Financial Sense Univ. ~ "Quick Look Report #20, Neural Net Prediction" by Stephen Rinehart 03/02/2006

QUICK LOOK REPORT #20 Neural Net Prediction on NYSE
by Dr. Stephen Rinehart
March 2, 2006

Background:

Quick Look Reports usually looks at a possible dominant trend in an Index, Equity or Commodity and some possible long-term (monthly/yearly) trends which could emerge from the dominant cycle(s) (the datasets will be either daily, weekly or monthly). Quick Look Report # 20 combines the cycles in various indices with a Neural Net Prediction (using EasyNN-Plus code) for the Weekly NYSE Index including variations in M3 money supply. As we know, M3 will no longer be reported after March 20^th, 2006 and for good reason. M3 and its 266-week cycle represent two of the top five variables for predicting the NYSE Weekly Index when the major NYSE cycles are combined with other key financial indices such as M3. The very preliminary results presented in this study represent hundreds of simulations looking at pruning the Neural Net to the fewest possible variables which may be involved in coupling with the NYSE.

The sensitivity in the NYSE Index Weekly closings from M3 is shown in Chart 1 for March 2006 thru 2008 based on the Neural Net described including M3 estimated YOY growth. Several different linear growth models were examined for M3 including 6%, 9%, 12% and 15% YOY increases in M3. This approach is considered to be a refined prediction technique instead of just using “cycles” provided one can find independent variables (M3) to insert into the Neural Net Prediction. To see if our recipes for long term predictions can be improved, we turn to some Neural Net estimates with our cycles and see if anything will start developing. Simply increasing M3 does not result in any linear scaling of NYSE as there are some complex interactions. There appear to some type of counter-intuitive behavior to be examined as an increase in M3 can result in sudden drops in the NYSE until you reach a relative “range-bound” behavior at higher growths rates of M3 (say 15% YOY). However, there does not yet appear to be a long-term stable economic configuration or orbit in the NYSE Index for any constant YOY increase in M3. It appears you have to constantly play games with M3 (up/down) to try and hold the NYSE Index range bound and expect some violent moves can occur suddenly. The values for the NYSE Index shown on the chart were scaled for the Neural Net and have to be multiplied by 10X to get the current NYSE Composite Weekly Index.

The sensitivity in the NYSE Index Weekly closings from M3 is shown in Chart 2 for 2009 thru 2011 based on the Neural Net described including M3 estimated YOY growth. Several different linear growth models were examined for M3 including 6%, 9%, 12% and 15% YOY increases in M3. It suggests that one can create the appearance of a long-term rally in M3 at levels around 9% YOY but the NYSE Index may be subjected to a growing instability (possibly debt service) which can result in a sudden drop in the NYSE (or economic crash).

There is no Bottom to This Line:

The estimates given in this study (and many other studies I made with Neural Nets over the past 15+ years) have clearly shown the NYSE price movements are strongly correlated to a combination of eight key variables which are: trend in M3, 266-week cycle in M3, Fed Fund Rate, the 10 year (or 30-year Treasury) Bond, CITI’s closing prices, the 760-week cycle in CITI (or FNM), the 510-week bond cycle in thirty-year bond (before its demise), and XOI (or XOM) Index. Yes, you can find other “financial surrogates” for some of these variables (such as FNM for CITI) but not for M3 and its major cycles (too important).
For a 6% YOY increase in M3, the top five correlating variables in the Neural Net for predicting the NYSE Weekly Index are CITI, XOI (Oil), CITI 760-week, XOI Index, Fed Fund Rate and a long Treasury bond cycle (384 to 510 weeks). However, as you increase M3 to 15% YOY, the top five variables become the Fed Fund Rate, The ten (or thirty) year Treasury Bond, the 510-week Treasury Bond cycle (use 384-weeks with Ten Year Treasury Bond).
The key to finding a successful Neural Net prediction or tool is to use as few of variables as possible (less then ten with financial predictions) as otherwise the Neural Net will lock-onto a set of small differences in many variables and not have any predictive capability (as many have discovered from overselling or trying this technique without understanding it technically).
To my knowledge, nobody has coupled cycles (from band pass filters) with Neural Nets and presented any results on the Internet on long range trends in the NYSE as a function of YOY growth in M3. Training the Neural Net repeatedly resulted in random validations of 97% within an error bound of less 7% to a target error of 7%. The average training error in all NYSE Neural Nets (with ten variables or less) was about 0.25% with a maximum error of about 9% at a couple of peaks (in 2000 and 2002).
In general, increasing M3 YOY from 6% to 12% suggests it can “smooth out” the major cyclical variations in the NYSE Weekly Index and it is even possible to create a rally (appears to correlate with 9%+ YOY increases in M3). If the increases in M3 were currently less than 6% YOY (around 4%), the NYSE was predicted to start dropping in 2006 and subsequent years with major cycles coming strongly into play at tops and bottoms.
There is a major transition which takes place between 12% and 15% YOY increases in M3 in the long run. At 15% YOY increases in M3, one tends to initiate a sharp rally in NYSE followed by a relatively “range-bound” or flat-line appearance to the NYSE. This could happen as money flows out of equities into bond markets due to high rates. However, there was a “sudden drop” predicted in the NYSE Weekly Index by the Neural Net at 15%+ YOY increase in M3 after several years. This requires a lot more investigation before anything definitive can be stated but clearly the implications of this little study suggest this maybe a deadly and dangerous experiment by the Federal Reserve Bank if we start approaching (or exceeding) 15% YOY increases in M3. It looks like a Catch 22 all the way.
There are a lot of questions and what ifs to be played with this Neural Net approach so stay tuned for future articles. Why does CITI come into play so strongly as one of the top five variables? How come gold prices are not significant? Closer looks are needed at Wal-Mart and GE which do appear in some top ten variables and parametric variations. Foreign indices can be added or we can focus on daily studies and on it goes. Currently, I am heading to Miami to see new grand-daughter and enjoy some time with family and old friends. How is your gold and commodities investing coming? The fun is just starting – stick around and we will crank it up another notch in “Let’s Make a Deal With M3!”

Hypotheses: Seven major cycles in virtually all the worlds’ indices/commodities determine our M3 financed lives far into the past (and may have for many decades). It appears some people you may know have been playing games on other people for a long time with M3. For the current recipes, just zip open your Fed Fund Meal Package (called an M3RE) and pour the magic Fed Cycles into a Neural Net and add-in M3 trends and cycles, Treasuries and Fed Funds Rate (stir constantly under a boil) and presto you have a delicious and tempting Witches Brew at 9%, 12% and 15% concentrations – dosages above these levels may prove fatal if improperly ingested. You can try and hide M3 but you cannot run away from its long-term consequences. What you gonna do when the Mayan (M3) Bear comes after you? Asta Lavista, Baby!

CONTACT INFORMATION
Dr. Stephen Rinehart
Lynn Haven, FL USA
Email

DISCLAIMER: The author is not a registered stockbroker nor a registered advisor and does not give investment advice. His comments are an expression of opinion only and should not be construed in any manner whatsoever as recommendations to buy or sell a stock, option, future, bond, commodity, index or any other financial instrument at any time. While he believes his statements to be true, they always depend on the reliability of his own credible sources. Of course, the author recommends that you consult with a qualified investment advisor, one licensed by appropriate regulatory agencies in your legal jurisdiction, before making any investment decisions, and barring that, we encourage you confirm the facts on your own before making important investment commitments.