors. Modeling, diligence, data analysis, deal materials. You'll leave with a prompt library, our AI for IB certificates and 27 CP credits. Details at wallstreetprep.com. Use code WTBD. Yeah, I mean, I'm just taking a step back. What you were just saying about the private markets, it still does apply in the public markets.

4:06You have different styles for different houses, you have hedge funds with a particular focus or strategy, more fundamental, long-only investors. and so what we're going to look at today is an example screen probably for a more fundamental investor but regardless of what your strategy is the so the purpose of the screen is so to filter your investment universe as you said you know the public markets they are vast you have thousands upon thousands of stocks that you could consider investing in you know based on any individual strategy. And so the screen kind of helps you to kind of filter down that universe to a list of stocks that, you know, maybe if you're an active investor, you can then take away and do some more detailed research on.

4:55So it kind of reduces your workload. In a more extreme form, you have quant funds who, you know, use screening effectively for building their portfolios. It's much more systematic where, you know, the output is effectively what you are going to invest in. and there's no add-on research that's needed so there are different styles different strategies but what we're going to look at is a very simple screen which Claude has built based on a couple of factors or styles I should say that are quite common for more fundamental firms but also for retail investors as well so the styles that we're going to look at today are value investments and also momentum investments.

5:37And I mean, we'll talk a little bit in detail about the sorts of factors that you can use for those styles. But effectively, if you're looking for a value investment, I mean, this is an investment style that was promoted by Benjamin Graham, and it's kind of well supported by Warren Buffett. But effectively, you're looking for stocks which are trading at below what is perceived to be their intrinsic value. So looking for effectively cheap stocks, but which are really quality stocks. And so you're looking for factors that help to identify those types of companies. So companies which trade on very low multiples of their book value or multiples of earnings that generate very high cash yield.

6:24So those are sorts of factors that you would expect in a value screen?

6:29Debs: So Debs, when we're talking a value screen, I mean, of course, that makes a lot of sense. We're looking for cheap stuff. How do we, in a screen like this, how do we determine what's cheap? Is it kind of looking at per company, say, all right, the industry average P multiple for a company in this sector is whatever it is, and we're looking for companies to trade below that. And then on a screen like this, how are we making, how do we determine if something is good value or is just a crappy company? That's a really good question. That's a great question. That's really the ultimate question. Holy grail of investing, shall we say.

7:09Debs: That requires a bit more work, right? You're not just going to take this list and say, okay, invest. So you can actually, what you've just suggested is absolutely something that firms do do in terms of screening on a relative basis compared to industry. So looking for stocks which are cheap within their industry. The screen we've got here is a bit more simple and it was generated by Claude and we will dive into it in a minute. I didn't suggest that I just prompted Claude in fact I'll just show you the prompt and it's a two-line prompt can you produce a value and momentum screen for the S &P 500.

7:45I've given instructions for where the data should come from fundamental data come from the SEC API which means it can auto refresh or at least I have to prompt it to refresh but i don't have to input any more data and then the market data should come from any free source which also offers an api and it should allow some stock scoring based on a user input for factor weighting so i can basically build the relative weights of the different factors but the actual mechanism for calculating the scores was done by claude and what we'll see when we dive into the screen is it basically ranks all of the companies based on each factor so pe price to book i think there's some ev multiples as well it ranks them across the whole s &p 500 and then gives them a factor score a z score if you like which just recalibrates it so it's looking at the spread across the whole range and effectively just ranks them and that allows you to compile different metrics all in the same uh kind of the same ranking system that you can then add up their scores and give a sort of compound score.

8:51But your question on the, what we refer to as value traps, what's just cheap and crappy versus cheap and an opportunity. Cheap and good. Yeah, is the exact reason why we tend not to just use value on its own. And one of the ways that you get around that is to combine it with momentum, because a momentum screen is one that's just looking for stocks that have some upward momentum in their share price. So it's where the market is kind of rallying behind those companies. And it's one way, it's not the only way, but it's one way of trying to ensure that you don't end up buying a value trap, that you have, you know, ended up, you know, that you have got a stock that has got some positive momentum behind it.

9:34The other thing I would just say about the value factors, an important part of trying to avoid just really crappy companies is the cash flow element. So we do tend to include a cash flow type yield, either a dividend yield, or I think this one has a free cash flow yield in it. Because then at least you are buying up companies, even if they're not high growth, which they are, you know, values not looking for growth, but they is looking for companies that genuinely are good quality. So having some kind of cash flow yield in there helps to get around that. So should we have a look at the actual results of Claude's inputs?

10:14Yeah, yeah.

10:15Debs: How did we do here? Well, I think, well, it's already, it's immediate response to my prompt was, this is a great build. So it's already congratulating me on my prompt. I mean, have you ever had something where Claude or GPT said, this is just a crappy idea, don't do this? Listen, Claude loves me. I know that. It tells me every day. Anyway, I shouldn't go into that. um basically uh right yeah so smart um so the first thing i would i would highlight all of our listeners i am not a coder i am not someone who has you know you know python coding skills or anything like that i've done a little bit of visual basic in my time uh but um what claude did was basically provide a screen that does pull in data from an api and i was required to do a little bit of work around pulling in the API instructions.

11:11So let me just show you where we started off.

11:15Debs: And the API here you're talking about is, I guess we need a couple, right? Because they're the market data APIs. And you said you're using Yahoo Finance. And then you are pulling any kind of non-market data, but the reported data direct from SEC. Absolutely. So what you need, particularly for the value screen, you need book values, you need earnings, you need cash flows. So that's what we refer to as fundamentals data. That's coming from the SEC API, so pulled directly from financial reports. And then the market data, which also is needed for the value screen, for the multiples, but for the momentum factors as well.

11:51Those come from Yahoo Finance. Originally, Claude did suggest using Stook, which I know people who use that, but for some reason it was blocking me. I don't know what is in my settings that it didn't like. So there's a bit of a pivot to using Yahoo Finance as a last minute thing. So yeah, so basically the first thing that Claude did was it generated this readme with some instructions for basically how this screen is going to work. Okay. So it was Claude that came up with the idea of all of the factors. I didn't give it any instructions for that. So it suggested, so the multiples, it's flipped those.

12:26So a high yield, a low PE, which is what we're generally looking for, scores well. Yeah. So we've got inverse of multiples. So earnings yield, book to price, EBITDA to EV, sales to EV, free cashflow yield. Yeah. So those are our factors for value. And then the momentum is just using the trends in the share price on a one month, six month, three month basis. So Claude has decided on those factors, which I'm reasonably be happy with. I would quite like to have a dividend yield in there. And one thing I think we probably will find is that, you know, when you have lots of factors which focus on the operational element, for example, EBITDA and sales, you tend to punish financial institution companies, so insurance and banks, simply because they just don't really have that data.

13:18There isn't really an EBITDA for banks and insurance companies. So that means they end up with a zero score. So there's always this concern when you're building a screen that you're going to end up with the factor that kind of distorts the results or naturally excludes a company that would otherwise be viable.

13:33Debs: I guess if you're looking across industries, because that's what you could do here is you say, I want to build a bank screen or insurance screen, and then you look at factors specific to just those industries or those companies. Absolutely. Now, one question I have here is in terms of the multiple calculations, do you know if Claude is calculating the multiples on its own by, say, pulling book value of equity from Edgar and using price data from Yahoo? Or is it pulling the multiple direct from Yahoo? Because I'm just thinking, you know, one of the things I spent a lot of time doing as an analyst when I was in M &A was just scrubbing all the numbers that go into the multiples that you present in a trading comp set or transaction comp set.

14:17Debs: And, you know, kind of thinking how much work has Claude done here on that side of things, if any? Because this is something that presumably you could actually prompt and instruct AI to do relatively well these days. My sense is that it's pulled a lot of the data on the fundamentals from the SEC. What I mean by that is things like EBITDA, earnings, book values that obviously would come from the SEC. But EV, I'm not sure. It would require a bit of digging. I'm not even sure if yahoo publishes ev values that seems quite a step for yahoo finance but um anyway yeah okay um yeah something that would need a bit of investigation um okay and then the next thing is we have the z score which is basically just the way ranking uh each you mean z oh sorry z score i'm in the u.s okay there we go um and then um then as a composite so there's build across all the factors to give them an overall score with rank one being the highest overall score.

15:17Okay. Okay. Then in terms of actually setting up the data, as the instructions say, it was a one-time setup. Now I used Power Query for this because it's something I've used maybe twice in my life before. So I was more familiar with that for building in the API than the alternative that applaud suggested um so basically what it did and you can see it really is paint by numbers it's giving me a list of instructions for what i do to get the api instructions into a power query okay so what we have down here and i know there's just a lot of hash uh hash refs here and they um basically each of these little query boxes is just a set of instructions it's basically some code that i then have to load into power query now i'm not going to open up power query on my device now I'll show you where it is but within the data and get data if you launch the power query editor on my laptop particularly when recording it'll take five minutes to open up but basically that opens up power query in excel that allows you then to paste the instructions into the query and it effectively then connects within the instruction it tells excel where to look for the api source you give it some kind of you know it basically gives all the instructions for what data item it's going to pull from the from the api and then the code also then tells power query where to download that data to and effectively what we end up with is this tab here of raw data this is the data that's come through um from uh this is i think the initial stuff is all from SEC.

17:04And then the stuff to the right is the stuff from Yahoo Finance. So effectively, it's just pulling all of that data through from the two APIs that I mentioned. Okay. And this is very classic.

17:18Debs: And presumably this is something that you set up once, I assume, and you can hit refresh if you want to. Yeah, so I go back into Power Query and just hit refresh data. in fact weirdly enough because i was wondering about whether to to show live how how it's all arranged within power query and i did take a few minutes for it to open um but as soon as i opened it said it did prompt to say do you want to refresh the data the data's now stale because i created the screen yesterday so um yeah it's super easy it's a bit of work up front but once you've done it you're away um so yeah okay um and then once you've got the raw data uh claude then puts this into a screen which has calculations taking the raw data there's you know a little bit of cleaning and this is kind of standard when you're building a screen you know you need to you know create a little bit of logic around you know if there's a zero value you want to make sure that if you're dividing by zero you don't end up with errors so there's a bit of cleaning that's all built into these formulas within the actual screen itself so here we have our factors all the way over to the right.

18:23And then those factors are basically built into your Z scores. So each factor gets the Z score. That's your ranking of results across the universe. And then that's combined into a value score, a momentum score, and then those are combined into a composite score. Now, in terms of that scoring, I have got a list of controls here. Now, this is where I have user override because I can say, well, do you know what? I don't really want to have, let's say, sales to EV as such a big driver of my outputs. I can replace that with 0.5 and that then automatically gives it a lower weighting. But then all the other factors have to compensate for that.

19:07So they get an automatically proportionally higher weighting. So that's how, you know, as a user, I have the ability to still control some of the outputs and just in terms of that relative weighting of each of the factors. And then what happens is we get to our dashboard that, again, all created by Claude, didn't need any prompting from me. I like the fact that at the top, it has a reminder of your tilt. that's what we when we're talking about portfolios um you know we talk about the tilt as being the leaning towards one particular theme or particular um type of factor or strategy um so it's saying here well the tilt is that you're mostly focusing on value here it's more 69 value 31 momentum and that's just reflects the relative weighting of the factors in my controls um so then as we go down we have the top ranked companies and so these are the ones which are scoring top of the board scoreboard for value momentum and then languishing at the bottom we have those companies there which have a negative composite score.

20:12Now at the moment this screen is just you know a dashboard that I could then take away and if I'm you know wanting to start my analysis I can start with that list at the top and do a little bit more research decide if they are genuinely companies that I want to then invest in. But actually, increasingly, even retail investors are using this for much more automated decision making. Now, I mentioned at the beginning that you have quants. They are professional investors who use systematic rules for investing in stocks. So once they've decided on their weighting, they'll do a little bit of back testing, which is basically checking if you work backwards in time, if they use this mix of weights and factors in their investment decisions has it genuinely produced historical good results you know what we refer to as alpha excess performance um in that with within that basket of stocks um and if they can prove to themselves that that gives them an you know an edge in terms of their investment returns they will then use those outputs for automated decision making it's

21:20Debs: what we refer to as a systematic strategy i always say it's so it's so interesting just because it's so different from the style of investing that I'm used to. I think about the process that I went through before making any investment in any particular company. And it was a months long process of fundamental analysis saying, okay, I might have an overall screen in terms of I like this industry. Thematically, I like companies that are positioned a certain way, that kind of thing. But ultimately what it comes down to is doing a lot of primary work on the particular business that you're considering an investment in.

22:01Debs: It's a months long process. You're at investment committee a few times to get final approval and sanction. And then you commit capital. And obviously it's different investment style, right? Because you're making a long-term illiquid investment. If something changes, you can't just trade out of it. You got to be really confident that you are going to be okay for the three to five years, wherever the time horizon is, that you're actually invested in this company. It's just so, it's so different. I find it, I do find it fascinating. Yeah. I mean, I get what you're saying. And I think to a certain extent for active investors, this is just the start of the process and the, you know, for them, qualitative analysis, understanding the company strategy, doing detailed research on, you know, the different business segments, the operations, and really digging down into the valuation and the investment thesis is still really important.

22:55It's just the fact that you don't physically have time to research the entire S &P 500 in that level of detail. So you have to have a filter as your starting point.

23:06Debs: Well, I guess it's that you also have, you can do all the fundamental analysis on a company that you want, but you've got all these other external factors that impact the price of that business, right? So you could say, all right, I'm just looking at the list we have here. You could say, I've done all the work on Hewlett Packard Enterprise, right? And I think they've got great products and they're well positioned, the markets they're serving, good relationship with the customers, whatever the kind of analysis you're looking at is. But then the market tides could change just on that company based on something that might not have anything to do with that company in particular.

23:45Debs: And all of a sudden, you've got a share price movement that you are, that regardless of how much fundamental analysis you've done, doesn't really matter. Yeah, absolutely. I mean, that's the fun of, you know, public markets investing is that you can have a great thesis. You can actually be, you can be 100 % correct in terms of your thesis, but we're completely wrong in terms of the actual investment result that is the reality you know we've you know the markets have had a wild ride you know we've had you know last year we had liberation day you know completely disrupting you know lots of you know really good you know investment ideas but you know that's the reality and that's kind of for me that is the you know not i don't want to say it's fun but that's i do enjoy the fact that you know what you know you can have one plus one equals two one day and then a completely different result the next day because an external factor...

24:36Yeah, one plus one equals negative five. The next day. And because, and that is, you know, it's that the world is constantly changing. It's not a contained system. It's something that's constantly evolving and you have to stay on top of that. And actually on that point, I think it's really interesting on the quant perspective is that with all the developments of AI, quant firms now have to be so much quicker in terms of their evolution of factors. You know, when I was an analyst, you would have quant funds which would spend you know weeks and months researching factors they would set up a portfolio that use those you know that those screens or those systems in their investment strategy and that would apply for weeks months and that you know that would work for a long time whereas you know even when I left the markets it got to the point where some of my clients were talking about you know the investment edge from a new factor it's eroded within weeks and I can imagine that now we're in days because you have AI constantly learning what works and what doesn't and then you know the rest of the market catches up so quickly so you know your edge disappears very quickly these days but I think the other thing that I find quite fascinating is the fact that this is now and it's something we mentioned last in last week's episode the fact that access to data is so cheap and easy and AI makes a lot of this you know process which was beyond a lot of retail investors a long time ago and it makes it so much more accessible and the fact that you know retail investors can and do now build their own screens they even now have the ability to connect the outputs of their screens to their trading platforms you can in theory trade automatically off the back of your investment screening, which, you know, it's a new development.

26:32And I strongly recommend that, you know, you put some good parameters in place if that's what you do. But, you know, it is, you know, changing world. The fact that you can, as an individual now do so much more, you have so much more power in terms of your ability to analyze and make investment decisions than we had even a year or six months ago.

26:52Debs: It's interesting. So I was at a client last week and there was a bit more markets content there. And one of the things that they were talking about was the democratization of private markets in the sense that you have a few more vehicles to, as a retail investor, get access to private markets than you used to. Right. We've talked a while ago about some of the, say, some of the semi-liquid private credit vehicles, like as an example. So are we now in the world where we've got some kind of democratization of quant trading through AI? Like maybe, I guess the question here is, you know, given the cost of AI, like you probably got to trade a reasonable sum of cash to actually justify all the tokens you're burning through coming up with your investment ideas here.

27:46Debs: Because presumably, if you're doing this relatively constantly and using, say, Claude or GPT's API instead of the monthly subscription service that you and I use for most of our stuff, presumably this gets pretty costly pretty quickly. Yeah, completely agree. And in fact, on top of that, there's two other advantages that institution investors and particularly quant funds have. Number one is the access to data. We are using free data sources. They all have proprietary data sources, you know, that they've built themselves or that they will have very expensive paid for data sources that, you know, as an individual we can't access.

28:26And then the trading cost point is also really important because, you know, as an individual, you're trading in small volumes. The relative cost is quite high compared to a quant fund, which is trading, you know, you know, they're trading billions of dollars of stocks each day. they can get at much cheaper pricing. And that's important, you know, particularly if you're in a relatively sort of fluid portfolio where you're trading regularly and it's not for long-term hold, then yeah, it's definitely something that, you know, the funds still have that advantage and I can't see that disappearing in the short term.

29:01Debs: Yeah, it's a good point even on just the trading commission. If you're using one of these tools to trade in and out of stuff frequently, like I'm not I'm not an active public market trader. I don't know what trades cost these days, but it's not nothing, particularly if you're a retail investor. So you got to make sure you've got a lot of conviction here before you actually use one of these tools to pull the trigger. It would be interesting just as a side project, just to come up with some kind of mock portfolio that Claude manages or GBT manages just to see how it does. I'm sure other people, I know other people on online have have done this so maybe we should review a few of those and see see what kind of results people are producing for sure and I know lots of people who are doing that already with a view to you know once they get to a point where they feel confident in their screens they plan to hook it up to their trading portfolios um that's a big yeah it's a big vote of confidence for Claude, isn't it?

30:02But yeah, absolutely. It's an interesting one. We can go away and have a think about maybe doing some analysis of that in a future episode.

30:11Debs: Yeah, no, 100%. But yeah, it's super interesting. I want to talk more about this just because, like I said, it's something that I don't have a ton of exposure to. It's a very different investment style than what I'm used to. And it's just always fun to learn a bit more. Great. So I hope you enjoyed our deep dive into screening using AI and a bit of an exploration into different investment styles and factors. That's all we have time for this week, but thanks for tuning in and see you next week.