Larger sectors take longer to evaluate

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.

Larger sectors take longer to evaluate

Postby Overload » Wed Feb 22, 2006 4:49 pm

I once took a search designed for the Nasdaq 100 Stocks and ran it against the S&P 500. Guess what? It took 5 times longer to find a comparable trading system than it took on the Nasdaq 100. That probably makes sense, but it’s an important point to remember. A 3-day search of the Nasdaq 100 takes 60 days on the Russell 2000 when looking for comparable systems.

But there is a reasonable solution to these time constraints when evaluating larger sectors. Rather than performing a very specific search, which is a reasonable effort with a small sector, you can instead perform a broad search with the larger sector. Put another way, a larger sector is filled with many more trading opportunities than a smaller sector. So performing a broad search can often create strong results much faster than a specific search, and you can thus take better advantage of the larger sectors unique opportunities.

Technically, it makes sense that a search used on a small sector should yield similar results when run against a larger sector. While this may be true, it disregards the fact that larger sectors simply take more time to evaluate. Thus, to take advantage of larger sectors while not spending excessive time searching them, one can adjust their search to accept a broader range of trading system possibilities.

In short, some OneClick searches work more efficiently on larger sectors than smaller, and vice versa.

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

Speeding up a search

Postby quantdev » Sun Nov 04, 2007 9:43 am

Could you be so kind as to let me know what the best way to speed up searches per hour is, other than reducing the number of symbols?

I have tried reducing the period, reducing the number of trading rules, using "always" rules, turning off scoring and multi-systems, switching from long/short to long strategies only--all with little effect.

I was especially surprised to find that the default four year search of the Nasdaq 100 (100,000 prices) seems to run much faster than, for example, a one month search of the Russell 2000 (40,000 prices)--despite having more than double the prices to calculate.

Neotick allows you to use multiple computers to optimize a single strategy. Is that a feature you might add in the future?
quantdev
 
Posts: 30
Joined: Mon Oct 29, 2007 9:18 am

Postby Overload » Sun Nov 04, 2007 10:06 am

There are no plans to link StrataSearch computers together at this time. But StrataSearch is multi-threaded and will take advantage of as many processors as you have available. So, from a hardware perspective, that is how you can best increase your processing speed.

The reason the Russell 2000 runs slower than the Nasdaq 100 in the example you provided is because there is overhead with symbol evaluation in addition to the prices themselves. In particular, variables need to be initialized for each symbol and synchronization with sector data or index references needs to be done with each symbol. Thus the same number of bars will run slower if they are spread across a larger number of symbols.

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

Postby taowave » Mon Nov 05, 2007 12:39 pm

Overload wrote:There are no plans to link StrataSearch computers together at this time. But StrataSearch is multi-threaded and will take advantage of as many processors as you have available. So, from a hardware perspective, that is how you can best increase your processing speed.

The reason the Russell 2000 runs slower than the Nasdaq 100 in the example you provided is because there is overhead with symbol evaluation in addition to the prices themselves. In particular, variables need to be initialized for each symbol and synchronization with sector data or index references needs to be done with each symbol. Thus the same number of bars will run slower if they are spread across a larger number of symbols.

Pete


Pete,is there a way to limit the portfolio sizes for MC calculations as well as other computations.For instance,I only care to look at statistics for 10-15 stock portfolios.Can I select that particular portfolio size only??
taowave
 
Posts: 584
Joined: Sat Dec 02, 2006 12:39 pm

Postby Overload » Mon Nov 05, 2007 1:15 pm

Can I select that particular portfolio size only??

No, but we're already looking at this as a possibility for the next major release.

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

Postby quantdev » Thu Nov 22, 2007 1:48 pm

The reason the Russell 2000 runs slower than the Nasdaq 100 in the example you provided is because there is overhead with symbol evaluation in addition to the prices themselves.


OK.I understand. So I tried to reduce that overhead by cutting the Lead up Days from 200 to 100. But that didn't seem to have much effect on the OneClick speed.

Is there any setting I can change that would significantly increase OneClick processing speed? (I am using a four processor server with 2 GB of RAM and RAID SCSI hard drives so it will be hard to improve on the hardware.)
quantdev
 
Posts: 30
Joined: Mon Oct 29, 2007 9:18 am

Postby Overload » Thu Nov 22, 2007 2:03 pm

The speed of a search is largely dependant on how the search is being run. In particular, the size of the sector and the evaluation period have the biggest influence. Including things like Alternate Data analyses or Parameter Shift analyses also take significant time to run, if they're included. As well, having a Monte Carlo Analysis as a field being evaluated in the Strategy Performance tab of the OneClick Setups creates some overhead as well.

From the results perspective, you can also save time by turning off the automatic rebuild of the Detailed Analysis. You can do this by entering the OneClick Setups, Results tab, and unchecking the box titled "Automatically Run Detailed Analysis on Latest Multi-System". You can then run the Detailed Analysis manually when needed.

All of the protective features create a bit of a catch-22. Most of us want all this curve-fit protection included in our searches, so that we don't have to spend as much time manually sifting through the results ourselves. But all the overhead of the curve-fit protection does make it run slower.

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


Return to OneClick Searches and Scoring

cron