It is very effective and simple to optimize strategy parameters using TSGenotic. Let’s review an example of parameters optimization of the simplest cross strategy of the two MA.

We can see 4 parameters in this strategy, which can be optimized: FastLength(9), SlowLength(18), SdFast(1), SdLow(5).

FastLength: from 0 to 30;
SlowLength: from 10 to 100;
SdFast: from 0 to 10;
SdLow: from 0 to 10;

Let’s prepare this strategy for optimization using TSGTEditor. Take a population of 30 individuals and run it through 300 generations (these genetic algorithm parameters can be a lot smaller for such simple strategies). Generally speaking, based on the experience of using genetic algorithms, we recommend using not less than 150 generations, but on the same hand, there is not point of setting more than 300. In most cases, any system is able to rich its maximum within 300 generations. Although it is only fair if the number of the individuals as well as other GA parameters are set correctly.

If population size is understated intentionally, then optimization process can stretch for a larger number of generations and sometimes this process might not converge one local maximum.
At the same time you should not increase population size too much, since it prolongs total optimization time.

So, let’s review the code obtained using TSGTEditor, which is ready for strategy optimization. Of course, this code can be written manually, though it is much faster and simpler to do it using the editor.

First section determines those parameters which will not be optimized, and also contains an important GTSteps parameter. This parameter is used for starting optimization within TradeStation. GTWork parameter is responsible for optimizer’s mode of operation. If GTWork equals 0, then optimization takes place, if GTWork equals 1, then those individuals that have already been calculated get recalculated again.

//--------------------------------------------------------------- //inputs section //--------------------------------------------------------------- inputs: GTsteps(1),GTWork(0); Inputs: Price(Close);

In the next section, there will be a description of genetic algorithm parameters as well as individuals necessary for optimizer’s functioning.

Here, we can see that our parameters turned into variables. That is the way it should be, since the parameters are not to be set by the user but the genetic optimizer.

Next section will execute initialization of TSGenotic optimizer. The optimizer receives the data about all parameters, their limits, types and steps.

This section is exercised at every execution on first bar on chart. The strategy requests the optimizer for parameter values of a current individual.

Then goes source strategy calculation.

Final section. Fitness, profit and drawdown calculation of the strategy as well as transfer of those values to TSGenotic optimizer will take place here.

In this case, fitness calculation is exercised based on coefficient of correlation of yield curve.

As is clear from the example, strategy structure for TSGenotic is rather simple. Although, we recommend using TSGTEditor for strategy preparation.

Start TradeStation.

Open the chart and the period you are interested in. For example, GBP/USD timeframe 5 minutes for 2 months.

Add prepared strategy on chart.

Set the MaxBarsBack parameter. (in this particular example, it should equal 150, in general, it is set up as maximum possible that might be required during optimization process, i.e. what maximum MaxBarsBack is required at any strategy parameters combination).

Further you should set GTsteps strategy parameter optimization from 1 to a number equal a product of number of individuals by number of generations. In this example, it is up to 9000. GTWork should equal 0.

Press “Ok”. Ignore possible message about incorrect MaxBarsBack settings. Press “Yes”.

The strategy in TS is ready for optimization.

Start TSGenotic.

There should not be any data in a table at this moment, except cases of continuation of optimization.

Press “Optimize” button in TradeStation.

There starts optimization process. As the calculation goes on, executed individuals will get filled with fitness, profit and drawdown values in the optimizer’s table.

In TSGenotic, you can see current results, sorted based on profit by pressing “Sorted results”.

Optimization process can be aborted at any moment by pressing “Abort” in TS or you can simply wait till the optimization process is complete.

As seen from the table, we could see good results at 180th step on the first line. We completed optimization, but the results have not improved. This is a normal occurrence. Since genetic algorithms are not based on direct enumeration, but the search of the maximum of some generalized parameters surface, best variants will appear at early stages of optimization. Although we recommend to always complete the optimization.

In order to realize that optimization has come to its logical end, you should analyze several latest executed populations. If all individuals have similar results and these results do not improve from one generation to another, it means that the optimization process can be considered complete.

I would like to remind you that we strongly recommend executing at least 300 generations before evaluating your results.

This is an equity curve chart of the test strategy that was optimized using TSGenotic. Optimization criterion (fitness) is a coefficient of correlation of yield curve.

Summing it all up, you can notice that instead of direct enumerating of 270000 variants, we have executed only 9000, that took us less than an hour. And despite the fact that we got good results on 180th step, optimization was still completed.