Economic Effects of Subprime, Part III: Leverage and Amplification

In the comments to my post Is the US Banking System Really Insolvent? Prof. Menzie Chin brought to my attention a wonderful paper: Leveraged Losses: Lessons from the Mortgage Market Meltdown (hereafter, “Greenlaw et al.”).

This paper has also been highlighted on Econbrowser under the title Tabulating the Credit Crunch’s Effects: One Educated Guess.

The source document is in several parts – to do justice to it, I will be be posting reviews of each section.

The initial post in this series Economic Effects of Subprime, Part I: Loss Estimates dealt with the authors’ methodology of estimating total losses of $400-billion on subprime securities. The second, Economic Effects of Subprime, Part II : Distribution of Exposure, looked at the way they allocated 49% of this loss to the “US Leveraged Sector”. In this post, I’ll be looking (very briefly!) at their “Section 4: Leverage and Amplification”.

I will admit, I had half a mind to skip it. I felt comfortable opining on their sections directly relevant to the securities markets, but economic predictions are another animal entirely. Fortunately, Willem Buiter stepped into the breach, with a highly entertaining polemic on VoxEU.

Greenlaw et al. review the procyclical nature of leveraging and eventually come up with the scorecard so far:

So far (up to late-January 2008) approximately $75 billion of new capital has been raised, compared to a cumulative running total of $120.9 billion for write-downs announced by banks and brokerage firms.

It does not appear that this figure of $120.9-billion is restricted to US banks and brokerages – and since the authors do not say, I suspect it is a world-wide figure.

At any rate, they go through a little arithmetic and conclude:

Our baseline scenario (marked in grey) is that leverage will decline by 5%, and that recapitalization of the leveraged system will recoup around 50% of the $ 200 billion loss incurred by the banking system. Under this baseline scenario, the total contraction of balance sheets for the financial sector is $1.98 trillion.

Section 4.4 does a little algebra to estimate the ratio of the decline in credit to end-users to the decline in total assets which, I am grateful to observe, is as fishy to Prof. Buiter as it is to me. Prof. Buiter observes:

The authors calculate/calibrate a value for the ratio of total credit to end-users (either the non-leveraged sector or just households and non-financial corporates) to the total assets of the leveraged sector (banks, the brokerage sector, hedge funds, Fannie May and Freddie Mac and savings institutions and credit unions). They then treat this ratio as a constant, which means that once they have the change in the value of the total assets of the leveraged sector, they know the change in credit to the end-users.

There are just too many ways to poke holes in the empirical argument. To start with, as noted by the authors) the credit variable used domestic non-financial debt, includes financing from non-leveraged entities and therefore does not correspond to the credit variable of the theoretical story.

My problem with this – which I think is the same as Prof. Buiter’s problem – is that the algebra treats the “leveraged sector” as being homogeneous … and it ain’t. Say, for instance, we have a Hedge Fund with $1 in investors’ money, levered up 10:1 to buy $10 of securities. Their balance sheet looks like:

Hedge Fund
Item Asset Liability
Securities $10  
Borrow   $9
Investors   $1

They are borrowing from a bank, which has the balance sheet:

Bank
Item Asset Liability
Loan to HF $9  
Deposits   $8.10
Capital   $0.90

Now what happens is the value of the securities falls to $9, the bank calls its loan and ends up owning the securities. The two balance sheets now look like:

Hedge Fund
Item Asset Liability
Securities $0  
Borrow   $0
Investors   $0

While the Bank’s balance sheet has changed to:

Bank
Item Asset Liability
Securities $9  
Depositors   $8.10
Capital   $0.90

So, with this particular example:

  • Aggregate leverage is unchanged: ($10 + $9) / ($1 + $0.90) = 10:1 = ($9 / $0.90)
  • the bank has been protected from the first loss on the securities since its claim was senior to that of the investors in the hedge fund.
  • Hedge fund investors have been wiped out
  • The bank’s liquidity has improved (since securities are more marketable than hedge fund loans)
  • There is no effect directly transmitted from the bank to the real economy.
  • There may be an effect on the real economy because the hedge fund investors aren’t so rich any more, but that’s second order

I just have all kinds of problems with this, Greenlaw et al‘s treatment of the leveraged sector as being homogeneous with effects on credit available to the real economy being a constant percentage of losses, regardless of where or how those losses are experienced.

I won’t look at their section 5.1, Correlations between GDP and Credit … I’m just not comfortable enough with economic thought. I’ll leave that task to Prof. Buiter:

More painfully, the authors seem blithely unaware of the difference between causation and correlation, or prediction and causation. What they perform is, effectively, half of what statistically minded economists call a Granger causality test but should be called a test of incremental predictive content. They run a regression of real GDP growth on its own past values and on past values of real credit growth and find that past real credit growth has some predictive power over future GDP growth, over and above the predictive power contained in the history of real GDP growth itself: past real credit growth helps predict, that is, Granger causes, real GDP growth. Lagged real credit growth is (barely) statistically significant at the usual significance level (5%).

When you do this kind of regression for dividends or corporate earnings and stock values, you find that stock values Granger-cause (help predict) future dividends. Of course, anticipated future dividends determine (cause) equity prices, so causation is the opposite from Granger-causation.

The authors are undeterred and treat the estimate of GPD growth on credit growth as a deep structural parameter.

The authors could be right about the effect of de-leveraging in the leveraged sector on real GDP growth, but the paper presents no evidence to support that view.

So, to sum up:

  • I’m suspicious of the authors’ loss estimates
  • I’m suspicious of the authors’ allocation
  • I’m suspicious of the authors’ calculation on the effect of losses on credit available to the real economy
  • Prof. Buiter is suspicious of the authors’ calculation of the effects of credit availability changes on GDP

All in all, the paper by Greenlaw et al. has turned out to be a typical product of brokerage house research departments:

  • Great Data
  • Interesting Ideas
  • Unsupportable conclusions

Leave a Reply

You must be logged in to post a comment.