userque & NUGT/DUST

I assume this means buy Dust at open sell Nugt at close?

Same nomenclature as before, just two signals instead of one. One for the open, one for the close.

System will buy dust at open (thus, selling nugt at open).

System will then buy nugt at close, thus selling dust at close. This call can change after the open is analysed by the system.

Same nomenclature as before, except I put the extra signal there in case one wanted to just hang on to nugt since it will likely be a buy at close...rather than daytrade dust.

Hope that made sense.:blink:
 
SPREADSHEET UPDATE:
0702152013.png

DISCLAIMER: Userque's trading signals are not affiliated with TSP Talk or Buy Low Sell High, Inc. The information is for educational purposes only! The information is not advice or recommendations. Userque does not give investment advice. Do not act on this data. Do not buy, sell or trade the funds mentioned herein based on this information. Userque may trade differently than discussed or posted in this forum.
 
A.I. SYSTEM PROGRESS REPORT

Tough week.

My first thought was that unexpected news was giving my system the flux. But, imo, it should be able to recover from that sort of thing in a day to two. So, painfully rejecting that idea, I searched for a shortcoming within the system itself.

The only thing I 'found,' is what I already knew. It has plenty of recent price action data to crunch; not so much going back a few or several days. If you recall, I recently implemented changes to give it this information, but it never chose to use the data and the implementation required a lot of computer resources. So I scrapped it, after all, the system was trading well.

(I could give my system any sort of data: moon phases, lotto numbers. It will try to find obvious and hidden correlations and use the data to generate signals. If it finds no correlations, it ignores the data. But I can't just throw the kitchen sink at it as each piece of information adds to its computational burden. But new data can't hurt its trading ability as long as it doesn't excessively bog down the computer.)

Anyway, I found a way to add older price history without adding great computational burden. I tried it out and the system liked the data. It used the data.

So now, it is generating a GDX signal for Monday. I will then use the previous model to also generate a signal for Monday for later comparison. Time permitting, I will go 'backwards in time' (proper backtesting) a few days to see how the new model would have performed where the previous model failed.

DISCLAIMER: Userque's trading signals are not affiliated with TSP Talk or Buy Low Sell High, Inc. The information is for educational purposes only! The information is not advice or recommendations. Userque does not give investment advice. Do not act on this data. Do not buy, sell or trade the funds mentioned herein based on this information. Userque may trade differently than discussed or posted in this forum.
 
07-06-15 DUST AT OPEN
07-06-15 NO CHANGE AT CLOSE



DISCLAIMER: Userque's trading signals are not affiliated with TSP Talk or Buy Low Sell High, Inc. The information is for educational purposes only! The information is not advice or recommendations. Userque does not give investment advice. Do not act on this data. Do not buy, sell or trade the funds mentioned herein based on this information. Userque may trade differently than discussed or posted in this forum.
 
SPREADSHEET UPDATED:
0706151345.png






DISCLAIMER: Userque's trading signals are not affiliated with TSP Talk or Buy Low Sell High, Inc. The information is for educational purposes only! The information is not advice or recommendations. Userque does not give investment advice. Do not act on this data. Do not buy, sell or trade the funds mentioned herein based on this information. Userque may trade differently than discussed or posted in this forum.
 
A.I. SYSTEM PROGRESS REPORT

Over the long weekend, I did a lot of testing, etc. Same engine; redesigned intake system--so to speak.

I found that I was able to improve performance by limiting input variables. I tested each input variable group by itself, one by one, to note their correlation to the desired output. The 'best' variables were included in the final model. This reduction in variables also led to drastically improved speed. It no longer takes hours to obtain a reasonable trading solution. It seems my system was suffering from its own version of neural network 'curve fitting.'

This endeavor was very labor and time intensive. I did the testing with the I Fund Model, but the results should apply to all of my models.

I've concluded that while it seems my system would have done 'less bad' last week with the above modifications; it appears that the recent markets are behaving in a way unlike they have in the past (and by 'past,' I mean the past that I have data for; through 2006 for NUGT/DUST). These 'new' patterns are always recorded and the system will learn from them...going forward; it just has to be a able to weather the current storm for as long as it lasts. We'll see.
 
Last edited:
Here, this will fix it...

I, RMI, do solemnly swear to no longer follow this system.

There, all fixed.
Have a great week!!


Sent from my (Daughter forcing me to use an) iPhone using Tapatalk...
 
Back
Top