Summary of Neural Network “Real Time” Equities Performance Prediction Results

January 7, 2019

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

New World Technologies, Inc.
www.nwtai.com

Michael Fouche
mfouche@nwtai.com

Table of Contents

Executive Summary. 3

Neural Network Stock Selection Design Process. 3

Stock Selection Process. 3

Historical Stock Performance Prediction Track Record. 3

Newsletter #3. 6

Discussion of Performance Results. 7

Newsletter #2. 8

Discussion of Performance Results. 9

Newsletter #1. 10

Discussion of Performance Results. 11

Final Comments. 13

 


 

Executive Summary

 

Neural Network Stock Selection Design Process

 

Historical stock data is obtained from Norgate’s Premium Data service for the purposes of this effort.  Three different sets of companies (typically at least one hundred companies in each set) are set aside for the next step.

In the first step, many Neural Networks are trained – using the custom-developed stock performance predictor software system, to recognize high performing stocks based on historical patterns.  The measurement of performance is percentage Return on Investment or ROI.  Thus if a stock is purchased for $10 and then is sold for $20, the ROI is considered to be 100% - this excludes transaction costs.

In the next step, the Neural Networks are tested, using custom-developed testing software, against random time intervals of companies in the 2nd data set (different companies from the first step).  The top performing Neural Networks are culled and set aside for the 3rd step.

The final step is to test the previously selected high-performing Neural Networks against a new set of companies (3rd data set) over a period of ten consecutive years.  “High performing” means that the aggregate of the stocks selected for that time period (one year) must outperform the market by at least a factor of 3 – this for every year of the 10-year period.

If this “team” of Neural Networks consistently outperforms the market every year, for the ten year period, then they are deemed suitable for predicting future “real time” performance.

 

Stock Selection Process


The system is designed to select several stocks – from the 3rd company data set, by consensus of a team of Neural Network modules.  The criteria is that at least three stocks must be selected by at least five Neural Network modules – that is that each of the Neural Network modules must agree that a particular company is going to be a high-performer in the future.  If the full criteria is not met – consensus of five Neural Network modules on the same three stocks (at least three), then it is considered that the probability of high performance for any of the stocks is relatively low.

High performance is considered to be when the stocks outperform the Dow Jones Industrial Average (or S&P 500) by at least a factor of 3.  This is what is used for measuring high performing stocks during the 10-year historical forecast test phase (“final step” discussed in the previous section).

 

Historical Stock Performance Prediction Track Record


For the 3rd newsletter, the selected three stocks achieved an aggregate ROI of 35% in 11 months and outperformed the Dow Jones Industrial Average by a factor of 3 for the same time period - the performance is shown in Figure 1.

For the 2nd newsletter, the selected three stocks achieved an aggregate ROI of over 100% in 8 months, and outperformed the Dow Jones Industrial Average by a factor of 5 – the performance is shown in Figure 2.

Figure 1 shows the performance of the stocks selected for the 3rd newsletter.  The left plot shows the percentage ROI for all three stocks and the average percentage ROI (colored green).  The right plot shows the average percentage ROI (colored green) versus the Dow Jones Industrial Average percentage ROI (colored red).

 

Figure 1 – Performance of Stocks Selected for Newsletter #3

 

Figure 2 shows the performance of the stocks selected for the 2nd newsletter.  The left plot shows the percentage ROI for all three stocks and the average percentage ROI (colored green).  The right plot shows the average percentage ROI (colored green) versus the Dow Jones Industrial Average percentage ROI (colored red) for the same time period.

 

Figure 2 – Performance of Stocks Selected for Newsletter #2

 

 

 

 

The 1st newsletter was entirely a mistake as I was in a hurry to release the first newsletter and bypassed the system criteria (at least three stocks by at least five Neural Network modules) – the result was that only one stock was selected (violated the three stock selection criteria) by a few Neural Network modules (violated the minimum five Neural Network module consensus criteria). 

Figure 3 shows the performance of the single stock (colored green), selected for the 1st newsletter, versus the Dow Jones Industrial Average percentage ROI for the same time period.  Note that eventually the stock broke even at just past the 15-month mark and caught up to the Dow Jones Industrial Average to generate a percentage ROI of 20% at just past the 18-month mark.  While the performance was “acceptable” - 20% ROI in 19 months is on par with the average mutual fund – it was not a high-performing stock which is what the system is about.  This was a result of bypassing the system selection process criteria.

 

 

Figure 3 – Performance of Stock Selected for Newsletter #1

 

 

 

 

 

 

 

 

Newsletter #3


The third newsletter was released on October 16, 2017 as shown below.  The three stocks selected were Astronics Corporation (ATRO), Energen Corporation (EGN), and EZCORP, Inc. (EZPW).

Discussion of Performance Results


These stocks, as an aggregate began to move almost immediately. 

The plot on the left in Figure 4 shows the percentage ROI, since the newsletter was released, for each stock and the average value of all three stocks (in green).  Note that the ROI reached 35%+ just past 11 months.

The plot on the right in Figure 4 shows the percentage ROI, since the newsletter was released, for the aggregate value of all three stocks as well as the Dow Jones Industrial Average percentage ROI.  Note that the aggregate percentage ROI of the selected stocks outperformed the Dow Jones Industrial Average percentage ROI by a factor of 3.

 

Figure 4 – Performance of Stocks Selected by Newsletter #3

 


 

Newsletter #2


The second newsletter was released on January 3rd, 2016 and then amended on January 13 to add an additional stock that had been missed by the software (a bug in the correlation process) – as shown below.  The three stocks selected were Famous Dave’s of America (DAVE), EZCORP, Inc. (EZPW), and AXT, Inc. (AXTI).

Discussion of Performance Results

 

These stocks, as an aggregate, did not move very much for the first 3 months – at that point two of the stocks began their upward climb – AXTI and EXPW.  The 3rd stock, DAVE, began languishing in negative territory and stayed there for most of the remaining investment period.  However, the gains made by EZPW (200+% ROI) and AXTI (100+% ROI) offset the minor losses incurred by DAVE such that the final aggregate percentage ROI between 8 and 9 months was over 100% - as shown below in Figure 5 in the plot on the left (ROI for each stock vs time).

The plot on the left in Figure 5 shows the percentage ROI, since the newsletter was released, for each stock and the average value of all three stocks (in green).  As mentioned above, the average percentage ROI reached 100% in just 8 months.

The plot on the right in Figure 5 shows the percentage ROI, since the newsletter was released, for the aggregate value of all three stocks as well as the Dow Jones Industrial Average.  Note that the aggregate percentage ROI of the selected stocks outperformed the Dow Jones Industrial Average percentage ROI by a factor of 5.

 

 

Figure 5 – Performance of Stocks Selected by Newsletter #2

 

 

 

 

 


 

Newsletter #1


The first newsletter was released on October 16, 2015 – as shown below.  Only one stock was selected (the system stock selection criteria was violated / bypassed) – Banco Santander (SAN).  

Discussion of Performance Results

 

As discussed in the Executive Summary – this newsletter should never have been released as the stock selection process was bypassed and only one stock was selected.  However, while the stock dropped significantly in the short term, it fully recovered in approximately 15 months and caught up to the Dow Jones Industrial Average at the 19 month mark – and generated an ROI of 20+% in 19 months, as shown below in Figure 6.  While the performance is about on par with a typical mutual fund, the high-performance capability of the Neural Network stock selection system was not demonstrated because the rules were not followed (stock selection process was violated).

 

 

Figure 6 – Performance of Stocks Selected by Newsletter #1

 

 

 

 

 

 

 

 

 

 

Figure 7 provides another view of the history of this stock, from the time frame that the newsletter was released until the point where the stock “caught back up” with the Dow Jones Industrial Average.

 

 

Figure 7 – Performance of Stock (SAN) Selected by Newsletter #1

 

 

 

 

 

 

 

 

 

 

 

 

Final Comments

 

I have not been consistent in releasing newsletters each month but will begin to do so in the next month.  In the past I have taken a lot of time, between newsletter releases, to completely modularize and test the three software modules (processing of Norgate’s Premium Data stock data code base, building and intermediate testing of Neural Network code base, and 10-year forecast test code base), and perform more rigorous testing.  In addition, my elderly parents have had a lot of health issues for the past three years and thus I’ve had to periodically focus my time on their needs.

However, while the techniques and software developed for this complex system have taken me years to learn and master, at this point I’m very confident in the system’s ability to forecast high-performing stocks on a consistent basis.  This confidence is based on the many 10-year forecast tests that have been run in the past for many different sets of stocks as well as the high-performing results of newsletter #2 and newsletter #3.