MineFund Independent Natural Resources Analysis & Analytics

Are Gold ETF Bar Inventories to Be Trusted?

ST. LOUIS (MineFund.com) -- Even before the World Gold Council’s benchmark exchange traded gold fund opened for sales it was heaped with several layers of conspiracy mongering. The worst charge is that exchange traded gold really has no bullion underpinning its paper.

Most of the conspiracies were offshoots of the now entirely demolished Gold Anti Trust Action Committee railings about a scheme to artificially suppress the price of gold. They were watered from time-to-time by competitors who correctly identified the threat SPDR Gold Shares (GLD) would pose to their businesses; not to put them out of work, but in siphoning away the lion’s share of the market.

Not all there

The phantom gold accusations have several forms which can be distilled as:

• the bullion inventory is fake and/or untrustworthy;
• investors don’t have unfettered ownership of the bullion underlying the shares;
• it’s not really gold because you can’t take delivery of physical product; and
• ETF investors have less rights than ordinary investors.

We confined our interest to the first allegation - that exchange traded gold funds do not possess “audited gold”. The other complaints are matters of individual investor preference and risk tolerance. So far we think the market has spoken unambiguously on those last three points.

“Unaudited gold” is a serious charge because it implies fraud. None of the accusers has gone so far as to allege actual fraud, but you’d need to be dimwitted to believe that that was not the underlying claim.

Some time has passed since the initial “gotcha” moments which revolved around duplicate bar numbers. All the duplication problems were satisfactorily resolved, but it was reasonably damaging to retail gold bug confidence because they had been softened up to expect the worst; it was literally a self-fulfilling prophecy.

Open audit

Since then the inventory of gold bars linked with an expanding number of ETF products has grown exponentially. Combined with the strict qualifications for London Good Delivery Bars, that provided us with an opportunity to subject the fake bullion claim to impartial statistical tests.

If the bullion inventory does not really exist but is artificially created and maintained there will inevitably be tell-tale statistical signals flagging tampering and manipulation. It's not too far removed from the sort of number crunching that revealed Bernard Madoff to be a crook long before the Securities & Exchange Commission could find a reason to lay a finger on his books.

By comparing multiple inventories, the likelihood of finding untoward activity rises dramatically. Put another way, if it really is just paper gold, then the issuer would need to go to extraordinary lengths to manufacture a plausible proxy for the claimed underlying metal. In fact, the costs of maintaining such a scheme over time would vastly exceed storage, insurance and audit costs, never mind the risk of being found out.

The testing parameters are nearly perfect for such an exercise:

• Quality is controlled by the London Bullion Market Association.
• Accredited melters and assayers are widely dispersed.
  • They mostly acquire their product from regional sources.
• The underlying product is tightly unitized:
  • London Good Delivery Bars must ideally contain 400 troy ounces of gold refined to a purity of 999.9 parts per 1,000.
  • The purity tolerance is strenuous:
    • assays of 999.5 and higher must test within ±0.05 parts per 1,000.
    • assays of less than 999.5 must test within ±0.15 parts per 1,000.
  • The minimum acceptable gold content is 350 troy ounces.
  • The maximum acceptable gold content is 430 troy ounces.
  • The minimum acceptable purity is 995 parts per 1,000.
  • Length, width and height are specified.
  • Each bar must be uniquely identifiable with four marks - serial number, refiner stamp, fineness, and year of manufacture.

We evaluated 87,662 gold bars weighing 35.2 million troy ounces and worth around $46 billion (see Table 1 above & 4 below. Click to see gold bar inventory detail). That represents all of the inventory underlying the ETF Securities funds, and roughly 60% of the SPDR Gold Shares inventory as of late October.

It is notable that the two organizations maintain different inventory reporting protocols even though they have common founders. That’s an important factor in looking for manipulation. A fraudster always seeks to mimic and cloak rather than differentiate because it avoids attracting attention.

Another telltale sign of untoward activity is a lack of transparency. Yet you can download the inventories at any time. In the case of GLD we did have to ask for a spreadsheet version rather than a PDF format, but it was quickly forthcoming.

That’s not the hallmark of market bandits; more so since the World Gold Council has always been a little skittish in dealing with your correspondent for the better part of a decade.

Key variances:

• SDPR Gold inventory weights are reported to thousandths; ETFS to hundredths.
• ETFS reports a bar brand, GLD does not.
• Refiner names have different nomenclatures in each organization.
• The inventories have different geographic characteristics.

Comparisons

To facilitate direct comparisons we had to standardize the bar brands, adding the full refiner name along with the city and country that each bar was manufactured in. That provided an enormous amount of additional power to our task because the data could be sliced in many more ways in the search for anomalies.

We first applied some common sense tests between the two inventories. The conclusion was that there were no reliable indications of tampering based on the randomness of the variations (see Table 2). The lack of "sameness" is also important since fraudulent activity is invariably a clone of the real thing. In datasets like these any replication at the aggregate level needs to be treated with suspicion.

The next step was similar, but this time comparing individual bars for the highest and lowest values by gross weight, fine weight and fineness. Again, nothing unusual stands out (See Table 3).

We followed up with a search for duplicate serial numbers. We did find several duplicate bar numbers, but only one of them was an obvious error. Not one of the apparent duplicates had duplicate gross or fine weights and were, therefore, not suspicious.

There are few organizations that could boast a defect rate of just 0.0015%.

Two Johnson Matthey bars manufactured in Hong Kong shared the serial number “AA22338” in the GLD inventory, which we queried with World Gold Trust Services. WGTS had the custodian, HSBC, pull the bars and confirmed that one of them was incorrectly identified and should have been “AA23338”.

The remaining duplicates were easily explained - both by knowledge of the various bar numbering systems, and by explanations provided by World Gold Trust Services (click to see the formal response to questions).

For example, Johnson Matthey London bars are numbered from “1” at the start of each year. GLD does not show the year of manufacture with the serial number which gives the impression of duplication. Similarly, American bars from the US Assay Offices & Mints have the same problem. In future, GLD inventory will add the date to the serial number.

A slightly different problem crops up with Russian and Uzbek Bars which can have duplicate serial numbers even when the year is concatenated with the serial number. The differentiator is a 2-character Cyrillic letter code.

The LBMA has issued guidance that Russian and Uzbek bars should be recorded with “an 8-digit bar number formed from the 4-digit year concatenated with the 4-digit bar number.” However, we expect the LBMA to amend this to add a Roman 2-character code in order to resolve the duplications that occur even with 8-digit serial numbers.

If you want to find fault, this would be the appropriate time to chide GLD for not being more thorough in the management of its inventory data, but given the low error rate it’s hardly material. However, a more important point to note is that the data is also a thorough validation of the LBMA's accreditation process.

Taking another tack in the search for duplicates, we isolated bars with exactly duplicate gross and fine weights.

There were 18,881 “gold content” duplicates in GLD, or 28.8% of the sample. Critically, no pattern emerges in those duplicates which span multiple refiners and weights. It is highly improbable that someone could artificially achieve the distribution observed in the duplicates. The number of duplicates is also satisfactory in terms of the probability given the distribution of data. Indeed, the ETF Securities inventory produced a similar result with 5,946 gold content duplicates representing 26.9% of the sample.

This is what can be reasonably expected given that more than two-thirds of all good delivery bars are likely to have a gold content between 392oz - 412oz.

Distribution

Distribution is a critical test that is a highly reliable way to uncover data tampering (see Figure 1).

In the case of the gold ETFs, you would need several super-computers hooked up in parallel to achieve normal distributions on a daily basis, but which were not identical or repeating since that would immediately arouse suspicion.

The greater problem is that it would require constant manipulation of the gold content which would have to be managed to a very fine level. That would show up in changing gold content in individual bars between inventory periods. One way to mask that might be to have a high turnover of bar numbers, but there is no evidence of that happening. A clever statistician might argue that each incremental fake bar can be assigned a gold content that maintains the overall distribution. We disagree.

It is highly unlikely that it can be achieved during high volume trading sessions without blowing everything up. Secondly, when the inventories are sliced down to, say city of manufacture, the distributions maintain integrity. The odds of achieving this without creating signals that would turn up in pattern recognition runs is implausible.

For example, for the sort of large scale fraud to be taking place which is implied, you would most likely see repeating patterns in the distribution of gold content across refiners, but more especially across funds. The opposite turns up - each fund has its own gold content “DNA”, and that DNA shows unique characteristics in several dimensions.

It's worth recalling that elaborate and sophisticated modeling of historical temperature data to prove global warming was ultimately proved to be false, if not an outright hoax, by forensic statistics. Billions of dollars were invested in propping up and camouflaging those dodgy datasets which were underwritten by organizations like NASA, NOAA, and a trove of respected universities employing the best minds in the field. Yet even they were unable to avoid having their manipulation unmasked by independent analysis.

It is highly presumptuous to believe that exchange traded commodity funds have cracked the formula for mass deception when the world's leading climate change lobby could not.

There are none of the fingerprints of design you’d presume to find from an artificial gold inventory dataset. Rather, there is the “conformity to randomness” expected when 1,096 tonnes of gold refined by 52 manufacturers in 43 locations and 20 countries are thrown together in a variety of custodial vaults.

© 2010, MineFund.com > Commodity Investment Analysis & Analytics

Be Sociable, Share!

•