PM Daily Market Commentary - 7/22/2015

davefairtex
By davefairtex on Wed, Jul 22, 2015 - 6:06pm

Gold fell today, dropping -7.40 to 1092.80 on heavy volume, while silver dropped -0.05 to 14.77 on moderate volume.  Gold mostly retreated today, making a low of 1085.60 before rebounding, while silver mostly traded sideways within a range.  The gold/silver ratio continues to improve - silver has been outperforming gold for the last four days.

No buyers yet for gold.  The dollar rallied modestly, and that didn't help, but based on how things are going right now, it would appear that gold is destined to re-test the 1080 lows.  My sense right now is it will most likely fail.

Silver once again looked better than silver.  Silver sold off, but rebounded and recovered most of its losses on the day.  Same conditions remain: silver must close above 15 to mark a swing low.

Miners did all right on a day when gold was having problems, with GDX down -0.28% on moderate volume, while GDXJ fell -1.29% on moderately heavy volume.  It looks like there are at least some buyers for the miners down here at these levels.  The fact that GDX made a new low today makes it much easier for GDX to print a swing low tomorrow; all it needs to do is close above 14.28, which isn't all that far away.

Platinum was dead flat on the day, printing a doji and closing unchanged.  Palladium was down -0.06%.  Copper was hit hard, dropping -1.96% and appears to be heading for a retest of the 2.38 low set two weeks ago.

The dollar rallied today, rising +0.30 to 97.72, finding support on the 9 EMA.  The dollar remains in a strong recent uptrend; on the flip side, the Euro appears to be struggling to break above its 9 EMA; if the Euro cannot move higher here, it likely means the dollar will continue to rally, which will continue to pressure gold and commodities.  Momentum continues to point to a dropping Euro; that's not bullish for gold.

SPX continued dropping today after printing its swing high yesterday, dropping -5.06 to 2114.15.  It was not much of a drop, and it appears that SPX found support on its 9 EMA.  VIX fell -0.10 to 12.12.  SPX appears to have a strong bid underneath it.  My guess is any correction will be mild, and will eventually lead to new highs.

Bond ETF TLT had another good day, climbing +0.62% and closing above the 50 MA for the second day in a row.  Another few days of this, and TLT will break out of its two month consolidation.

The CRB (commodity index) fell -1.28%, a relatively large drop, setting a new low for this cycle.  CRB is just a few percentage points from its previous low, set back in March.  Commodities overall look weak, which is continuing bad news for PM.

Contributing to this is WTIC (oil), which dropped -1.40 [-2.77%] to 49.21, dropping below $50 for the first time since early April.  Oil equities are being sold hard - they don't like the falling oil prices very much at all.  WTIC's break below 50 leaves it vulnerable to a potential retest of the lows set in March: 42.41.  That would be an ugly outcome, especially for related oil equities.  Oil services have already made new multi-year lows, and the E&P stocks are not far behind.  It appears that oil stocks were bid up on the expectation that the lows in oil would be brief; this recent drop has not been kind to the oil stocks at all.

Commodities continue to be weak, gold is looking to retest 1080, silver is holding up relatively well, as are the miners.  The commodity complex probably holds the swing vote here; if commodities can't rally soon, we probably make new lows across the board in PM - and its likely a strong dollar would also cause that same outcome.  This is why although prices are very low right now, we must wait for the market to show us that it can reverse direction before jumping in.

Note: If you're reading this and are not yet a member of Peak Prosperity's Gold & Silver Group, please consider joining it now. It's where our active community of precious metals enthusiasts have focused discussions on the developments most likely to impact gold & silver. Simply go here and click the "Join Today" button.

4 Comments

Nate's picture
Nate
Status: Platinum Member (Offline)
Joined: May 5 2009
Posts: 572
Armstrong

Armstrong's view on gold.

http://www.armstrongeconomics.com/archives/35296

davefairtex's picture
davefairtex
Status: Diamond Member (Offline)
Joined: Sep 3 2008
Posts: 5058
no buyers again today

PM took off in London, but traders took that as an opportunity to sell.  Miners are now making new lows.  No buyers yet.  Oil dropped, copper dropped, commodity index overall dropped.  No good news.

Armstrong has a "weekly reversal" for gold at 1084.  That is, gold needs to close at or above 1084 on Friday then he thinks we get a 2 week bounce.  Otherwise - we will get another leg down.

This is why we wait for the market to demonstrate a reversal before jumping in.  Low prices can always get lower.  And then they can sell off some more.

I'm telling this to myself as much as to the rest of you!  I surely thought we'd have bounced by now, but that's just not happening.

HughK's picture
HughK
Status: Platinum Member (Offline)
Joined: Mar 6 2012
Posts: 760
Neural networks: Care to comment, Dave?

Hi Dave and all,

After a quick perusal of some posts on Socrates over at casa de Armstrong, I recalled that you have mentioned that you have developed or are developing some neural networks.  If you have any inclination to share more about how difficult this was and how much coding you knew or had to learn, I, for one, would be very interested.  

And here in the Amazon Queen of Digital Natives, speaking of her experience making a neural network as a high school student in order to help detect breast cancer.

davefairtex's picture
davefairtex
Status: Diamond Member (Offline)
Joined: Sep 3 2008
Posts: 5058
machine learning & markets

Simple answer: its not easy.  You have to understand both how markets work, and how machine learning works.  A simple thing like throwing prices of everything at the network and expecting it to predict the future won't work.  I'm not going to reveal my particular secret sauce, but I was able to train my network to incorporate a group of technical indicators and come up with an answer.  And the answer makes sense.  I look at the indicators and I understand why the net gives me the answer it does.  The nice thing about it is, it takes the emotion out of chart reading.    But its a tool, not a magic crystal ball.

Interestingly, the question you ask the network - and the way you structure the question - is as important as the way you structure the inputs you feed it.  And for markets, neither are particularly easy.

A much easier task was a recent experiment I tried: peer to peer lending.

The Lending Circle is a peer-to-peer lending company which provides data on the performance of every loan originated.  In a weekend, I was able to train a network to cherry pick a particular grade of loans (LC grades loans by "A", "B", "C", ... "F"), such that (at least on historical data) my returns over time were more than double that of simple random selection.

The keys to solving the problem are:

  • phrasing/structuring the question properly
  • collecting a set of inputs - larger the better - that most likely will be instructive, and then formatting them properly
  • running your simulation and seeing if the network converges
  • iterating over your inputs - adding, deleting, combining, or otherwise transforming them so the machine learns better.

Lending Circle data was relatively easy, since you can ignore the time dimension.  The loan either goes late and eventually charges off, or it remains current and eventually gets paid off.  Its pretty binary.  Happy loan/unhappy loan.

I'm not running out and loading up on Lending Circle loans, however.  My imputed returns were ok, but it looked like a whole lot of work to get myself (perhaps) 6% ROI annually... and if we assume the debt bubble will pop sooner rather than later, buying the equivalent of relatively illiquid subprime debt with a 3 year duration is probably not top of my list of assets to own...

It was cool, however.  The (historical) loans my network picked had a chargeoff rate of 1.8%, while the broad group of that same grade of loans had a rate of 9.72%.  But the network only selected 1 of 9 loans to fund.  True cherry picking.

As for coding requirements - you just need to be able to manipulate large data sets and reformat them into a configuration file that your machine learning package (free on the net) can understand.  Perl or python, with some shell scripts to run everything.  If you don't know how to process csv files using some kind of software, its gonna be a long process to do by hand.

But learning how to use that machine learning package can take a while. 

Last point.  In trying to read about Armstrong's machine, one of the things he does is correlate everything with everything, to see what sorts of things are interacting.  Correlation is a relatively simple algorithm, but getting the data is the hard bit, and then running the correlations and presenting them in a meaningful way is hard too.  Armstrong talks a lot about what his machine says and that it discovers things on its own.  Simply put, if you correlate (say) gold to a bunch of other things, and then you sort the correlations, you get to discover things too: you get Silver, inverse dollar, the Toronto Stock Exchange, Crude oil, and the DJ Utility Average, in that order.  Negative correlations are: NIKKEI, DJIA, USD.CNY, the 10 year treasury yield, and the Mexican stock exchange (!) in that order.

But correlations change over time.  And how do they work with capital flows?  And his energy models?  And his predictions in time?  He has it all working together.  Correlations are an important input, and in some sense they provide the "discovery" (and/or) learning aspect, but I have no idea how all his stuff fits together.

I'd sure like to know though.  :-)

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
Login or Register to post comments