JP Morgan Markets Its Latest Doomsday Machine (or Why Repo May Blow Up the Financial System Again)

Readers of ECONned will be very familiar with the name of Gary Gorton, author of ‘Slapped in The Face by the Invisible Hand’, which explores the relation of the so-called shadow banking system to the financial crisis. His work is pretty fundamental to understanding some of the mechanisms which made the crisis so acute. Now he’s done an interview, which I would like to have a growl at; but first, he has some basic points about shadow banking, useful later in this rather long post. Gorton explains repo thus:

You take your $200 million to the bank, to Lehman Brothers, say. You deposit it, so to speak, overnight so you can have access to it the next morning if you want to. They pay you 3 percent. And you want it to be safe, so they give you a bond as collateral. But Lehman earns the interest on the bond, say, 6 percent.

..and then “haircuts” (an extra margin of security in case that bond isn’t so safe after all):

There may be a haircut. If you deposit $100 million and they give you bonds worth $100 million, there’s no haircut. If you deposit $90 million and they give you bonds worth $100 million, then there’s a 10 percent haircut.

…and then “rehypothecation”:

If you put a dollar in your checking account and the bank has to keep 10 percent of it on reserve, they lend out 90 cents. Somebody deposits that 90 cents, the bank can lend out 81 cents (because of the 10 percent reserve requirement) and so on. So you end up creating $10 of checking accounts for $1 of demand deposits, assuming there’s a demand for loans…And that can happen in repo as well because if you’re Lehman and I’m the depositor, and you give me a bond as collateral, I can use that bond somewhere else. So there is a similar money multiplier process.

…and finally the link to regulated banking:

And shadow banking very importantly is not a separate system from traditional banking. These are all one banking system.

So much for the preamble. The other point you need to know: we are talking about an unregulated banking system that at its height was just as big as the regulated banking system, yet coupled to it, and apparently more profitable, though, as we now know, much riskier. Now we get to the part of Gorton’s interview that I’m not so happy with:

In summary, I would describe shadow banking as the rise to a significant extent of a very old form of bank money called repo, which largely uses securitized product as collateral and meets the needs of institutional investors, states and municipalities, nonfinancial firms for a short-term, safe banking product.…It’s a valuable innovation.

The bit in bold is where I raised my eyebrows. The truth of the bolded claims depends on the yet-to-be-discovered solutions to repo’s core problem, formulated thus by Gorton:

Of course, the problem with repo and shadow banking is that they have the same vulnerability that other forms of bank money have. We can talk at great length about what that vulnerability is, but loosely speaking, it’s prone to panic. Looking back at history, think about how long it took to devise a solution to the first banking panic, related mostly to demand deposits. That was in 1857. It wasn’t until 1934 that deposit insurance was enacted. That’s 77 years where we’re trying to understand demand deposits and figure out what to do.

The situation that we’re in now, seriously, is one where we are back in about 1860: We’ve just had a big crisis, and we’re trying to figure out what to do. We can only hope that it doesn’t take 77 years to figure it out this time.

That doesn’t sound like a safe banking product to me. Next comes some irritating pussyfooting:

Nobody wants to be given collateral that they have to worry about. And the mechanics of how repo works is exactly consistent with this. Firms that trade repo work in the following way: The repo traders come in in the morning, they have some coffee, they go to their desks, they start making calls, and in a large firm they’ve rolled $40 to $50 billion of repo in an hour and a half. Now, you can only do that if the depositors believe that the collateral has the feature that nobody has any private information about it. We can all just believe that it’s all AAA.

Gorton is polite, and that can mislead. Impolitely: for “private information”, read “knowledge that the collateral is wildly overvalued”, or ”aware that the collateral is backed by assets with massive gearing to fraudulent loans”. The system’s gatekeepers (originators, ratings agencies and credit insurers) had an “agency problem”, (impolitely: issued cows, or rated cows, or insured cows, for money), so the fraud-backed collateral got past them.

Gorton is vague about how this “information insensitive” collateral is to be created. Presumably the options are to have reliable ratings agencies, or some other gatekeeper on the collateral, or a very deep-pocketed credit guarantor (that’s you, dear reader), or all three. But he doesn’t quite spell it out; perhaps that’s just the interview format. What we get instead is this:

We want all securitized product to be sold through this new category of banks: narrow-funding banks. The NFBs can only do one thing: just buy securitized products and issue liabilities. The goal is to bring that part of the banking system under the regulatory umbrella and to have these guys be collateral creators.

Well, I don’t immediately see why a narrow funding bank, thus described, is a more reliable generator of quality collateral than a narrow rating agency, or a narrow monoline insurer, or for that matter, Gorton’s old client, AIG. Why would an NFB be any better than any of those organizations in filtering out low quality collateral, given the demand for collateral? The NFB has exactly the same agency problem.

And are we really sure we know what ‘good collateral’ is? Gorton’s formulation of the problem isn’t quite accurate: it’s the stability of the haircuts that matters, not the reliability of the ratings. In truth, the ’08 crisis in repo was ended by explicit and implicit government backstops. By that time: haircuts on pristine US treasuries had gone from ¼% pre crisis to 3% mid crisis; on investment grade bonds, from 1.5% or so to 10% plus. As we will see later, when there is lots of rehypothecation, those moves would matter just as much as the annihilation of triple-A CDOs. More on this later: identifying a different problem with repo is the meat of this post.

Also missing from Gorton’s picture: how much of the need for repo collateral is simply driven by the increase in OTC derivatives. According to another of Gorton’s papers, there was $2Trillion of derivatives collateral in 2007; $4Trillion in 2008. So is that why Gorton simply assumes that repo “has” to grow: to provide collateral for the OTC derivatives market? So why, then, does the OTC derivatives market have to grow? One would like to see the connection between ETFs and repo worked out by someone, somewhere, too.

Despite my whining, do have a read of the interview:  here it is again. There’s plenty more, and plenty of it is good – why Dodd-Frank is a big miss, how few data are available on the enormous shadow banking system. That’s what happens when you don’t supervise financial innovations: you don’t know what they do, you don’t know how they work, and you don’t know what went wrong. If you are in full geek mode, you can download the papers that underlie the interview here and here (I must get round to that). Metrick and Gorton write about haircuts here. That lot should keep you going…

Now I’m going to double back to something that Gorton skirts: the interaction of repo haircuts, rehypothecation and the credit multiplier. Recall his interview:

If you put a dollar in your checking account and the bank has to keep 10 percent of it on reserve, they lend out 90 cents. Somebody deposits that 90 cents, the bank can lend out 81 cents (because of the 10 percent reserve requirement) and so on. So you end up creating $10 of checking accounts for $1 of demand deposits, assuming there’s a demand for loans. Now, that money multiplier process is very important because it means that the amount of endogenously created private bank money in checking accounts is 10 times the size of the collateral, so to speak, of $1 of government money. So, in a traditional banking panic, if everybody wants their $10 back, there’s only $1. And that’s the problem.

What you have here, in the equivalent language of repo, is a 10 per cent haircut, with unlimited rehypothecation (so that you can just keep reusing the collateral to raise more and more liquidity, haircutting away until the amount you can still pledge isn’t worth bothering with), and a credit multiplier of 10. To get a general picture of how the credit multiplier, haircuts and rehypothecations tie together, we now need a tiny spot of mathematics.

An aside: one of the peculiarities of mathematical economics, as opposed to mathematics, is the relative frequency of “theorems”. In mathematics, theorems are as rare as unicorn droppings, things of near-holy awesomeness; in mathematical economics, by contrast, they occur horribly frequently, like depictions of unappealing sexual acts in the oeuvre of the Marquis de Sade.

So I should probably try to get people to give this shoddily presented and deeply unoriginal formula,


some kind of grand title: Smith’s Unrestricted Rehypothecation Theorem, perhaps. What does it mean? It describes the relation between the credit multiplier under unrestricted rehypothecation, Cm¥, and h, the haircut, which is a value between 0% and 100%; k keeps count of the number of rehypothecations. Any charges levied for the rehypothecation are assumed to be negligible (I won’t keep saying this, but bear it in mind – it means that the credit multiplier is never quite as big as I say it is, though pretty close, because the charge for a rehypothecation is not huge). As you see, with unrestricted rehypothecation, you just invert the haircut to get the credit multiplier. That is the big picture.

So, as with Gorton’s deposit example above, if the repo haircut is 10%, the ultimate credit multiplier is 10. Which is to say: if you could rehypothecate for ever, liquidity amounting to 10 times the amount of underlying collateral would be created, if the haircut was 10%. Once again his polite example might be a bit misleading. Consider instead the effect of a haircut of just 1% – then the ultimate credit multiplier is 100. With zero haircuts and no rehypothecation charge, the credit multiplier would be infinite…

For anticlimactic comparison, Singh and Aitken estimate that the credit multiplier in force at the height of the bubble was about 4: there were $1Trillion of rehypothecable hedge fund assets, transmuted by the magic of rehypothecation into $4Trillion of pledgable collateral at banks. No cause for concern then? Well, that depends how much you enjoyed the ’08 liquidity crisis; and unfortunately, much higher levels of rehypothecation may be just around the corner, which is a worry.

To show why, we need another theorem, Smith’s Restricted Rehypothecation Theorem this time, I suppose. It’s just as banal as the other one:


This gives you the total credit multiplier Cmr, when you have a finite number of rehypothecations. The number of rehypothecations is given by r; h is the haircut again. As r tends to infinity, the first term on the right hand side of the equation tends to zero, giving you, in the limit, the Unrestricted Rehypothecation Theorem.

Now we can work out, for a range of haircuts, what number of rehypothecations it takes to give a credit multiplier. I’ll use the example of the IMF haircut table that Yves exhibits in ECONned (figure 9.4 there), assume an initial credit multiplier around 4 per Singh and Aitken, and use the rehypothecation formula to show the impact of the increased haircuts. The table is a bit rough and ready, but it gives an idea. The collateral has been rehypo’d on average just over three times, giving a credit multiplier of 4. Given the assumed proportion of each collateral type, the loss of liquidity amounts to $750 Billion for every $1Trillion of collateral (this is August ‘08, is before the crisis got really acute).


The estimate of the amount of each collateral type involved is pretty much a guess. If only we knew! But I hope it doesn’t matter much: the table does illustrate the point that haircuts increasing from a single digit number to 10 or more, on widely held collateral (in this example, investment grade bonds, or Prime MBS), can have just as big an effect on liquidity as total wipeouts on CDOs, or big haircuts on ABS (rightmost column, in bold).

With lots of rehypothecation, it gets worse. To get a better idea of how haircuts, rehypothecation and the credit multiplier work together, it’s time for a picture of Dragon Country.


This is my two equations, graphed. Some more explanation:

  • Haircut: the 1.0 (bottom right hand corner) is of course 100% , in other words, there is no repo, and the credit multiplier is 1, so there is no effect on credit. I’ve assumed the charge for rehypothecation is negligible.
  • The thick black curving line shows the theoretical maximum credit multiplier when there is an infinite number of rehypothecations. On the basis that a 1000x credit multiplier is absurd enough, I stopped at 0.1% haircuts, though 0% haircuts have supposedly been used in repo.
  • The unimaginable top left hand corner won’t fit on the graph: a haircut of zero and an infinite credit multiplier.
  • The thin green line shows the credit multiplier for various haircuts when there are just 4 rehypothecations. You can see from the graph that this gives to a credit multiplier of around 4, for a range of haircuts from 0-20% or so.
  • The just about detectable blue curve, above the green one, shows the credit multiplier when there are 20 rehypothecations: already enough to move the credit multiplier to worrying levels when the haircut is less than 20%, and when there is only a 0.25% haircut, to an absurd 17x.
  • I’ve assumed that Q4 ‘08 is nasty enough for all of us, and that therefore an overall credit multiplier of 4 is as much as we want; so that’s where I’ve put the horizontal red line.
  • The red area is Dragon Country, where low haircuts and lots of rehypothecation result in huge credit multipliers, and very great (exponential-like) sensitivity to increases in haircuts.
  • I’ve used a logarithmic scale on the y-axis to cram the whole thing in. Dragon country would be impressively vast on a linear scale.
  • The graph shows you something else Gorton doesn’t really emphasize: the only reason to like a small haircut is to maximize the amount of liquidity you create via repeated rehypothecation.

Have I just put forward one of those daft theoretical constructs beloved of economists and technocrats? I think not, for a couple of reasons.

First, “infinite” rehypothecation just sets out a limiting case that exhibits some unfortunate but representative dynamics. It doesn’t have to be that bad to be bad. In particular, the graph highlights the critical relevance to banking stability of very small repo haircuts (and by extension, collateral ratings) and the concomitant large credit multipliers. If those ratings are volatile, haircuts are volatile, and your banking system is unstable. That combination of small haircuts and large credit multipliers may be exactly what we saw in the run-up to the crisis of Q4 ’08. It did seem to be the sudden doubts about the value of “AAA” rated CDOs that caused the initial spasm of funding difficulties before March ’08 and the Bear implosion.  But even unimpeachable collateral was tainted somehow. How did that happen? What’s to stop it happening again? Now add some counterparty doubts, courtesy of Lehman, with hedge funds deleveraging and then pulling their assets from Prime Brokers, which stops rehypothecation in its tracks, and by Q4 ’08 you are in a proper crisis. Even if the UK has better bankruptcy processes next time, they will get their first proper test live, in a crisis. There is no particular reason to expect a hedge fund to feel more confident about its UK Prime Broker next time we get a Q4 ‘08.

Second, the doubts that have inhibited rehypothecation have been more about rights in bankruptcy than about the very idea of rehypothecation; but it’s rehypothecation that is the bogey. That’s not necessarily how the US lawmakers saw it back in 1934, when they capped rehypothecation in the US, as described by John Hempton, but the 1934 law helped a lot, anyway. Banks operated in London to get around the US restriction; then rehypothecation was a massive factor in the complexity of the Lehman bankruptcy, which dragged lots ($15-$45Bn, depending on how the bankruptcy plays out) of hedge funds’ rehypothecated assets into the mess; surviving hedge funds toned down their agreements to London rehypo, in a rush. But I suspect John is jumping the gun when he announces the end of the City of London; the availability of unrestricted rehypothecation is just too darned convenient, if you can get someone to do it. One notes that two years on, there is no UK reform of rehypo; yet there is consultation on reform of UK bankruptcy processes for investment banks, which might encourage hedge funds to agree again to unrestricted rehypothecation.

It also happens that JP Morgan, originators of those not unmixed blessings, Value-At-Risk and Credit Default Swaps, are thinking hard about how to get rehypothecation going in the grand style. They know a volume business with a cheap government backstop when they see one; they are on a marketing push, and presumably they have the systems and processes that go with it.

JPM is very keen to assure us and potential clients that the right business model (they think they have it) will be bankruptcy proof. So – unrestricted rehypo might breathe again, if enough London hedge funds can be reassured  by the UK treasury and by JP Morgan.

The JPM technologists are still at work, too. Rehypothecation is getting slicker, at least for banks whose custody and treasury systems aren’t a hopeless Augean mess of underinvestment, rubbish outsourcing deals, and unmaintainability. Again, see this JP Morgan offering, for an example of what they think they can do. Presumably they think that with an implicit Government backstop, it’s OK to rehypo to the max; they have a derivatives business to support, too.

That would be a Doomsday Machine: iterated rehypothecation, huge credit multiples multiples, low haircuts, and then – sudden increases in haircuts, due to some credit shock or other. Then add some derivatives margin calls arising from the same shock…

If JPM pull it off, there will be a big credit multiplier and a big area of Dragon Country for us all to visit. Replaying the ’08 repo crash with 20 rehypothecations rather than 4 gives a system that has $17 trillion of liquidity in it, pre-crash, for every $1 trillion of collateral. Apply the crisis haircuts and $8 trillion of liquidity vanishes. That is Doomsday. All it takes is some solvency doubts; the quality of the collateral makes no difference. Indeed, because of the small starting haircuts, the US Treasuries make a larger contribution to the liquidity loss, if the number of rehypothecations is larger.

Still, the rehypo graph above, and the availability of new rehypothecation systems, and JPM’s business model, charging some fraction of a bp for each rehypothecation, do suggest ways to make sure the Doomsday Machine doesn’t blow us all up, as long as regulators get a grip.

So how does one wall off Dragon Country? The ultimate objective must be a UK version of the 1934 Securities Law, which capped rehypothecation quite effectively in the US. Pending that, or as well as, some or all of the following:

  • The number of times a given piece of collateral can be rehypothecated is critical. Regulators might consider restricting it.
  • Haircuts also matter, especially when they are very small. Regulators might consider increasing the minimum haircut on rehypo’d assets. If haircuts of less than 10% were not permitted, then the credit multiplier could not exceed 10 for any repo collateral, no matter how much rehypothecation there was. If it was 20%, the credit multiplier could not exceed 5.
  • Regulators might consider eating some of, or a lot of, JPM’s lunch, by taxing each rehypothecation. The proceeds would build up a fund that provides the liquid assets needed in a crisis…something like an FDIC for repo liquidity.

But I wouldn’t hold your breath; the UK authorities’ response so far suggests that they, like JP Morgan, think unrestricted rehypothecation is a thoroughly good thing. I disagree.

Originally published at naked capitalism and reproduced here with permission.

One Response to "JP Morgan Markets Its Latest Doomsday Machine (or Why Repo May Blow Up the Financial System Again)"

  1. Derek   January 18, 2011 at 7:07 pm

    Hi,Could you explain how haircuts decrease the ability to rehypothecate? You say that haircuts keep chipping away until the amount left that could be rehypothecated is negligible but how does that occur?From my understanding, Goldman could take repo out some MBS to JPM with a 15% haircut. Which means, Goldman would $85 if the MBS was worth $100. So now JPM is in possession of a bond worth $100. They too could repo this out at a 15% haircut…also giving them $85. What is going wrong that the amount it not getting cut down?