Topic: Strategy Generation

Hello - can you quickly describe how FSB generates strategies.  The reason I ask is that I had what looked to be a reasonable strategy using 2 opening logic conditions, so I locked them in and ran the strategy generator just to see if it could improve upon it.  I ran the generator 6 times - twice with a working time of 5 minutes, twice for 30 minutes, and twice for 60 minutes.  The generator came up with entirely different scenarios each time.  I'm guessing there could be a difference when using different working times due to the fact that more combinations can be tested.  However, why would I get different results using the same time frame?  Thanks much.

Re: Strategy Generation

The Generator uses random algorithm. Theoretically, if you run the Generator for an unlimited time, you will receive the optimal strategy for the data set. However, we don't need this. Such a strategy will be over-optimized for the current data.

We intentionally made the generator exploring different strategy rules at each cycle. In that way we prevent repeating results from the calculations. If the calculation route was fixed, the generator would repeat equal calculations and results each time. So, for example, every time you start, you would receive $12500 at 1st minute, $14700 at the 2nd min....

Re: Strategy Generation

Thank you.  Based on your response, would it be accurate to say that the longer the working time, the more optimized the resulting strategy will be to the current data set?  Would that mean it is better to select shorter working times in order to avoid that?

Re: Strategy Generation

Each cycle of the generator is independent and has equal chance to produce a particular strategy. The Generator creates a random strategy and checks if its profit is among the top 10 generated profits. If the profit is satisfactory, the Generator checks if the strategy satisfies the limitations as count of ambiguous bars, max drawdown or win-loss ratio. If this conditions are fulfilled, the Generator accepts the strategy. We can say that longer work of the generator will produce more and more profitable strategies.  Unfortunately there is no an easy way to determine if a strategy is over-optimized or not.
If we find a clear criterion for determination of the rate of over-optimization, we'll include it in the Generator's limitations and we'll be sure that the shown strategies are acceptable.

For now, we first generate a strategy and after that we are trying to determine its robustness according to our experience. The Analyzer tool can help: menu Analysis - Over-optimization report.

Re: Strategy Generation

Thank you so much for your responses.  One more thing and then I'll stop bothering you. 

If I understand the theory behind the over-optimization report correctly, there should not be a dramatic change in outcomes for small changes in parameters.  So, if I run an over-optimization report and the outcomes do not materially differ for each incremental change from 1% to 20%, then it would seem that the risk that a strategy is over-optimized is small.  Is that correct? 

I have attached a report which, if my thinking is correct, reflects a strategy that has a low risk of over-optimization.  Can you let me know if I'm right (or wrong) about this? 

Lastly, the charts generated with the over-optimization report show 2 lines, one red and one purple.  Can you tell me the difference between the two?

Thanks again.  FSB and FST are great learning tools!

Post's attachments

EURUSD Over-optimization Report.xls 136.5 kb, 11 downloads since 2011-09-03 

You don't have the permssions to download the attachments of this post.

Re: Strategy Generation

The report itself without the strategy is not enough informative.

It seams the strategy indicators contain 4 numerical parameters:

Index Parameter name
  1      MA period
  2      Multiplier
  3      Initial Stop Loss
  4      Take Profit

The second table of the report shows the values used during the test:

Values of the Parameters                
+20%     173     3.76      4038      3061
   0%      144      3.13     3365      2551
-20%      115      2.5       2692      2041

From here I see that you left the generated values. I would prefer more round values like MA of 150, Multiplier of 3.0 .. TP of 250 pips...  but this is not a problem.

Next table shows Account Balance calculated  by using the deviated values.
We see that the Initial SL doesn't affect the balance. Why?
Probably because it's too far from the entry.

You can start reducing the Initial SL until the balance starts changing. In that way you'll find a more safe SL.

Almost the same is for the Take Profit.   It seams that only one position closes at Take Profit. Or only one position is affected by the change.

From this I see that SL and TP are not used effectively. I don't thing this strategy is robust enough to be live traded.


About the lines on the chart, each line shows change of one of te parameters. Unfortunately there is no a legend so far.