A quick GLD Dark Bullion Post (updated)

During the last week we mapped the Dark Bullion in GLD, similar to our SLV study. I've always been curious about what was happening with the inventory during August 2011 during that magnificent price peak, and here is an interesting look at that month. If you haven't already seen the mechanic and terminology, please have a read of SLV Database #5. The black band is bullion which was removed from the bar list, but which we confirm has been added back at a later time. Additionally, there is some color to show which of the inventory has a 'history' of being dark, for both visible and lost. The scale (measured in Fine Oz) has been adjusted to highlight the movements, but bear in mind the dark band only represents a small percentage of the total inventory held.

August and September 2011, clearly showing that the inventory from
the big million-ounce removals were actually returned to the GLD
inventory at a later date.

What I've tried to cobble together here is the relationship between price and inventory, with the price offset by 3 days to try and match the trade+settlement lag (of 3 days). At this low level, showing just 2 months worth of data on the screen, the weekends and days become significant. Here's the same graph, but showing the limit of our resolution - with each column representing a bar list document that we have processed, which may reveal a bit more detail compared with the stacked area chart.

Same graph for August + September 2011, but with stacked column,
to show the source data better. A full 5-column stack represents one
bar list for every day of the week. Sometimes we don't see a new
bar list issued every day.

The main take-away point here is that with both of those steep inventory declines in August 2011, most of the bullion removed was actually returned at a later date. I find it interesting that the overall amount of metal in the vault didn't change a whole lot during this time and if we ever get a substantial uptick in the GLD inventory in 2013, we may be able to do the same kind of dark bullion identification for the 'shedding inventory' (but until we get more bars added we won't have that critical 2nd piece of info). This particular trough formation is unique, we don't see enough volatility in the data to see anything similar elsewhere.

As we move towards better charts we might be able to visualize the exact entry and exit points of each bar, but it still remains a challenge to display that kind of information density. Unfortunately the low percentages of dark bullion means we've found no specific correlation just yet. Additionally, the annoying time lag involved (because we need two points to prove a bar was dark) limits any ground-breaking analysis on a short time frame. That's all for the minute as I've got a lot of data to check. There's a lot of ways to slice and dice this data.

----- updated: 2013 June 02. VtC asked for confirmation about the time lag between trade settlement and bar list information. Just for the record, we typically get GLD trade information from this spreadsheet, while the bar list PDF (settlement) is here. We took the time in May to chart the relationship between the bar list totals (for Fine Oz) vs. the inventory specified in the spreadsheet, this is what it looks like:

GLD time lag between CSV and PDF (@17th May 2013)


milamber said...


I didn't realize I was a week behind in my STFU reading until this am. Quick question about GLD inventory ( and this may have already been answered)

In 2013 I think there have been 5 or 6 days where GLD added inventory. All other days have been decreases or inventory stayed the same.

Can you classify those adds yet?

Thanks again for all the hard work


Anonymous said...

Warren, fantastic work again. You are writing that they settle T+3. Are you sure this is true on the gold side, too? I vaguely remember that LPCMC said they allocate/dealocate on the same day , if possible within an hour.

Second question concerns the pukes (you may have answered it somewhere, but I forgot). When GLD loses inventory, what we mortals see is the decline of net inventory on their spreadsheet. But you also know the gross movement, i.e. that some bars are added and others removed on the same day. The last thing I Remember is that Bron insisted this cannot happen because the trustee would net the trades before allocation, but I said I'd like this confirmed in the data.

(note: haven't checked my email)

Bron Suchecki said...

VtC, the clearing is done same day between BBs, but normal gold market settlement is T+2. If I remember from the GLD prospectus they work on T+3 because that is the settlement period the stock market works on even though the gold market could settle earlier.

Re the gross movements, I still think the custodian would net the redemption and creation requests, but given the existence of 9999 bars in the ETFs, the custodian may take the opportunity to take out 9999 bars and put 9950 bars in, as they can melt the 9999 bars into kilobars to sell to the East.

I don't think the custodian, if there are no redemptions or creations, can just decide to swap some 9990 for 9950 bars.

Will be interesting to see if gross or net occurs.

I assume your interest in this VtC is in relation to the coat check theory?

Warren James said...

Hi VtC, yes we're absolutely certain about the time lag being three days. I've updated the post to show you a chart which matches the reported fine oz from the CSV vs. the PDF. The relationship is pretty clear, although there are some anomalies which pop up from time to time.

re: simultaneous add and remove, there are actually some events where a simultaneous add and remove have occurred, however in most cases this has been because of a change in signature (e.g. fineness or serial adjustment mechanic), in fact I used that mechanic to clean up the GLD data. I'll see what I can do to get the remaining items out for you, but just so you know the situation is not the norm so it's expected to be a data anomaly more than a legitimate signal. I will email you this week with those details as you are interested.

Hi there Milamber, yes you're right, there have been only 6 'add' events in 2013 so far. We did chart every single one of these for the last couple of years, along with the percentage of dark bullion detected. I've uploaded the spreadsheet here for anyone interested. Please note that because of the nature of the dark bullion measurement, these figures will always change as we get new data - so this one I will try and keep updated. p.s. anyone who detects a pricing signal in this data, please share it with me!!

Warren James said...

NOTE: I should have clarified for Milamber - the number of 'Add events' for the CSV and the PDF do not match, even with the 3-day offset. Part of the reason for this may be because the spreadsheet is produced at a different time of day compared with the PDF, but also because the PDF is sometimes issued more than once per day (our system keeps the last-most-copy for a particular day).

The discrepancy can be seen in the chart above, but specifically this translates to:

1. 4th-Jan-2013 CSV Add Event (+58,100 oz) was not detected at all in the bar list data.

2. 20th-Mar-2013 we detected a PDF Add Event (+280,591 oz) which was not represented in the CSV file.

3. 9th-May-2013 CSV Add event (+87,027 oz) was not reflected in the bar list data, unless you want to say that showed up in the 13-May PDF increase of 38,738 oz.

4. On Monday's bar list we are hoping to see the added bars from the 29th May 2013 CSV Add Event of 29,002 oz (but I don't think it will show up).

It's my opinion that the discrepancies aren't particularly significant and can be explained by bookkeeping lag or wrong data. i.e. it's possible that a mistake was made in the bar list reconciliation which was then fixed the next day, OR that our system (which checks every 24 hours) didn't download an updated bar list which might have been produced towards the end of the day. I could configure the system to download every hour in order to capture every detail but I would wrestle with my own OCD on that one.

Bron Suchecki said...

Maybe just need to ensure the time at which you do your each 24hr check is toward the end of their daily settlement processes, which I'd guess would be end of day London.

I am still mistified at the multiple bar lists during the day.

Warren James said...

@Bron, because their process uses a spreadsheet, creation of the bar list is still be a manual exercise and therefore subject to (a higher amount of) human error. Contrast that to SLV with it's 7.00am automated report.

But yep, it does occur -> more that one bar list specifying for the same day of settlement. I always figured that the first one was less authoritative than the second and so the system was configured to discard the other one ... so unfortunately we don't even know how many times this has occurred in the past. I just wish they would put more internal faxes in place of the bar list ! :)

Warren James said...

I should hasten to add that (from memory) the duplicate-day-document was a rare occurrence.

I might be able to check the file system for any deleted 'duplicates' files.

I do regret now not putting in an 8-hour check mechanic. It's possible that some of the occasional gaps in my bar list data are a result of the list being published late, outside the 24-hour check. But there are definitely some days the bar list is not updated too ...

(I guess that's part of this data challenge - working with partial sets of information).

Bron Suchecki said...

It shouldn't matter if you miss a bar list given the number of lists over the years and given the number of bars in total. A small error factor.

costata said...

Hi Warren,

Can't thank you enough for all the hard work on this project. I'm not usually inclined to rely solely on gut-feeling but I can't shake the feeling that this project is inching toward a major breakthrough.


Warren James said...

Thanks Costata.

For anyone keeping track, the 29,000 oz addition to GLD last Thursday, showed up in yesterday's bar list document. Addition of 73 bars but curiously enough, ZERO of these were from dark bullion (i.e. 0%). That's weird because it's a different result from previous add events. One would suppose that out of so many tonnes removed, that statistically they would add some back (background effect is 30% rate of return). However it might just mean they had a pallet spare which was 'near the door'

The added bars are from Krasnoyarsk and Johnson Matthey Canada - they appear to be recently produced bars - e.g. the serial number for the (21) Russian bars start with '2012' denoting the year of manufacture.

ALL 73 Bars added are 9999 fineness. Still trying to figure out the significance of this and the correlations, but it's weird (for comparison, 36% of GLD's holdings are 9999 fineness).

Warren James said...

p.s. I still need to check those Russian bars to see if they were known with a different serial number (sometimes the year is prefixed and sometimes it is not, resulting in a different signature). If anyone has a theory regarding 9999 bars, I'm very keen to know (and in return I can share database extracts).

answer2me said...

Pictures of serial numbers on some deep bullion.


Bron Suchecki said...

The interesting thing is our dealers are saying the premium on kilobars collapsed when gold moved up above $1400, and when did that happen – on Thursday the 30th when you had net buying. Classic case of Westerns buying on a rise and Asians holding off.

So why did the BBs add back in 9999? Maybe they had some 9999 400oz bars lined up to be melted into kilobars for the Asian market, so with Asian demand down it doesn't pay so they allocate it into the ETF. However I would still have thought they would have preferred to add in 9950 bars rather than 9999 bars - once in the ETF the BBs can't get them out unless ETF holders become are net sellers.
The fact that the addition is zero dark and all 9999 gold is very anomalous.

costata said...


Are these the first Russian bars added recently?

Warren James said...

@costata, yes and no. Less than 1% of the bars in GLD are of Russian refinery origin. In terms of 'add' they are not particularly significant. Here are the totals from 'Add events' over the last three years in GLD, for your reference:

DatePublished | Added Russian Bars
2011-06-20 = 722 (records start)
2011-07-20 = 52
2011-07-22 = 14
2011-08-02 = 14
2011-09-14 = 5
2012-05-10 = 2
2012-09-25 = 10
2012-12-27 = 174
2013-03-21 = 10
2013-03-26 = 10
2013-06-03 = 21

p.s. I checked the Russian serial numbers it seems these aren't 'changed signature' bars, so my 0% dark bullion count still stands.

p.p.s. @answer2me, thanks for the pics - those are the old US melt 'bricks' and unfortunately there's not enough info on them to correlate anything, but these have gone into my digital file.

costata said...

Thank you Warren.