Tuesday, 13 May 2014

Algos: Data in One End, Money Out the Other?

Kiki and I are working on the next set of flobots as we continue our processs of greater automated diversification.

As part of my due diligence, I had a look around the web for what is going on in the world of algos. Quite a shock. There are many websites offering algos for sale. Some offer some form of "training" to go with them but many are just "Algo For Sale" with no trading logic disclosed. This is quite a frightening prospect for me because I am at a loss as to how people can make money with a dark locked down black box.

The issue is that an algo is NOT "data in one end and money out the other". I wish it was.

There are several issue for me:
  1. Psychologically, its very difficult to leave a closed black box turned on after a couple of losing trades
  2. An algo is not suited to all markets all the time. 
  3. The inputs of an algo will need to be tuned as markets change and evolve.
The issue #1 above is more than just a psychological one I guess. Without knowing why a trade is being entered, its difficult to see when issues #2 and #3 have come into play.

Markets change in both volatility and character. For example, our index markets have changed as the Fed and the ECB became more active in non standard operations due to the dangers to the world economies. There are other changes due to the lower volumes we have been experiencing in recent years. All this means is that if you had bought an algo that was designed for the markets as they were before the above, that algo is unlikely to be profitable now. Take that further, if I was to buy an algo designed for the current markets, is it likely to be profitable as the Fed stops its QE operations and th ECB invokes new and as yet undisclosed non standard operations?

My typical tweaks to algos have been to change bar size or bar type, and to change an indicator input to smooth out chop.

How do you know when an algo needs a tweak? You don't need to "know". Your testing has shown you how often you need to test your algo in the same way as you did when you created it. Walk Forward Analysis tells you how often you need to retune.

The thrust of this post is to say that one should learn how to fish rather than buying fish, fish which may already have gone off or have a very limited shelf life.

Lastly, the learning curve for creating an algo can be much, much shorter than learning how to trade consistently profitably. And you should know on the balance of probabilities that you will be successful BEFORE you risk your money.

No comments:

Post a Comment