no results for OC-percent of perfect search

StrataSearch will find what you tell it to, but what should that be? Will the same criteria work across different Sectors? Different time periods? Here we discuss the ins and outs of OneClick Searches.

no results for OC-percent of perfect search

Postby averdo » Mon May 09, 2011 10:47 am

Hello,

I want to search for a good strategy for trading in the German DAX index.

I created a sector with only the Dax index as a symbol in it and started OC - percent of perfect (with static trading strategies) and an evaluation period of Jan 2000 until May 2011.

After 20 hours and 400,000 runs there still is no result visible when I select the Combination Results view. I can however see many results of an earlier autosearch that was stopped due to too many results.

I just read on this forum that you should use OC option with "single system search" in stead of "percent of perfect". Does that mean that OC-percent of perfect will not work at all with only one symbol? Should I first clean/purge all the results from the results database?

Any tips are highly appreciated.

Alfred
averdo
 
Posts: 11
Joined: Mon May 09, 2011 4:48 am

Re: no results for OC-percent of perfect search

Postby Overload » Mon May 09, 2011 11:09 am

There are several issues here:

First, the collection of trading rules used for a larger sector might not be appropriate for a single symbol search. For example, there may be trading rules that evaluate the sector separately, or require things like minimum price or minimum liquidity that might not apply to an index symbol. For this reason, you might want to change the AutoSearch Setting on the General tab of the OneClick Setup (Percent of Perfect) to "Single Symbol Search". On the Optimization tab of the OneClick Setup, you should do the same thing. This will ensure the trading rules used in your search are correct for a single symbol system.

Second, keep in mind that all OneClick Searches are configurable. Because of this, they are optimized toward certain characteristics. In particular, on the Strategy Performance tab of the OneClick Setup, you might want to change the "Number of Trades" setting to a lower value. By default it is set at 50, but this might be too high for a single symbol search over a 10 year period.

After making the above changes to your Percent of Perfect setup, I recommend clicking the Clear Results button on the General tab of the OneClick Setup. That will clear the results for that search and start it over from scratch.

Finally, if your results database recently filled due to too many results, I would recommend switching to a new Results Database. Enter the File > Results Database menu and enter a new Database name. After clicking OK, you should be ready to start your revised search in the new Results Database.

There may be other issues here, but the above steps should at least get you going on the right track.

Pete
Overload
 
Posts: 2248
Joined: Wed Nov 30, 2005 12:14 pm

Re: no results for OC-percent of perfect search

Postby averdo » Mon May 09, 2011 2:39 pm

Hi Pete,

thanks for your quick reply. I followed your advise and it's running fine now. I believe creating a new database did the trick.

I do have one more question, in the results-trades are only long trades, but in both the OC-setup and in the trades-setup I selected to search for both long and short trades. What could be the reason for not getting any short-trades?

Thanks and regards,
Alfred
averdo
 
Posts: 11
Joined: Mon May 09, 2011 4:48 am

Re: no results for OC-percent of perfect search

Postby Overload » Mon May 09, 2011 5:32 pm

You should understand that allowing the OneClick Search to look for short systems does not guarantee that it will find any that are more profitable than the long systems it finds. The OneClick Search looks for the trading systems with the highest score, regardless of whether they are long or short. So it is possible you will find one or the other, but not both.

I suspect that your Evaluation Period is heavily weighted toward a bull market rather than a bear market. If that is the case, it would definitely explain why you are finding long systems but no short systems. To find an equal number of each, you would need to have an Evaluation Period that has equal periods of bull and bear.

As a test, you might also try making a copy of your OneClick Search and searching for short systems exclusively. You can then examine what it finds, and see how they compare to the long systems. Depending on the data in the Evaluation Period, the Strategy Performance filters being used may not be equally effective at selecting the best long and short systems. In other words, you may need unique settings to search for each side of the market.

Pete
Overload
 
Posts: 2248
Joined: Wed Nov 30, 2005 12:14 pm

Re: no results for OC-percent of perfect search

Postby averdo » Tue May 10, 2011 4:13 pm

Hi Pete,

thanks, I used the period Jan 2000 until May 2011, so plenty of opportunities for short trades I would say. I stopped the search after 30 hours and just started a new OC for the same period but now with only short trades. Will let you know what it will find. I will try a dynamic search after that.
averdo
 
Posts: 11
Joined: Mon May 09, 2011 4:48 am

Re: no results for OC-percent of perfect search

Postby Overload » Tue May 10, 2011 6:28 pm

Okay. Be aware that a Dynamic Search does not provide you with an actual trading system, but instead simply tests your search criteria over a specified period. If that does not make sense to you, then you may want to read a little more about Dynamic Trading Strategies in the StrataSearch Help prior to attempting to run one.

Pete
Overload
 
Posts: 2248
Joined: Wed Nov 30, 2005 12:14 pm

Re: no results for OC-percent of perfect search

Postby averdo » Fri May 13, 2011 8:17 am

Hi Pete,

I completed another OC on the DAX index for the period jan 2000 - May 2011 but this time with only short trades. The results are pretty good with more than 40% average annual profit for the short strategies and 50% for the long strategies. I assume the first OC with both long/short only came up with long strategies because there were more years with long opportunities than short opportunities. The long strategies hardly had any trades in the bear markets of 200-2001 and 2008 where the short strategies had most of their trades in this period. I exported the trades of both OC's in an Excel sheet and the overall results of the combined systems are much better.

I now want to make a multi-system in SS with the both the Long and Short strategies of the two OC's so I can test them together on other indexes as well. I have created a new multi-system and now want to add these strategies to it. However, I can't find/select the individual strategiesin the setup-multi-system memu. Do I first need to save the individual strategies from the OC's somehow?
averdo
 
Posts: 11
Joined: Mon May 09, 2011 4:48 am

Re: no results for OC-percent of perfect search

Postby Overload » Fri May 13, 2011 11:56 am

By default, the strategies from a OneClick Search are "hidden". This is a Strategy Setup setting that keeps such dependant strategies from cluttering up the Strategy Listings.

To "unhide" the Strategies, enter the Setups > Strategy Setups menu and check the "Show Hidden" box. You should then be able to locate the strategies from your OneClick Searches. They will begin with "OC:", followed by the name of the OneClick Search. You will then need to update each of them individually and uncheck the "Hide Strategy" box on the bottom-right corner of the General tab. When they are no longer hidden, you should be able to create a Multi-System setup with them.

Keep in mind that these strategies (and therefore your Multi-System) may change if you were to continue running either of your OneClick Searches. If you want to create non-changing Strategy Setups, you can right-click on your OneClick Result in the Combination Results Listing and select "Create Multi-System". Make sure you uncheck the "Individual Strategies Should be Hidden" box. When complete, the strategies can then be combined into your single Multi-System, and they will not change even if you were to continue the OneClick Searches.

Pete
Overload
 
Posts: 2248
Joined: Wed Nov 30, 2005 12:14 pm

Re: no results for OC-percent of perfect search

Postby averdo » Mon May 16, 2011 10:02 am

Hi Pete,

thanks, I'm starting to see that this really is a great programm with many features still to discover. Your tips made it easy to create a combined multi strategy with long and short strategies. I took the 3 best long strategies and 2 best short strategies. I promised to give you some feedback about the results.

For the DAX index with an evaluation period of 2000-2011 the results are pretty good with an average annual profit of 40-50% and 75% win/loose trade ratio and excellent standard deviations for both yearly and monthly performance. The average trade performance is approx 3%. It has 15-20 trades per year which is just fine for me.

My concern is that if I run this strategy against other indexes (I only want to trade indexes), like the S&P, DJ and Dutch AEX the results are not that impressive. Although the qty of trades is about the same and win/loose ratio is approx 70% the average trade performance is only 1 to 1.5% with more than 50% of the winning trades having a performance of less than 1% which I'm not happy with. I'm looking for larger moves and not necessarily many trades because the transaction costs in my case will be approx 1% per completed trade

I first thought my DAX results could be to much curve fitted for the DAX index. I ran new Single Symbol OC's for the S&P, DJ and Dutch AEX index for the same evaluation period 2000-2011 with the same strategy performance settings that I used in the DAX OC to see if I could better results that way. The funny thing is that the OC results for these other indexes are signifincatnly lower than that one for the DAX. The qty of trades and average trade profilibility is about the same but the average trade performance is still only 1% to 1.5% with the majority of trades lower than 1%.

I'm not sure what to conclude from this. Is the DAX just a better index to trade or am I missing something? My goal is to have a robust system that trades between 5 to 20 times a year and has a good win/loose ratio and average trade performance of more than 2.5% consistenltly over time and index.
averdo
 
Posts: 11
Joined: Mon May 09, 2011 4:48 am

Re: no results for OC-percent of perfect search

Postby Overload » Mon May 16, 2011 12:26 pm

The big question is this: Could the patterns you evaluated in the S&P, DJ and Dutch AEX be recreated at some point in the future by the DAX?

As you suggested, this is partially a question of curve-fitting. To be ultimately robust, a single system must be effective against every stock, index, currency or derivative. Most traders, however, have discovered that such a system doesn't exist. Or, if it does, the returns are so poorly optimized for the intended market that they are hardly worth trading.

What system traders have to do, then, is find the proper balance of robustness versus optimization. Does the trading system need to work against every single price stream in the world? Maybe not. But perhaps it should work effectively for the price streams that have a significant chance of being seen in the near future. It is up to the trader to decide what should be included in this back testing. Almost without question, the entire history of the stock (or index) should be included in the back test. But whether or not to include data from alternate stocks (or indexes) is up to the trader.

To find a balance, many traders make use of a "Walk Forward Analysis". Using this technique, the trading rules remain the same when trading different stocks. However, the parameters of these trading rules are adjusted over time based on the most recent price movements. The thought is that individual stocks can take advantage of adjusted parameters based on the current conditions. But just as importantly, a winning system can work just as effectively against other stocks by using parameters optimized for those other stocks under their own current conditions. In other words, the trading rules remain the same, but the parameters are adjusted according to the most recent conditions of the current stock.

For more information on using a Walk-Forward Analysis in StrataSearch, open the Help > Contents menu, Advanced Features chapter, and see the page on Walk-Forward Analysis.

Hopefully this helps.

Pete
Overload
 
Posts: 2248
Joined: Wed Nov 30, 2005 12:14 pm

Re: no results for OC-percent of perfect search

Postby averdo » Wed May 18, 2011 8:09 am

Hi Pete,

I ran a FWA for the top 4 strategies out of my created multi strategy for the period 2000-2011 with a new strategy frequency of 1 year and a rolling evaluation period of 182 days. These strategies were showing very consistent and positive results in my multi strategy throughout the whole 2000-2011 period, but the results of the FWA for them were much worse. They showed only a very few trades or even no trades at all. The trading performance was also awful.

Could this be caused by incorrect settings in my FWA setup?

By the way, I could not create a FWA OC for the strategies of which I deleted the original OC and are only available in my multi strategy. If I selected this multi strategy from the combinations results list and try to create a FWA for one of these strategies I get the message that this is not possible because there are no variables for the strategy.
averdo
 
Posts: 11
Joined: Mon May 09, 2011 4:48 am

Re: no results for OC-percent of perfect search

Postby Overload » Wed May 18, 2011 11:53 am

I ran a FWA for the top 4 strategies out of my created multi strategy for the period 2000-2011 with a new strategy frequency of 1 year and a rolling evaluation period of 182 days. These strategies were showing very consistent and positive results in my multi strategy throughout the whole 2000-2011 period, but the results of the FWA for them were much worse. They showed only a very few trades or even no trades at all. The trading performance was also awful.

Could this be caused by incorrect settings in my FWA setup?

Just as a OneClick Setup itself needs to be adjusted for the sector and evaluation period it's being run against, a WFA needs to be adjusted as well. Keep in mind that your search period is just 182 days, which probably won't produce a lot of trades when used against a single-symbol sector. For this reason, you should definitely examine the filters on the Strategy Performance tab of your OneClick WFA. For example, you might want to make sure that there isn't a requirement to produce too many trades in that lead-up 6 month period.

If the nature of your system is such that it only produces 3 or 4 trades every 6 months, then a WFA may not be helpful. There simply might not be enough trades produced to optimize the parameters based on the most recent conditions. An effective WFA really needs to have enough trades available to create a worthwhile optimization for each interval. If there are only a few trades, that may create a rather weak optimization.

However, if you're confident your WFA settings are adjusted properly, there is also the chance that your system fails the WFA test. That doesn't mean the system isn't usable. The WFA is only one test, and many traders don't follow the WFA at all. But at the very least it should be a warning, and you should examine other ways to evaluate how robust your system is. Parameter Shifts, for example.

By the way, I could not create a FWA OC for the strategies of which I deleted the original OC and are only available in my multi strategy. If I selected this multi strategy from the combinations results list and try to create a FWA for one of these strategies I get the message that this is not possible because there are no variables for the strategy.

That is correct. To create a WFA, StrataSearch needs to go back and reference the original trading rules used. It is from these trading rules that it identifies the range of variables that can be tested. If the original OneClick is deleted, it can no longer create that reference. In other words, there won't be any variables to optimize, and the WFA won't be able to do anything more than run the same strategy (with the same parameters) for all intervals.

Pete
Overload
 
Posts: 2248
Joined: Wed Nov 30, 2005 12:14 pm


Return to OneClick Searches and Scoring

cron