I don't have a fully developed framework for it. I just view trend, cross sectional momentum, carry etc as smart beta factors. If you are long SPY (buy & hold) you are harvesting the equity risk premium (simple beta). With stuff like Trend you want to enhance this simple approach (=smart beta).
Maybe it's not pure risk premia harvesting but an alternative shape of it.
It's just my line of thinking, I'm curious what you think about it.
I don’t fully buy this. “You need a why” sounds rigorous, but most trading whys are just educated guesses. Unless the mechanism is directly observable, it’s usually a post-hoc story. Data can be overfit, sure - but narratives can be overfit too. I’d rather trust a simple pattern that survives serious robustness checks than a beautiful explanation with weak stats.
You're framing this as a choice between a beautiful explanation with weak stats and a simple pattern with strong stats. That's a false choice. Ideally, what you're after is the 1-2 combo of a plausible why AND supportive data.
Reality is messy though. So the strength of your "why" and of your data exists on a spectrum. It's rarely as clean as you want it to be. You can't directly observe the thing that explains the edge. It exists at the margins, alongside thousands of other things happening in the market at the same time. You have to weigh the evidence, but mechanism sits higher in the hierarchy than statistics. The mechanism-data combo is asymmetric too: I've traded things on a sound mechanism with little data (some of my best trades have looked like this), but I will pass on solid stats with no realistic mechanism every time. The stats are the weaker evidence by far, and there's a good reason for that.
Financial data is horrendously noisy and there's never enough of it. A real edge will have the numbers on its side, but so will countless non-existent ones. Mine enough patterns and plenty will pass not only vanilla statistical tests, but "serious robustness checks" on luck alone. The robustness tests that combat multiple testing are similarly flawed (you can't realistically keep track of everything you tried, and you have even less chance of tracking the hypotheses that were tested by others but not published or talked about. But I digress...)
The point is, the why is how you tell the real edges from the rest. If you can't see the why, the fix is to upgrade your thinking, not reach for a robustness test.
On post-hoc stories, I think we agree. A "why" retrofitted onto a pattern to justify trading it is just a story. Ideally you form the hypothesis before you look at the data, but again reality is messier than that... sometimes you hear an idea from someone and go looking, or you notice something interesting in the data while investigating something else. That's fine. The market doesn't give out prizes for being the purest scientist. The question to ask is whether the why stands up on its own: would it make sense if the backtest didn't exist? Does it point at real market actors with real constraints, or with real objectives other than maximising their expected return? The good edges do, and they're inevitably simple and obvious (examples: leveraged token issuers document exactly how and when they rebalance; wealth managers rebalance on schedules and according to mandates that everyone knows). No mental gymnastics required... just a reasonable mental model of the market and the maturity to not bullshit yourself.
I realised as I was writing this that I often assert that a pattern without a mechanism is of little value without ever having shown why with any rigour. I tend to assume it's obvious, which is a bit arrogant now that I think about it. I'll write a piece on that if there's interest.
I think this is one of those discussions without one universally correct answer. I don’t dismiss the "why" at all - as a retail/systematic trader who can easily fool himself with data, I actually rely more on edges connected to well-established factors or frictions, especially momentum/trend, but also things like rebalancing, forced flow, liquidity shocks, etc.
A good "why" definitely improves confidence. It also helps understand the likely decay profile, capacity, and when the edge might stop working. So I agree it matters a lot.
Where I'm less convinced is that mechanism should always sit above statistics. To me, it depends on effective sample size, degrees of freedom, and portability.
If the sample is small, the setup has many parameters, and it only works on one market, then yes, I'd want a very strong mechanism. But if the edge has many reasonably independent bets, few degrees of freedom, survives costs/regimes, and is portable across assets or related implementations, then I need less of a perfect "why" to include it modestly in a diversified portfolio.
So I see it more as a tradeoff: a strong mechanism can compensate for limited data, and strong/portable data can compensate for an imperfect mechanism. Neither side is absolute proof.
If I've got a mechanism I'm not totally sold on, and the edge shows up strongly and persistently in the data, I might take a punt on it too... sized modestly, and heavily dependent on context. So I'll give you "strong data can compensate for an imperfect mechanism", within very tight limits! And in reality, I'd be revisiting the mechanism until I understood what I was missing. But I would definitely hold the line between “imperfect” and “absent”. Thinking that strong data compensates for the lack of a mechanism is dead wrong, and the weighting between the two is nowhere near symmetric: I'd trade a strong mechanism on zero data (I have, and some of those have been my best trades). I can't imagine trading a strong statistical effect with no mechanism at all, at any size, because I'd have no realistic way of knowing whether it was an edge or a mirage, and no way of noticing when it died. I don’t see any world where that’s a good bet.
So maybe one question gets to the bottom of it: would you trade a purely statistical effect, with no plausible "why" in sight? If the answer's no, we've been agreeing for the last three comments, and we're just haggling over the exchange rate between mechanism and data. Mine prices the "why" at a very hefty premium.
I would classify Trend etc as risk premia and not edge (inefficiency)
That's interesting! What risk does it represent in your framing? I haven't really thought about it in that way before.
I don't have a fully developed framework for it. I just view trend, cross sectional momentum, carry etc as smart beta factors. If you are long SPY (buy & hold) you are harvesting the equity risk premium (simple beta). With stuff like Trend you want to enhance this simple approach (=smart beta).
Maybe it's not pure risk premia harvesting but an alternative shape of it.
It's just my line of thinking, I'm curious what you think about it.
I don’t fully buy this. “You need a why” sounds rigorous, but most trading whys are just educated guesses. Unless the mechanism is directly observable, it’s usually a post-hoc story. Data can be overfit, sure - but narratives can be overfit too. I’d rather trust a simple pattern that survives serious robustness checks than a beautiful explanation with weak stats.
You're framing this as a choice between a beautiful explanation with weak stats and a simple pattern with strong stats. That's a false choice. Ideally, what you're after is the 1-2 combo of a plausible why AND supportive data.
Reality is messy though. So the strength of your "why" and of your data exists on a spectrum. It's rarely as clean as you want it to be. You can't directly observe the thing that explains the edge. It exists at the margins, alongside thousands of other things happening in the market at the same time. You have to weigh the evidence, but mechanism sits higher in the hierarchy than statistics. The mechanism-data combo is asymmetric too: I've traded things on a sound mechanism with little data (some of my best trades have looked like this), but I will pass on solid stats with no realistic mechanism every time. The stats are the weaker evidence by far, and there's a good reason for that.
Financial data is horrendously noisy and there's never enough of it. A real edge will have the numbers on its side, but so will countless non-existent ones. Mine enough patterns and plenty will pass not only vanilla statistical tests, but "serious robustness checks" on luck alone. The robustness tests that combat multiple testing are similarly flawed (you can't realistically keep track of everything you tried, and you have even less chance of tracking the hypotheses that were tested by others but not published or talked about. But I digress...)
The point is, the why is how you tell the real edges from the rest. If you can't see the why, the fix is to upgrade your thinking, not reach for a robustness test.
On post-hoc stories, I think we agree. A "why" retrofitted onto a pattern to justify trading it is just a story. Ideally you form the hypothesis before you look at the data, but again reality is messier than that... sometimes you hear an idea from someone and go looking, or you notice something interesting in the data while investigating something else. That's fine. The market doesn't give out prizes for being the purest scientist. The question to ask is whether the why stands up on its own: would it make sense if the backtest didn't exist? Does it point at real market actors with real constraints, or with real objectives other than maximising their expected return? The good edges do, and they're inevitably simple and obvious (examples: leveraged token issuers document exactly how and when they rebalance; wealth managers rebalance on schedules and according to mandates that everyone knows). No mental gymnastics required... just a reasonable mental model of the market and the maturity to not bullshit yourself.
I realised as I was writing this that I often assert that a pattern without a mechanism is of little value without ever having shown why with any rigour. I tend to assume it's obvious, which is a bit arrogant now that I think about it. I'll write a piece on that if there's interest.
I think this is one of those discussions without one universally correct answer. I don’t dismiss the "why" at all - as a retail/systematic trader who can easily fool himself with data, I actually rely more on edges connected to well-established factors or frictions, especially momentum/trend, but also things like rebalancing, forced flow, liquidity shocks, etc.
A good "why" definitely improves confidence. It also helps understand the likely decay profile, capacity, and when the edge might stop working. So I agree it matters a lot.
Where I'm less convinced is that mechanism should always sit above statistics. To me, it depends on effective sample size, degrees of freedom, and portability.
If the sample is small, the setup has many parameters, and it only works on one market, then yes, I'd want a very strong mechanism. But if the edge has many reasonably independent bets, few degrees of freedom, survives costs/regimes, and is portable across assets or related implementations, then I need less of a perfect "why" to include it modestly in a diversified portfolio.
So I see it more as a tradeoff: a strong mechanism can compensate for limited data, and strong/portable data can compensate for an imperfect mechanism. Neither side is absolute proof.
I think we agree more than I first thought…
If I've got a mechanism I'm not totally sold on, and the edge shows up strongly and persistently in the data, I might take a punt on it too... sized modestly, and heavily dependent on context. So I'll give you "strong data can compensate for an imperfect mechanism", within very tight limits! And in reality, I'd be revisiting the mechanism until I understood what I was missing. But I would definitely hold the line between “imperfect” and “absent”. Thinking that strong data compensates for the lack of a mechanism is dead wrong, and the weighting between the two is nowhere near symmetric: I'd trade a strong mechanism on zero data (I have, and some of those have been my best trades). I can't imagine trading a strong statistical effect with no mechanism at all, at any size, because I'd have no realistic way of knowing whether it was an edge or a mirage, and no way of noticing when it died. I don’t see any world where that’s a good bet.
So maybe one question gets to the bottom of it: would you trade a purely statistical effect, with no plausible "why" in sight? If the answer's no, we've been agreeing for the last three comments, and we're just haggling over the exchange rate between mechanism and data. Mine prices the "why" at a very hefty premium.