Screwtales: Fed Gold and German Tanks

Satisfied that GM Jenkins has the skinny on gold price movements, I've been busy in the other cellar trying to finish a new database article. A few readers asked me to look at the National Geographic documentary titled 'Inside Americas Money Vault' (link link), specifically regarding bars shown in the first section of the clip - Bron Suchecki added a few more observations which made a small article worthwhile so here it is, for anyone interested.

With the new clip, we attempted the same as the recent 'Bob Pisani Revisited' article - to see if the database can tell a story about the gold bars we see on the camera. This is normally cut and dried - there is either a match in the data or there is not. There are plenty of great visuals in the clip, at high resolution.

Unambiguous and clear Serial number - "Johnson Matthey" #163466 (2011)
In the database (specifically the WAREHOUSE_GOLD table), there isn't any match for that particular six-digit serial number. Same applies for all the other bars we identified clearly - 163408, 163437, 163466, 163481, 163497.

So our database can't say anything about the bars specifically - how about the batch? At this time, I wish to deploy a technique put forward by Jeanne d'Arc, who suggested that estimates about bar volumes could be modeled on estimation of wartime German Tank quantities (using observed numbers and probability). A decent write-up in Wikipedia exists so I won't re-explain it here - but suffice to say it should be possible to infer quantities of bullion production from a refiner by looking at their serial number sequences, mathematically modeled if required.

Happily, we don't need to delve too deep to get a handle on the problem scope. The bars appear to be from a well known run of Johnson Matthey bars - specifically 6-digit serial numbers between 100000 and 199999. In the database we have records for 55,032 of these bars (@400 oz each), most of which are currently sitting in GLD (and a handful in ETF Securities). If we assume that Johnson Matthey are simply incrementing the number each time they produce a bar, then we might assume a total of 99,999 bars for that set. It means we know the location for about 55% of those bars for that batch, and the presumed produced remainder, remains unaccounted for. So if you thought GLD was huge already, then consider the missing 45% represents about 18 million gold ounces, which is most likely out there in some form-or-other. Simple observations like this provide pure stumbling blocks for the 'shortage of metal' meme (yep, another argument against that old chestnut).*

If the Germans used Microsoft Excel, they probably
would still have had sequentially numbered tanks.

So back to the FED ... the curious thing about the records is that we do see a bunch of 163XXX serials. Our records have the bar 163394, then there is a gap - the next in the sequence jumps to 163500. Bron did the hard work of counting the bars as they get stacked and unstacked (multiple times) and concluded there are approximately 86 bars. We know that 5 of the bars fall within the missing number range so it's not too much of a stretch to assume the TV is showing the entire set of 1634XX bars.

I want to kick the tyres on some of the charting software that I've been working on for my other article**. Does anyone remember the days before solid state drives, where a computer disc had to be defragmented? This is the best analogy for describing where the refinery production ends up. Here's a custom graph showing the data points we do have, and the ones we don't. Working on the assumption of (163497 minus 163408 = 89 bars, approximately 3 tonnes) for the fed, their delivery shows up as a yellow line in the middle area. The grid is condensing the information density to show 10 bars per square, with 100 squares per horizontal line (i.e. total of 1,000 bars per row with 100 rows deep). The intensity of the color indicates how many bars we have records for that 10 bar block. e.g. the first square represents bars '16300X' of which we have the full 10 bars, the intensity is 100%. If we have only 5 bars for that square then the color ~ 50% and so on. The color represents approximately the 'Year we first saw the bar in the data', pink/red = 2010, purple = 2011 and blue = 2012, green = 2013.

Top row represents bars 100000 to 101000, pattern repeats down till we get to 199999 ***
There are a few different ways to interpret the picture. The gaps represent bar sequences that we don't have records for. Perhaps they are in ZKB, or some other fund that doesn't publish a bar list. Or a central bank, or a private holding. The date layering indicates that the majority of the records started showing up in 2011, which makes perfect sense since we started to download the files in June 2011. What this means is that the date of appearance alone is not an indicator, but the composition of the picture gives us additional clues. The presence of the pink/red bars in the middle area suggests that most of these bars could have been manufactured as early as 2010, with the last section being exclusively 2012 manufacture. My own rough guess for the production vintage is circled below.

Thanks for listening, and be sure to let people know we can't find any evidence of the FED's gold being double-counted in an Exchange Traded Fund. p.s. why is the FED getting gold delivered anyway? I thought The Bernank said it was not money ... I suppose it's "do as I say, not what I do ...". Regards, Warren

* To give some idea of the size significance of the six-digit Johnson Matthey bars, 55,000 bars is approximately 25% of all known gold bar signatures we have on record. That's huge.
** If you stare long enough at the graphic, like I know Duggo will, you may see the next Pope
*** Apologies for the state of the chart, this was the first output we've ever had from the software so it's quite raw at this time.


AdvocatusDiaboli said...

great approach, great work.

I wonder if refiners do/would provide any further information about their serial numbers or outputs (volumes)?
Have they ever been asked or in case they rejected detailed information, is there some official answer why they might not publish those information?
Greets, AD

Biosci said...

Warren, this chart is brilliant. To my eye it carries three main points: first, and most obvious, is that you have data for a HUGE fraction of this block of bars.

Second, bars travel in consecutive bundles. There are fairly staggering statistical implications of this, not least that you might be able to impute transaction sizes.

Third, that you might be able to make some informed prediction about the rate at which bars are being removed from circulation. The white blocks get bigger as you go forward in time; determining whether this is selection bias (there's been more opportunity to observe older bars) or real signal (production is going elsewhere, outside of your monitoring) would be very interesting.

Two quick questions: (a) are your dates correct? Should purple and red be switched? and (b) Do you have any info on the rate at which bars are melted and recast? That is an alternative explanation for some of the gaps.

Brilliant work. Analyses that beget more questions are always the best ones.

Anonymous said...

Wow - I can't believe you actually used my German tank idea..! Chuffed :-) :-) :-)

Great article, Warren.

I suspect that your colourful chart - far from revealing the next pope - will more likely than not eventually settle down into the perfect image of a grinning prosimian. At that point, the PM world will know that it has been bested by the STFU.


Anonymous said...

I take home that there are vintage 201x bars in the FRBNY vault that holds
1) a tiny portion of the U.S. gold
2) some 6000 tonnes of gold that belongs to foreign governments, central banks, and international organizations

In my article on "Central Bank Gold Leasing", I plotted the inventory under (2). It has been declining until about 2000, then constant except for a 400t loss around 2008/9. I don't see any reason why fresh bars would enter component (2) of the inventory.

So we conclude that the fresh bars belong to (1), correct?

Warren, this is f***ing brilliant!

PS: That vault holds *only* official gold, they don't store anything for private entities (they have always had to go across the street to JPM).

Warren James said...

@AD, according to Bron Suchecki refineries are very guarded about their output numbers and volumes because of the competitiveness of the industry - to the point where some bar serial numbers don't appear to contain any sequence or is otherwise codified.

@Biosci, thanks. Yes, the dates are correct, the 2010 bars are from Goldmoney records ... what we're seeing here the outer edge of what's visible with the data - there's a limit of reading error involved and the best we can do is to get aggregate visualizations. If we had every document starting from 2006 then the resolution would be a lot clearer. As we gather more data points we'll get a better picture, which is why the last part of the chart is very clearly distributed. Yes, we now have a mechanic for tracing appearances and disappearances of bars - it is incredibly complex and is the subject of an article I've been writing since last year.

@Jeanne, yep :) Your idea was way ahead of its time - we are only now starting to process the 'sequences' and we hope that the tank estimation formula will help on the other 'missing' figures. Nice work.

@VtC, yeah, weird huh? . btw, apologies I am swamped with work and can't process your GLD enquiry but stay with me - my plan is to publish the Dark Bullion findings on SLV and then discuss the Dark Bullion calculations on GLD internally (you're included in that) ... I promise it will be worth your time.

Bron Suchecki said...

First, excellent work. I'm just hoping the refiners and bullion banks never find out about this as they may feel frequent bar listing publication gives away too much of what goes on.

Refiners generally will not say what serial numbers were made in what year as it does give away production volumes to competitors.

Re consecutive bundles, for gold in the wholesale market it is usually traded in tonnes, so maybe each square should be 32 bars? Ten per square probably gives better visibility however. Anyway, so three squares in a row is one tonne.

Re the amount of white space, I think the more oportunity to observe older bars accounts for less white space in the early bar numbers compared to the later bar numbers. The interesting conclusion from the fact that there is less white space in the early bar numbers is that bars hang around. If JM's output was "consumed" in jewellery or coin or small bar manufacturer by others, then there would be a lot of white space in the early bar numbers.

If bars sit around in London just changing who owns them, then over time we should have those bars appear and disappear from the ETF lists and fill up our white space.

As time progresses, if we find a sustained section of white space then we could conclude that around the time of the production of those serial numbers there was a lot of non-London investment demand for metal.

The amount of information this data can reveal I don't think we realise yet.

AdvocatusDiaboli said...

"Refiners generally will not say what serial numbers were made in what year as it does give away production volumes to competitors."
hmmm, but at least the Münze Österreich AG publishes in their annual report, exactly how many of what products they sold in detail. I dont know if they publish this info voluntarily or if it is obligatory. So when it comes to retail business, I guess these infos are more helpful for competitors than plain serial numbers?

Anonymous said...

A glimpse into a fascinating but unobtainable World of Gold bars.

As it's raining in Antibes and just before visiting my French accountant I have decided to contribute to your Freegold section. The title is:-

What is FOFOA?

Louis Cypher said...

Excellent, excellent stuff Warren. I really like the hard drive defrag layout. With the recent outflows from the ETF's it should be informative to see how long it takes those bars to come back to the ETF's.
As Bron points out the picture is becoming clearer and the mechanics of their operations will be laid bare for the world to see.

Biosci said...

One way to confirm the German tank model is to check the starting serial number for each block of bars for which you have data. If they are uniformly distributed on [0-9] it would fit the model that you're seeing random samples from the population of blocks from [100000-199999]. If there's a bias toward, e.g., 0 or 1, it might indicate stops and restarts at higher round numbers in the serial numbering, suggesting those white blocks might not exist instead of simply being invisible to this study.

I'm a Bayesian in real life but, lacking any domain knowledge whatsoever, I'm forced to fall back onto these frequentist measures...

Warren James said...

@Biosci, I should have pointed out that a colored block indicates only that we've seen the bar at least once (a JM bar in that number sequence). There are also over 6000 bars in the > 199,999 range and also below 100,000 there are a bunch of 5-digit serial numbers we could apply the same charting process to.

The 10-bars per square was chosen for presentation purposes. Representing the 'exists but don't know the current location' is a graphical challenge of the highest degree but I'm happy to say I've got it nailed (in my next article).

If you wanted to do some number crunching on the full set of serial numbers for all known Johnson Matthey bars (108,000 bars), I can send it to you. Regards, Warren

costata said...

Once again Warren, kudos for your sleuthing and hard work on this project.

I noticed this then discuss the Dark Bullion calculations on GLD internally (you're included in that) ... I promise it will be worth your time.

Any chance of an invitation to sit in on this discussion?

Warren James said...

Hi Costata, yes sure - I'll need your contact details though, I don't have them on file.

best contact is

Robert said...

Yes, I can see the next pope hidden in the graphic. Stare at it long enough and it becomes obvious!

Biosci said...

On a lighter note, this is an absolute must-read:

Take The Zero-Hedge Test

Slow Loris Larry said...

@ Costada

Was there a URL hidden somewhere in your suggestion:

I noticed this then discuss the Dark Bullion calculations on GLD internally (you're included in that) ... I promise it will be worth your time.

Bron Suchecki said...

Some additional thoughts about the fact that bars seem to "hang around":

1. When investors sell their allocated BBs buy it off them and then short futures (explanation of the big short?) as the BBs expect/know that investors will come back eventually and the BBs want to have the physical inventory to meet that demand. Even better if there is contango as the futures market "pays" the BBs to carry the inventory.

2. There is a lot of turnover by investors in the physical market by that meaning there is a big pool of more speculative investors who hold 1000oz bars and trade it.

3. There are a lot of investors holding unallocated which the BBs cover/back with at lot more 1000oz bars than people think ie the OTC London market is less fractional than belived.

Warren James said...

So this week Zero Hedge and TFMetals report have picked up on the National Geographic video (yep, it's the same clip). Reminds me of the time Silverdoctors breathless announced they found Bob Pisani's mystery bar, 9 months after we identified it.

Who cares ... anyway, further to the Johnson Matthey Sequence, there is evidence to suggest the 5-digit number sequence (which later morphed to six digits) started in 2003, so this gives us an idea of the overall layering of that stack. It suggests that the yearly bands are much older than we see here, and may give us a chance to date this very large sequence. If I get the chance I will expand out the Johnson Matthey Sequence Graphic to show the serial numbers prior to 100,000, as well as the 20,000 or so which appear after 200,000. Regards, Warren