In previous instalments of this series, I’ve provided some information on why you should build your own trading system, with the last part providing a simple trading system.

The big question for many traders is “How do I get a system that gives me an edge?”.  Following the well worn approaches of moving averages and other lagging indicators will not get you there.  You can be sure that the world’s best traders are not going over the same old ground.

I have mentioned some advanced tools and algorithms: what are they, what do you use them for and how can you easily access them?

The latest tools draw from artificial intelligence and non-linear mathematics.  You don’t need to have a degree in maths or computer science to use these.  However you do need to understand how they can help you.

I’ve already discussed the benefits of Loess smoothing in a previous article.  This is a way of getting an edge over lagging moving averages.

The first new type of tool is a genetic algorithm.  These tools use a simulation of Darwinian evolution, by generating random rules, testing their performance, retaining the best rules (survival of the fittest), then bringing these forward to the next generation and combining characteristics to continually “evolve” better rules.  The system randomly provides some mutations to see if this can lead to a better rule.

In summary, genetic algorithms generate rules from data by rapidly trying out different combinations - they are a search tool.  They are similar to Excel’s solver, but far more powerful.  Genetic algorithms don’t do anything that trying every single possible combination of inputs doesn’t do, but they are far quicker, especially as the complexity of the data increases.  Simulated annealing is a similar process.

My favourite accessible genetic algorithm tool is Pi Blue’s OptWorks for Excel.  It is very simple to use.  If you have an Excel spreadsheet, you can pretty much let OptWorks loose on it and evolve some rules.  Of course, you also need to backtest the rules against data that has not been used to build the rules.  Also, you can’t just put in a high, low, close and open, and expect a winning strategy, you need to lay out the data so that the system can find some rules.

Neural networks are often seen as a panacea for traders, but many traders don’t understand what they are and what they can do.  A neural network is a computer model of biological neurons.  It learns by modifying the connections between neurons so that the result it generates matches that from a sample set.  The benefit of a neural network is that you can give it examples, then it can learn from the examples and identify trading patterns from real life data.

Neural networks don’t generally explain their rules - they simply learn generalised patterns.

Again, you need to be conscious of using a proper backtesting strategy, ensuring that you have plenty of data so that your network doesn’t just learn from a small data set.  There are a number of neural network programs around.  My favourite is Tradecision.  This one hides the complexity of neural networks from you and lets you focus on your trading strategy.  It has a complete set of integrated tools so that you can check out your equity curve, Sharpe ratio and performance of your strategy.

The developers of this package take their work seriously, and actively develop it.  It can do automated trading. In my view, this is the premier neural network tool for traders.

This tool allows you to select some traditional strategies, then the neural network builds rules from a combination of them.  It has its own language built in.

Another way is to use maths tools to explore data. If you want to play with some advanced mathematics, you’ll need some programming expertise.  I recommend the Python language.  It has an add-in called Orange, which is quite useful for generating decision trees from data. My favourite technique in this tool is called Bayesian analysis which I generally find to be superior to neural networks.

You could also use Matlab (a commercial tool) or R, which is a statistical tool for exploring data.

These are some of the tools that you can use to improve the performance of your trading models.  There are no magical solutions - you still need to manage risk and ensure that your models stand up to real life conditions, but they can give you an edge over other traders.  Of course, a simple model will generally be more robust than a very complex, curve fitted model.

The best way is to start experimenting with them on some historical data and understand the capabilities and limitations of these tools.

(No Ratings Yet)

 Loading ...

Tags: backtesting, build your own trading system, genetic algorithms, neural networks, Trading Systems

This entry was posted on Sunday, May 4th, 2008 at 1:19 pm and is filed under Trading Systems. You can follow any responses to this entry through the RSS 2.0 feed. You can skip to the end and leave a response. Pinging is currently not allowed.