Google+ Followers

Thursday, April 17, 2014

Systemic Risk of Hedge Funds?

The WSJ “Real Time Economics” blog carried a post Monday, April 14, headlined “Hedge Funds Help Fan Financial Crises: SF Fed Paper” that referred to an article in the San Francisco Fed Letter released earlier in the day.  The article is written by a visiting scholar, Reint Gropp, and summarizes a paper he and others have written that I think I found online ( I will refer to the authors collectively as “Gropp” for expediency). If that is not the actual paper, it is a draft or earlier iteration thereof.

Briefly summarized: the researchers construct their own VaR (value at risk) model for four different kinds of financial intermediaries – commercial banks; investment banks; insurance companies and hedge funds.  Using “daily data”, they derive VaR estimates for each sector (and a control group of REITs, commodities and non-financial stocks) over 2,000 trading days from 2003-2010, which they further divide into periods of tranquility, normality and financial stress.  They then run regressions showing correlations, etc., among the various sectors during these periods, to see whether risk appears to spill over in a persistent pattern from any of the sectors to any of the other ones.   They conclude that, during periods of “tranquility” or “normality,” increases in VaR of the HF universe produce very small (8-9 bps) changes in the VaR of investment banks; otherwise, not much changes.  But in periods of financial stress, they find much greater (71 bps) correlation and impact running from HFs to IB’s and also that the spillover tends to run its course over approximately a two-week period and thus is not detectable when measurements are done using less-than-daily data.

Conversely, they perceive that insurance companies play little role in transmitting risk, as their returns are found to be negatively correlated with the returns of other financial institutions.  Finally, they find very little spillover between commercial and investment banks. The conclusion that hedge funds are the  largest amplifier of risk has implications in relation to the scope of regulation for hedge funds, which are relatively lightly regulated compared to the other sectors.

Their conclusion has an intuitive appeal, in that hedge funds are commonly seen to be more risky and active that the larger institutions and conversely, insurance companies are seen to be the most conservative. As Gropp explains,

Why are the spillovers from hedge funds during financial crises so much bigger, and why do they seem to increase more than those from other financial institutions? Hedge funds are opaque and highly leveraged. If highly leveraged hedge funds are forced to liquidate assets at fire-sale prices, these asset classes may sustain heavy losses. This can lead to further defaults or threaten systemically important institutions not only directly as counterparties or creditors, but also indirectly through asset price adjustments (Bernanke 2006). One channel for this risk is the so-called loss and margin spiral. In this scenario, a hedge fund is forced to liquidate assets to raise cash to meet margin calls. The sale of those assets increases the supply on the market, which drives prices lower, especially when market liquidity is low. This in turn leads to more margin calls on other financial institutions, creating a downward spiral. Another example is investment banks that hedge their corporate bond holdings using credit default swaps. If hedge funds take the other side of the swap and fund the investment by borrowing from the same bank, the spillover risk from the hedge fund to the bank increases. These types of interconnectedness may underlie some of the spillover effects in our study.


The paper appears to have crunched on a very sophisticated level through massive amounts of data, producing an analysis that would take a reader a very long time to investigate thoroughly[1].  But I also have two huge reservations.   The quotation above carries the seeds of one of them.   Most of the statements in it are actually wrong to the extent they purport to describe characteristics that are unique to hedge funds; that is, they are not true “if and only if” the subject is a HF.  For example, when the paper states “hedge funds are opaque and highly leveraged”, that is only a partly true statement.  Opaque – yes, but not that much more so than the other institutions, and not as much as you think (and also, as I realized while I wrote this post, it cannot be true of the HFs whose data they rely on – that is, the HFs they analyze cannot be opaque to the extent the paper relies on information about them!

Sure, to most observers and probably regulators, most individual hedge fund trades are opaque, yes.  But there are reporting requirements concerning equity stakes in public companies that provide disclosure on the largest equity positions, and various other ways in which hedge funds’ positions become disclosed, such as shareholder activism, being a member of an ad hoc committee in a bankruptcy, or talking up a position in some conference.  As well, other informal disclosures occur: I often found that traders had a good sense of which hedge funds had been taking positions in a distressed situation.  A prime broker would normally know reasonably thoroughly the positions of its client HFs.  Similarly, when a company goes out for a “drive-by” bond offering, or an equity raise via a private placement, the investment bank(s) running the deal invariably have a very good idea who is interested and who is not, because they have teams of sales people calling on asset managers all the time and staying up to date on what their interests may lie. Finally, to take a position in a financial asset outside of the securities exchanges, an investor must often enter into a contract in which its identity is necessarily divulged:  in loan trading, for instance, when a loan passes by assignment, the admin agent will have to sign off, and the borrower in a non-default context will also, and the identity of the buyer and seller must be disclosed.  So too in derivatives, the counterparty, often a financial intermediary, knows who it is contracting with on an ISDA form.  And, just to reinforce the point, is the “opacity” of an HF portfolio all that different from the opacity of the portfolio and trading books of the largest commercial and investment banks?  From a public investor’s perspective, I don’t think it’s all that different.  An observer of the markets may well be able to name a greater proportion of the positions held by, say, Bill Ackman’s hedge fund, or Dan Loeb’s, than those held by Goldman Sachs.  At the regulatory level, there is a difference, I admit, although I question how much actual or practical insight the regulators truly have over those institutions’ books, given their failure to apprehend any of the insolvencies in 2007-09.  As I said, the statement “hedge funds are opaque…” is indeed partly true, but just partly.

Moving on, what does the statement “hedge funds are … highly leveraged” mean, especially in comparison to the other kinds of institutions Gropp studies?  Although I do recall one memorable anecdote to the contrary[2], many of the hedge funds that I have worked with did not have any permanent leverage at all, because they held leveraged loans, HY bonds, distressed securities, ABS or other debt securities as to which the risk of illiquidity was too high to get into a margin situation in the first place.  But even assuming there are a lot of hedge funds with leverage, what makes them “highly” leveraged compared to commercial banks and investment banks from 2003-2010?  I doubt there was any hedge fund that had a leverage ratio higher than the commercial and investments banks in that period.  I would be surprised if any hedge fund had more than a 4:1 debt/equity ratio, and I would expect the average among the levered funds is less than 2:1, whereas the largest commercial and investment banks have leverage ratios in the 10:1 or higher range, depending on how one counts trust preferreds and other hybrids.  Especially for HFs that are mainly taking positions in equities, Reg U and other rules make it very difficult to do so on a basis as leveraged as a money center bank’s balance sheet is leveraged.

Tying this back to their research: the hedge fund universe that Gropp works with in his paper consists, the 2013 paper says, of 47  of the largest and most liquid such funds which comprise a “Hedge Fund Equally Weighted Index” which is one of the few sources for daily data on hedge fund performance.  But the researchers do not seem to know, and probably it is not disclosed, which, if any of those, are levered and to what extent and did it differ from day to day.  So I think the description of HFs relevant to the paper as “highly leveraged” is not supported in a scientific manner.

Moving on through Gropp’s explanation of his research, the two examples he gives of ways in which hedge funds might amplify risk are loss-and-margin consequences, and hedging credit risk via CDS’s.  But note that, again, these are not at all unique to hedge funds.  As I recall, when the subprime mortgage started, a lot of hedge funds weren’t long that asset class on borrowed money, they were short subprime MBS and indices tied thereto.  This is important because the 2013 paper states unequivocally “The subprime and financial crisis of 2007-2009 spread from mortgage-backed securities and CDOs to commercial banks and on to hedge funds and investment banks.” Think of “The Big Short”, or John Paulson being short the ABACUS vehicle in the Fabrice Tourre lawsuit.  The institutions that were long subprime were investment banks (think Merrill); the GSE’s; commercial or investment banks at home and especially abroad; andinsurance companies (AIG, the various bond insurers like MBIA and FGIC, etc).   And most of all the dozens of originators themselves, like AHM and so on.   There were certainly some mortgage funds, like the Bear Stearns’ funds, that were long subprime, but was the HF universe net long subprime?  I would be skeptical (it’s also an interesting taxonomy question relative to the research, how one should classify a HF managed by an investment bank).  So, both generally and specifically with respect to the financial crisis of 07 onwards, I doubt “loss-and-margin” consequences are particularly unique to hedge funds, especially in reference to subprime-mortgage-related assets. 

The other example Gropp gives, hedging credit risk through CDS, is again, not unique to hedge funds; for example one of the biggest individual players in CDS is the mega-billion PIMCO Total Return fund.  Further, the idea of transmitting risk through CDS raises the question of how matched the intermediary’s book is – it may be the case that the value or risk of one side of a CDS position goes up in value, but whether that intermediary’s overall CDS book loses value or has increased risk exposure depends on whether there is an offsetting position, among other things. I am not even sure how accurate it is relative to the HF sample the paper studies, because the hedge fund index they study appears, as I discuss below, to be heavily weighted toward investments relating to public equity markets, not corporate credit strategies, which are a clear minority of the strategies encompassed by that index.

So that is the first reservation I have about the study, to what extent are HFs that unique in relation to the risks and characteristics identified by the researchers as compared to the other kinds of financial intermediaries studied.

The other major reservation I have is that the study is a construct of constructs with potential for measurement errors or questionable assumptions and choices at each level.  That is, value at risk, as calculated by the authors is, obviously, a construct or a model, as it is for everyone who assays such a calculation.  But on top of that, the underlying data sets their VaR model is analyzing are themselves not the actual assets and liabilities of the subcategories but proxies for those assets and liabilities and thus potentially inaccurate reporters of the underlying value at risk, especially on a daily basis.  As well, there are a variety of financial market sectors that don’t appear to be analyzed in the paper that might have been relevant.

For example, as noted above, the set of  “hedge fund” data comes from a Hedge Fund Equally Weighted Index maintained by Hedge Fund Research.  Its methodology is described here:  Fwiw, the “strategies” that are “equally weighted”  in the index are Equity Hedge; Event-Driven; Macro/CTA; and Relative Value Arbitrage; HFR maintains indices in each of those strategies and the HFRX is just the sum of the NAV of the four individual indices.  Each of the underlying indices is comprised, HFR says, of funds that, in the aggregate, have the highest statistical correlation the aggregate performance of all funds with that strategy.  So the index is itself a statistical representation.

I am not going to go into a lot of detail about the underlying index, as the scope of this post is just some high-level observations; plus, I am not pursuing tenure as a professor of finance, nor billing by the hour as an advocate for HFs so someone else is welcome to push the analysis deeper.  The keeper of the index does indeed report it on a daily basis, which I find a bit curious as I don’t know of any HFs that disclose daily NAVs.  I searched the index manager’s website a few times to see if I could confirm it was receiving daily NAV data and not making its own estimates, but could not find any statement one way or the other on the subject.   I have to take them at their word, but this is a cool article from professors at the University of Maryland who tried to create daily VaR measurements for HFs using the same index that Gropp appear to be working with; they have mixed results although their conclusion that intra-month volatility is much higher than month-to-month volatility is similar to the Gropp conclusion.

Mesirow Advanced Strategies put out a paper in 2011 entitled “Understanding Hedge Fund Indices” that  contains a short, user-friendly discussion of some of the issues with hedge fund indices, including the one used in Gropp’s paper.  It also has some eye-popping charts that show wide variances in performance among the various indices that amply illustrate the caution needed in drawing conclusions from them.  A venerable alternative investment firm called Pictet also has a paper available on the Web entitled “Hedge Fund Indices: How Representative Are They?” from which I culled this little quote: “less than 1 per cent of the hedge fund industry reports to all databases, highlighting the unrepresentative nature of hedge fund databases.”  I am sure the keepers of the HFRX would disagree, but the point is, there are intelligent voices suggesting that all HF databases be taken with at least a small grain of salt.

A further complicating factor is that a lot of HF assets are not valued on a Level I basis, but may be Level II or Level III valuations that contain greater human guesswork (link for an explanation of these terms: ) which introduces further potential for measurement errors in the data the authors study.

Finally, the identity of the components of the index are not disclosed in the Gropp paper or on the HFR website; they are only available to subscribers.  The Gropp paper references an appendix that supposedly goes into more detail, but every time I pasted the link into my web browser, I just got a “server error” message, so I could not investigate further.  But the main point is I can’t tell how US-centric they are, which seems to be reasonably important vis a vis the paper’s overarching topic of the regulation of US financial intermediaries.  .  .

The Gropp paper compares the VaR of the HF index to three baskets of equities of various large, publicly traded US-centric commercial banks, investment banks and insurance companies. To a certain extent, I question a VaR comparison between the NAVs  of HF’s and the equity prices of these other kinds of institutions, as equity prices of financial stocks are not equal to their respective NAVs, but are determined by secondary trading.  As well, all these other types of institutions have substantial operating, income-generating businesses in addition to holding portfolios of financial assets.  So, there is something of an apples and oranges comparison here, although I don’t think too much needs to be made of it; the geographical issue I mentioned above is perhaps more worth pondering.  

The Gropp paper states that its rosters of commercial banks and insurance companies is taken from a list compiled by Viral Acharya and others in a paper called “A Tax on Systemic Risk”, but, when I checked that paper for the list, I found the description in the Gropp paper did not quite match the Acharya paper (Gropp: 26 commercial banks and 31 insurance companies; Acharya: 29 commercial banks and 36 insurance companies).  I have no idea what changes were made, or whether they were explained by the link that did  not work.

Turning to the commercial bank subset, assuming it is the Acharya set, it contains the large money center banks which had substantial capital market businesses, like JPM, Citi, B of A (Gropp acknowledges that their classification as “commercial banks” is imperfect).  But this set also has numerous regional banks with no capital market business as well. 

The insurance sector  list oddly contains Countrywide Financial. That oddity is compounded by the fact that none of the largest mortgage insurers – Fannie, Freddie, MGIC – show up on the list (MBIA, AMBAC and AIG do, though).  So how accurate a list is that?

In contrast to the large number of constituents in the insurance and commercial bank sets, it is noteworthy that the investment bank category in the Gropp paper is composed of only 8 institutions (not all of which I can identify; compare the Acharya paper which lists 10 but those 10 include NYMex, Schwab, T. Rowe Price and ETrade, yet omit Jefferies, so I just don’t know how accurate these categories are).  Further, at least two of the investment banks in the Acharya list collapsed (Bear Stearns and Lehman) and one (ML) was merged out of existence, during the period studied.  It jumped out at me that the IB sample, as best as one can understand it, seems to be a little small to confidently draw conclusions from; appears to be  much more tied to equity markets than the other sets; and also a large proportion of the constituents seem to have been the subject of one-time events during the period of study, making a very noisy sample as well. 

Another aspect of “noise” in this data is how much was unknown in real time, but came out later, in the form of fines, penalties, damages, settlements etc, about the amount of contingent liabilities that various financial institutions other than HFs had during the period in question, which leads one to wonder, how accurate were the equity prices of the underlying assets and liabilities of those institutions?

A lot of other financial entities don’t appear in any of the sets Gropp studies, to the extent the paper tracks the Acharya paper’s list.  Fannie, Freddie, MGIC, as mentioned, and also Amex, Annaly, Blackrock, Capital One, CIT, Franklin Resources, Legg Mason, TD Ameritrade.  The exclusion of the mortgage –related entities utterly baffles me since the researchers state “The subprime and financial crisis of 2007-2009 spread from mortgage-backed securities and CDOs to commercial banks and on to hedge funds and investment banks.” I would have thought it would have been essential in the context of that thesis to study correlations between mortgage-centric entities and financial institutions, but no.  Also on the subject of things not studied for correlation, it struck me as odd that whole sectors of the capital markets, like HY indices, leveraged loan indices & MBS indices that are at least as credit-driven as anything the Gropp paper studies were not examined for correlations.

A last observation on the underlying data relates to what sounds like an overstatement in regard to the period studied.  The paper covers 7 years roughly.  Recall that one of its self-described principal innovations, compared to prior analyses, is to break that period up into periods of “tranquility”, “normality” and “distress”.   While I cannot get the link to the backup data to work, so I cannot be sure of what I am about to say, I gleaned from the paper that its repeated references to “periods of financial distress” are really just references to the 2007-09 period, taken as a whole.  As I said before, that was a very noisy period with all sorts of things going on – collapses, bailouts, shotgun weddings --  that had never happened before.  And, in any event, it’s just one period!  So I wind up skeptical about the prescriptive significance of finding some correlations between HF NAV changes and a small set of financial intermediaries in a single period that was complicated by many one-time events.  I don’t see how anyone could ever determine whether this was a phenomenon capable of recurring, or just a one-time confluence of factors, or an artifact of the assumptions and choices the authors made in generating the paper.  It seems impossible to replicate the conditions of the period to test the hypothesis.. 

Often a correlation exists between two data sets because both are displaying the influence of a third variable.  For example, if Mary and Herb live in Scarsdale and work at banks in Manhattan, and Mary leaves her house every morning to catch the 7:16 from Scarsdale, and, Herb leaves his house every morning to catch the 7:34, there is a high degree of correlation between their schedules, but no causation even though Mary consistently precedes Herb.  Their schedules are determined by exogenous variables, namely the schedule people who work in the banks and take Metro North to get there have to keep. 

Here for example, the correlation between the VaR metrics for HF and IB in times of financial distress could simply show, not that one caused the other, but that both categories held assets that were more similar than they were to the portfolios of commercial banks and insurance companies.  That is, IB’s assets may have been more HF-like than CB’s or insurance companies assets were – for instance, it jumps to mind that they may have had more HY and equities as a proportion of total assets than CB’s and insurers did.  The HY part of that conjecture would help explain why the authors found the correlation greater on the downside – being a debt instrument, HY can only go up so much, so starting from a non-distress point, (which is where the Gropp study starts, in 2003, a bull year), HY tends not to provide much return beyond the coupon, while in a distress environment it can fall several multiples of the coupon.

When the authors find that negative changes in HF VaR appeared to lead changes in the VaR of a portfolio of IB equities, that could just be because the HF VaR reflects daily marking to market of the underlying assets (or the index manager’s estimation thereof), undiluted by other factors that may affect the stock prices of IB’s, such as secondary market technical factors, or the market’s evaluation of the advisory and other operating businesses of the IB’s. 

Another possibility could be that hedge funds, as they saw a broad financial deterioration sweeping the developed world, looked to hedge their exposure by finding shorts, and the hugely overleveraged balance sheets of certain brokerage firms were prime candidates, better than insurance companies or FDIC-insured banks .  They are hedge funds after all.  I know I got many calls from HF clients in 2007 asking for an explanation of how SIPC receiverships worked in reference  to a generic or hypothetical brokerage firm where they had repo’d their cash balances or otherwise had exposure from other balances. And everyone remembers David Einhorn’s very public short bet against Lehman Brothers throughout 2008.  So there was definitely worry among them about the health of some brokerage firms.  And recall that the investment banking sector of the Gropp study is only 8 firms, so it is very susceptible to one or two members of the group driving the results in a certain direction. 

All of the above seem plausible to me, yet none of the above would justify any sort of heightened regulation of HFs.  I tend to think that most HFs tend to carry less leverage than money center intermediaries do.  As always, the most appropriate financial-sector-specific regulation that needs to be in place is having enough equity capital in the system to buffer it against the level of loss that might arise from a specified level of financial distress, and to apply the capital requirements broadly throughout the financial markets so they cannot be evaded .  But again I suspect the HF universe would be broadly in compliance with the kind of capital adequacy requirements now imposed on banks and the like.  I'm not necessarily averse to the concept, as I don't work in or for HFs anymore.  All I am trying to do is use my experience to pose some hopefully intelligent questions on the topic, fwiw.


[1]           In part because the authors kept much of the data crunching out of the paper itself, and in a statistical abstract, the link to which unfortunately did not work the three times I tried it.
[2]           In early 2004, I had drinks with a small group that included the number 2 guy at a well-known hedge fund that was focused on below investment grade bonds. I asked him how his fund had done in 03, and he said, proudly, “up 81%” which was incredible for the billions of AUM they had.  My next question, in all innocence, was “did you have any leverage?” and he replied, “yes, 2.7 to 1”.  When I got into the office the next day, I looked up the performance of the relevant HY index, I think it was Lehman, for 03, and, amazingly, it was exactly 30%.  Of course that was unlevered, so what that meant was the brilliant hedge fund had, as far as asset class and security selection go, been just a market performer, and its entire outperformance was due to the leverage ratio.  Of course that is only one anecdote and I have others about guys running even larger amounts of AUM in the same sector with zero leverage, fwiw (but of course they weren’t getting 81% returns).