Kevin_in_GA
4,599 posts
msg #135887
- Ignore Kevin_in_GA
modified |
5/13/2017 7:26:09 AM
I have really only looked at XIV - they come with a few preloaded stocks (AMZN, AAPL, GS, TSLA, FB) but that is not really where I typically trade.
XIV is certainly not typical in its outperformance of the S&P over the last 5 years, so there is a bias toward long systems that can do well. I would think that the best test would be to find stocks that exhibit different price action characteristics - high flyers, stuck in a sideways channel, trending down, big gaps up/down - and see how well the NN systems perform against a simple buy and hold.
For fun I will run a bunch of system input variations against IJR (this had a big decline in early 2016, a huge run up after the election, but has been in a sideways channel that can't break out above 71 for most of this year).
|
Kevin_in_GA
4,599 posts
msg #135891
- Ignore Kevin_in_GA |
5/13/2017 8:00:26 PM
So I ran about 20 different NN input settings on IJR - bottom line is all system beat buy and hold and had decent but not outstanding win percentages (typically 60-80% ROI over the last two years, 70-80% win rates). That sounds good, but I noticed that most of the losses (a large percentage of them) occurred in the OOS sample, which means the results for the system trading forward from its training dataset were not very impressive. One needs to remember that there was a clear change in the stocks behavior around that transition (12/2016) into a sideways choppy trading range, which might have been the issue.
|
mahkoh
1,065 posts
msg #135894
- Ignore mahkoh |
5/14/2017 7:54:37 AM
I've watched a couple of videos about the software, in one of them they mentioned that it is optimized for a one year lookback period with three months out of sample. Also, if there is a significant change in stock behavior it would be advised to relearn the model.
From what I understand it looks for pattern recognition and assumes history will repeat itself.
I'm wondering how the results would relate to a one click strategy in Stratasearch on a single symbol with the same lookback and out of sample periods.
|
Kevin_in_GA
4,599 posts
msg #135898
- Ignore Kevin_in_GA |
5/14/2017 3:33:33 PM
You know - I had forgotten how good SS can be at these types of systems, and how easy One-Clicks are to set up and run.
I'm running one for the period 1/2/2013 - 1/2/2017 using the "Highest Sharpe Ratio" One Click setup. Very nice results so far, especially in avoidance of significant drawdowns like one saw in XIV during August of 2015.
|
redrobin
10 posts
msg #136102
- Ignore redrobin |
5/29/2017 10:59:13 AM
I wonder if Neuromaster yahoo download still working, it is broken at ss.
|
Kevin_in_GA
4,599 posts
msg #136104
- Ignore Kevin_in_GA |
5/29/2017 3:51:41 PM
Both programs can now use Google Finance as their EOD data source. I have not reinstalled Stock Neuromaster Pro because I am getting a new MB and more RAM installed this week so my primary computer has been out of commission for a while.
|
mahkoh
1,065 posts
msg #136138
- Ignore mahkoh |
5/31/2017 2:49:20 PM
Looks like you may have a problem there with your XIV system. Google price is 36.96 for XIV, TVIX is way off too.
|
Kevin_in_GA
4,599 posts
msg #136227
- Ignore Kevin_in_GA |
6/7/2017 8:11:50 AM
Software is definitely pulling in the correct data - maybe the source is something other than Google, but whatever it is it seems to be correct.
|
mahkoh
1,065 posts
msg #136241
- Ignore mahkoh |
6/7/2017 2:18:36 PM
Historical data from Google Finance in the Stratasearch stream is also correct, however if you try to pull data through Google sheets all Credit Suisse ETF's have price issues. Don't know what's up with that.
|
Kevin_in_GA
4,599 posts
msg #136244
- Ignore Kevin_in_GA |
6/7/2017 7:00:47 PM
I have yet to get data properly loaded under the most recent update to SS. Not sure what the issue is, but it never works. I am updating the few stocks I watch manually each night.
|