Altucher versus the Rest
James Altucher’s book, Trade Like a Hedge Fund, was recognized by Barron’s a couple of years ago as the best investment book. There are many strong features of this work. I like it and enjoyed reading it. I will look at some of the specific strategies later and offer some comments. In the (unlikely) event that Altucher ever reads these, I am sure that he will accept the commentary in the constructive spirit in which it is offered. After all, that was one of his stated reasons for writing the book!
Here are some of the things that I like about the work. First, he uses Wealth Lab, the excellent development tool we use. When I read his comments, they remind me of my partner, Vince, with whom I have worked for nearly fifteen years. Vince is a career Navy scientist who has recently retired. He helped to bring the best technology ideas to the service of our country, leading an interdisciplinary team. While he was a civilian, his work was highly classified. When Vince and I joined up we made a deal with the Navy. They would not do investment resesarch and I would not build any ships. Vince is a guru on Wealth Lab. He was doing forecasting using neural networks and genetic algorithms for years before they became popular. He bought the source code from leading developers and did his own adaptations.
Second, Altucher recognizes the risk of "curve-fitting" as he calls it. Others may say backfitting, post-dicting, or data mining. These are all traps for getting Fooled by Randomness. Altucher does two things to fight this problem:
- He uses very few variables. This leads to more robust models.
- He tests the results over a range of values. This makes sure that he has not picked the single value or filter that makes the system work.
Vince does both of these things as well as some others, but that is a subject for the future.
My point here is to compare the very good work of Altucher with most of the Street research we see. Here is how most researchers approach the problem.
First, they use ALL of the data. Why? Because, like Everest, it is there. Why not? There see no need to consider whether the information is relevant to the problem.
Second, they begin with the data, not the theory. This is a major flaw. Altucher has a theory in mind at the start of his analysis.
Finally, most researchers, partly because they are not working from a theory, do not worry about whether they have specified their models correctly. What has been left out? Does it all make sense?
Altucher scores well on all of these points.