Are there any guidelines for minimizing selection bias when deciding on which assets and algos to include in a strategy? There's the temptation, of course, to select only those with the best backtest, but that bias already starts to overfit the strategy. So what to do? What things should one look for when assessing whether to include an asset?
This is a good question with no really good answer.
The mere process of developing a system already causes selection bias to some degree. The obvious way to avoid further selection bias is trading all assets for which the system is theoretically suited, even when the backtest shows that they are unprofitable. But normally no one would have the nerves to do that. We don't either with the Z1 and Z2 systems.
There is also a mathematical algorithm to eliminate selection bias. It is described in the book by Aronson in the Zorro book list. We cannot implement this algorithm in Zorro because it's under license, but users could use it for their private trading.
I'm moving this topic to the automated trading forum for further discussion.