The last few months of removals has started cutting into GLD stock originally accumulated more than two years ago (i.e. this is shown where the corner meets the inventory curve), whereas that wasn't the case in February. For our 'overall' measurement, the dark yellow line has moved DOWN (from dotted line) since last month, so we can say that over 1,000,000 ounces of gold removed in the last month, was taken from stock present prior to June 2011. But while this is interesting, it is no particular conspiracy as it just reflects the fact a lot of gold is being de-listed.
|Also bear in mind this chart summarizes a lot of detail. Don't read into the figures too much.|
Speaking of summaries, here's an updated info-graphic showing GLD activity for July. There was an 'add event' in the bar list data (with most of those 'new' bars originating from ETF Securities stock) but the addition can only be detected from the bar list figures - if you're only looking at the trade settlement spreadsheet (link) then you won't see an increase in GLD inventory for July.
|I'm still experimenting with the layout and information content. Feedback welcome.|
"... Anyway, I call bullshit. To be consistently redeeming the exact same or multiples of the same number of tendered shares is statistically impossible, akin to getting struck by lightning repeatedly." - Turd Ferguson (TFMetalsReport.com)His angst on this particular issue started with three particular withdrawals each for approximately 1.5 tonne (on TFMetalsReport here and here), with another similar just recently in the first week of August. I have no knowledge of 'tendered shares', and as such I'm not going to tackle his carefully crafted statement. What I am interested in however, is the propensity for 'redemptions' to occur in tonne or half-tonne blocks. To begin deconstructing this, let's look at the data:
|Bar List data Changes for July 2013 (added/removed)|
Is it common for redemptions to use exact amounts? Using the database I have assembled a rather crude copy of Victor's Histogram, which highlights (in blue) the removals mentioned by Turd. I hope this demonstrates that statistically, when charted against the backdrop of their fellow data points, these seemingly suspicious amounts actually have nothing remarkable about them.
|Chart showing number of bars added/removed between mid-2011 and mid-2013.|
Some limit of error reading does exist since the data needs a bit of cleaning up.
[Total source table size is over 65,000,000 rows.]
Save your breath on threatening me ... I'm after a much bigger prize and I did promise 'a different view'. One of the things which stands out in the data is that removals are often from arbitrary sections of the bar list. A single day's removal events are often constructed of multiple contiguous blocks in the bar list data. We know this because our database records the line number and page number - we also know for a fact the data is based on Excel, so removals are basically the removal of rows inside a spreadsheet. Treating each individual segment as a separate 'bullion removal' event, changes the graph!
|A different view of GLD! Pallet based removals!|
Showing all the events side-by-side. Interesting that the very large removal, was
still one continuous block in the spreadsheet data. Many entries are multiples of 80.
The 1.5 tonne items show up in the data as separate bullion blocks, I've left them highlighted blue so that you can decide for yourself whether they are still noteworthy. As you can see, blocks of 80 and 40 bars are incredibly common in this view, that is 1 pallet and ½ pallet respectively (and 80+40 bars = approx 1½ tonnes). Bron Suchecki has forwarded me an explanation as to how this would occur in relation to vault operations and I'm quite happy with it (the explanation) **; later in part 3 we will chart these individual 'redemption blocks' as our 'bars selection of interest', revealing their own unique stories.
UPDATED 13th August: I've had feedback about whether the charts above are histograms or not. They aren't, although they look like it and was inferred. The ambiguity was meant to reflect the lack of precision available, for example the very large 'contiguous blocks' don't show us pallet-level information (though the smaller ones do). The y-axis is 'number of bars', the x-axis represents an individual event, with the highest amounts on the left, smallest amounts on the right. The graph is meant to demonstrate as a quick visual only. Here's a bit more clarity on the last graph, you can see that many of the events come in multiples of 80 ... e.g. there are 14 individual events of exactly 160 bars (2 pallets).
|Close up, showing example of multiples of 80|
|Actual database output isolating the Turd entries.|
Does the separate half-tonne pallets indicate these
were separate customers/recipients/participants? Not sure.
NOTES: These graphs DO have limit of reading errors, approximately ± 3% because some 'changed signatures' are still showing in the older data as removals. The 'contiguous blocks' analysis (the last graph) is a brand new analysis set, so it's still really primitive and I expect to gain better detail over time. This is also subject to it's own limit of reading error if they do weird things with the spreadsheet records (e.g. shuffling entries). The 'contiguous blocks' chart does not currently incorporate any dark bullion measurements.
We welcome criticism, in particular pointing out any flaws with regards to analysis (because it helps refine it), just be polite.
** Updated 12th August --> See Bron's fine comment below as a great example of how the 1 tonne batches arise in a custodian environment. This is a refined version of the information he shared originally with me.
Bar List Random News:
missing 871 pages - similar to the October 2009 anomaly noticed by Rob Kirby* but this time it went completely unnoticed by the metalsphere. I analyzed the missing bars but couldn't find a pattern, the header repeats itself on page 294 so it looks to me just like a 'print-to-pdf' error. The mistake was not present in the bar list the following day.