How Hard is It to Develop a Viable Algorithmic Trading System? (Part 1)

The title of this blog post poses an interesting question that is very difficult to answer.

In his book Building Winning Algorithmic Trading Systems (2014), Kevin Davey claims it takes 100-200 trading ideas to yield one tradable system. In my view, this is a daunting statement because from what I’ve read on trading system development and from the challenges I have faced with coding, this could be a lifetime! The claim is a bit vague, though. What exactly constitutes a “trading idea?” If each parameter value counts as a different trading idea then it might take only a few trading strategies to come up with a “tradable system.” This phrase is vague too. Does “tradable system” mean a system that would be acceptable to me given my particular requirements for profit and my subjective function?

Last year I saw a forum post that got me thinking on this topic. Somebody wrote:

> Any strategy which is publicly available cannot by definition be
> profitable, since nobody would play against it and be the loser
> to make it profitable.

I disagree with his conclusion and I especially think “by definition” is too strong.

However, this does make me wonder what the institutions can do with publicly available strategies. I believe these trading strategies are low-hanging fruit for institutional quants who have tons of computing power and lots of experience doing this work. Being in the public domain, these “big boys” can find the ideas easily, run them through their system development mills, and quickly determine whether Edge exists. If so then they will trade these systems. Edge will run out when enough money is deployed to trade these systems, which will likely happen when enough institutions get on board. For this reason, once in the public domain they probably won’t stay profitable for long.

I will continue this discussion in my next post.