If you’ve been a subscriber to Edge Alchemy for any length of time, you know that mechanism-first thinking is a fundamental requirement of running a serious systematic trading operation.
But what is “mechanism-first” thinking?
Simply, it’s putting the reason that a trade should pay you front and centre. It’s elevating the edge’s reason for existing above the research and definitely above the backtest.
It’s asking the question “Why would the market pay me, on average, to do this trade?” And treating that question like the first-class citizen it is.
Why is the mechanism important?
Imagine you’re doing what every beginner trader with some level of technical acumen does - you’re backtesting trading rules and using the shape of the equity curve as feedback as to whether the rules “work.” When it doesn’t work the way you wanted, you tweak some parameters, add a stop loss, maybe a profit target, try it on some different markets, throw in some machine learning, add a time filter… until the equity curve says it works.
Then you put some real money behind it. Maybe it works out for a while, but eventually you find yourself in a drawdown, and you start wondering if the strategy is “broken.” The only feedback you have is the strategy’s P&L, so you go and bootstrap some backtested equity curves to figure out if this drawdown is statistically unusual (which would be a reasonable thing to do if you actually knew the distribution of the process generating the returns… but sadly you can’t know this, and it’s changing underneath you anyway).
As the drawdown continues, you start to wonder if the strategy needs some more work. So you go and change some parameters and add some filters that would have sidestepped the current drawdown.
But while all this is going on, your anxiety doesn’t go away. Because deep down, you know that the P&L is a lousy indicator of whether a strategy is actually based on something real. After all, why would you get paid for optimising parameters?
Backtesting is very good at a handful of things.
It’s the best tool I know for understanding the real-world costs and frictions of trading a given set of rules. It’s also an excellent tool for understanding implementation trade-offs - things like balancing trading costs with fidelity to a set of target weights.
Used carelessly, it’s also really, really good at identifying false positives - sets of trading rules that happened to produce a positive equity curve over some set of historical price data.
One of those things is not only useless, but also quite dangerous. It leads to the anxiety-inducing, life-choices-questioning situation described above. It’s what happens when you confuse backtesting with research.
Good research is driven by mechanism. Why would doing a thing in the markets pay me? Who’s willing to trade at bad prices for them? Why would they do such a thing?
Real mechanisms - at least, the ones that we solo operators have a chance of harnessing - are nearly always grounded in someone making a deliberate choice to do a trade for purposes other than maximising short-term trading returns:
Rebalancing to a mandate
De-risking because the boss told you to
Paying to insure against an unacceptable risk
Selling something for less than its expected value because it’s risky
Paying more than something’s worth because its payoff profile is like a lottery ticket
The people doing these things are happy to do them. They don’t rely on someone doing something inherently stupid. Of course, people do stupid things in the market too, but the competition to trade with those people is usually too fierce for solo operators like me to stand a chance. So I don’t play those games.
So I start with the mechanism. A simple, logical reason that I can get paid to do a thing in the market. Armed with the mechanism, I then go looking in the data. Not to prove myself right, but to prove myself wrong as quickly as possible.
It’s much easier to prove yourself wrong than it is to prove yourself right with this stuff. You can prove yourself wrong using simple data analysis experiments based on free or cheap data with known imperfections. You can move fast. And if you can prove yourself wrong fast, you get to the good stuff sooner.
If I fail to prove myself wrong, I can continue with the simple data analysis until I really understand the mechanism as well as I possibly can. Eventually, I understand enough to decide to put some money behind it. And that’s where I pick up a backtest.
And I don’t need to torture it to find the strategy rules. They drop out of the research quite naturally, and at this point, they should be obvious. The backtest tells me what it would have cost in the past to harness the effect using this particular set of rules, and it helps me understand important implementation considerations, like “what’s the least amount of trading I can do to actually harness this effect?”
And then when I actually get it in the market with some real money behind it, those inevitable drawdowns don’t have the same effect. Of course, no one likes a drawdown. But if I understand the reason the trade should pay, that it’s based on something real, I can accept those drawdowns as a cost of doing business and manage them accordingly. They’re no longer the cause of anxiety that comes with trading something you don’t understand. If I’m really concerned, I can go and re-run the research and try to figure out if something in the mechanism has legitimately changed. I’m not beholden to the realised P&L of the thing as my only feedback.
Of course, nothing is as neat and tidy as I made it sound here. I made the research sound like a linear path from idea to research to backtest to trade, but in reality it’s much more iterative and full of tangents and dead ends. But the general principle holds - mechanism-first thinking is the basis for a serious systematic trading operation.
On the 26th of September, I’m running a workshop with my friend Erik from Outlier Trading on building trading strategies with mechanism-first thinking. If you’d like to come along, you can reserve your spot here. If you can’t make it, it’s worth registering anyway, as you’ll get the replays.
Starting with the mechanism before the backtest makes sense. It gives you something to check when the equity curve starts looking too good.
Mechanism-first also changes how failure should be monitored. Once the rationale and implementation are frozen, a drawdown becomes evidence to interpret against the mechanism, not an invitation to rewrite rules around recent pain. That separation between research, implementation and later custody is what keeps a plausible idea from becoming an endlessly retuned backtest.