Welcome to the deep dive, the show where we try to cut through all that financial noise to uncover, well, the hidden signals. Today, we're really digging into how experts are actually figuring out what truly drives stock returns. You know, going way beyond just the daily headlines. Our mission really is to take you on this fascinating deep dive. We want to look at how researchers and practitioners are identifying the real predictors of stock performance. Exactly. We'll start with some uh major academic work that sorted through hundreds of these market characteristics, right? And then we'll pivot. We'll look at how trading volume, specifically this interesting interplay between options and stocks, can reveal some pretty surprising insights. And finally, we'll look at how cutting edge technology, you know, fintech, is actually applying these kinds of ideas to generate actionable intelligence. Yeah, the goal is to take these dense data heavy insights and pull out the most important nuggets for you. Give you a bit of a shortcut to being wellinformed about these subtle forces shaping the markets. Hopefully, you'll discover some surprising facts, maybe even have a few aha moments about where that elusive edge in the markets might actually lie today.

 So, for this deep dive, we're drawing from three really powerful sources. There's a key academic paper that challenges some conventional wisdom on predictors. Then, there's a groundbreaking study on the predictive power of option and stock trading volumes. And finally, we've got an in-depth look at how an innovative financial tech company approaches market analytics. So, uh, let's get started. Okay. All right. First up, imagine like a veritable zoo. That's the term the economist John Cochran used. Hundreds of supposed stock market predictors all kind of shouting for attention. Yeah, it's a mess. And Cochran basically asked back in 2011, okay, out of all these things people claim predict returns, which ones actually provide independent information for average US monthly stock returns? That is, it's like trying to find the signal in an overwhelming amount of noise, right? Which pieces of the puzzle are actually unique when So many look alike. Absolutely. And a team of researchers, Green, Hand, and Zang, they really tackled this headon. Their approach was uh incredibly rigorous. Looked at 94 different firm characteristics all at the same time using something called FMA McBTH regressions, which is basically a way to run lots of tests to isolate what's really driving returns, not just finding accidental patterns. Exactly. And crucially, they were careful not to give too much weight to Micro cap stocks, you know, those tiny companies, right? They only make up like 3% of the market value, but can skew results. Precisely. And maybe even more importantly, they adjusted for what's called data snooping bias. Ah, yes. The risk of finding patterns just by chance because you looked at so much data, like seeing shapes in clouds. That's the one. They worked hard to make sure their findings weren't just statistical flukes. Okay. So, they do all this sophisticated analysis using data from 1980 all the way to 2014. What did they actually, you know, find? Did anything really stand out? Well, the big reveal was just how few characteristics actually had independent predictive power, especially for the non-micro cap stocks, the ones most people invest in. How few are we talking for the broad market value weighted analysis? Only six characteristics out of 94. Wow. Only six. That really cuts through the clutter. It really does. It shows how much noise we often mistake for a real signal. So, when they kind of pulled their different analytical approaches, is what did they end up with for those non-micro cap stocks? They identified what you could call a dirty dozen. Just 12 independent characteristics that genuinely seem to matter for non-micro caps. And what kinds of things were in that dozen? Anything surprising? Well, some you might expect like booktomarket ratio, cash levels, but also more dynamic stuff like changes in the number of analysts covering a stock, earnings announcement returns, even short-term momentum, okay? And things like R&D spending relative to market cap. return volatility, share turnover, even how often a stock had zero trading days. Kind of interesting, right? Very interesting. And what's really striking you mentioned is how these relate to the standard models. Yeah, this is fascinating. 11 of these 12 characteristics were not part of the commonly accepted benchmark factor models, you know, the Carheart model, VMA French, the Q factor model. So the models many people rely on to explain returns based on size, value, momentum, they were missing mostly the independent predictors. pretty much it definitely raises the question do we need completely new frameworks or at least much broader ones to really understand what drives returns that's a big question and didn't the researchers find something else happened partway through their data period oh absolutely this is maybe the most critical finding they discovered a sharp really significant shift around the year 2003 a shift what kind of shift the economic importance even just the statistical significance of these characteristics it just plunged after 2003 plunged how much for non-micro cap stocks the number of independent predictors dropped from those 12 before 2003 down to just two after 2003 only two from 12 that's dramatic it is and for these non micro caps predictability based on characteristics became statistically insignificant after 2003 the signals basically vanished for the bulk of the market wow so what happened in 2003 what does the paper suggest caused this Well, the timing lines up really well with some major changes in the trading environment. Think regulatory changes like regulation FD which aim for fairer disclosure and Sarbain Oxley. Okay. SOX market structure changes too like decimalization making price increments smaller and the rise of auto quoting software and I guess the big one computerized quantitative trading hedge funds using algorithms. Exactly. The paper conjectures that all these things likely boosted market efficiency. They lowered what economists call costly limits to arbitrage. Meaning It got cheaper and faster to exploit any tiny mispricing precisely. So these subtle predictive signals got arbitrageed away much much quicker. The market effectively got smarter or at least faster at incorporating that information like that leaky dam analogy technology plugged the leafs almost instantly.

 Okay. So if the traditional characteristics kind of lost their punch after 2003 because the market adapted, where else might we look for signals? Are there other maybe less obvious things driving returns? That's the perfect leadin to our next source. It looks at a completely different area, something maybe less intuitive, the relationship between options trading and stock trading. Right. The option to stock volume ratio or OS ratio. This idea has been around a bit explored by R Sch Schwarz and Brahmanyam originally. That's right. They proposed that how much options trade compared to their underlying stock can tell you something important, maybe something hidden. And the specific study we're looking at by Johnson and Salow. So what did they find about this OS ratio? They found a really consistent negative relationship between the OS ratio and future stock returns. Negative relationship meaning high options volume relative to stock volume is bad news for the stock price. Basically, yes. In plain terms, firms with low OS ratios tended to outperform the market going forward while firms with high OS ratios tended to underperform. And did they quantify this? Was it a significant effect? Oh, absolutely significant. They constructed a strategy. Go long the stocks with the lowest OS ratios and short the stocks with the highest OS ratios. That strategy yielded an average weekly riskadjusted return of.34%. 0.34% per week. Per week, which if you annualize it, comes out to a pretty staggering 19.3%. Wow. Okay. 19.3% annualized just from looking at relative trading volumes in options and stocks. That's well, that gets your attention. But why? What's the underlying reason this works? That's the fascinating part. The paper argues it's primarily driven by short sale costs in the stock market itself. Okay, explain that. What do short sale costs have to do with options volume? Well, think about it. If you have negative information about a stock, you might want to short it. But shorting can be expensive or difficult. You have to borrow the shares, pay lending fees, face potential recall risks, right? There are frictions involved. Exactly. So, if you have negative news, it might actually be more attractive, cheaper, or easier to express that view in the options market. For instance, you could write call options or buy put options. Ah, so you bypass the hassle of shorting the stock directly. Precisely. Which means that a surge in options volume, this high OS ratio, can be a subtle indicator that informed traders are betting against the stock, perhaps because shorting the stock itself is costly or constrained. It's like a backd dooror for negative information flow. That makes sense. It's using the options market as an alternative channel. But how practical is this? Do you need special data feeds to know who's trading or why? That's the really neat part. The key innovation here is that the signal relies entirely on publicly available total option and equity volumes. So you don't need fancy directional data like figuring out if trades were buyer initiated or seller initiated? Nope. Just the total volumes reported are enough to construct the OS ratio and capture this effect. That makes it potentially very practical. Are there conditions where this signal works better or worse? Yes. The study found the predictive power is strongest when those equity short sale costs are indeed high. Makes sense, right? Right? That's when traders have more incentive to use options and also when option leverage is relatively low. It suggests the kind of nuanced interplay between these market frictions and how information flows. And it's not just about predicting general price drifts, is it? Didn't they look at earnings announcements, too? They did. They found that the OS ratio in the week before an earnings announcement was negatively correlated with the actual earnings surprise. So, high options volume predicted a negative earning surprise and Low volume predicted a positive one. Exactly. It suggests traders are using options to position ahead of earnings based on private information. And interestingly, most of that information seems to get baked into the price right at the announcement. There wasn't much predictability left in the days after the announcement based on the prior OS ratio. So the signal really captures that pre-announcement information flow. Very interesting. Yeah, it shows how options markets can sometimes reflect information before the equity market fully catches on. Okay, so we've seen academic theory wrestling with predictors, finding many faded after 2003. Then we saw a powerful signal emerge from the options market driven by frictions like short sale costs. Now, how does all this translate into the real world? How are firms actually trying to apply these kinds of datadriven insights? That brings us to our third source, looking at a company called Exponential Technology or Xtech, right? X-Tech led by Morgan Slade who has a deep background in AI and high frequency trading. from some major institutions. They're really at the intersection of theory and practice and their main product, Indigo Panther. What's the core idea behind that? Indigo Panther is designed to figure out the direction of active trading specifically distinguishing between institutional investors, market makers, and retail traders just by analyzing the standard US consolidated feed trade data. So trying to see who's buying and who's selling even within the public data stream. Essentially, yes. The core assumption is that institutional net flow The net buying or selling by large institutions generally reflects views on a stock's fundamentals. These are often informed players making big bets. But isn't their trading impact usually slow because they have to trade huge volumes carefully? Exactly. That's the inefficiency Xtech tries to exploit. The impact is often slow, inefficient because of the sheer size and the limited liquidity for massive orders. Institutions try to hide their tracks, break up orders. They leave a subtle footprint as XTEC calls it. Precisely. Xtech uses advanced algorithms, presumably machine learning, to detect these subtle patterns, not just the trades themselves, but maybe the order flow, cancellations, how trades are fragmented across exchanges, essentially the digital breadcrumbs left by these big players trying to be discreet. And does their research actually confirm that this institutional flow leaves a detectable, predictable trail, that the market isn't perfectly efficient in absorbing it? It seems so. Their analysis found that over 71% of stock symbols showed significant autocorrelation in the cumulative institutional net flow they calculated. Autocorrelation meaning the flow tends to persist. If institutions were buying yesterday, they're likely buying today. Correct? It's not random. And furthermore, they looked at something called the Hurst exponent, a measure of long-term memory or trendiness. The median value they found was 65. And what does 65 mean in that context? Well, a value of 0.5 suggests randomness like a coin flip. Anything above.5 suggests persist. or momentum. So 65 indicates a pretty strong tendency for these institutional flows to trend, not just be random noise. It's evidence against the strong form of market efficiency. So they found evidence of inefficiency. Have they actually built trading strategies based on this? Yes, they've developed and back tested several flow-based strategies. One example they highlight is a daily rebalancing T+1 open to open mean reversion strategy applied to S&P 500 stocks from 2010 to 2025. Mean reversion. So betting that stocks with strong inflow today might revert slightly tomorrow and vice versa. That's the idea. And the back test results were quite impressive. An 8.55% annual return, a sharp ratio of 1.08 and a cumulative return over that period of about 242%. That's strong performance and you mentioned it was robust. Yeah, they showed the results held up reasonably well even when factoring in different levels of transaction costs which is crucial for real world applicability. And it's not just mean reversion, right? They also looked at momentum. Correct. They also demonstrated flow momentum reversal strategies. They mentioned acronyms like TSM and MSTR which also shows significant cumulative returns in out of sample tests beating benchmarks. So they seem to be capturing different types of flow dynamics capturing alpha across different time horizons too. It seems exactly that's another key point. Their research suggests these flow signals can provide alpha or edge across a whole spectrum of time horizons from very short-term stuff like 1 to three day mean reversion or even predicting prices in intraday auctions all the way to to medium-term like identifying trades that resemble the size of quarterly 13F filings by institutions playing out over maybe 10 to 45 days and even longer term multi-quarter momentum factors driven by persistent flows. That's incredibly versatile. It sounds like this kind of flow indicator could be used in many different ways. For sure. The potential applications they mentioned are really broad. It could give risk managers real-time market color on positioning. Help traders interpret stock behavior around on big events like earnings or macro data releases. Definitely improve volatility forecasts, potentially even help predict activist investor campaigns, you know, 13D filings or identify stocks involved in M&A based on unusual flow patterns. So, it's about generating actionable intelligence for various market participants and time frames. That's the idea. It really exemplifies taking these deep data insights and applying serious technology, AI, machine learning to build sophisticated real world tools. It's a big step from just identifying an academic anomaly. Wow, that was quite a journey covering a lot of ground from academic theory to cutting edge tech. So, just to recap for everyone listening, we started by exploring how rigorous academic research took that zoo of potential stock predictors and found that especially after 2003, very few actually held independent predictive power for most stocks, market efficiency seemed to really increase. Mhm. Then we pivoted to the options market and saw how a relatively simple measure, the option to stock volume ratio can actually reveal powerful hidden information about future returns likely driven by frictions like short sale costs in the equity market. That negative relationship low OS correlating with outperformance was quite striking. And finally, we looked at a practical application with XTAC, how they're using advanced analytics to decode institutional trading flows from public data, finding persistent patterns, and building strategies that seem to generate alpha across various time horizons from days to quarters. Right? I think what this whole deep dive really shows you is that even if Markets seem random on the surface. Underneath are often measurable patterns, inefficiencies waiting to be found. Whether it's spotting that subtle footprint of institutional trading like Xtech does, or understanding the information revealed by leverage and frictions in the options market, datadriven approaches are constantly finding new ways to understand and maybe even predict market behavior. It definitely highlights the ongoing evolution. So, here's something to think about. We saw how quickly the market seem to adapt into diminish the power of those traditional characteristic anomalies after 2003. And now we see constant innovation in data analysis and AI from firms like Extim edge. What is this ongoing race between information discovery and market efficiency tell you? How fast does today's alpha become tomorrow's standard market noise? That's the multi- trillion dollar question, isn't it? And maybe more personally for you listening, how might your own understanding of market signals need to evolve in a world where that edge, that alpha is constantly being discovered, competed for, and often eventually arbitrageed away. Something to definitely ponder.

Learn how to subscribe to macro data sources and forecasts: https://www.exponential-tech.ai/

Learn more about XTech Products: https://calendly.com/xtech-marketing-leads/brief-introduction-to-exponential?from=slack&month=2025-08