Taking our limited repo example from Part 1 a step further still, in real-world operation a bank might put up any number of securities – including any it might have just that day obtained full title too – to secure financing all at once. Thus, there are groupings of securities over which the bank has varying degrees of control – some outright purchases, some shorted, some borrowed, others themselves part of complex iterations (like transformation) – pledged in one or several SFT’s, not just repo but also any other funding subset via derivatives.

In most realistic cases, it’s a blending of assets against a blending of liabilities.

If stretching our example a bit further, a bank has pledged four types of assets to secure just a single funding operation then the relationship (and, yes, co-relationships) of those assets becomes a new parameter by which the bank must consider.

Putting up, say, an off-the-run Treasury note, an on-the-run Treasury bill, an agency MBS, and some kind of lower quality junk corporate, altogether to some extent overcollateralized, as the basis for just one SFT this leaves the bank exposed to any changes in the context of those collateral pieces. What happens, for example, if the market for off-the-run Treasury notes becomes even more uncertain and thus illiquid than it already is?

Or if lower-quality jump corporates suddenly are in danger of being perceived as no better than pure junk?

This would pressure the better parts of the collateral grouping, among many groups, to hopefully make up any (overcollateralized) difference without demanding much more of either. If the over-collateralization gets used up entirely, then the bank would have to reach into its portfolio of assets for only one or the other to post, on net, in bespoke as well as triparty repo just to maintain the same level of liabilities.

And if it didn’t have more of on-the-run Treasuries or liquid agency MBS? Then that’s getting into Bear Stearns/AIG/Lehman territory; those who ended up getting the dreaded call from the triparty repo custodian no longer willing to extend the same daylight overdraft generosity for fear of being left on the hook from some unsettled or imbalance between the bank’s assets for liabilities (too many of the former unable to secure enough of the latter).

In the example immediately above, we’d expect that collateral for our bank would begin to concentrate in the more liquid and usable formats; less off-the-run Treasuries, fewer corporates, leaving mostly on-the-run UST’s (bills) and hopefully the same for agency MBS to conduct the same level of funding.

And if this happens for the one bank, we’d very much expect to find the same thing happening across the entire system; collateral becoming relatively more concentrated into whatever’s left as the most usable.

Conversely, we’d also expect to find that during periods of risk-taking or especially reflation (risk-taking done for suspicious reasons) there’d be a higher level of collateral dispersion. It’d make sense to use all kinds of collateral across various SFT’s in order to more efficiently match assets and liabilities, particularly when higher-return riskier assets could fetch relatively decent funding terms.

In yet another case of (like TIC) of regulatory authorities collecting information but having no idea what they are looking at or what to do with it, the Federal Reserve Bank of New York obtains a wealth of data on triparty repo which is reported directly via call reports from JP Morgan and Bank of New York. The data isn’t sampled; this is it, all of it, everything that goes on in this segment.

However, “all of it” refers only to triparty repo rather than the entire repo market which is far larger still. In other words, much perhaps a majority (there is some debate) of repo and then a whole wide range more of SFT’s in derivatives aren’t included anywhere in these figures (or really any others). We’re still subjected to limitations regarding the totality of the phenomena, only these limitations are somewhat (hopefully meaningfully) less.

Among the parameters, FRBNY collects are those like “number of repos” and “number of observations.” The former is simply the raw number of individual repo transactions reported by the triparty custodians in any given time period (monthly as presented here, up through April 2021). The other, number of observations, refers to how many other data points which might be included with a repo transaction: such as the number of different types of repo collateral used in them.
 

If we put those two parameters together (chart immediately above), coming up with the number of collected observations per reported number of repo transactions, what we find is almost exactly what I just described. During reflationary or “risk-on” periods, the relative number of observations per repo rises, indicating how the repo system as a whole is putting up a wider variety of collateral on average per transaction.

Conversely, during dollar shortages – of which collateral has played a huge part – that average tends to decline if only somewhat.

One reason why:

When we look at the overall total dollar repo volumes (collateral used, at par before haircuts) this has tended to fall during QE (QE2 being the exception). And it’s been true for each of the two main instruments which QE targets: UST’s as well as MBS.

Putting that together with the interpretation of collateral dispersion embedded within the average number of observations per repo, this tells us that, indeed, QE tends to make the system more dangerous by removing those forms of collateral which are always treated as among the highest quality. The rising average number of observations indicates that repo (and SFT) participants have to adjust to less available collateral by using more of other types.

This along with reflationary risk-taking sets up for an increasingly risky backlash.

These things then collide during dollar shortages (Euro$ #n) when there’s less available collateral but also other forms become less usable. You can easily see above when the volume pledged in UST’s (Euro$ #’s 2, 3, 4) and agency MBS (Euro$ #’s 2, 4) go up at the same time the average number of observations drops slightly.

Collateral bottlenecks.

During Euro$ #4, the total volume of triparty repo surges with nearly all of that increase backed by just UST’s and MBS; everyone herded more and more into the narrowed list of acceptable, workable collateral formats. That’s why the volume goes up but the average number of observations actually declines; as market prices had already dictated, there was not “too many” Treasuries.

Remembering complementarity, what’s not obvious or much apparent in this data is what must have been going on outside of our experiment, even if our experiment reaches further into the shadows than most. In still trying to seek out the totality of the phenomena, we know the world was experiencing an increasingly problematic dollar shortage (given to us in real-time by market prices) that ended up in surging repo use if via an increasingly narrowed list of collateral types.

In other words, if we stuck only to what one or two of the parameters available to us here (such as total volume), or, more worrisome, attempted to marry these figures with only the mainstream monetary focus on the changes in the level of bank reserves, either way we’d be left scratching our heads; none of this would make much sense.

Instead, combining what we see here with other documented reactions to the same hidden things we can better develop the “single picture” emerging from these shadows: repo as lender-of-last-resort (rising volumes) which can only be conducted/supported by a narrowing set of available collateral. It makes the blending of assets and liabilities far more difficult, at least more uncertain, which is itself feedback into the dollar shortage uninterrupted by the level of bank reserves.

And that leads us into the post-March 2020 environment. As with other QE periods, and this is by far the most QE of any of them, the triparty repo data indicates the same stripping collateral only this time there really hasn’t been the same adjustment to it when collateral dispersion typically has increased. The average number of observations has been more or less consistent for more than a year.

This would seem to indicate less that’s like reflation than otherwise (consistent with low yields for the higher quality collateral instruments compared to other reflationary periods).

But while the data appears to describe conditions somewhat unlike prior reflationary periods, it does still indicate lower levels of ongoing repo stress (including, perhaps counterintuitively, lower overall repo volumes). In other words, still a nod toward collateral concentration but less overall (or emergency) dependence on triparty repo as a stricken system’s ultimate backstop.

Better than early 2020, but not all that much better. Quasi-reflation this time?

Again, we have to be aware of the limitations of even this more penetrating data which, for once, potentially tells us more about collateral than just about any other datasets. It is very often collateral conditions and the changes in them which do appear far better at explaining the world around us, contributing better to a more unified if still scattershot global money, finance, and economic picture – all regardless of the systemic level of bank reserves. 

Totality of the phenomena focuses our attention almost everywhere else other than bank reserves.