Considering seasonal trend analysis when building strategies

67 views
Skip to first unread message

mandelmus

unread,
May 22, 2013, 1:15:25 PM5/22/13
to adaptrad...@googlegroups.com
In response to fury's post ...

I've got a similar theory I'm testing out ... except I'm using Seasonalysis.com (or TradeMiner, as recommended by Quilln on Stockfetcher) to identify symbols with bullish and bearish historical seasonal trends.  Then, I build "Long-only" strategies for symbols with bullish historical seasonal trends and "Short-only" strategies for symbols with bearish historical seasonal trends.  But, I've just started this analysis, so I'm still working out the details.  Alternatively, I could build long/short strategies and only trade "long" signals during the bullish periods and only trade "short" signals during the bearish periods.

Seasonal trend analysis appears to be helpful, but by itself, does not appear to be sufficient enough as a stand-alone trade signal ... kind of the same concept as using a 50-day or 200-day moving average as a base requirement and then building from there.

Here's a screenshot from today's "Seasonal Pattern Screening" results.  I chose a 42-day bullish trend pattern because anything greater than 30 days seems to be more predictive/protective of a longer-term bullish trend and I want to give more time for the Builder strategies to work.  If I chose symbols with 7- to 14-day patterns, there might not be enough time for the Builder strategy to enter and exit a position during that short window.



In trying to achieve a higher significance, and based on some of Acrary's System Development suggestions, I almost always include the metrics to "Maximize No. Trades" (usually one of the higher weights) and "Minimize Ave Bars", "Minimize Ave Bars Wins", "Ave Bars Loss" (low-to-middle weight).  Then, I set "Conditions for Selecting Top Strategies" to "Ave Bars <= 5.00", "Ave Bars Wins <= 5.00", "Ave Bars Loss <= 5.00".  The theory is that we should have more confidence in a strategy's persistence the more historical trades the strategy has produced.

This is a work-in-progress, but in case anyone is curious, here is the current list of "Build Objectives", "Build Conditions", and "Conditions for Selecting Top Strategies" I use with most of my builds ...

Build Objectives
Maximize Significance, weight 2.000
Maximize Corr Coeff, weight 1.000
Maximize No. Trades, weight 3.000
Maximize Kelly f, weight 1.000
Minimize Ave Bars, weight 1.000
Minimize Ave Bars Wins, weight 1.000
Minimize Ave Bars Loss, weight 1.000
Maximize Account Return, weight 1.000
Minimize Drawdown, weight 1.000
Minimize Ave MAE (%), weight 1.000
Minimize Max MAE (%), weight 1.000
Minimize Ave MAE, weight 1.000
Minimize Max MAE, weight 1.000

Build Conditions
< always empty >

Conditions for Selecting Top Strategies
Significance >=  90.000% (OOS)
Significance >=  90.000% (In-Samp)
Corr Coeff >=  0.9000 (OOS)
Corr Coeff >=  0.9000 (In-Samp)
Ave Bars <=  5.00 (OOS)
Ave Bars Wins <=  5.00 (OOS)
Ave Bars Loss <=  5.00 (OOS)
Ave Bars <=  5.00 (In-Samp)
Ave Bars Wins <=  5.00 (In-Samp)
Ave Bars Loss <=  5.00 (In-Samp)
Kelly f >=  25.00% (In-Samp)
Kelly f >=  25.00% (OOS)
Account Return >=  0.00% (In-Samp)
Account Return >=  0.00% (OOS)
No. Trades >=  1 (In-Samp)
No. Trades >=  1 (OOS)

*I used SysExporter to quickly capture this text from Builder

Obviously, these conditions select for higher frequency trading systems -- not designed for buy-and-hold strategy selection.













Reply all
Reply to author
Forward
0 new messages