“When I look at any potential trade, the first question I ask is: who’s losing money on the other side of this, and why will they keep doing it?
That’s the game. Edge comes from structural constraints, stuff other participants can’t or won’t do because of mandate restrictions, capacity limits, or operational awkwardness. If you can’t answer that question, you don’t have an edge.”
Incredible stuff and truly eye-opening for an aspiring trader. Thank you for this article.
Great piece, Kris! I’m convinced that anchoring LLMs to curated knowledge bases—such as peer-reviewed papers and books—is the key to establishing a 'positive bias.' Standard self-supervised training often struggles with the sheer volume of low-quality data on the open web; prioritizing verified literature provides the necessary grounding that raw internet data simply lacks.
Thanks! I think you’re on to something. Curated knowledge bases help. But the deeper issue, as you suggest, is that the LLM has no way to distinguish signal from noise in its training data. And with trading specifically, the signal-to-noise ratio is terrible compared to, say, coding where Stack Overflow answers get ruthlessly corrected and open source code gets tested by thousands of people. There's no equivalent correction mechanism for trading content, so the LLM speaks with the same confidence about both. That uniform confidence over wildly uneven quality is what makes it really dangerous.
I actually tested this exact idea recently. Built a vector store of embeddings from my own course material (which deliberately avoids the conventional wisdom traps) and asked a third-party LLM to only answer using that corpus. Even with that constraint, it would confidently recommend things like “validation via an out-of-sample backtest” as if that sort of nonsense is standard good practice. The model's priors are so strong that the conventional wisdom comes through even when you anchor it to better material. It may be solvable by a better embeddings model and/or better prompting, or by an LLM trained specifically on the curated corpus (rather than an existing model using it in a RAG setup).
That experiment is strong and a killer point about the model priors—it’s wild how aggressively the "conventional wisdom" overrides even a curated RAG setup.
Since the truly clean models will probably stay proprietary and behind institutional walls anyway, it really just reinforces your usual philosophy: the indie "sweet spot" is still finding the niche strategies institutions can’t touch. While LLMs can definitely speed up the mechanical research and strategy engineering, it lands us right back at your core thesis—the actual signal stays human-led. Fair point.😀
Curious how you actually identify “who is losing and why”? Edge for me is strictly quantified through statistical positive expectancy with a measurable system: if it works consistently, why/how can we verifiably put a ‘reason’ on it?
Thanks, and great question. It gets at something I think is really underappreciated.
You mention quantifying edge through statistical positive expectancy. I'd push back a little on that. Or rather, provide a bit more nuance. Yes, a real edge will have the stats on its side. But so will countless non-existent ones, simply because of the low signal-to-noise ratio in financial data and the finite samples we're working with. The stats alone can't tell you which is which. You need the 1-2 combo of a plausible "why" and supportive data.
Neither is sufficient on its own (usually).
So really there are two approaches, and only one of them is something you'd have any confidence building a serious trading operation around:
Approach A (pattern-first): Scan data for patterns with positive expectancy, backtest them, and if the stats look good, trade them. The "why" is optional or reverse-engineered after the fact.
Approach B (mechanism-first): Start with a hypothesis about who is trading at prices that don't reflect expected returns, and why they'll keep doing so. Then look for evidence in the data.
Both can produce systems with positive expectancy. But Approach B is FAR more likely to work out for a few reasons (not necessarily in order of importance):
1. It's a much better filter. Most patterns in financial data are noise. If you start from "I think X is happening because of Y structural reason," you massively reduce the space you're searching. Fewer false positives.
2. It moves you forward. Every time you hypothesise something and test it, you learn something, whether you end up with a trade or not. You often end up with your next lead and your next hypothesis. You don't get that feedback loop if you're just data mining for favourable statistics. There's nothing to link it back to market reality. You never develop the intuition that we call "trader smarts."
3. It tells you when the edge is gone. If you know the mechanism (say, fund managers forced to rebalance at month-end), you can monitor whether that mechanism still exists. If you only know "this pattern had a positive expectancy from 2005-2023," you have no way to distinguish a normal drawdown from an edge that no longer exists.
4. It keeps you in the trade. If you've landed on something without a "why," you'll never really trust it. Deep down you know that price patterns aren't predictive of forward returns (why would they be?), and that will cause you endless anxiety when you try to trade something you don't understand.
I actually have a concrete example. I traded a bond seasonality strategy for years, but the mechanism I had (window dressing) always felt a bit hand-wavy. When it had a rough patch in 2024, I pulled it. It came roaring back in 2025, about 15% at Sharpe 1.8 after costs. I left most of that on the table because I didn't trust the "why." A Pro member who's an ex fixed-income PM recently gave us a much stronger structural explanation, and I now have far more confidence in the trade.
So to answer "how do you identify who is losing and why":
You think about the participants in the market you're trading. What are they forced to do? What are their incentives? What constraints do they face? (This is what I mean by "trader smarts"). Index rebalancing, regulatory requirements, benchmark tracking, liquidity provision, hedging obligations. These create flows that aren't driven by expected returns, and that's where the opportunity is.
This is actually the core of what we teach in Bootcamp - https://robotwealth.com/trade-like-a-quant-bootcamp/ - building that mechanism-first research process from scratch. If you want the full framework, that's the place to get it all in one place.
It's not that the stats don't matter. They do, a bit, I guess. But the mechanism is what gives you confidence in the stats and tells you what to do when things get rough. I'd 100% trade something based on a clear mechanism that I didn't have enough data to calculate good stats on (have done this a number of times). I'd be much more reluctant to trade something whose stats looked great but whose mechanism didn't make sense (have passed on these in real life too). The point is, there's a hierarchy to all this talk of evidence. And stats definitely ain't at the top.
I've been thinking about writing a proper cornerstone piece on this approach. Your question is a good nudge to actually do it.
Really appreciate you taking the time to write that out, one of the more thoughtful explanations I’ve read on this. You obviously have a lot of experience with deploying and executing this stuff and, I’m still relatively “fresh” in my trading… so I’m coming at this more from a place of trying to expand my knowledge and how I’m currently thinking about markets.
I started very much in the “mechanism-first” camp, trying to understand what’s happening, who’s doing what, and why. But over time (especially coming from an economics/stats background), I’ve drifted more toward thinking about trading as a problem of designing and validating a distribution, rather than explaining the underlying cause.
A few thoughts on your points:
1. On patterns and noise
I completely agree that most patterns are noise. But for me, patterns aren’t really there to predict the market, they’re just a tool to create something repeatable and measurable. A a pattern/structure is essentially just a way of defining entries/exits so I can observe and measure a distribution of outcomes in a consistent manner. The “edge” then comes from how that distribution behaves (and how it’s managed), rather than the pattern itself having inherent predictive power (I see your point why most content out there is garbage because it says these patterns are the way forward alone).
2. On learning and feedback loops
I agree that pure data mining without context is dangerous. But I’d push back slightly in that I think the most reliable learning comes from statistically significant samples. Otherwise it’s very easy to observe a handful of events and construct a compelling story around them. For me, the feedback loop is less “did my hypothesis about participants hold?” and more “does this event/structure produce stable, significant results across samples and conditions when paired with risk management (ones that are measurably better than a random outcome?”
3. On knowing when edge is gone
This is where I’ve leaned heavily into modelling. If I have a system with known return characteristics (mean, variance, drawdown profile, pay-off, win rate, etc), I can simulate expected paths (bootstrapping / Monte Carlo) and compare live performance to that distribution. If I’m significantly outside expected bounds, I can start questioning whether it’s edge decay or execution. So in that sense, I feel like you can diagnose edge deterioration without necessarily knowing the underlying mechanism.
4. On confidence and staying in trades
Completely agree, without having a way to keep believing in a system when it gets tough, nothing else really matters. But, I get confidence from the distribution itself. If my modelling tells me there’s, say, a 95–99% probability of long-term profitability given my current stats and execution, that’s what keeps me aligned during drawdowns. Additionally, I can see there’s a drawdown range that is expected (normal) and there’s a range that’s not (linking back to measuring edge decay), so when I experience that normal drawdown it’s simply a function of the distribution I made - nothing to “fix”.
So I guess where I land (at least currently) is:
Edge = the ability to consistently produce a positive distribution of returns
Mechanisms = potentially useful, but not strictly necessary if the distribution is robust
In the end, I can’t argue that it would be better if I knew why things were happening… but I think it will come with more time and experience (I think it’s much harder to learn why something is happening and executing it as opposed to creating a robust distribution).
Really appreciate the reply again, this is exactly the kind of discussion that helps me learn and think more, will keep reading!
Kris I enjoyed your article and most importantly it drew on real experience.
I find myself getting all hot under the collar when I hear people say they plugged Claude into X and it is running for 24hrs trading. Or some variation of that clickbait.
A month ago I launched backtesting and trading platform that I see as a framework on top of all these functions. After 25yrs of active trading on every side of the fence I wanted to build something that was bias aware. It's an area of speciality for me and no matter how smart people are they continue to make the same mistakes.
The platform a few weeks ago was not all about AI although it incorporate it.
I have just started a series of posts dealing with exactly what you are arguing against.
I suspect based on my experience and what I am learning that you are probably partially right.
All the parts that you speak about in terms of gaining knowledge about the edge or the experiments that didn't work and taught you something I resonate with. I am also old school in many of my core beliefs but I think there are clear opportunities where Agentic AI can play a bigger role. I am working my way through this understanding as we speak.
“When I look at any potential trade, the first question I ask is: who’s losing money on the other side of this, and why will they keep doing it?
That’s the game. Edge comes from structural constraints, stuff other participants can’t or won’t do because of mandate restrictions, capacity limits, or operational awkwardness. If you can’t answer that question, you don’t have an edge.”
Incredible stuff and truly eye-opening for an aspiring trader. Thank you for this article.
You’re very welcome! Glad it was useful.
Great piece, Kris! I’m convinced that anchoring LLMs to curated knowledge bases—such as peer-reviewed papers and books—is the key to establishing a 'positive bias.' Standard self-supervised training often struggles with the sheer volume of low-quality data on the open web; prioritizing verified literature provides the necessary grounding that raw internet data simply lacks.
Thanks! I think you’re on to something. Curated knowledge bases help. But the deeper issue, as you suggest, is that the LLM has no way to distinguish signal from noise in its training data. And with trading specifically, the signal-to-noise ratio is terrible compared to, say, coding where Stack Overflow answers get ruthlessly corrected and open source code gets tested by thousands of people. There's no equivalent correction mechanism for trading content, so the LLM speaks with the same confidence about both. That uniform confidence over wildly uneven quality is what makes it really dangerous.
I actually tested this exact idea recently. Built a vector store of embeddings from my own course material (which deliberately avoids the conventional wisdom traps) and asked a third-party LLM to only answer using that corpus. Even with that constraint, it would confidently recommend things like “validation via an out-of-sample backtest” as if that sort of nonsense is standard good practice. The model's priors are so strong that the conventional wisdom comes through even when you anchor it to better material. It may be solvable by a better embeddings model and/or better prompting, or by an LLM trained specifically on the curated corpus (rather than an existing model using it in a RAG setup).
That experiment is strong and a killer point about the model priors—it’s wild how aggressively the "conventional wisdom" overrides even a curated RAG setup.
Since the truly clean models will probably stay proprietary and behind institutional walls anyway, it really just reinforces your usual philosophy: the indie "sweet spot" is still finding the niche strategies institutions can’t touch. While LLMs can definitely speed up the mechanical research and strategy engineering, it lands us right back at your core thesis—the actual signal stays human-led. Fair point.😀
Hi Kris, great read, thanks!
Curious how you actually identify “who is losing and why”? Edge for me is strictly quantified through statistical positive expectancy with a measurable system: if it works consistently, why/how can we verifiably put a ‘reason’ on it?
Thanks, and great question. It gets at something I think is really underappreciated.
You mention quantifying edge through statistical positive expectancy. I'd push back a little on that. Or rather, provide a bit more nuance. Yes, a real edge will have the stats on its side. But so will countless non-existent ones, simply because of the low signal-to-noise ratio in financial data and the finite samples we're working with. The stats alone can't tell you which is which. You need the 1-2 combo of a plausible "why" and supportive data.
Neither is sufficient on its own (usually).
So really there are two approaches, and only one of them is something you'd have any confidence building a serious trading operation around:
Approach A (pattern-first): Scan data for patterns with positive expectancy, backtest them, and if the stats look good, trade them. The "why" is optional or reverse-engineered after the fact.
Approach B (mechanism-first): Start with a hypothesis about who is trading at prices that don't reflect expected returns, and why they'll keep doing so. Then look for evidence in the data.
Both can produce systems with positive expectancy. But Approach B is FAR more likely to work out for a few reasons (not necessarily in order of importance):
1. It's a much better filter. Most patterns in financial data are noise. If you start from "I think X is happening because of Y structural reason," you massively reduce the space you're searching. Fewer false positives.
2. It moves you forward. Every time you hypothesise something and test it, you learn something, whether you end up with a trade or not. You often end up with your next lead and your next hypothesis. You don't get that feedback loop if you're just data mining for favourable statistics. There's nothing to link it back to market reality. You never develop the intuition that we call "trader smarts."
3. It tells you when the edge is gone. If you know the mechanism (say, fund managers forced to rebalance at month-end), you can monitor whether that mechanism still exists. If you only know "this pattern had a positive expectancy from 2005-2023," you have no way to distinguish a normal drawdown from an edge that no longer exists.
4. It keeps you in the trade. If you've landed on something without a "why," you'll never really trust it. Deep down you know that price patterns aren't predictive of forward returns (why would they be?), and that will cause you endless anxiety when you try to trade something you don't understand.
I actually have a concrete example. I traded a bond seasonality strategy for years, but the mechanism I had (window dressing) always felt a bit hand-wavy. When it had a rough patch in 2024, I pulled it. It came roaring back in 2025, about 15% at Sharpe 1.8 after costs. I left most of that on the table because I didn't trust the "why." A Pro member who's an ex fixed-income PM recently gave us a much stronger structural explanation, and I now have far more confidence in the trade.
So to answer "how do you identify who is losing and why":
You think about the participants in the market you're trading. What are they forced to do? What are their incentives? What constraints do they face? (This is what I mean by "trader smarts"). Index rebalancing, regulatory requirements, benchmark tracking, liquidity provision, hedging obligations. These create flows that aren't driven by expected returns, and that's where the opportunity is.
This is actually the core of what we teach in Bootcamp - https://robotwealth.com/trade-like-a-quant-bootcamp/ - building that mechanism-first research process from scratch. If you want the full framework, that's the place to get it all in one place.
It's not that the stats don't matter. They do, a bit, I guess. But the mechanism is what gives you confidence in the stats and tells you what to do when things get rough. I'd 100% trade something based on a clear mechanism that I didn't have enough data to calculate good stats on (have done this a number of times). I'd be much more reluctant to trade something whose stats looked great but whose mechanism didn't make sense (have passed on these in real life too). The point is, there's a hierarchy to all this talk of evidence. And stats definitely ain't at the top.
I've been thinking about writing a proper cornerstone piece on this approach. Your question is a good nudge to actually do it.
Really appreciate you taking the time to write that out, one of the more thoughtful explanations I’ve read on this. You obviously have a lot of experience with deploying and executing this stuff and, I’m still relatively “fresh” in my trading… so I’m coming at this more from a place of trying to expand my knowledge and how I’m currently thinking about markets.
I started very much in the “mechanism-first” camp, trying to understand what’s happening, who’s doing what, and why. But over time (especially coming from an economics/stats background), I’ve drifted more toward thinking about trading as a problem of designing and validating a distribution, rather than explaining the underlying cause.
A few thoughts on your points:
1. On patterns and noise
I completely agree that most patterns are noise. But for me, patterns aren’t really there to predict the market, they’re just a tool to create something repeatable and measurable. A a pattern/structure is essentially just a way of defining entries/exits so I can observe and measure a distribution of outcomes in a consistent manner. The “edge” then comes from how that distribution behaves (and how it’s managed), rather than the pattern itself having inherent predictive power (I see your point why most content out there is garbage because it says these patterns are the way forward alone).
2. On learning and feedback loops
I agree that pure data mining without context is dangerous. But I’d push back slightly in that I think the most reliable learning comes from statistically significant samples. Otherwise it’s very easy to observe a handful of events and construct a compelling story around them. For me, the feedback loop is less “did my hypothesis about participants hold?” and more “does this event/structure produce stable, significant results across samples and conditions when paired with risk management (ones that are measurably better than a random outcome?”
3. On knowing when edge is gone
This is where I’ve leaned heavily into modelling. If I have a system with known return characteristics (mean, variance, drawdown profile, pay-off, win rate, etc), I can simulate expected paths (bootstrapping / Monte Carlo) and compare live performance to that distribution. If I’m significantly outside expected bounds, I can start questioning whether it’s edge decay or execution. So in that sense, I feel like you can diagnose edge deterioration without necessarily knowing the underlying mechanism.
4. On confidence and staying in trades
Completely agree, without having a way to keep believing in a system when it gets tough, nothing else really matters. But, I get confidence from the distribution itself. If my modelling tells me there’s, say, a 95–99% probability of long-term profitability given my current stats and execution, that’s what keeps me aligned during drawdowns. Additionally, I can see there’s a drawdown range that is expected (normal) and there’s a range that’s not (linking back to measuring edge decay), so when I experience that normal drawdown it’s simply a function of the distribution I made - nothing to “fix”.
So I guess where I land (at least currently) is:
Edge = the ability to consistently produce a positive distribution of returns
Mechanisms = potentially useful, but not strictly necessary if the distribution is robust
In the end, I can’t argue that it would be better if I knew why things were happening… but I think it will come with more time and experience (I think it’s much harder to learn why something is happening and executing it as opposed to creating a robust distribution).
Really appreciate the reply again, this is exactly the kind of discussion that helps me learn and think more, will keep reading!
Kris I enjoyed your article and most importantly it drew on real experience.
I find myself getting all hot under the collar when I hear people say they plugged Claude into X and it is running for 24hrs trading. Or some variation of that clickbait.
A month ago I launched backtesting and trading platform that I see as a framework on top of all these functions. After 25yrs of active trading on every side of the fence I wanted to build something that was bias aware. It's an area of speciality for me and no matter how smart people are they continue to make the same mistakes.
The platform a few weeks ago was not all about AI although it incorporate it.
I have just started a series of posts dealing with exactly what you are arguing against.
I suspect based on my experience and what I am learning that you are probably partially right.
All the parts that you speak about in terms of gaining knowledge about the edge or the experiments that didn't work and taught you something I resonate with. I am also old school in many of my core beliefs but I think there are clear opportunities where Agentic AI can play a bigger role. I am working my way through this understanding as we speak.
Great article!