June 30, 2017 Result. Mkt price of target is 105.26 as I write this at 2:17pm. Network output is another +1, so now three +1's in a row => strong signal to go long. This is an algorithmically generated signal, the output of this simple network, and of course it could be wrong. The Hinton Diagram shows the network output as the top white box in "Unit Output" display, -1 is shown as black, and zero is mostly grey. The 40-box row shows the internal network "neurons", which can take on values anywhere between -1 and +1. The actual numeric output is shown on the centre cyan coloured screen, with part of the boolean case table shown to the left.

June 25, 2107: Data to June 23, 2017, as of Friday. Note that the network has generated a boolean target value of -0.533 which is suggesting (if we round to a boolean -1), a downturn in price of greater than 1% from the June 22nd value, to the value that will prevail as of the close of June 22 + five trading days => June 29, 2017. June 22 close was 107.06. So, June 29, 2017 expected close, is 107.06 - (0.01 *107.06) => 105.99 or lower. What is interesting, is that this forward estimate is only 20 cents off the dividend value, of 1.27/shr, so the NN is providing an expected value that is not unreasonable. And the boolean table does not make a price forecast, it only says it expects the price to be down 1% or more on the target forward day, and only at the close.


This is a collection of some of the more interesting economic phenomenon observed recently, and forms a key part of my research work on markets and efforts to develop AI-based Augmenter tools. 

The picture at right, is of the Probability Calculator - here shown running on Samsung Tab-A tablet. It tells you - for a given level of capital, and a chosen level of acceptable risk, how big a bet should be, and what the expected outcome is. Everyone has this type of tool now, and there are few asymetric opportunities for arbitrage. But they do sometimes still come up.

[June 25, 2017] -  The Xerion-based Neural-Network experiment, using the boolean jump-delta table is producing interesting results. The approach looks as if it might be useful. The information it provides is clear and actionable. Of course, one needs a lot more data to say anything with any legitimate scientific validity at this point. I will need to evaluate the results in conjunction with a methodology for interpretation, that can be tested against various runs of randomness. I will need to see the network produce consistantly better-than-random results, before I can assert that it is providing value. And of course, the human world is also changing, so I may find that I am trying to hit a moving target, perhaps unsuccessfully.

The NN-AI results, provided by the boolean-table driven neural-network, have to be looked upon with suspicion. The AI might just be getting lucky.

But having pointed out this important caveat, the results do look interesting, and are explicitly keeping me in a trade that I probably would have exited. Here is the data-screen from the new AI-box, (with the nice wide-screen, which gives me more visual real-estate.). The target security goes ex-dividend on Monday (June 26, 2017), so of course, it will show a negative delta of 1.27. Despite the actual security price moving up, the AI has shifted its view down below the zero-line, on the most recent data, and I find this curious and a little surprising. This is what it is, of course, supposed to do, but I rather suspected that it would not do this, and would be "surprised" by the down-turn on Monday, which I know will show up in the data (I have not tried to use prices that were "adjusted" for dividends). The raw price data is current to June 23, 2017, and the most recent price-data built into the boolean jump-delta table by MAKECASEBOOL program, is to close of June 22, 2017. And the NN-AI has flipped to negative, with the MarketNet network generating a -0.533 value for the June 22, 2017 target. (See both the "compareMarketNet" output, and the GNUplot "plotValues" graphic.)

If you were using the NN-AI as the "old, wise fellow" it is conceptually designed to be, you would have sold your position in the target-security on Friday, somewhere in the upper 107 level (say, 107.70). Mkt close of 107.38 on June 23rd (Friday), means that this strategy looks already to have been successful, as the target will open near 106.11 on Monday morning, all else being equal.

My objective now is to produce an evaluation methodology that lets me assess both NN-AI effectiveness, and expected profitability. The idea is to see if there really is an *edge* being created here, or am I just curve-fitting randomness. We will try not to be fooled here.

[June 22, 2017] - Update on the Neural-Network results.  Four photo's below.  Using photos is just about always the best way to explain a process, or describe an event.  These results are *very* preliminary - but remember, life is short, and it is also what usually happens while you are making plans.   We need to move fast now, no?   So, here are some quick thoughts on how the AI is doing...



June 22, 2017: Summary of the current state of NN-AI output. The on-screen text is beside the most recent boolean output. The TSM (Time Series Manager) screen provides historical price quotes which are being used to make boolean table which NN-AI is using.

This is today's action, June 22, 2017, (as of earlier this morning.). Target securty is now up almost 2 (up 1.89% in Cdn$ terms, and 2.41% in US$ terms. The postive results from the AI (and the desire to get my dividend!), kept me in my position. Like it or not, the researcher is *always* part of the experiment.

This is an older, curve-fitting model. It is interesting, but the NN-AI is better technology, I suspect.

This is the Long-Wave picture. This particular business has been in operation a long time, and paid it's dividend throughout the Great Depression of the 1930's. The so-called 2008 "Great Recession" was a trivial burp, compared with what the 1930's looked like. But even the 2008 "burp", took our target's price down 60%. Equities - and life - are risky business. But if the process is good, and one stays focused, and does act unwisely, good decisions are usually rewarded.

(Older results... before NN-AI developed... ) Chart of example raw price-data series, with 20-day simple moving average, showing breakout and run up to a "resistance" level.

Neural Network output chart (Actual & Predicted) for old D-Mark data (late 1990's), after 90,000 interations, using steepest descent, fixed step search. Basically, just curve-fitting the time-series. Not really very useful, but shows success of the Xerion backpropagation algorithm running on the Linux platform. Chart produced in GNUplot, viewed via Ghostscript.

This is just a chart of today's action in the Cdn-Dollar. It's like the output of a square-wave generator - zero rise-time, a bit of noise on the top flatline, and then back to where it started, 100% mean-reversion. Floor-traders on the old (and very sophisticated) commodity exchanges many years ago, would call this "Shake Da Money Tree!", or sometimes "a gun-run", (from "gunning for stops"). where the market would zoom in one direction, take out all the stop-loss orders, and then zoom right back to where it started from. This is not a random process or phenomenon.

Long-waves of Gold Prices in $US, April 1, 1968 to June 9, 2017. This picture is as much about US economic policy, as it is about the valuation of a physical commodity. Reminds me of a Cheryl Crow song, "Run Baby, Run..". If you look at long-wave of oil-prices, it gets even more wild.

Here is same Gold Price series (US $), for last 175 days, (to June 9, 2017), with a 7th-order non-linear least-squares curve fit. If you look at enough of these price charts, you end up becoming an "Elliott Wave" aficionado. But the key thing about gold, is that "Gold ist gold" - as the world changes, gold does not. It simply remains as gold. The Chief Economist of the Bank of Montreal was recently quoted as pointing out that as measured in ounces of gold, the prices of houses in Toronto (Canada's largest city), although high in Canadian-dollar terms, are in fact *lower* than they were at their peak, in 2005, when priced in ounces of gold. Douglas Porter's May 26th note indicates that in 2005, the average house price in Toronto was 655 oz. of gold. In May of 2017, this was down to 540 oz. of gold. What of course has happened, is the value of the Canadian dollar has shifted down sharply, because of policy actions on the part of the Canadian Government and the Bank of Canada. They elected to substantially deprecate the value of the currency. Devaluing currency to boost economic activity is a very old (and effective) economic strategy, but it always results in a wave of inflation, which first shows up as rising asset prices. Of course, markets often move counter to the known trends, as wise players play the other player, rather than just the economic trend that all players know will ultimately play out over time.

The long-wave picture of oil prices (WTI Cushing-Hub, Jan. 1986 to June 12, 2017 in US$) is one of the most interesting economic times series charts ever, in all of economic history, even going back to Roman times. At $USD 120/bbl, everyone knew oil - the most important common commodity we all use - was priced too high - but many traders lost badly, if they shorted at $120, only to watch the market climb to $140/bbl. We must *always* remember, that the *VERY BEST CURE FOR TOO-HIGH PRICES, IS TOO-HIGH PRICES!*. Markets actually work *really* well. They do their job with the beautiful efficiency of a mass/energy conversion device - in that amazing, beautiful year of 2008, oil was repriced - by global consensus of all the market players on the entire planet - from $140 USD/bbl, to $40 USD/bbl. Now just think about that for a minute. Think what actually happened. The whole world agreed to take $100 USD/bbl off the price of oil - a commodity we *all* need to heat our homes, and power our cars and aircraft. Think about it. The world got together, and re-priced oil *down* $100/bbl - by agreed action - no military force, or government law was needed. Markets work. They work better than the governments and ABC's (Agencies, Boards and Commissions) of the world. And they do this amazing thing, without war or violence. And, when oil got too expensive *again*, at the end of 2014, the world oil market RE-PRICED OIL DOWN AGAIN - knocking $70 USD/bbl off the price. This chart should be used to teach Econ. 100 in every University course, in every school, in every nation, from Lake Geneva, to the Finland Station. ;)