Re: Advice for Developing Strategies with Adaptrade

[mandelmus]

go back at least a year and read our posts, really ... there are a ton of great tips out there.  keep in mind that mike added many improvements over the past month, that take care of many of our earlier requests. 

make sure you use the correct settings for the time frame you want to trade ... for example, no need to "exit end-of-day" for a daily, weekly, monthly time frame. 

when building intraday strategies, consider only feeding builder the most volatile custom sessions (this can be done in tradestation), like the first 2-3 hours of trading every morning ... otherwise, builder will struggle to fit the morning, noon (flat), and afternoon trading patterns into a single strategy -- not the best request. 

builder creates amazing strategies when using certain types of range bars on tick data -- the strategies even backtest almost perfectly in tradestation -- but, tradestation has so many quirks, it can't trade the same code properly live, and i don't know when/if mike would consider addressing any tradestation work-a-rounds at this time ... there are so many threads on the tradestation forums discussing all the work-a-rounds, etc.

consider reading all mike's articles on his 2 websites (builder and futures) ... download his sample files.  consider using mike's sample gpstrat files and settings as the foundation for your own strategies.  read builder's help file and faqs for tips.

start your strategies very basic and add on just 1 metric at a time ... i rarely use more than 4 metrics since many are redundant

howard bandy wrote several books on designing quantitative trading systems ... http://www.amazon.com/Quantitative-Trading-Systems-Howard-Bandy/dp/0979183804 ... apparently, he discusses step-by-step approaches using builder

make sure you set the correct "initial capital" and "round-trip commission" amounts.  initial capital of 100,000 will allow a lot more leverage than 10,000.  Don't expect a strategy developed using 100K initial capital to scale perfectly down to 10k -- it's probably not going to work.  use exactly how much you plan to trade.  set your commissions as realistic as possible.  i've tossed out a bunch of good strategies because i forgot to reduce the "r/t commissions" value for that particular market/symbol.  i've also accepted a bunch of bad strategies because i forget to increase the "r/t commissions" value. 

builder allows you to build across multiple markets.  we were all excited when this feature came out, but you need to know when to use it.  there are times when it is better to set up a single gpstrat file for each market.  a few months ago i compared builder's ability to create strategies with single stocks of the dow30 vs. across all stocks of the dow30.  although builder created a bunch of good strategies across all stocks, i noticed that many of the component stocks did not perform well.  next, i ran each stock separately -- combining the individualized strategies performed better than the "across all stocks" strategies.  maybe that's due to some curve-fitting, so it'll be up to you to decide whether you'd rather trade the single symbol strategies or the "across all markets" strategies.

[mandelmus]

metrics i like to sort by in the performance window include ... "Significance", "Corr Coeff", "Account Return" ... i like significance because it seems to identify good performing strategies that are backed by (powered by) more trades

build objectives i like to use include ... "account return", "significance", "no of trades", "drawdown"

keep in mind that builder assesses only the single "most fit" strategy at each (breeding) generation cycle ... i think, it would be nice if builder would also assess (and breed) the top n strategies based on "top strategies" conditions or something like that

[Mark Knecht]

On Wed, Feb 27, 2013 at 2:15 AM, mandelmus <gmb...@gmail.com> wrote:

howard bandy wrote several books on designing quantitative trading systems ... http://www.amazon.com/Quantitative-Trading-Systems-Howard-Bandy/dp/0979183804 ... apparently, he discusses step-by-step approaches using builder

FYI - Bandy's books do not use Builder. He uses AmiBroker.

What the books do focus on a bit is the use of metrics to create a fitness function that the user trusts. I've found in Builder that focusing on the choice and weighting of metrics with small sample sizes, once tested to ensure the strategies I like best are at top of the list, generally translates pretty well to large sample sizes. 

Said another way, focus on metrics & fitness first. Run small populations over fewer generations and ensure Builder isn't throwing away things that looks good. Then run larger populations with more generations looking for tradable solutions. 

HTH

[Howard B]

Greetings --

I am a fan of Michael's work and writings.  In my book, Quantitative Trading Systems (QTS), I describe the importance of, use of, and design of, objective functions.  While the examples in my books use AmiBroker to illustrate the points I am making, the concepts are broadly applicable.  The custom objective function I describe in the appendix to QTS is an illustration of defining the desired features of a system, then allowing the search process to select system logic and parameters that satisfy them.  That "sort of" works in AmiBroker, but Builder is much more capable.

I have posted some articles on my blog site that might be interesting to the Builder community.
http://www.blueowlpress.com/WordPress/

Also look at the free chapters of my most recent book, Mean Reversion Trading Systems, for an outline of a platform independent systems development process that I believe helps in the development and management of robust trading systems:
http://www.meanreversiontradingsystems.com/book.html

Best regards,
Howard

--
You received this message because you are subscribed to the Google Groups "Adaptrade Builder" group.
To unsubscribe from this group and stop receiving emails from it, send an email to adaptrade-buil...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Steve Carkeek]

Great reply Mandelmus, would you please elaborate on:

 

builder creates amazing strategies when using certain types of range bars on tick data -- the strategies even backtest almost perfectly in tradestation -- but, tradestation has so many quirks, it can't trade the same code properly live, and i don't know when/if mike would consider addressing any tradestation work-a-rounds at this time .........

 

Thanks,

 

Steve

On Wed, Feb 27, 2013 at 2:15 AM, mandelmus <gmb...@gmail.com> wrote:

[Michael R. Bryant]

I recently finished reading Dr. Bandy’s latest book, Mean Reversion Trading Systems. More good stuff. He and I definitely share the same basic, quantitative approach to trading strategy design. I particularly liked the section on transformations of indicators, particularly the percent rank transformation. For me, the most interesting strategy was the “regime change” strategy, which I’d say is more of an approach for multi-strategy trading than a particular strategy in that it can be applied to any type of strategy.

 

Also, if you’re looking for some good build targets for swing trading systems, particularly for ETFs, you’ll find some good recommendations there. There’s a lot more in the book, of course, but those are some of the highlights that come mind.

 

Mike Bryant

[mandelmus]

oops, I knew AmiBroker was in there somewhere

[mandelmus]

first, I searched the top 100 stocks (by average daily volume) looking for symbols that have a lot of intraday tick data ... there are probably better ways to do it, but I couldn't figure it out, so I opened each chart in TradeStation ... I was surprised at how few symbols had enough intraday tick movement ... next, I displayed each chart with each standard and advanced range bar type (using the last 20 days of average true range as a guide to the range value setting), exported the data from tradestation (via the data window), imported the range data into Builder, and then ran builds ... next, i took the strategies I liked the most and applied them to tradestation ... after some tweaking, I was finally able to match tradestation's backtest performance with builder's ... however, when i ran the code live (in demo, of course), tradestation would just try to trade multiple times with every new tick ... i tried looking for fixes for this on the tradestationforums and discovered that there are many special code adjustments that need to be made in order to handle trading with range bars on tick data (i'll see if i can find those threads) ... anyway, if i start messing with builder's code, i'm not sure if i can expect to maintain the same results in tradestation

here are some descriptions of the various range bar types available in tradestation ... and the key points when backtesting ... read all of these
http://help.tradestation.com/09_00/tradestationhelp/charting/range_bar_chart.htm
http://help.tradestation.com/09_00/tradestationhelp/charting/advanced_chart_types_-_strategy_back-testing_automation.htm

Strategy Back-testing & Automation

Strategies can be effectively back-tested and automated on Range Bar charts when using an appropriate Range value and a '1 Tick' underlying interval. Automating strategies on this chart type using higher underlying intervals will yield real-time results that are inconsistent with back-test results. For additional information on back-testing and automating strategies on Advanced Chart Types, see Advanced Chart Types - Strategy Back-Testing & Automation.