050 – Samir Varma - When Academic Finance Theory Fails
Where Real Edge in Quant Trading Actually Comes From
Do not watch this podcast. This is Part 1 with Samir Varma, and in Part 2 we go into great detail about his quantitative trading. In the Collective, he gives our members some specific instructions on how to measure risk differently – this stuff isn’t fluff. But in Part 1, I got derailed into quantum physics, determinism, AI, Asimov’s three laws of robotics and more. One of my favourite shows – but the first show I’ve done that isn’t about trading! It’s the warm-up you need to make the most of Part 2 though, and if I didn’t publish it, I’d be depriving a great many of you who will no doubt find this stuff as fascinating as myself! Still, if you only have time for strict ‘trading content’, fair warning, skip this. Let me know your thoughts…
Subscribers get a community, bonus content, strategy library, mentorship & more.
Samir’s Substack has some wonderful material, and today’s write up in inspired by his article “The Emperor Has No Alpha – What Data Mining Reveals About Academic Finance”. Forgive me for drawing heavily from it, but it’s great, and you should read it in full here:
In his article he references a paper from the Federal Reserve Board: “Does Peer-Reviewed Research Help Predict Stock Returns?”
The answer – “not really”. What a scientist like Samir reminds us of is that “elegant theories must predict, not just explain.”
In the Fed paper, 29,000 accounting ratios were brute-force data mined for statistical significance and compared to 212 peer-reviewed return predictors from academic studies. The result? Both approaches retained around 50% of their predictive power out of sample.
In physics, theory often leads the way. Scientists predict something should exist, then go looking for it. As Samir puts it, “Theory constrains the search space. It tells you where to look”.
Academic finance likes to imagine it works the same way.
A predictor is published. Then comes the explanatory framework. Maybe it works because investors are being compensated for risk. Maybe it works because of behavioral reasons. Maybe it reflects information asymmetry, attention effects, limits to arbitrage, or some other respectable academic mechanism. The theory is supposed to help us separate genuine phenomena from statistical rubbish.
But the Fed paper would suggest that theory does not seem to do that job very well.
That is the uncomfortable punchline. And for traders, it matters.
Categorizing by theoretical foundation: risk-based, mispricing-based, and agnostic, the paper found that the only category showing any real sign of outperformance is the agnostic research: the papers that refuse to take a theoretical stance.
That is a serious result. It suggests that, in markets, elegant explanations may be far less useful than people think. Worse, they may create a false sense of confidence. How many of them are created to sell a hedge fund?
Oh, and the other theory that did actually out-perform? Momentum!
Data Mining Is Not the Villain
In quant circles, “data mining” is often used as shorthand for bad research. Overfitting. Curve-fitting. Random pattern hunting. Statistical hallucination dressed up as alpha. Fair enough. Bad data mining exists. Plenty of it. Most of it ends badly.
But the deeper point is this: all empirical research is, at some level, data mining. The real issue is not whether you search the data. The issue is whether you search it with rigor.
If useful information exists in prices, accounting data, flows, or market structure, a computer can help uncover it. This shouldn’t a surprise. You do not need a grand theory before you are allowed to notice that a pattern is there. The market leaves fingerprints. Some are noise. Some are not. Your job is to figure out which is which.
In his paper Samir argues that data mining could have uncovered many of the major themes in asset pricing before they were formally published. Investment, accruals, external financing, earnings surprise—many of these broad areas could have been identified by searching the data first, with theory arriving later as explanation rather than discovery.
That should change how a trader thinks.
It means the academic paper is not always the origin of insight. Often it is the formalisation of something the data was already trying to say. The theory may tidy it up, give it a name, and provide a narrative respectable enough for publication. But the pattern itself may have been available long before the journal article arrived wearing a tie.
That is not an argument against academia. It is an argument against passivity.
If a trader waits for peer-reviewed permission to investigate an idea, he may already be late.
Academic Research Still Matters
Now, let us be clear. None of this means academic finance is useless. It’s incredibly helpful. Samir mentioned to me he has probably read 3 papers a day for many years.
The academic literature remains essential because it gives traders a map of what has been observed, what has been tested, and how others have tried to explain recurring effects. It gives you language, structure, categories, and a body of work to interrogate. It tells you what others think they have found. And often, that is enormously valuable.
Now, if only you had a tool to accumulate hundreds of great academic papers into a searchable library, summarise their trading potential, produce the strategy code even, and quickly test them before proceeding with deeper research? What if you could work collectively with others testing, sifting, letting the best ideas rise to the top? Funny, I had the same thought, and this is why I’m about to drop the Algo Vault for Collective annual subscribers. It’s exactly that.
Samir, like any good scientists reveals the crucial shift we need to make in our approach to these theories: once you understand what the academics are saying, your job is not merely to admire it. Your job is to figure out “why they’re wrong”.
That is where the trader’s work begins.
Not because academics are fools - often they are very smart. But because they are playing a different game. They are trying to explain patterns in a publishable framework. You are trying to make money going forward, after costs, after slippage, after crowding, after decay, and after your own psychology starts whispering that this time it is different.
Those are very different objectives.
Stand on the shoulders of giants, but you better believe you need to ‘get to work’ yourself.
An academic explanation may be directionally useful and still be wrong in the ways that matter most to a live trader. The risk framework may miss the actual driver. The behavioral story may be too neat. The statistical result may weaken after publication. The implementation assumptions may be unrealistic. Or the entire thing may only survive in sample because of specification choices no practitioner would ever use in live trading.
That is why understanding the literature is necessary but not sufficient. You must know the “what.” Then you must attack the “why.”
Why should this persist?
Why did it work?
Why did it stop working?
Why did the paper measure it this way?
Why would a live trader actually get paid for bearing this exposure?
Why would this effect survive after being published, copied, and crowded?
That line of questioning is where genuine edge begins.
Now, even by answering the ‘why’, we may not need an elegant theory, we just nee to have tested it till we threw up, so that the ‘why’ is ingrained so deeply in us that we are comfortable we can trade it. The Agnostic / ‘unexplainable’ theories can have a ‘why’ that lives precisely in the fact that they ‘don’t fit neatly’ with academic theories.
Samir quotes Peter Brown, co-CEO of Renaissance Technologies, said in Gregory Zuckerman’s The Man Who Solved the Market:
“If there were signals that made a lot of sense that were very strong, they would have long-ago been traded out. There are signals that you can’t understand, but they’re there, and they can be relatively strong.”
It’s a quote I’m happy to repeat. Samir adds “The patterns that persist are precisely the ones that resist explanation.”
PS: Don’t forget we interviewed Zuckerman way back when:
That is the point, isn’t it?
The signals that are easiest to explain are also the easiest to crowd. The ones that fit neatly into an academic model are the ones most vulnerable to being noticed, packaged, published, marketed, and arbitraged to death. By the time a signal becomes intellectually elegant, it may already be commercially extinct.
Meanwhile, perhaps the weird stuff survives longer precisely because it resists neat explanation. The pattern looks uncomfortable. The rationale is murky. The narrative is weak. Which means fewer people pile in with conviction, fewer funds leverage it aggressively, and fewer allocators convince themselves it is a durable gift from the heavens.
Ugly signals may live longer because they do not flatter the human need for coherence.
But hey, this is just my theory, you should test it. It probably doesn’t hold up.
Theory Does Not Trade for A Living. Process Does.
The real lesson here is not that theory is worthless. It is that theory is not a substitute for research discipline.
What actually seems to predict persistence is not theoretical elegance but statistical strength. The stronger the in-sample evidence, the better the odds that something real is there.
That is why the serious quant has to be maniacal about process.
In fact, one advantage of looking at older papers, is that there is more out of sample time for you to check them against. This is a common issue with the peer review process; it doesn’t generally do robustness testing out of sample.
You need to test ideas hard. Across time. Across markets. Across parameter ranges. Across regimes. Before publication and after publication if possible. Before costs and after costs. Under realistic assumptions, not fantasy ones. You need to know what breaks the idea, what strengthens it, what dilutes it, and what its true behavioural profile looks like when the equity curve gets ugly.
That is where conviction comes from.
Papers are full of great ideas, plausible explanations for phenomena are important, but conviction comes from doing the work yourself and understanding what the work is actually telling you.
This is especially important because many traders do not really understand the ideas they trade. They understand the headline. They understand the sales pitch. They may even understand the backtest summary. But they do not understand the mechanism well enough to survive the periods when the strategy inevitably disappoints.
That is fatal.
If you do not understand the strategy deeply, you will not trust it when you most need to. And if you do not trust it, you will override it, reduce it, abandon it, or morph it into something else at exactly the wrong moment.
Your Edge Must Fit You
The final filter is not just whether a signal is statistically valid. It is whether it is congruent with the person trading it.
Samir says “that durable competitive advantage needs to be 100% congruent with your own personality”.
That line is more important than most people realise.
A strategy is not just a return stream. It is an experience. It has a rhythm, a temperament, a pattern of pain. Some strategies win infrequently but explosively. Some grind steadily then suffer sharp reversals. Some require patience bordering on monasticism. Others demand speed, decisiveness, and emotional resilience in the face of constant noise.
If the strategy clashes with your nature, it will eventually beat you, even if the backtest is valid.
A risk-averse person may not be able to hold a sharp momentum system through its inevitable drawdowns. A slow, patient thinker may sabotage a short-term strategy by second-guessing constant signals. A trader who craves narrative certainty may abandon a robust but ugly model simply because he cannot explain it elegantly enough to himself.
That is why your edge cannot be borrowed. It cannot just be copied from a paper, a fund manager, or a clever person on the internet. Samir says it so well, “Your edge must be yours. Not borrowed. Not copied. Not theoretically optimal for some abstract investor. Yours”.
That does not mean inventing something from nothing. It means making it your own through understanding, research, and repeated testing until the logic and the pain profile are both something you can actually live with.
The Trader’s Real Job – Practical Take-Aways
I’ve said it before – Trading is a business. Writing academic papers is not. Academic insight remains essential because it tells you what has been seen and how people have framed it.
But your real job is to test it yourself, understand it deeply, and figure out where the academic story is incomplete, fragile, or simply wrong.
That is where edge lives.
Not in theory alone.
Not in data alone.
In the disciplined confrontation between the two.
And then, one level deeper still, in building something you can actually trade.
Because in the end, the best strategy in the world is useless if you do not understand it, do not trust it, or cannot execute it.
That is why the work matters. Test it. Break it. Rebuild it. Strip it back. Understand it until it is no longer just an idea from a paper, but a method you could hold through stress. Empirical evidence is your only friend. Here’s the only ‘process’ you really need: test until you throw up, then test some more.
One last quote from Samir. “The data mining result in this paper is liberating in this sense. It means you don’t need to genuflect before academic authority. You need to understand the academics, yes. But then you need to find what works for you, validated by data, congruent with your psychology.
That’s a much harder problem than reading papers. It’s also the only problem that matters.”
Research widely, always be skeptical, don’t assume complexity makes more money, mine the data vigorously, and, if it holds statistical significance out of sample, then devise the theory.
Get in Touch with Samir
And in his extended conversations with me he mentioned the groundbreaking work of his colleagues in quantum mechanics, you may want to check that out.
If you loved Asimov and want to read about how the three laws of robotics were developed after his death, check out Roger MacBride Allen, who wrote three novels (Caliban, Inferno, Utopia), approved by Asimov shortly before his death, exploring the creation of “No Law Robots” and “New Law Robots” — a direct interrogation of what happens when the Three Laws framework is dismantled or reformed.
Samir recommended this book as an excellent introduction into quantum physics:
Waves on an Impossible Sea by Matt Strassler