Help with a parameter

Strategies might be attractive on the surface but fail moving forward. Why? Here we discuss the qualities of effective trading strategies, and the many traps one should avoid.

Help with a parameter

Postby Jaded » Sat Nov 03, 2007 3:11 pm

OK.. Here is what I want but I am not sure if I am doing it right.. In the strategy tab of the Oneclick I want to add a criteria to compare the monte carlo average annual return to the average annual return of the strategy. Here is what I am trying:

($F_AvgAnnReturn / $F_MCAvgAnnReturn) * 100

Best Rated Value: 101
Worst Rated Value: 50 (Reject Strategies box checked)

.. I have ran 50,000 runs through this and it rejects everything. Am I doing something wrong? Any tips or a formula for doing what I want would be appreciated. Basically I want the Monte Carlo Average Annual to be at least 50-75% of the Average Annual of the Portfolio-picked stocks if that makes sense.

Let me know!
Jaded
 
Posts: 151
Joined: Mon May 14, 2007 6:49 pm

Postby Overload » Sat Nov 03, 2007 3:59 pm

First, I think your formula might be need to be reversed to:

($F_MCAvgAnnReturn / $F_AvgAnnReturn) * 100

Second, while I believe this may be a helpful filter to add, it is going to make it a lot more challenging for systems to be approved. So it may take significantly longer to find a system that passes this filter.

My experience evaluating stocks has also been that it is rare to find a system where the Monte Carlo annual return exceeds maybe 25-30 percent. So if you happened to have another filter requiring the AvgAnnReturn to be at least 50%, you may have a hard time finding a strategy that passes this combination of filters. In short, such dependant filters may need to be adjusted downward after adding this Monte Carlo filter.

Pete
Overload
 
Posts: 2246
Joined: Wed Nov 30, 2005 12:14 pm

Postby Jaded » Sat Nov 03, 2007 4:56 pm

I've tried the formula both ways and it had the same results.. everything was being rejected by that filter so I just switched to:

($F_MCAvgAnnReturn + $F_AvgAnnReturn) / 2

and have it rating from 0 to 100 on that instead of the F_AvgAnnReturn.

I am not sure if this is helpful to me or not.. Ideas?
Jaded
 
Posts: 151
Joined: Mon May 14, 2007 6:49 pm

Postby Overload » Sat Nov 03, 2007 7:07 pm

I think the formula I suggested more closely reflects what you wanted to do (i.e. "Monte Carlo Average Annual to be at least 50-75% of the Average Annual of the Portfolio-picked stocks"). Neither your new formula nor your old one will do that.

Again, you may need to give such a tight filter plenty of time to work. And you may need to look at the other filters to ensure they do not need to be readjusted after adding this Monte Carlo Filter. There is only one reason why you would not get results, and that is that the filters are too tight.

Here's a tip. If you look at the OneClick Setups, Strategy Performance tab, you can see the Reject Count column. This helps you identify which of your filters are rejecting the strategies. Click the Reset Count button if you've recently made changes, and after letting it run for a while, you'll be able to see which of the current filters is too tight.

Pete
Overload
 
Posts: 2246
Joined: Wed Nov 30, 2005 12:14 pm

Postby Jaded » Sat Nov 03, 2007 7:11 pm

Overload wrote:I think the formula I suggested more closely reflects what you wanted to do (i.e. "Monte Carlo Average Annual to be at least 50-75% of the Average Annual of the Portfolio-picked stocks"). Neither your new formula nor your old one will do that.

Again, you may need to give such a tight filter plenty of time to work. And you may need to look at the other filters to ensure they do not need to be readjusted after adding this Monte Carlo Filter. There is only one reason why you would not get results, and that is that the filters are too tight.

Here's a tip. If you look at the OneClick Setups, Strategy Performance tab, you can see the Reject Count column. This helps you identify which of your filters are rejecting the strategies. Click the Reset Count button if you've recently made changes, and after letting it run for a while, you'll be able to see which of the current filters is too tight.

Pete


I have been resetting the count after making changes.. Already a step ahead on that one.. I re-added the

($F_MCAvgAnnReturn / $F_AvgAnnReturn) * 100

and had it only reject below 20, and it is producing results now. I have it setup to account for 10% of the rating between 20 and 100, so it will favor systems with higher monte carlo to annual return figures.

We'll see how it goes!
Jaded
 
Posts: 151
Joined: Mon May 14, 2007 6:49 pm

Re: Help with a parameter

Postby Jaded » Wed Jul 06, 2011 2:52 am

I know this is an ancient topic, but if anyone else is wondering how to find systems that more closely match the monte carlo simulation... I believe I came up with a good filter for strategy setup:

Definition: ( ($F_MCAvgAnnReturn - $F_AvgAnnReturn) / ( ($F_MCAvgAnnReturn + $F_AvgAnnReturn) / 2) ) * 100
Importance: 0
Best Rated Value: -40 [Reject outside of]
Worst Rated Value: 40 [Reject outside of]

and a second filter with the following definition and same parameters:
Definition: ( ($F_AvgAnnReturn - $F_MCAvgAnnReturn) / ( ($F_AvgAnnReturn + $F_MCAvgAnnReturn) / 2) ) * 100
Importance: 0
Best Rated Value: -40 [Reject outside of]
Worst Rated Value: 40 [Reject outside of]

If I am correct, this will only accept strategies that are +/- 40% of the average annual return of the strategy itself.

Any recommendations or thoughts on this? I imagine that a strategy return that is closer to the monte carlo return is a more robust system.. I'm not sure why anyone would want to make sure the portfolio-picked stocks are much higher than the monte carlo, since the monte carlo would be a better metric for testing a strategy against anyway... thoughts?
Jaded
 
Posts: 151
Joined: Mon May 14, 2007 6:49 pm

Re: Help with a parameter

Postby Overload » Wed Jul 06, 2011 8:46 am

It's a good question whether one would want the portfolio selection to be better than the Monte Carlo. There are actually good reasons why one might want that. In fact, the Rank field is provided specifically to create a better portfolio selection, and many users put some emphasis on the Rank field. However, focusing too heavily on the portfolio selection does create a significant risk of curve-fitting, so there are drawbacks. But the two do seem to be in conflict. You can either put emphasis on the Rank field and portfolio selection, or you can search for systems where the Monte Carlo more closely matches the portfolio selection. But it doesn't seem that you can do both at the same time.

One approach I've always favored is to find a system where the Monte Carlo is favorable, and then focus on the Rank afterwards in a separate step. By using this approach, the curve-fitting associated with focusing too heavily on portfolio selection is avoided, but the Ranking can be used as a sort of icing-on-the-cake when the general system has already been approved.

Pete
Overload
 
Posts: 2246
Joined: Wed Nov 30, 2005 12:14 pm

Re: Help with a parameter

Postby Jaded » Wed Jul 06, 2011 11:16 am

Wow.. I didn't even think that the Monte Carlo simulation wasn't taking the rank into consideration..this makes more sense now why one would want their portfolio selections to be higher than monte carlo.. Thanks for the clarification.
Jaded
 
Posts: 151
Joined: Mon May 14, 2007 6:49 pm

Re: Help with a parameter

Postby Jaded » Tue Jul 19, 2011 9:30 pm

How does this look...

Name: Monte Carlo near Avg Annual
Definition: 100 * ( 1 - ($F_AvgAnnReturn / $F_MCAvgAnnReturn) )
Importance: 0
Best Rated Value: 30
[x] Reject Strategies outside this value
Worst Rated Value: -60
[x] Reject Strategies outside this value

If my poor mathematical mind is thinking correctly, this will force the oneclick to only accept strategies that are within 30% (either direction) of the MC Avg Annual return.. Does that look right to you?
Jaded
 
Posts: 151
Joined: Mon May 14, 2007 6:49 pm

Re: Help with a parameter

Postby Overload » Wed Jul 20, 2011 10:17 am

No, I don't think that's right. I often put these calculations into a spreadsheet just to test them out, and yours didn't match up. You should try the following:

Name: Monte Carlo near Avg Annual
Definition: ($F_AvgAnnReturn-$F_MCAvgAnnReturn)/$F_AvgAnnReturn*100
Importance: 0
Best Rated Value: 30
[x] Reject Strategies outside this value
Worst Rated Value: -30
[x] Reject Strategies outside this value

Pete
Overload
 
Posts: 2246
Joined: Wed Nov 30, 2005 12:14 pm

Re: Help with a parameter

Postby Jaded » Wed Jul 20, 2011 10:28 am

Overload wrote:No, I don't think that's right. I often put these calculations into a spreadsheet just to test them out, and yours didn't match up. You should try the following:

Name: Monte Carlo near Avg Annual
Definition: ($F_AvgAnnReturn-$F_MCAvgAnnReturn)/B1*100
Importance: 0
Best Rated Value: 30
[x] Reject Strategies outside this value
Worst Rated Value: -30
[x] Reject Strategies outside this value

Pete


What is B1 in the definition?
Jaded
 
Posts: 151
Joined: Mon May 14, 2007 6:49 pm

Re: Help with a parameter

Postby Overload » Wed Jul 20, 2011 10:38 am

Oops. Like I said, I test this out in Excel, and I apparently didn't replace all my macros. B1 should be $F_AvgAnnReturn, and I'll go back and edit my post.

Pete
Overload
 
Posts: 2246
Joined: Wed Nov 30, 2005 12:14 pm

Re: Help with a parameter

Postby Jaded » Wed Jul 20, 2011 11:03 am

Perfect.. I wonder if this could make it into the Curve Fitting documentation somehow. I think that rejecting strategies that are too far outside of the Monte Carlo returns would be beneficial in producing strategies that are more likely to happen going forward.. Just my 2 cents :)
Jaded
 
Posts: 151
Joined: Mon May 14, 2007 6:49 pm

Re: Help with a parameter

Postby Dacamic » Thu Jul 21, 2011 12:23 pm

Jaded wrote: ... I think that rejecting strategies that are too far outside of the Monte Carlo returns would be beneficial in producing strategies that are more likely to happen going forward ...

Over the years I have increasingly relied more on Monte Carlo returns and less on portfolio returns. In fact, I don't currently use the latter as a filter in my One-Click searches.

In the early days of StrataSearch, we didn't have the option to perform automated searches using statistics from the Detailed Analysis. Instead, the Monte Carlo statistics were available only by manually running a Detailed Analysis for each strategy we wanted to review at that level. In that era, I did use portfolio-based returns as a metric for filtering in high-volume searches ... because that was what was available.

As you noted, though, selecting strategies based upon portfolio returns seems more susceptible to false positives than using Monte Carlo returns. After OneClick was added to StrataSearch, which is when Monte Carlo statistics became accessible in high-volume searches, I eventually decided to avoid the potential pitfalls of relying on portfolio returns by not using them as a filter.

An advantage of your idea (comparing portfolio returns to Monte Carlo returns) is that it will allow portfolio-based statistics to be more reliable as filters. In other words if the portfolio returns are "right", then the other portfolio statistics might be "right", too. I need to think about this some more; nonetheless, it's an interesting idea.
Steve
Dacamic
 
Posts: 457
Joined: Wed Nov 30, 2005 12:40 pm


Return to Curve-Fitting and Other Pitfalls

cron