davesaint86
725 posts
msg #140310
- Ignore davesaint86 |
12/22/2017 7:34:47 AM
Kevin - For your NM tests did you also run out-of-sample data? If so, for you 12/10/07 forward tests what setting did you use?
Thanks,
Dave
|
Kevin_in_GA
4,599 posts
msg #140312
- Ignore Kevin_in_GA |
12/22/2017 9:00:44 AM
No OOS done on these current filters - my plan, based on discussions with the software developer, is to update the NN models after each trade.
|
davesaint86
725 posts
msg #140324
- Ignore davesaint86
modified |
12/22/2017 11:19:21 AM
When you say 'update' do you mean re-learn a model after every trade?
Thanks,
Dave
|
Kevin_in_GA
4,599 posts
msg #140325
- Ignore Kevin_in_GA |
12/22/2017 12:38:45 PM
Yes and no - the model is re-learned but retaining the previous model settings as a starting point. Thus it begins with a very similar structure and advantage (maybe) over a naïve learning model.
|
shillllihs
6,045 posts
msg #140327
- Ignore shillllihs
modified |
12/22/2017 12:54:45 PM
Dgaz will be a monster short, down 5% headed for 29
Soxs headed to 13 something
covered Labd for about 9%
Uslv cleared ma20 shorted a little 10.50 something, that ma20 seems critical, may need to go higher.
Yang should hit 5.60s
Sdp headed 22ish
|
shillllihs
6,045 posts
msg #140339
- Ignore shillllihs
modified |
12/22/2017 11:08:11 PM
Kevin,
As I was chilling in the hot tub, I was thinking the exact same thing, and there it was,
I've reduced my Xiv exposure to 25% and have been adding short, mostly commodity etfs.
Are you doing units and what kind of drawdown can you expect?
Also, have you checked or is there any way to check if MA20 is any kind of threshold?
Great work.
|
Kevin_in_GA
4,599 posts
msg #140344
- Ignore Kevin_in_GA |
12/23/2017 11:16:24 AM
I really can't put things like the MA(20) into the system, as that is not how the NN systems work - they may include MAs as part of the inputs into the models, but I cannot be sure. In the advanced settings you can add in moving averages but I doubt they would be able to improve on the current system results (not one less than 95% win rate and all are quite solid ROIs).
Drawdown - this software does not generate an equity curve, so seeing the drawdowns or calculating the Sharpe ratio needs to be done manually. I considered doing that to help determine optimal allocation weightings, but in the end I think I'll just trade a set amount to each signal as they are generated. Over time as the results come in that might be an area for further refinement but for right now I plan to keep it as simple as possible.
|
shillllihs
6,045 posts
msg #140350
- Ignore shillllihs |
12/23/2017 2:17:03 PM
I know the losses are negligible, but if I were to guess, most if not all have occurred
shorting the long right?
|
davesaint86
725 posts
msg #140354
- Ignore davesaint86 |
12/23/2017 6:36:14 PM
Kevin,
Your wrote, "Yes and no - the model is re-learned but retaining the previous model settings as a starting point. Thus it begins with a very similar structure and advantage (maybe) over a naïve learning model."
Are you using Advanced settings is that what you mean about retaining the previous model settings as a starting point or are your referring as the previous model settings like weekly versus daily and the start date of the model?
Thanks,
Dave
|
Kevin_in_GA
4,599 posts
msg #140356
- Ignore Kevin_in_GA |
12/23/2017 8:46:54 PM
Batch -> Learning -> AfterLearn (if you own the software).
Let's try to keep Shills thread clean on these types of software questions. You can email me if you are an owner of the software and I will do what I can to help. If not, I will be posting the systems signals in the other thread. Kevin
|