Saturday, April 17, 2010

Some Further Comments on the Leverage Cycle

In a previous post I discussed a paper by John Geanakoplos on the Leverage Cycle (due to appear in the NBER Macroeconomics Annual later this year). I presented this paper in the Columbia finance reading group last Thursday, and have posted my slides in case anyone is interested in taking a look.  
The paper is considerably more accessible to the general reader than most of the recent theoretical literature on the financial crisis. It avoids the standard theorem-and-proof format, and consists instead of a sequence of elaborate numerical examples that fit together like a jigsaw puzzle. It also contains a number of very interesting ideas and insights, many more than I was able to discuss in my earlier post. 
One of the key features of the Geanakoplos model is that the same set of physical assets can serve as collateral multiple times for loans of different maturities. For example, housing serves as collateral for long-term loans in the mortgage market, while the loans themselves (after securitization and tranching) can serve as collateral for short-term borrowing in the repo market. Geanakoplos shows that the extent of leverage in the long-term market will endogenously be such as to allow for a positive probability of default, and is interested in the effects of bad news in this market (interpreted as an increased likelihood of eventual default) on the market for short-term loans backed by financial rather than physical assets.
Among the main insights in the paper is the following: a decline in the expected terminal value of the physical assets will result in a far greater decline in the prices of the financial assets that they back. This happens for three reasons. Most obviously, there is a decline in fundamentals. But the effects of this are amplified because the initial prices (before bad news arrives) reflect the beliefs of the most optimistic market participants, who borrow from the pessimists in order to buy their asset holdings. In other words, the marginal buyer is (endogenously) very optimistic and this is reflected in the market price. The decline in fundamentals not only wipes out these highly leveraged optimists, but also substantially reduces equilibrium leverage in the market. As a result, the decline in the financial asset price is far greater than any market participant's expectations concerning the terminal value of the physical asset.
The careful reader will note that incomplete markets and maturity mismatch play a critical role in this argument. One of the most interesting aspects of the model is that both market incompleteness and maturity transformation arise endogenously, and the asset prices at various points in the tree of uncertainty are all correctly anticipated. The results are driven not by irrational exuberance or systematic biases, but by heterogeneous preferences and beliefs, and changes over time in equilibrium leverage. It's a precise, rigorous and carefully constructed interpretation of recent events, based on work that was done well before this crisis erupted.
In response to my earlier post on this paper, David at Deus Ex Macchiato agreed that the work is important, but added:
What astonishes me however is that this is in any way news to the economics community. Ever since Galbraith’s account of the importance of leverage in the ‘29 crash, haven’t we known that leverage determines asset prices, and that the bubble/crash cycle is characterised by slowly rising leverage and asset prices followed by a sudden reverse in both?
This is a good question. A lot of the less formal work in this area never made it into the canonical models taught to successive cohorts of graduate students in economics. Geanakoplos doesn't mention Galbraith explicitly, but he does mention Minsky and Tobin, who themselves were surely familiar with Galbraith's work on the crash. Implicit in David's question is the accusation that the training of professional economists has become too narrow, and on this point I believe that he is absolutely correct.


  1. Does this borrowing create net leverage? If you have Asset A, Loan B against Asset A, and Loan C against Loan B, it looks like the party with long exposure to loan B and short exposure to Loan C nets out to zero exposure. If the asset declines in value, their loan collateralized by the asset also declines in value; but so does the loan backed by that loan.

    This scenario does introduce counterparty risk, but haircuts are often more than enough to compensate for this. (i.e they create a scenario in which a given dollar of net debt requires more, not less, capital).

    The scenario that can create more total risk is when debt is owned by an entity, which borrows against the whole portfolio of debt (i.e. a traditional bank, or a CDO). In this case, a decline in the value of an underlying loan is not immediately reflected in the decline of the debt backed by the portfolio; instead, the portfolio's price reflects the probability of default for the whole portfolio. That can mean--recently, did mean--that a deteriorating portfolio can be overpriced, causing future losses and making those counterparty risks a serious problem.

  2. I hastily put together the same explanation based on information theory and you might glance at it.

    A very hasty post and unedited, but it contains the essentials.

  3. Byrne, in the Geanakoplos model the only counterparty risk is in the long-term (mortgage) loan market: equilibrium haircuts in the repo market are large enough to ensure loan repayment in every state. The phenomenon he is interested in is not the creation of new net leverage but rather the excess volatility (relative to beliefs about fundamentals) of asset prices.

    Matt, I looked at your post but couldn't really follow your argument. My fault, not yours - I just don't know enough information theory or quantum mechanics.

  4. NIT:

    The lawyer in me despairs at the statement "the same set of physical assets can serve as collateral multiple times for loans of different maturities," although the example mostly clears it up.

    When truely identical collateral is used for multiple loans, the legal system priorities them (1st, 2nd, 3rd liens, etc.) with elaborate rules for doing that, and the only time those rules don't work is when you have a kind of mortgage fraud not common in the current crisis (elaborate document forgery and title company fraud).

    The physical assets are, of course, not the collateral for the second order loans, the mortgages are, and that distinction matters for many purposes.

    Most importantly, it is perfectly possible to have a crisis in the financial assets without having a default rate crisis in the underlying mortgages, although the reverse is not true. You would see this, for example, if a sudden easy money policy at the Fed created a surge in inflation and confounded the interest rate bets made on fixed or temporarily fixed rate mortgages. Millions of people would receive notices that their loans had been sold, but that would be the end of it.


    The real criticism of the leverage model in the work cited in your original post is that it works too well. It doesn't explain why the housing bubble and subsequent collapse was overwhelmingly a regional phenomena, rather than a national one. We didn't have a national housing bubble; we had a housing bubble in California, Arizona, Nevada, Florida and to a lesser degree a few other states. This is particularly striking, for example, across the Georgia-Florida line. Georgia did not have a significant housing bubble or crash, while Florida did, despite the facts that the two states have many other similiarities and both had boom economies. Conversely, both Florida and California saw big bubbles in all of their metro areas despite their largely independent economies (so far as metro areas in the same state can be independent).

    One way to account for this is that California and Florida were perceived as non-recourse mortgage markets, while the Arizona and Nevada can be seen as spillover bubbles of California real estate money, rather than primary bubbles.

    The heads I win, tails you lose component to the purchaser level decisions in California and Florida were critical to the bubble phenomena that happened there, and the quirkiness of those state's mortgage laws helps explain why metro NYC financial services and legal professionals, and metro Washington DC and metro NYC area regulators, who represented investors in mortgage backed securities didn't see what was happening sooner.

  5. Andrew, thanks for your comment. Perhaps I should have been more careful with my choice of words but (as you note) I think its clear from the post that it is claims on housing that serve as collateral in the repo market and not the housing itself. As an economist, I see this as housing serving indirectly as collateral; as a lawyer you see the loans and the homes as distinct assets.

    Regarding regional variations, I think that the best discussion of this is Krugman's "Flatland versus Zoned Zones" post back in 2005:

    It doesn't explain all the variation obviously, but he has follow up posts on desert bubbles etc. that fill out the story. There are also regional variations in laws and regulations (Texas has been examined a lot lately for having avoided the worst excesses). But an abstract model of the leverage cycle can't accommodate these institutional features. The question is whether it adds any insight at all, and I think it does.