The human algorithm

In Michael Lewis’ latest book, “The Undoing Project,” he writes about a specific behavioral experiment performed on a set of first year residents and accomplished oncologists. In the experiment, the scientists asked the accomplished doctors to tell them how they make a decision regarding whether a patient has cancer from looking at an x-ray. The doctors all tended to give the scientists a 10 point checklist with a 1-10 rating for each of the 10 points, add up the points and you can accurately determine whether it’s cancer or benign. The scientists proceed to give a set of x-rays (the outcomes of which are known only to them) to the doctors and the residents, asking them to determine whether each is cancer or not. They also give the doctor’s checklist to the residents to use.

The oncologists who supplied the rubric in the first place show almost zero ability above random to accurately determine whether the x-ray was cancer or not. They didn’t follow their own rubric, suffered from an astounding amount of representative heuristic, and failed to do their job well. Meanwhile, the first year residents were able to score far higher accuracy rates on average and therefore would have been able to help their patients. They were simply acting as the human measurement component of an algorithm.  In short the residents followed the rules while those who created the rules didn’t follow them at all.

Leigh Drogen, CIO of Starkiller Capital writes a very elegant article in LinkedIn about how investment managers and traders pretty much fall into the same trap and his main point is that we would all be better performers if we acted more as a human algorithm rather than impulsive, inconsistent creatures that we are.

No argument from me.

But there is a very subtle but important difference on what it means to be a “human algorithm” depending on the type of strategy you trade. If you are trading an “insurance” strategy like I do in my trade room every day your single biggest risk is adverse selection. In the insurance model you make money almost all the time, but occasionally suffer a few large losses. You are trading for income rather than return and as with any insurance business your single most important function is to avoid loss as much as possible. So to be a truly effective “human algorithm” you need to control FOMO. The motto in my room is “Never Chase. Retrace.” This can be excruciatingly painful to practice in high momentum markets as you sit on your hands while others are making bank. But if you don’t master this one skill you will never succeed with this particular strategy.

On the other hand if you are trading the have-a-hunch-bet-a-bunch lottery model then the “human algorithm” must act in a completely opposite way. You have to seek out risk anywhere and anytime it appears because you never know when the trade could turn into a 10R return while enduring three, four, five, ten consecutive stop outs on false entries. You must in effect train your “human algorithm” to be like the young Kevin Bacon in Animal House as repeats “thank you sir may I have another” as upperclassmen mercilessly paddle his butt. That’s a pain of a different kind altogether.

So the real question isn’t that we should all strive to be a “human algorithm” in our trading, but rather that we should ask what kind of algorithm we want to be. The more honest we are with the answer, the better trader we will be.