[thoughts] On Predicting when Bullion will Return to GLD (premise)

[just a quick post] I've been chasing a lot of things lately, but primarily I've been trying to unlock patterns in the GLD bars list data. The 'dark bullion effect' is easy to measure but difficult to understand. Below is an infographic that I've been meaning to release regarding the 2013 calendar year. It is significant because for the most part of the year, most of the new 'additions' came from bars which had already been seen in the inventory. This is not specifically surprising by itself considering that so much had been removed, but it just demonstrates there is a really large stock of gold in London and that it doesn't necessarily all vanish instantly to China. In terms of visibility, interpreting the gold market through the ETF volumes alone is like trying to figure out how a department store operates by studying it through the retail window display.

In any case, there are a bunch of patterns which are consistent - for example the 'rate of return' for bars which we have measured over the years. Basically if a bar will return then there's a 90% chance that it will do so in the first three months. Note: as per. above, not all gold makes it 'back' but some does.

I've often said that models which have no specific predictive ability are virtually useless, so I want to start predicting some things on the public record and we'll see how we go. For the barlist on the 3rd of April 2014, some 80 x AGR bars were removed from the list and I think we'll see these back again in the list by July 30th. The serial numbers are:
208277, 208278, 208280, 208283, 208284, 208285, 208286, 208287, 208288, 208289, 208290, 208291, 208293, 208294, 208296, 208297, { sequence gap } 208362, 208363, 208364, 208366, 208367, 208368, 208369, 208370, 208371, 208372, 208373, 208374, 208375, 208376, 208377, 208378, 208380, 208381, 208382, 208383, { sequence gap } 208452, 208453, 208456, 208457, 208459, 208460, 208461, 208462, 208463, 208464, 208466, 208467, { sequence gap } 208516, 208517, 208518, 208519, 208520, 208524, 208525, 208526, 208527, 208528, 208531, 208533, 208534, 208535, 208536, 208539, 208540, 208543, 208544, 208546, 208548, 208550, 208551, 208553, 208554, 208556, 208557, 208559, 208560, 208561, 208562, 208564

If you are reading this, I do apologize because this article won't assist you directly with respect to your trading or day-to-day living - it's just purely me writing about a piece of the puzzle which is vexing me, and putting forward a prediction about flow, and hoping there might be some discussion about it. I'll write a follow-up piece mid-year along on whether the prediction was wrong or right, with more discussion about why I think those bars specifically will return.



Bron Suchecki said...

Now thats a prediction, right down to specific bar numbers.

I'm guessing the reason you think those bars will return is because you've seen those bars come and go a few times, indicating that they are up the back of the vault and don't physically move?

Warren James said...

None of the bars which have been delisted in the last week, have a dark history - but statistically some of the bars which 'leave for the first time' will be identified as dark later, and I think this is one of those sets. The current loose criteria is:

1. Block of whole pallet
2. Early inventory (not recently moved), therefore (yes) more likely to be up the back of the vault.
3. These bars seem to have been selected from the spreadsheet, rather than their physical location. i.e. their physical layout is slightly distributed so less likely to be a vault movement.
4. Not a Johnson Matthey bar (those ones tend to be removed frequently).
5. Not .9999 bars.

One of the problems with my theory is that there doesn't seem to be any specific affinity to the gold (e.g. no support for coat check theory), which suggests the dark movements we see are purely coincidental - i.e. a statistical anomaly more than anything else. However, with this I am attempting to put some boundaries about what can be generally expected.

If the bars noted above, return on a 30-day multiple then this is also significant.

Bron Suchecki said...

I notice there is a number of gaps in the bar numbers, like 208279 or 208563.

Do these numbers exist in the bar list? If so then definately someone selecting bar with the purpose of trying to get to a specific number of ounces (to equal a redemption request).

Warren James said...

Good point Bron. Yes, there are stacks of bars between the numbers in that sequence:

48 bars numbered 208298 - 208360
21 bars numbered 208397 - 208445
40 bars numbered 208468 - 208515.

What I can tell you is that the records 'removed' physically just happen to be the first 80 rows of 'AGR' bars (of which there are total of 3972). Page 47 of the April 2nd document.

So that confirms my theory of a spreadsheet-based removal for those records, not a vault-based removal. Someone has set the data filter to AGR bars and selected the first 80 rows. It suggests that the order of the spreadsheet records were changed since the bars were originally added (which makes sense because those bars are pre-HSBC) But it leaves us with more mystery - what the heck does that mean?

btw, it does seem there are specific types of removals emerging - two are classified here just agove. Another distinct type of removal is the .9999 bars.

Bron Suchecki said...

I wonder why, if the redemption was never going to be physically shipped from the vault, did they choose AGR bars rather than just any sequential list of bar, whatever the brand?

Warren James said...

I think it's because out of the 56 brands identified, 'AGR MATTHEY' appears at the top of the list alphabetically. i.e. those 80 bars seem to be the first of a 'set'.

Bearing in mind that the bars weren't 'removed' from the first 46 pages of the spreadsheet document (i.e. the selector ignored the first 2000 bars) this may give some clues as to 'don't touch / okay to move' inventory allocation, bearing in mind we guess there are other columns in the spreadsheet.

Because of the value of clues like that, I may have to go through and classify each removal in these terms to get the bigger map - but that's a lot of work, even for my personality type ;p Establishing the metrics like these helps the process but I think it's a completely new science.

Manly said...

HSBC Gold Vault: Thought for the Day

Can a sporting analogy (or two) help determine the size of the HSBC London gold vault? Possibly, but it's not as clearcut as you might think.

In the recent past, various sporting venue analogies have been used to convey the size of the HSBC vault.

The London vault space has been variously described as "not quite as big as a cricket pitch" and "about the size of a football field".

At first glance you might think "great, so it's about that size then', and then go back to your daily routine. However, further analysis reveals that in the world of football and cricket, size isn't everything.

Notwithstanding that a 'cricket pitch' is the 22 yard strip between the wickets, the attributed quotation was presumably referring to the 'cricket field', however, it is worth noting that there is no standard shape of a 'cricket field', let alone standardised dimensions.

While not wanting to get into mind bending ICC technicalities (I will leave that to aficionados of The Gabba), suffice to say that the field could be circular or oval with a variable diameter of between 450 and 500 feet on the 'long' side (sometimes giving 16,000 sq yards).

Turning to the 'football field', it is not clear whether this analogy was intended for a US audience or non-US audience. So it could mean 'American' football, or soccer or rugby, or even an outside chance of compromise rules Aussie-Gaelic football (but I presume it meant American or soccer).

Again, in soccer, there is no standard size 'field'. The sidelines (touch lines) have to be between 100 and 130 yards (110 to 120 yards for international matches), while the goal lines (end lines) must be between 50 and 100 yards (70 to 80 yards) in international matches. This could result in over 7000sq meters or over 1.75 acres apparently (I haven't checked the calculation).

The American field is thankfully standardised, being 120 by 53.33 yards or 6400 sq yards.

So where does that leave us? As clear as mud you might say...and you'd be right.

Seriously, it does though give a flavour for permissible dimensions. I'm thinking rectangular based on Bob Pisani's video and I can't see it being as big as a soccer pitch, again given the video 'tour'.