Speaking as a former M&A financial advisor and valuation nerd, historical financial data is very close to worthless for any valuation work, except perhaps for vaguely connecting the dots to your proprietary, forward looking financial model which is based on a deep understanding of a particular company and industry.
This reads to me like garbage in, garbage out... just like 99.9999% of current resources on financial data.
There's definitely an opportunity to disrupt this backwards "financial data" industry, but it's not going to be done by slapping LLMs and RAGs onto stale data in 10-Qs and 10-Ks.
While I agree with your statement and recognize that, for now, Perplexity has only introduced a financial information platform comparable to Google Finance or Yahoo Finance, the true value of any forward-looking financial model is rooted in the depth of the qualitative research supporting it.
Building a useful forward looking financial model mostly involves qualitative analysis. This means thoroughly examining the company's and competitors' 10-Ks and 10-Qs, digesting industry reports, understanding the company’s business model, breaking down the underlying mechanics of the income statement, balance sheet, and cash flow statement, identifying the core processes driving value creation, forming solid hypotheses on how the business will evolve, etc.
I believe Perplexity, as an advanced answering engine, provides a strong foundation for supporting this kind of in-depth research and hope to see the platform evolve into this direction.
I am always surprised companies create these distractions in their product. Agree on all your points. Little to no value in this tool based on the data they listed. All of these things already exist and for anyone doing work of any value, you would already have these datasets at your fingertips.
Historical financial data is used for calculating correlations between assets, historical volatility, stress testing, VaR etc. All very useful if you follow any sort of risk management when constructing a portfolio.
That's pretty much just stock prices, not 10-Qs and 10-Ks. And it's already available elsewhere. For those use cases, you need databases and data sources, not LLMs and RAGs.
Is there a "backwards" financial data service you'd recommend instead?
I've been researching a bit and everything I could find was basically APIs that charged you by the number of API request calls to extract the dataset bit by bit. First the tickers, then the aggregates... and so on.
They care about anything that helps give insight into the future. Its still obtainable data but that is where the edge comes from, taking unique data and applying your way of thinking (models) to it. This could be anything and everything that relates to the company/industry you are looking at. You take lots of different bits of information and distill it into a model. If you are in the automotive space, maybe you care how US policy is looking for China.
To put it differently, historical financials, K/Qs are all data points that have been commoditized and you can pull it instantly.
It depends on whether you're publicly sharing your valuation work or confidentially advising a client about valuation.
The former is what equity research analysts do at major investment banks (like Morgan Stanley, BAML, JP Morgan, etc.) and boutique / middle-market research firms (Stifel, Cantor Fitzgerald, Raymond James, Guggenheim, etc.). Google has led me to this LinkedIn post with a long list of equity research firms which seems accurate and credible after skimming it briefly: https://www.linkedin.com/pulse/most-comprehensive-list-sell-...
The latter is what I used to do as an M&A advisor. Basically the "I want to buy company X, how much should I pay?" or "I may be interested in selling / people are reaching out expressing interest in my company, how much am I worth?" type of scenarios, plus some other more complex but not necessarily more fun things like merger of equals and what have you. In these situations, the analysis tends to be more of a "let's look at it from all angles" which usually gets distilled into a one-page summary nicknamed "football field" showing ranges of values according to various methodologies. Things like DCF, discounted equity value, relative trading multiples (also called "comps"), LBO (also called the "floor valuation") all get featured and are the standard metrics on which an advisor's view is supported.
As an advisor, I never came up with my own projections for valuation, because that would open the door for litigation so it's just not done. Instead, we point to what "the street" is saying, by taking the consensus view (often some filtered average/mean of equity research projections usually provided by Bloomberg, FactSet or Capital IQ for liability reasons, but which may be "handspread" in some situations by actually pulling the specific numbers from the latest available research reports from several analysts and calculating the mean).
I hope that helps answer your question but happy to answer any follow-ups too since I'm always glad to share what I know about the topic
That’s super helpful - so basically what other deals are going on in the market and how they’re being structured?
Which makes sense - seems like Perplexity will always cater towards your average retail investor, there’s just too much complexity and cost to acquiring the type of data a professional cares about. Seems like that will need a to be its own platform.
> That’s super helpful - so basically what other deals are going on in the market and how they’re being structured?
That's one of the rows of the football field. Actually one I forgot to mention, often called "precedent transactions". You can see some examples of these slides if you scroll down to "Why these slides are made" here https://www.alexanderjarvis.com/investment-banking-slide-exa...
But when I mentioned relative valuation or trading comparables, I meant looking at similar public companies today and what their implied valuation is based on their current share price (more about that here https://news.ycombinator.com/item?id=41862295)
Those relative valuation methodologies are different so-called "intrinsic" valuation methodologies like DCF or discounted equity value, which just look at the company you're valuing and do some math "in a vacuum" to get some implied value per share.
> Which makes sense - seems like Perplexity will always cater towards your average retail investor, there’s just too much complexity and cost to acquiring the type of data a professional cares about. Seems like that will need a to be its own platform.
I agree, but there's definitely a lot of room for automation in the professional space. Unfortunately I can only tackle one startup at a time so that's like ~third on my list of revolutionary ideas that I'll get to one day ;-)
Warren Buffett is famous for reading the annual reports, basically all of them and starting from the back where you tend to get the embarrassing stuff in small print.
That's a lot of "this won't work" without very much "here's what does work" leading me to conclude this is bluster and ego.
I always hear these finance people dick waving about how crap everyone else's methodologies are without examples of their own. This leads me to conclude it's all snake oil anyway.
So I'm asking, speaking as a former M&A financial advisor... What _does_ work?
I'm not a former M&A guy but observing people in the market, figuring out the fundamentals better than the other guys and investing to maximise your returns can work and people like Buffett, Watsa, Soros and the late Simmons could do well enough to give billions away to charity. A few gotchas
- It can be a bit of a zero sum game and the Buffett's and Soros make outsized returns because joe public makes lower ones.
- Investing like that is hard - Buffett would basically spend all waking hours studying the stuff when younger. Just reading a How to Invest Like Buffett article doesn't cut it.
- Most "investment professionals" make money on fees from clients and so will recommend what sells to clients - typically the hot thing of the day - rather than what's the best investment which is often cheap because most people think it's dull / doomed / unrespectable.
As a quant myself, we don't try to predict the market, at least not the way that people normally talk about predicting, and we certainly don't move the market in our favor.
At least what my firm does, is we look at the current state of the market at any given time point, and test whether the current state of the market satisfies our model of an efficient market. If it does, then there's no action to take, if it doesn't then we determine what kind of violation is present and jump in to close the gap.
So a very trivial example would be to take two ETFs, like QQQ and TQQQ. As a simplification a model of an efficient market would have at any moment in the day the change in price of TQQQ = 3x the change in price of QQQ.
We then observe the actual state of the market and if the actual change in price of TQQQ matches our model, then there's nothing to do. If it doesn't, then either TQQQ is under priced or it's overpriced or QQQ is underpriced or it's overpriced (or our model is just wrong or some outlier). Depending out what the condition is we buy x dollars worth of TQQQ and sell 3x worth of QQQ or do the opposite.
There's no real prediction here, we simply have a model of what an efficient market looks like, we scan the market for violations of that model, and then we perform an action to bring the market back to an efficient state.
The model I presented above is incredibly simple and just for illustrative purposes, but in a nutshell, that's our job. We have literally hundreds of models for an efficient market and for every model we have algos that test whether the market satisfies our model, and when the market deviates from our model the algo produces a signal which other algos act.
So basically, you’re seeking super low-risk arbitrage opportunities of low-moderate complexity, but like, really high throughput and with really low latency trading?
Exactly, over the past 15 years of doing this I can basically recall every single day that my firm had a net loss, with Brexit was the biggest one. Most of the losses were due to technical failures/bugs/networking issues, very few one of them were due to issues with the model.
And yes, high throughput and low latency are critical aspects of our trading and they are factored into the model as well, in that for every deviation we observe from our model need to measure how long such a deviation is likely to last and we only trade on those which are likely to last long enough for the trading algo to complete.
The one consistent method I know of would be high-frequency trading to front-run orders, which involves maintaining a moving target of state of the art infrastructure (both hardware and software), including a relationship at the markets so you can get an ultra high-speed connection (I'm certain there are rules with this to make it fairer, but I would assume not everyone will be provided a connection simply due to physical limits).
But I also assume that's not the type of thing parent comment is asking about - Any rational actor with an opportunity to do this would already be doing this after all.
There are a number of wealthy investors who don't tell other people about their methodology. They're either very lucky, criminals, or have beaten the market.
There's wealthy, and there's Wealthy. As the market itself tends to trend upwards, you can absolutely ride that into wealthy, no crime or extreme luck required.
nothing. There is no way to predict the stock market. Even the way you're probably thinking of. Even the ones in the replies to this comment. Even the really basic ones and the really advanced ones.
Tbf, I’ve had friends in finance describe the Millennium fund (now Medallion fund I think?) as operating in a very, very specific niche of the market. Probably helps that it’s limited to a very select group. Its extraordinary impressive for sure, but I don’t think it is widely applicable or comparable.
Am I missing something, or is this just a normal stock browser like Yahoo Finance? I don't see anything remotely new in the article or the site (https://www.perplexity.ai/finance/NVDA) and it doesn't seem to have anything to do with AI. It's a nice-looking feature but I don't think it's newsworthy.
The benefit is that it gives you some suggested topics of investigation, including references to where it found that information. For example, I looked at one stock I'm interested in and it suggested there was a potential short squeeze in the making, and the references pointed out that that was 7 months ago. So it was summarizing it thinking that was a current state. It also suggested summaries about how changes in classification of the business might impact the stock price.
I'll probably use this in some of my investigations, but definitely need to look at the citations.
It’s probably the traditional way a glowing article gets written: a reporter repeats something that a PR rep sent to their inbox cause they needed something written to hit a deadline.
This is why good books on PR will always tell you to write your press release like an article, so that a lazy journalist can post it verbatim. (Have gotten this result a couple times from press releases I wrote when I was running my previous company!)
>At this stage of the game, though, Perplexity Finance needs much more refining before I can call it a real competitor to the existing stock analysis programs, such as Stock Rover, WallStreetZen, and TradingView.
Somewhat related: Last night I fed 3 companies proxy statements to Notebook LM and had it generate a "podcast" on them. One of which was a 100 page document, which I read the bulk of, and the podcast was pretty good, though it did misrepresent one 2023 statement as being a current statement. Definitely a tool I'm going to be using more.
It would be much more interesting (and challenging) a platform to automatic trend recognition through news, while it could only be valid if personal, due to obvious bias any owner can introduce...
This reads to me like garbage in, garbage out... just like 99.9999% of current resources on financial data.
There's definitely an opportunity to disrupt this backwards "financial data" industry, but it's not going to be done by slapping LLMs and RAGs onto stale data in 10-Qs and 10-Ks.