This is a mobile optimized page that loads fast, if you want to load the real page, click this text.

Developing a mechanical system from scratch

Jump to new

Correct - no stop losses and equal position sizing. I'm not trying to avoid draw-downs. I embrace draw-downs, due to equity market draw-down occurrences, consequently supporting my positive expectancy (i.e. everyone is trying to avoid draw-downs, by chasing high sharpe ratios). Also, there are strong risk-reward relationships - the higher the risk the greater the reward.
 
soso said:
No stop loss at all? You never exit losing positions? Tx.
Click to expand...

I'm implying that I don't set automatic stop-losses. My exit criteria is a 30% drop in price from entry or a confirmation that the trend is down (signal based).

As my records show, I exit more than enough losing positions.
 
Same momentum system on the Nikkei. IT works better over time than the Dax.

Win rate is above 40%, average trade is acceptable and the avg. win to loss is good. Max Drawdown is 10%. I was quite surprised at how well it worked recently on OOS data.
 

Attachments

  • NinjaTrader Cumulative Profit Report, 1_1_2009 - 8_17_2016nkd.jpg
    91.8 KB · Views: 12
  • NKDX_closes TF hold thru close.JPG
    174.9 KB · Views: 14
Similar momentum system on the SPI, hasn't taken a trade since March
 

Attachments

  • NinjaTrader Cumulative Profit Report, 1_1_2007 - 8_17_2016SPI.jpg
    92.5 KB · Views: 9
Wouldn't a more robust idea be to build a system that trades reasonably well in all of the markets? Other wise if one market changes overtime you are going to be in trouble. Its like having a system that trades CBA and not ANZ

I am not saying they aren't any good, Just a thought..
 
These systems I'm using here are all two variants of a single system. The difference is a simple signal switch. The key to the successful back tests is to find the sweet spot in the time of day, then test that on alot of OOS data.
 
CanOz said:
These systems I'm using here are all two variants of a single system. The difference is a simple signal switch. The key to the successful back tests is to find the sweet spot in the time of day, then test that on alot of OOS data.
Click to expand...

I replied to this whilst in the doctors office...to elaborate a bit, its my belief, that certain times of day hold the key to successful, robust results. These time periods need to make fundamental sense as well. For example, the closing few hours of the US session is known for its reversals as traders take profits. The Eu morning on the Euro has typically been bullish and the afternoon bearish as the US opens....these types of things. So do some data snooping, but only to determine key time periods that make logical sense. If the thesis is valid it should test well on at least 3x the amount of OOS data.
 

This is an example of a SPI system and the same system on the ES.
 

Attachments

  • NinjaTrader Cumulative Profit Report, 1_1_2012 - 8_17_2016SPI Early AM Fade.jpg
    94.8 KB · Views: 11
  • NinjaTrader Cumulative Profit Report, 1_1_2012 - 8_17_2016ES Afternoon Fade.jpg
    91.8 KB · Views: 9

Greetings --

Trading systems are models plus data. The sole purpose of the model is to identify patterns in the data that precede profitable trades. In my opinion, there is no requirement -- no reason why we should expect -- the patterns for one issue to be the same as the patterns for any other issue.

Maybe bank stocks are similar enough to each other, as the pattern goes, that one model will find profitable patterns for several bank stocks. Or maybe not even that. But I would not expect that same model to find profitable patterns in a country ETF, for example.

Finding models that detect profitable patterns in any one issue is hard enough. Requiring that it find profitable patterns in many issues -- or even just several issues -- is a very high bar.

Thanks for listening,
Howard
 
CanOz said:
This is an example of a SPI system and the same system on the ES.
Click to expand...

Beware of relying on in-sample testing. In-sample results are always good -- whether the model learned to recognize profitable signals or just fit to the idiosyncrasies of the data. It is only out-of-sample testing -- using never before seen data that is more recent than the in-sample data -- that provides useful estimates of future performance.

Best,
Howard
 

Howard, you're either in a different time zone to your home or burning the midnight oil...

I sort of expected the system to work on the ES in this case as the concept of the system was based on the US action. I was pleasantly surprised to see it 'works' on the ES, as i had not tested it until the post by Roller.

In regard to in sample and OOS i always test my concepts on a brief period (in sample) and then follow up testing with OOS tests back roughly 2x the initial period and then forward roughly on 1-2x the in the sample period.

Any markets that have insufficient data to conduct this test regime on, then i shelve it and come back several months later and test again, case in point would be the HKFE markets where i only have three years of data.
 
I'm pretty much finished setting up my strategies and ready to leave them running whilst i am away. I'll tally up the results when i get back from behind the wall...I will continue to test until around the first of October when i would like to start trading them on my live accounts.
 

Attachments

  • All systems.JPG
    30.4 KB · Views: 171
Got my ATR% Filter working....as planned.

Here is the Dax result from 2009 until current, without and with the filter.
 

Attachments

  • DAX_NinjaTrader Cumulative Profit Report, 5_5_2009 - 8_18_2016no filter.jpg
    108.2 KB · Views: 13
  • DAX_NinjaTrader Cumulative Profit Report, 5_5_2009 - 8_18_2016W filter.jpg
    105.2 KB · Views: 11
Some results with an additional short term volatility filter and with and without trailing stops....
 

Attachments

  • DAX3NR-NinjaTrader Cumulative Profit Report, 5_5_2009 - 8_18_2016no filterW trail.jpg
    80.5 KB · Views: 6
  • DAX3NR-NinjaTrader Cumulative Profit Report, 5_5_2009 - 8_18_2016no filterno trail.jpg
    81 KB · Views: 7
  • DAX3NR-NinjaTrader Cumulative Profit Report, 5_5_2009 - 8_18_2016 filter.jpg
    100.2 KB · Views: 5
  • DAX3NR-NinjaTrader Cumulative Profit Report, 5_5_2009 - 8_18_2016no filter.jpg
    101 KB · Views: 6
Arghh, arrived in China on Friday....already missing my desk. Lots of time to read this time though with my young fella more independent now. Got some good systems books by Kaufman, chan, Williams etc.

Funny enough my systems aren't taking any trades. Still short the nk from Friday but that's it, no new signals. Everything connected as normal though.

It's hot and humid here in Shanghai....the building keeps going, construction everywhere!
 

Have you ever tried King Grade Dragonwell green tea? Exy, but wow.
 
I don't think so but I never know what I'm drinking....Jenny just gets it from her family. Is it from hangzhou?
 
Recent returns on my multi system portfolio, last 100 days in most cases. In a couple of cases where the systems trade less frequently there are no trades at all.

Overall, certainly worth continued testing. These results are just the back tests on recent unseen data, not live results or live results on simulated accounts.
 

Attachments

  • recent results.JPG
    104.6 KB · Views: 37
You must log in or register to reply here.

Similar threads

L
Technical Trading Software
Replies
0
Views
2K
Lyndon
L
November 2024 DDD
2 3
Replies
53
Views
5K
ducati916
Trading systems/methods guidance
Replies
14
Views
6K
rnr
R
W
Trading system template for Amibroker
Replies
0
Views
3K
whacky
W
Ozempic: Magic Pill or Devils Bargain?
Replies
10
Views
2K
sptrawler
Menu
Log in
Register
Cookies are required to use this site. You must accept them to continue using the site. Learn more...