Posts Tagged genetic

Never one for rhetoric; I presented a similar case, to those I have described in the prior posts, at a talk I did last year. Using a similar model we followed how that can evolve from a base set of criteria and environments to something more complex. The following two snap shots are from a control where we work at random and an evolved model a few generations on.

Here we are using £2 at random

Here we use an evolved model

I reckon I could replicate the performance of the two at an pre-determind date. But some of the performance will always be reliant on prevailing market conditions, but it’s the gap in performance that is the most interesting. The second benefits from picking the ‘right’ market, direction and exit models. It was only by studying why one performed better than the other that I was able to fully understand the reason and why that was important. You can then use than to hone other strategies.

Of course there an infinite amount of variables, so you need to point it in roughly the right direction. If you don’t know the direction, then do a binary chop to get there quicker. You also have the issue of course that the market can and will change. We recently saw the introduction of cross matching across many sports and that has changed the ‘environment’ to the point where I am sure some strategies have started to fail. This in turn will alter the behaviour of those people who are trying to avoid becoming dinosaurs. Sometimes however accidents in evolution produce strange results. On more than one occasion I’ve made an error on a spreadsheet and it has failed to do something I wanted. That in turn though created some unusual payoffs which lead me down paths I hadn’t considered exploring. Just another path on the road of evolution.

So while using a genetic algorithm is a useful trick, evolving doesn’t stop, it continues ad infinitum. If you stop evolving you  may appear in the betting exchange equivalent of the natural history museum in a few years time.

So how did my view on evolution, natural selection and Darwin translate into market strategy?

There are many ways to apply it, but first you need to start. Your first strategy is to have no strategy, just poke around at random. This gives you a control sample to work with, a base metric from which you can build your activity. You should regularly revisit your control to ensure the base you are working from is still the same as when you started.

Next you need to decide the environmental factors. In the case of the market, when do you open a position, when do you close and what do you use to judge that? Could it be the market type, the price, the time? You could have a number of factors that tell you if a market is ‘right’ or is adequate for your participation, these define the environment in which you will work.

Next you need to decide exactly what your entry and exit points are. Start simple and build out from there. It will be fairly obvious, fairly quickly where things are not working and you can ‘kill’ off these experiments. After each life cycle of experiments carry forward the successful ones and add in some new variations for the next generation. It’s surprising how quickly you ended up with a much better understanding and a much better strategy. It’s not just the entry and exit that are variables the environment is also, don’t forget that.

It would be nice to say ‘That’s it’s’ but that is rarely the case. You have to remember that the market itself evolves and people in the market occupy certain niches. You need to find a niche but also compete with people who may already be present in that niche. Your activity in the market will also shape the environment, so that in itself produces a response from the environment. One failure of market participants is to fail to evole a strategy, every strategy has its moment but many people find, often by accident, something that works; from that point it starts slipping away. The general reason is that your activity has probably forced others to evole and so the whole things rolls on again but also a lot of participant never work out why what they did works. I am always restless for understanding.

More reading: -

http://en.wikipedia.org/wiki/Genetic_algorithms