Latest / Deep Values: Value Investing Deep Dives / [ROE] AI Algorithms Hack Traditional Value Investing
Transcript
- 0:00You know that classic image of Warren Buffett?
- 0:03Oh, absolutely. The Oracle of Omaha.
- 0:05Yeah, exactly. He's like sitting in this quiet,
- 0:08dimly lit office out in Omaha, far away from all the noise of
- 0:12Wall Street. Right, just surrounded by these
- 0:15towering stacks of paper annual reports.
- 0:18Yeah, reading through corporate filings for, well, for hours on
- 0:21end. And he uses decades of sheer
- 0:23intuition and pattern recognition to just find a
- 0:27hidden gem in the stock market. It's incredible.
- 0:30It is. We treat value investing almost
- 0:32like this, this romantic art form, you know, it's like the
- 0:35ultimate human endeavor. Right, because it relies
- 0:38entirely on a financial 6th sense that takes a lifetime to
- 0:42cultivate. Totally.
- 0:43I mean, the traditional philosophy going all the way
- 0:45back to Benjamin Graham and David Dodd in the 1930s is
- 0:48really about finding an imperfect relationship between a
- 0:51company's historical financial metrics and its future stock
- 0:54price. You want to buy a dollar for
- 0:56$0.50? Exactly, you are always looking
- 0:59for a margin of safety. So, OK, imagine taking that
- 1:02deeply human, intuitive art form, ripping it out of the
- 1:05hands of legendary stock pickers, and just feeding it
- 1:09into a machine learning algorithm.
- 1:11Which sounds kind of crazy. It does, but that is exactly
- 1:15what we are diving into today. Data scientists are essentially
- 1:18hacking traditional value investing.
- 1:20They really are. So today, for you listening, our
- 1:23mission is to explore how modern algorithms are quantifying
- 1:28intuition, and why the financial numbers you rely on every day to
- 1:33evaluate a company might be completely lying to you.
- 1:36We are really compelling collision of old school
- 1:39philosophy and cutting edge data science to look at today.
- 1:42Yeah, we've got a great stack of research.
- 1:43We do. We are drawing from an academic
- 1:45paper by RJ Sanic on predictive modeling in systematic value
- 1:50investing. Which is fascinating.
- 1:52It is. We're also looking at a
- 1:53specialized real estate journal that analyzes how financial
- 1:57leverage artificially manipulates profitability
- 1:59metrics, right. And then to tie it all together,
- 2:01we have a forensic accounting brief from new constructs.
- 2:04OK, let's unpack this. Because the first major hurdle
- 2:08in classic value investing is that it is just infuriatingly
- 2:13subjective, extremely subjective, right?
- 2:15Like if you read a book like Buffettology, which is Mary
- 2:18Buffett's breakdown of Warren Buffett's methods, you are hit
- 2:22with all these qualitative metrics.
- 2:23A soft metrics, yeah. Buffett looks for a company with
- 2:26an an identifiable consumer monopoly or a business that
- 2:31demonstrates strong earnings. Well, strong is not a
- 2:35mathematical formula, exactly, and monopoly is not a standard
- 2:38accounting metric. You could just, you know, scrape
- 2:41from a database. That is the core problem, and it
- 2:43brings us right to RJ Sack's research, because Sack actually
- 2:46takes Buffett and Clark's highly subjective value investing
- 2:50metrics and defines them in purely quantitative terms.
- 2:53It translates them. Yeah, but I struggle with this
- 2:56premise a bit. Oh really?
- 2:57Why is that? Well, I mean, can an algorithm
- 3:00truly capture something as abstract as a consumer monopoly
- 3:03or franchise power? It's a fair question.
- 3:06It feels like trying to translate a beautiful piece of
- 3:08poetry into cold algebra. You know?
- 3:11Like are we just stripping the soul out of the strategy and
- 3:14pretending we still have the same thing?
- 3:16I get that you might lose some of the poetry, sure, but Sack
- 3:20argues the trade off is absolutely worth it.
- 3:22OK, how so? Let's look at how he actually
- 3:24does this. He doesn't just assign a random
- 3:27number to a feeling, right? He translates a concept like
- 3:30consistently upward earnings trends into specific geometric
- 3:34mean calculations over like an 8 year period.
- 3:38Oh wow. OK.
- 3:39Yeah, he takes the abstract idea of a company having a durable
- 3:42competitive advantage and turns it into a strict threshold of
- 3:45historical gross margin stability.
- 3:47So he is forcing the algorithm to mathematically prove the
- 3:50things Buffett just intuitively senses by looking at A.
- 3:53Balance sheet Precisely by quantifying these metrics, AC
- 3:57enables them to be used in systematic algorithmic stock
- 4:00icking. Which changes the game entirely.
- 4:02It really does, because when you do that, you remove human bias.
- 4:05You completely eliminate the emotional error of falling in
- 4:08love with a stock story. Or panicking during a market
- 4:11downturn, which we all do. Exactly.
- 4:15You turn the legendary Oracle into a cold, repeatable process.
- 4:18I mean, I understand sacrificing the poetry for discipline, but
- 4:22people have tried systematic value investing before.
- 4:25They have, yes. Like it isn't a new concept to
- 4:27just write a list of mathematical rules and let a
- 4:30computer pick the stocks. They have tried it and
- 4:33traditional systematic value investing has a massive flaw.
- 4:37OK, what is it? Well, look at Wesley Gray and
- 4:40Tobias Carlisle's well known quantitative value strategy.
- 4:44It's a brilliant strategy, but it uses static variable
- 4:47coefficients. Meaning the rules are completely
- 4:50rigid and they never change. Right, A static model assumes
- 4:54that the relationship between a specific financial metrics, say
- 4:57like the price to earnings ratio and the stocks future market
- 5:01outcome never alters. It assumes the rule works the
- 5:04exact same way across every single business cycle.
- 5:06Which is wildly unrealistic if you have spent like any time at
- 5:10all watching the market because the metrics that predicted a
- 5:13winning stockduringthe.com boom of the late 90s would get you
- 5:17absolutely crushed in the 2008 financial crisis.
- 5:21The economy is a living, breathing ecosystem.
- 5:24You know it changes. And that rigidness is the exact
- 5:28problem SAC introduces predictive modelling to solve.
- 5:31OK, instead of static rules, predictive modelling uses
- 5:35machine learning algorithms to dynamically update these
- 5:38relationships every single time new financial statement data
- 5:42becomes available. As wild and.
- 5:44Out of the several models SAC tests, the one that showed the
- 5:47most predictive power was a random forest algorithm.
- 5:50OK, so for anyone who isn't a data scientist, a random forest
- 5:53algorithm sounds incredibly intimidating.
- 5:55It does sound like a sci-fi concept, but think of it this
- 5:58way. A single decision tree is just a
- 6:00flow chart. Like a yes or no thing.
- 6:02Right. If earnings are up, go left.
- 6:04If debt is high, go right. But a single tree can be easily
- 6:07fooled by weird, noisy data. So what does the forest do?
- 6:11A random forest algorithm builds thousands of these distinct
- 6:15decision trees, gives them slightly different slices of the
- 6:18data, and then merges them all together to take a collective
- 6:21vote. Oh, I see.
- 6:22Yeah, it provides a much more accurate and stable prediction
- 6:24because it just averages out the individual errors.
- 6:27That makes a lot of sense. So Sacks set this random forest
- 6:31to a 50% cut off threshold, meaning it had to be highly
- 6:35confident to make a pick. And he didn't just feed it
- 6:38standard financial data. He combined the classic
- 6:41quantitative value variables with the quantified buffetology
- 6:45variables we just discussed. He fed the machine both the
- 6:47standard math and the translated poetry.
- 6:50I love that and the results of his back test are really what
- 6:53make this research stand out. They are phenomenal.
- 6:55He ran the model over a 26 year period from 1990 to 20, 2016.
- 7:00He trained specific variations, Model 11 and Model 12, to
- 7:04predict whether a stock would generate alpha over a five year
- 7:07and 10 year horizon. And they drastically
- 7:10outperformed traditional systematic strategies.
- 7:12But I mean in the finance world anyone can over optimize a back
- 7:16test to make it look like they beat the S&P 500.
- 7:18Oh sure, which is exactly why we have to look at the academic
- 7:21benchmarks. Right.
- 7:22The models generated statistically significant
- 7:26returns above both the Fama fringe 3 factor and Carhartt 4
- 7:30factor models. OK, let's break those down,
- 7:32because throwing around Foma French is a great way to lose
- 7:36someone's attention. Fair enough.
- 7:38Why do those benchmarks actually matter?
- 7:40Well, in finance we have to prove an algorithm isn't just
- 7:44getting lucky, right? Like if the overall market goes
- 7:47up 20% in a year, a monkey throwing darts at a newspaper
- 7:50could make 15%. Exactly.
- 7:52Everyone looks like a genius in a bull market, right?
- 7:55So Foma, French, and Carhart are models that strip away obvious
- 7:59market trends. They adjust for things like
- 8:01general market risk, the tendency for small cap stocks to
- 8:05outperform large cap stocks, the tendency for cheap value stocks
- 8:08to outperform expensive growth stocks.
- 8:10No, I get it. It's simple price momentum,
- 8:12right? Exactly.
- 8:13So if your algorithm beats those models, it means you aren't just
- 8:17riding a general market waste. You're actually generating
- 8:20unique standalone value, yes. The machine is genuinely finding
- 8:25hidden gems. That is incredible.
- 8:27It is, but what's fascinating here is SAC ran into a highly
- 8:33nuanced statistical hurdle while proving this.
- 8:36And it is a fascinating problem that plagues quantitative
- 8:40finance. What's the problem?
- 8:41Well, when you run massive financial back test with
- 8:45thousands of variables across decades of data, you run into
- 8:48the problem of false positives. Wait, if it's scanning decades
- 8:52of historical data, isn't it basically cheating?
- 8:55Sort of, yeah. Like of course it is going to
- 8:56find a pattern looking backward. How do we know the pattern is
- 8:59real and not just a coincidence? That is the exact danger of data
- 9:02mining. Imagine you test 10,000
- 9:05completely random variables to see if they predict the S&P 500.
- 9:08OK. You might test whether the
- 9:10amount of butter produced in Bangladesh correlates with the
- 9:12tech sector in California. Just totally unrelated things.
- 9:15Right. And just by sheer mathematical
- 9:18luck, you will find a few patterns that look incredibly
- 9:21predictive on paper, but they are completely fake.
- 9:25So how do you fix that? In statistics, you normally use
- 9:28something called the Benjamin E Hochberg procedure to control
- 9:31that false discovery rate. OK, so it acts as a like a
- 9:34mathematical penalty? Exactly.
- 9:36It punishes the model for taking too many wild guesses, ensuring
- 9:40the patterns it finds are actually rooted in reality, not
- 9:43just random noise. You've got it, and Sack actually
- 9:47proposed a brand new statistic in this paper termed the SAC
- 9:50ratio. Oh, he named it after himself.
- 9:53Class. Naturally, but it extends that
- 9:56Benjamin E Hochberg procedure specifically to help
- 9:59inferentially identify false positive observations in these
- 10:03massive financial back tests. So it acts as a built in lie
- 10:06detector for the algorithm just to ensure it isn't just finding
- 10:09shapes in the clouds. That's a perfect way to put it.
- 10:11But OK, here's where it gets really interesting.
- 10:13We have built this perfect emotionless lie detecting stock
- 10:17picking we have. But here is the fatal flaw in
- 10:21this entire setup. A predictive model, no matter
- 10:24how advanced, is only as good as the raw ingredients you feed
- 10:28into it. The data, right?
- 10:30What happens to our brilliant random forest algorithm if the
- 10:34standard accounting numbers it relies on are actually
- 10:37manipulating the truth? Well that is the Achilles heel
- 10:40of quantitative strategy. Yeah, if the underlying
- 10:43corporate data is legally manipulated or misleading, the
- 10:47models predictions will be dangerously wrong.
- 10:50So let's look at how a company can legally manipulate its
- 10:52profitability on paper, which brings in the study from the
- 10:55Indonesia Auditing Research Journal.
- 10:57Right, the real estate study. Yeah, they looked at the
- 11:00property and real estate sector. It is a highly cyclical,
- 11:03extremely capital intensive industry.
- 11:05Very much so. And they focused on the
- 11:07relationship between financial leverage, which is taking on
- 11:10debt and return on equity or Roe.
- 11:12Which is a huge metric for investors.
- 11:15The research shows a robust positive correlation.
- 11:18When real estate companies take on more debt, their Roe shoots
- 11:21up. To understand the mechanics of
- 11:23how that trick works, we really have to look at the formula for
- 11:27Roe. OK, let's do it.
- 11:28Return on equity is calculated by taking net income and
- 11:32dividing it by shareholders equity.
- 11:34Let's walk through a real world scenario so we don't get lost in
- 11:37the fractions. Let's say you and I start a real
- 11:40estate company. Together, we put in $1 million
- 11:43of our own money. OK.
- 11:44So that is our equity. Right.
- 11:46We buy a building, we rent it out and at the end of the year
- 11:49our net income is $100,000. So our Roe is 10%.
- 11:53Yeah, a 10% return is solid, but let's say you want to
- 11:57aggressively expand. As one does, right?
- 12:00Instead of finding more investors and issuing more
- 12:02stock, which would increase that equity denominator, we go to the
- 12:06bank. We take out a loan.
- 12:07Exactly. We borrow $9 million, we buy 10
- 12:10more buildings and assuming the market is good and the
- 12:13properties perform well, our net income might jump to $500,000
- 12:18after paying the interest on the loan.
- 12:20Wow. OK.
- 12:21But our equity, the money you and I actually put in is still
- 12:24just $1 million. Yes.
- 12:26So now you divide that $500,000 net income by our 1,000,000 in
- 12:32equity? And our Roe just skyrocketed to
- 12:3550%. Exactly.
- 12:36We didn't necessarily become better operators.
- 12:38We just use the bank's money to artificially pump up the return
- 12:42on our own money. That is wild and if you are
- 12:45someone who just looks up a stock on a free finance website,
- 12:49sorts by highest Roe and hits by this research is a massive red
- 12:54flag. Oh, a huge red flag.
- 12:56You might be looking at a company that is just drowning in
- 12:58debt. The study explicitly points out
- 13:01that while leverage amplifies gains on paper, it exposes the
- 13:05company to extreme downside risks during market
- 13:08fluctuations. Right, because debt doesn't care
- 13:10about a recession. Exactly.
- 13:12If property values drop or interest rates rise, that
- 13:15massive debt load still has to be serviced.
- 13:18The equity can be wiped out rapidly.
- 13:19The study also contrasts this with earnings per share, or EPS,
- 13:23right. They found the impact of
- 13:25leverage on EPS is far more multifaceted and nuanced.
- 13:28EPS is heavily influenced by operational efficiency and
- 13:31market cycles, not just, you know, capital structure
- 13:33trickery. So if a machine learning
- 13:35algorithm is blindly scanning the market for a high Roe, it
- 13:39might just be loading up on highly indebted real estate
- 13:42companies right before a massive housing crash.
- 13:45Completely convinced it is buying great value stocks.
- 13:49If we connect this to the bigger picture, the real estate debt
- 13:52problem perfectly illustrates the core warning from our final
- 13:55source. The new constructs brief.
- 13:58Right, The forensic counting brief from new constructs flat
- 14:00out says that because leverage artificially inflates Roe, it is
- 14:04an incredibly dangerous metric for both human investors and
- 14:08machine learning models to rely on.
- 14:10It's just bad data. Tandard counting data is
- 14:12fundamentally flawed if you were trying to find true intrinsic
- 14:15value. And the new contracts research
- 14:17argues that the alternative, like the true driver of a
- 14:20company's valuation is return on invested capital or ROIC.
- 14:24So how does ROIC actually solve the debt manipulation problem we
- 14:29just talked about? Well, ROIC looks at the return a
- 14:32company makes on all the capital invested in the business, both
- 14:35debt and equity. So it removes the illusion
- 14:37entirely. It does go back to your real
- 14:39estate example. ROIC would look at the full $10
- 14:41million deployed. The 1,000,000 in equity plus the
- 14:459 million in debt. Exactly.
- 14:47It measures how effectively management is allocating the
- 14:50total pool of capital, completely neutralizing the
- 14:53accounting trick of keeping the equity denominator small.
- 14:56That sounds like the perfect metric, but I'm guessing there's
- 15:00a catch. There is always a catch.
- 15:01You cannot just look up a pristine unmanipulated ROIC
- 15:05number on a standard stock Screener, can you?
- 15:08No, you cannot. Calculating a clean ROIC
- 15:12requires true economic earnings. Yeah, OK.
- 15:15And to get those, you have to dig incredibly deep into the
- 15:18financial footnotes and the MDNA.
- 15:20The management's discussion and analysis section exactly which
- 15:23is where companies bury the inconvenient truths.
- 15:25Oh, absolutely. They hide off balance sheet
- 15:27debt. They obscure operating leases.
- 15:29They bury unusual write. Offs and they mask changes in
- 15:32their accounting assumptions deep in like page 140 of a dense
- 15:36legal text. Right.
- 15:38And no human being, not even a massive team of Wall Street
- 15:41analysts, can efficiently read and decode the footnotes for
- 15:45thousands of publicly traded stocks every single quarter.
- 15:49It is physically impossible to keep up with the sheer volume of
- 15:52text. It really is, which is where the
- 15:54technology circles back to save the strategy.
- 15:58The New Constructs paper explains that to feed accurate,
- 16:01unmanipulated data into the kinds of predictive models RJ
- 16:05SAC built, you require robo analysts.
- 16:08Robo analysts, OK. These are systems utilizing
- 16:11natural language processing to instantly read, parse, and
- 16:14translate the dense text to financial footnotes into clean
- 16:17economic data. So the AI is reading the
- 16:19footnotes for us. Exactly.
- 16:21Natural language processing allows the machine to read
- 16:23context. It isn't just looking for
- 16:24keywords, it actually understands.
- 16:27When a company is trying to hide A liability at a convoluted
- 16:29sentence that. Is amazing.
- 16:31It extracts the true economic reality of the company,
- 16:34calculates the clean ROIC and then feeds that pristine data
- 16:38into the dynamic random forest algorithm to pick the stocks.
- 16:42Wow. It is literally a complete end
- 16:45to end evolution of the value investing process.
- 16:48It is. The machine does the reading,
- 16:50finds the truth hidden in the fine print, calculates the real
- 16:53metrics, and dynamically updates its investing rules based on the
- 16:57current economic cycle. It's beautiful in a very nerdy
- 17:00way. It really is now for you
- 17:02listening if you were trying to figure out how to actually track
- 17:05all this to access more advanced researches, applications and our
- 17:09latest published personal intelligent multi agent Deep
- 17:12value intelligence. Welcome to explore and join Deep
- 17:15Values AI community and attract total market temperature.
- 17:19Welcome to Search and download iOS App Buffett Indicator from
- 17:22Apple App Store. And I do have to mention, just
- 17:24as a standard requirement disclaimer, all analysis is
- 17:28based on publicly available sources and may be subject to
- 17:30revisions or differing interpretations.
- 17:33All contents from this channel is for research and
- 17:35entertainment purposes only and does not constitute any
- 17:38investment or financial advice. Conduct your own due diligence
- 17:41before making investment financial decisions.
- 17:44Don't forget to like this video. Much appreciate it.
- 17:47Thanks for that. So what does this all mean?
- 17:49We started today with the image of the romantic, intuitive value
- 17:53investor sitting in a quiet office.
- 17:56The. Classic Buffett image.
- 17:57Right. But we have seen that value
- 17:59investing has dramatically evolved.
- 18:02It moved from subjective human intuition looking for a consumer
- 18:05monopoly to static mathematical rules that couldn't adapt to
- 18:10changing markets. And now it has evolved into
- 18:12dynamic machine learning ecosystems.
- 18:14Exactly. Ecosystems that require pristine
- 18:18robo analyzed data prioritizing true metrics like ROIC over
- 18:22easily manipulated numbers like Roe.
- 18:25The translation of poetry into algebra isn't the loss of the
- 18:29art form you know. It is the ultimate refinement of
- 18:31it. I think you're.
- 18:32Right. But this evolution raises a
- 18:34critical issue for you, the listener, to consider long after
- 18:38we wrap up today. Oh, what's that?
- 18:40Well, Benjamin Graham's original concept of the margin of safety
- 18:43relied on the fact that other investors in the market were
- 18:45making emotional, irrational mistakes.
- 18:48Right. It relied on the fact that other
- 18:50people simply hadn't read the footnotes.
- 18:52Exactly human error left value on the table for the smart
- 18:56disciplined investor to find. So if predictive algorithms can
- 18:59successfully decode Warren Buffett's intuition, dynamically
- 19:03adapt to market cycles without emotion, and instantly use
- 19:06natural language processing to parse every footnote to avoid
- 19:10accounting traps. What happens to the stock
- 19:13market's margin of safety when every investor has access to
- 19:16this exact same artificial intelligence?
- 19:19Oh wow, does the advantage disappear entirely?
- 19:21Or does the market simply evolve into a new Uncharted game?
- 19:25When the X-ray machine is in everyone's office, the hidden
- 19:28fractures aren't hidden anymore. Precisely.
- 19:30The market might just become a battle of whose AI can read the
- 19:33footnotes A millisecond faster. It's a brave new world.
- 19:36That is a fascinating thought to leave on.
- 19:38Thanks for joining us on this deep dive, and we'll see you
- 19:40next time.