DONALD MACKENZIE
Of all
the contested boundaries that define the discipline of sociology, none is more
crucial than the divide between sociology and economics. Despite his
synthesizing ambitions, Talcott Parsons played a critical role in reinforcing
this divide. The economy, argued Parsons and Smelser (1956: 7) is a ‘differentiated
subsystem of a more inclusive social system’. Conventional neoclassical
economics could, Parsons believed, quite appropriately be applied to that
subsystem. The technical core, so to speak, of the workings of market economies
was the business of economists, not of sociologists.
In more
recent years, a revived economic sociology has rebelled against this
intellectual division of labor, which Stark (2000) calls ‘Parsons’ Pact’. A
range of authors—amongst them White, Granovetter, Fligstein, Podolny, and
Callon—have proposed a variety of ways of conceptualizing social processes not
as ‘surrounding’ economic life but as being at its core (White 1981, 2001;
Granovetter 1985, 1990; Podolny 1993, 2001; Fligstein 1996, 2001; Callon 1998).
This chapter seeks to contribute to this post-Parsonian economic sociology not
by proposing a new approach but a new (or almost new) topic for sociological
investigation: arbitrage.
Arbitrage is trading that
exploits price discrepancies, for example differences between the prices of
the same asset at different geographical locations, or between the prices of
similar assets at the one location. There is a sense in which arbitrageurs are
the border guards, in economic practice, of the Parsonian boundary between
economics and sociology. Suppose that the prices of two similar financial
assets temporarily diverge for reasons that
This
chapter is a revised version of a paper (MacKenzie 2003) published in Economy
and Society. The case study of LTCM was
supported financially by DIRC, the Interdisciplinary Research Collaboration on
the Dependability of Computer-Based Systems (UK Engineering and Physical
Sciences Research Council grant GR/N13999), and my ongoing research in social
studies of finance is being supported by a professorial fellowship from the
United Kingdom Economic and Social Research Council (RES-051-27-0062).
are ‘sociological’ rather
than ‘economic’: investors’ irrational preferences, enthusiasms, or fears;
legal constraints (often ultimately moral in their roots: see Zelizer 1979) on
market participants such as insurance companies; regulatory impositions
(perhaps driven by political ideologies); and so on. Arbitrageurs can then
profit by buying the cheaper of the assets, and short selling the dearer
(financial terminology such as ‘short selling’ is defined in the glossary in
Table 3.1). Their purchases tend to raise the price of the
Table 3.1. Financial Terminology
Arbitrage Basis point Future
Haircut
Implied volatility
Libor (London interbank offered rate)
Option
Short selling
Swap
Swap spread
Value-at-risk
Volatility
Yield
Trading that seeks to profit from price
discrepancies A hundredth of a percentage point A standardized contract traded
on an organized exchange in which one party undertakes to buy, and the other to
sell, a set quantity of a particular asset at a set price on a given future
date When money is borrowed to buy securities such as bonds, and these are
pledged as collateral for the loan, the haircut is the difference between the
amount of money lent and the market price of the securities
The volatility of a stock or index consistent with the
price of options on the stock or index The
average rate of interest at which banks with the highest credit ratings are
prepared to lend each other funds
A contract that gives its purchaser the
right, but not the obligation, to buy (call) or to sell (put) an asset at a
given price on, or up to, a given future date (the ‘expiration’)
Selling an asset one does not own, for
example by borrowing it, selling it, and later repurchasing and returning it
A contract to exchange two income streams
The difference between the fixed interest rate at which interest-rate swaps can be entered into and
the yield
of a
government bond of equivalent maturity denominated in the same currency A
method of estimating the exposure of a portfolio of assets to potential losses
The extent of the fluctuations of the price
of an asset, conventionally measured by the annualized standard deviation of
continuously-compounded returns on the asset
The yield of a bond is the rate of return it
offers at its current market price
cheaper
asset, and their sales to lower that of the dearer, thus helping to restore
equality. The consequently plausible assumption that pricing discrepancies
will be eliminated by arbitrage allows the development of elegant and
influential economic models of markets. Arbitrage-based reasoning is, for
example, central to the work that has won Nobel Prizes in economics for three
of the five finance theorists so far honored: Merton H. Miller, Robert C.
Merton, and Myron S. Scholes.
Arbitrage is thus seen by
economists as making it possible for financial markets to be efficient even in
the presence of investor irrationality and other social or psychological
‘factors’:
Neoclassical finance is a
theory of sharks [i.e. arbitrageurs] and not a theory of rational homo
economicus... [Arbitrageurs spot [price discrepancies], pile on, and by their
actions they close aberrant price differentials... Rational finance has
stripped the assumptions [about the behaviour of investors] down to only those
required to support efficient markets and the absence of arbitrage, and has
worked very hard to rid the field of its sensitivity to the psychological
vagaries of investors (Ross 2001: 4).
Furthermore,
finance theory is itself drawn on by modern arbitrageurs, so arbitrage is a key
issue for the ‘performativity’ of economics: the thesis that economics creates
the phenomena it describes, rather than describing an already existing
‘economy’ (Callon 1998).1 To the extent that arbitrageurs can
eliminate the price discrepancies that finance theory helps them to identify,
they thereby render the theory performative: price patterns in the markets
become as described by the theory.
Despite
the centrality of arbitrage, there has been little empirical study of it by
economists and, for all the flowering in recent years of the sociology of the
financial markets, almost none by sociologists. The only extant sociological
study focusing directly on arbitrage is by Beunza and Stark (see Chapter 4,
this volume), which is primarily descriptive: it does not, for example, investigate
the capacity of arbitrage to eliminate price discrepancies and thus maintain
the boundary between ‘the social’ and ‘the economic’. That investigation, in
contrast, is the goal of this chapter. It focuses on the hedge fund, LongTerm
Capital Management (LTCM).2 LTCM was highly skilled: it emerged from
the celebrated arbitrage group at the investment bank Salomon Brothers, a group
headed by John Meriwether, by common consent the most talented bond trader of
his generation. LTCM, set up and led by Meriwether, had available to it the
best of finance theory: amongst its partners were the Nobel laureates Merton
and Scholes. It was hugely successful: at its peak, it deployed what is almost
certainly the largest single concentration of arbitrage positions ever. And
yet, in August and September 1998, in one of the defining moments of the
economic history of the 1990s, adverse price movements drove LTCM to the brink
of bankruptcy (it was recapitalized by a consortium of the world’s leading
banks, coordinated by the Federal Reserve Bank of New York).
The crisis of LTCM has
provoked widespread comment—for example, books by Dunbar (2000) and Lowenstein
(2000)—and even features in a novel (Jennings 2002). Typically, popular
commentary advances two accounts:
1. The partners in LTCM were
guilty of greed and gambling (consciously reckless risk-taking).
2. The partners in LTCM had
blind faith in the accuracy of finance theory’s mathematical models.
More
informed discussion (e.g. President’s Working Group on Financial Markets 1999)
avoids blaming individuals’ alleged character flaws, and instead advances a
third hypothesis:
3. LTCM was over-levered—too
high a proportion of its positions were financed by borrowing, rather than by
LTCM’s own capital.
This
third hypothesis, however, explains at most LTCM’s vulnerability to the events
of August and September 1998: it does not explain those events. The most common
explanation of them is:
4. On August 17, 1998, Russia
defaulted on its ruble-denominated bonds and devalued the ruble. This triggered
a ‘flight-to-quality’ in the financial markets—a sudden greatly increased
preference for financial assets that were safer (less prone to default) and
more liquid (more readily bought and sold).
That
there was a flight-to-quality in August and September 1998, and that the
Russian default triggered it, cannot be denied. The hypothesis of this chapter,
however, is that superimposed on the flight-to-quality, and sometimes cutting
against it, was a process of a different, more directly sociological kind:
5. The success of LTCM led to
widespread imitation (White 1981, 2001; Fligstein 1996, 2001), and the
imitation led to a ‘superportfolio’ of partially overlapping arbitrage
positions. Sales by some holders of the superportfolio moved prices against
others, leading to a cascade of self-reinforcing adverse price movements.
This
chapter draws upon sources of information of four kinds. First is a set of
‘oral history’ interviews conducted by the author with partners in and
employees of LTCM. These initial interviews were then followed up by further
exchanges in person, by electronic mail, and by telephone. The second source of
information is interviews conducted with other key individuals, not affiliated
with LTCM, who were also active in the markets within which LTCM operated.
These interviews give additional insight into the market processes surrounding
LTCM, and make it possible to check for any ‘exculpatory’ bias in the views of
LTCM insiders. These first two sources then permit reliable published sources
on LTCM to be distinguished from unreliable ones (the only consistently
reliable, detailed source is Perold 1999), and these form the third source of
data drawn on here. The fourth source is the price movements of key parts of
LTCM’s portfolio in the months of its crisis, August and September 1998:
readers interested in how these movements serve as a quantitative test of this
chapter’s hypothesis should consult MacKenzie (2003: 367-70).
An
economist might object that a study of LTCM is not really a study of arbitrage.
In finance theory, arbitrage is conceived as involving no risk and demanding no
capital (it can be performed entirely with borrowed cash and/or securities).
These are, indeed, precisely the assumptions that make arbitrage’s capacity to
close price discrepancies unlimited. LTCM’s activities, in contrast, involved
risk (even in ‘normal’ times, not just in 1998), and demanded at least modest
amounts of capital. The response to this economist’s objection is simple
(Shleifer and Vishny 1997): much ‘real-world’ arbitrage involves risk and
demands capital. Certainly, there is a spectrum in this respect—there are some
arbitrages, typically of evanescent ‘mispricings’, that are very low risk—but
LTCM’s activities are reasonably characteristic, in terms of their risks and
their capital demands, of a large class of arbitrage trades, including some of
great theoretical significance, such as the arbitrage that enforces
Black-Scholes-Merton option pricing, the single most influential model in
finance theory (MacKenzie and Millo 2003).
This chapter has four
parts. After this introduction comes a section describing LTCM’s arbitrage
trading and its risk management. Then comes a section on LTCM’s 1998 crisis, which,
after briefly discussing the other explanations, draws on the interview data to
flesh out the ‘superportfolio’ explanation. In the conclusion, I return to more
general issues of the sociology of arbitrage and its bearing upon the relations
of ‘economy’ and ‘society’.
LTCM,
which began trading in February 1994, was a hedge fund based in Greenwich,
Connecticut. It also had an office in London and a branch in Tokyo, and its
primary registration was in the Cayman Islands. Its offices were not
ostentatious (its Greenwich head office, for example, was a modest, low-rise
suburban office block), and in terms of personnel, LTCM was of limited size: by
September 1997, 15 partners and around 150 employees. These people, however,
managed a considerable body of assets: in August 1997, LTCM’s assets totalled
$126 billion, of which $6.7 billion was the fund’s own capital. While most
hedge funds cater for rich individuals, they were the source of less than 4% of
LTCM’s capital, which came mostly from financial institutions, particularly
banks (Perold 1999: A2, A22).
LTCM’s
basic strategy was ‘convergence’ and ‘relative-value’ arbitrage: the
exploitation of price differences that either must be temporary or that have a
high probability of being temporary. Typical were its many trades involving
swaps: by the time of LTCM’s crisis, its swap book consisted of some 10,000
swaps with a total notional value of $1.25 trillion (Anon 2000). A swap is a
contract to exchange two income streams, for example fixed-rate and
floating-rate interest on the same notional sum. The swap spread is the
difference between the fixed interest rate at which swaps can be entered into
and the yield of a government bond with a similar maturity denominated in the
same currency. Swap spreads can indicate arbitrage opportunities because the
party to a swap which is paying a floating rate of interest while receiving a
fixed rate is in the same situation as someone who has borrowed money at a
floating rate and used it to buy a bond which pays a fixed amount of interest.
If there is sufficient discrepancy between the terms on which swap contracts
can be entered into and on which positions in bonds in the same currency and of
similar maturities can be financed, arbitrage may be possible.
Several
features of swap-spread arbitrage go to the heart of LTCM’s strategy. First is
leverage. LTCM swap-spread trades were highly levered: that is, were
constructed largely with borrowed capital. High levels of leverage were
necessary if the small price discrepancies LTCM was exploiting were to yield
adequate profits, and did not necessarily imply huge risk (as much subsequent
commentary suggested). The risks of swap-spread trades, for example, are rather
limited. Bond prices and the terms upon which swaps are offered fluctuate
considerably, particularly as interest rates vary. LTCM, however, almost always
neutralized that latter risk by constructing ‘two-legged’ trades, in which the
effects on one leg of a change in interest rates would be cancelled out by its
equal-but-opposite effect on the other leg. The chief market risk of
swap-spread trading is of the spread temporarily moving in an unfavorable
direction, but if that were to happen the arbitrageur can simply continue to
hold the position and wait until such time as it became profitable to liquidate
it. Indeed, if necessary the position can be held until the bond matures and
the swap expires. That feature was the essence of convergence arbitrage: if
held to maturity, a convergence arbitrage position has to make a profit, whatever
the market fluctuations along the way.
If the
risks were limited, the profits from LTCM’s swap-spread and similar arbitrage
trading were impressive. Between February and December 1994 LTCM’s returns
before fees were 28.1% (unannualized); after management and incentive fees were
deducted, investors received 19.9% (unannualized). Gross returns in 1995 were
59.0%, and returns after fees 42.8%; in 1996, the corresponding figures were
61.5% and 40.8%.3
Although
LTCM was active in the US and Japanese markets, it also had particularly heavy
involvement in European markets. In the 1990s, financial deregulation in Europe
proceeded apace, but arbitrageurs such as LTCM initially found much less
competition than in the United States or Japan: ‘the Japanese banks...were the
ones who were terribly interested in setting up proprietary desks. The European
banks were still a bit hesitant’ (Kaplanis interview). LTCM scrutinized the
‘yield curves’ for European government bonds (see Figure 3.1), along with the
corresponding swap curves, looking for the ‘bulges’ and other anomalies that
might indicate arbitrage opportunities. If LTCM was confident it understood the
reasons for anomalies—frequently
Years
to maturity
Figure 3.1. A
Hypothetical Example of a Yield Curve (highly schematic). Yield curves
usually (but not always) have the upward slope shown here
|
they
were matters such as regulatory requirements on insurance companies to purchase
bonds of particular maturities—it would seek to exploit them by trades
carefully constructed to neutralize the risks of interest-rate fluctuations or
of changes in the overall steepness of the yield curve.
As well
as diversifying geographically, LTCM also diversified from bonds and
interest-rate swaps into other asset classes. Some relative-value trades
involved pairs of shares, such as Royal Dutch and Shell Transport (Perold 1999:
A9). Shares of Royal Dutch are traded in Amsterdam and the corresponding
American Depository Receipts trade in New York, while shares of Shell trade in
London, but the two sets of shares represent equivalent rights of ownership of
what is essentially a single company (Royal Dutch/Shell) and thus equivalent
claims on a single income stream. However, they often trade at different
prices, for example, because the way dividends are taxed leads investors to
prefer one or the other. in a situation like this, arbitrage can be attractive
if the difference between the prices of the two sets of shares is expected to
narrow, to widen, or to change direction. LTCM could profit from an expected
change in relative value while being protected from fluctuations either in the
overall stock-market or in the performance of Royal Dutch/Shell.
Another
equity-related position, taken on in 1997, responded to an anomaly developing
in the market for equity index options with long expirations (see Table 3.1 for
the meaning of ‘option’). Increasingly, banks and other financial companies
were selling investors products with returns linked to gains in equity indices
but also a guaranteed ‘floor’ to losses. Long-maturity options were attractive
to the vendors of such products as a means of hedging their risk, but such
options were in short supply. The price of an option is dependent upon
predictions of the volatility of the underlying asset, and market expectations
of that volatility (implied volatility) can be deduced from option prices using
option pricing theory. In 1997, however, the demand for long-expiry options had
pushed the volatilities implied by their prices to levels that seemed to bear
little relation to the volatilities of the underlying indices. Five-year
options on the S&P 500 index, for example, were selling at implied
volatilities of 22% per annum and higher, when the volatility of the index
itself had for several years fluctuated between 10% and 13%, and the implied
volatilities of shorter-term options were also much less than 20% per annum.
LTCM therefore sold large quantities of five-year index options, while hedging
the risks involved with index futures and sometimes short- expiry options
(Perold 1999: A7, A8).
Not all
LTCM’s trades were successful: ‘We lost a lot of money in France in the front
end [of the bond yield curve]’, says LTCM’s Eric Rosenfeld (interview).
Nevertheless, as noted above, extremely attractive overall returns were earned,
and the volatility of those returns was reassuringly low. Most of LTCM’s
positions were almost completely insulated from broad market movements. The
firm had only limited involvement in areas where the chance of default was
high, such as in high-yield (junk) corporate bonds or ‘emerging markets’, such
as Russia, Thailand, and Argentina. Risks were carefully calculated and
controlled using the ‘value-at-risk’ approach standard in the world’s leading banks
(Meriwether interview). In the case of the dollar swap spread, for example,
historical statistics and judgements of likely future values led LTCM to
estimate that the spread had an ‘equilibrium value’ of around 30 basis points,
with a standard deviation of about 15 basis points per annum (Rosenfeld
interview). Using those estimates, it was then possible to work out the
relationship between the magnitude of possible losses and their probabilities,
and thus the ‘value-at-risk’ in the trade.
When a
firm holds a large number of positions, the estimation of the probabilities of
loss in individual positions is less critical to overall value-at-risk than
estimates of correlation between positions. If correlations are low, a large
loss in one position is unlikely to be accompanied by large losses in others,
so aggregate value-at-risk levels will be modest. In contrast, if correlations
are high, then when one position goes bad, it is likely that other positions
will also do so, and overall value-at-risk will be high. LTCM’s positions were
geographically dispersed, and in instruments of very different kinds. At the
level of economic fundamentals, little if anything connected positions such as
on the spread between US government bonds and mortgage- backed securities, on
the difference between the prices of the shares of pairs of companies such as
Royal Dutch and of Shell, on the bulges in the yen yield curve, on the chances
of specific mergers failing, and so on. LTCM was aware that its own and other
arbitrageurs’ involvement in these diverse positions would induce some
correlation, but nevertheless the observed correlations, based on five years of
data, were very small: typically of the order of 0.1 or lower.
The
standard deviations and correlations that went into LTCM’s aggregate risk model
were, however, not simply the empirically observed figures but deliberately
conservative estimates of their future values. The observed standard deviation
of the US dollar swap spread, for example, was around 12 basis points a year,
while, as noted above, the risk model assumed it would be 15 (Rosenfeld
interview). Past correlation levels, likewise, were ‘upped’ (Meriwether
interview) to provide a safety factor: despite observed correlations being 0.1
or less, LTCM was ‘running analyses at correlations at around 0.3’. The
consequence of this conservatism was that while the firm’s risk model suggested
that the annual volatility (standard deviation) of its net asset value would be
14.5%, in actuality it was only 11% (Meriwether interview). Both figures were
considerably less than the risk level of 20% that investors had been told to
expect (Perold 1999: A11).
Of course, such statistical
analyses of risk assumed the absence of catastrophic events in the financial
markets. LTCM’s key members were well aware of the possibility of such events.
So LTCM also ‘stress tested’ its portfolio, investigating the consequences of
hypothetical events too extreme to be captured by statistical value-at-risk
models, events such as a huge stock market crash or failure of European
economic and monetary union (EMU). As well as investigating the consequences of
such events for market prices and for LTCM’s risk capital, it also
calculated—and set aside—the funds necessary to cope with a sudden increase in
‘haircuts’ (see Table 3.1) in a situation of stress. When an event could have
particularly catastrophic consequences, LTCM either turned to insurance—it
bought insurance against bond default by the government of Italy—or balanced
its portfolio to minimize consequences, as in the case of EMU failure.
The
partners in LTCM, therefore, believed themselves to be running the fund
conservatively, and in the modest volatility of its returns they had evidence
for the correctness of this belief. After the fund’s crisis, it was commonly
portrayed as wildly risk-taking, but I have found almost no one inside or
outside LTCM who can be proved to have expressed that view prior to the crisis.
Gambling—consciously reckless risk-taking—does not explain LTCM’s 1998
disaster. Nor does the second hypothesis advanced in the commentary: blind
faith in mathematical models. Models were much less critical to LTCM’s trading
than commonly thought. Many of the pricing anomalies it sought to exploit could
be identified without sophisticated modeling, and although models were
important in how its trades were implemented and in assessing the risks
involved, all those involved knew that models were approximations to reality
and a guide to strategy rather than a determinant of it. LTCM’s traders had
often themselves developed the models they used: no one was more aware than
they of the models’ likely deficiencies. The way in which the standard
deviations and correlations in the most important model of all— LTCM’s overall
risk model—were increased by explicitly judgement-based ‘safety factors’ is
indicative of that.
The
third posited explanation of LTCM’s crisis—over-leverage—is almost
tautologically correct. If LTCM had been operating without leverage, or at
lower levels of leverage, the events of August and September 1998 would have
placed it under much less strain. However, LTCM’s levels of leverage were
comparable to those of the leading global investment banks (Perold 1999: C11,
C12; President’s Working Group on Financial Markets 1999: 29) and, in any case,
blaming LTCMs crisis on leverage is like attributing a plane crash to the fact
that the aircraft was no longer safely in contact with the ground: it
identifies the source of overall vulnerability but not the specific cause. That
cause was the financial crisis of August and September 1998, and in particular
the way in which the adverse price movements of those months exceed LTCM’s, or
anyone else’s, expectations. Conventionally, the 1998 crisis is regarded as a
‘flight-to-quality’: an increased relative preference for assets with low risk
of default, and/or an increased preference for more liquid assets, in other
words those that can more readily be bought and sold at or near prevailing
market prices.4 The interviews drawn on here, however, suggest a
rather different, more directly sociological process. Meriwether’s group at
Salomon and at LTCM earned remarkable profits, and were known to have earned those
profits. This encouraged others—in other investment banks, and increasingly in
other hedge funds—to follow similar arbitrage strategies. Others were being
told: ‘LTCM made $2 billion last year. Can’t you?’ (Meriwether interview). For
example, LTCM’s success meant that it rapidly became largely closed to new
investors, and in January 1998 a new fund, Convergence Asset Management,
‘raised $700 million in a single month purely from disgruntled investors denied
a chance to buy into LTCM’ (Dunbar 2000: 197).
LTCM
tried hard not to reveal its trading positions. For example, it would avoid
using the same counterparty for both legs of an arbitrage trade. However, as
one trader and manager not connected to LTCM put it: ‘(t)he arbitrage
community...are quite a bright lot, so if they see a trade happening—and the
market gets to find out about these trades, even if you’re as secretive as
Long-Term Capital Management—they’ll analyze them and realize there’s an
opportunity for themselves’ (Wenman interview).
LTCM’s
basic strategy—convergence and relative-value arbitrage—had to be disclosed to
potential investors and thus could not be hidden, and others seeking to follow
that strategy would often be led to take similar positions to LTCM’s. It
‘doesn’t take a rocket scientist’ to discover the kinds of arbitrage
opportunities being pursued (Rosenfeld interview), especially when discovering
one leg of an LTCM trade through being a counterparty to it would greatly
narrow the range of possible other legs. Some of LTCM’s trades were well-known
to market insiders before LTCM became involved: the Royal Dutch-Shell trade,
for example, was the ‘classic European arbitrage trade’ (Wenman interview), and
the relationship between Royal Dutch and Shell shares had even been discussed
in the academic literature (Rosenthal and Young 1990).
As a
result of conscious and unconscious imitation, many of LTCM’s positions became
‘consensus trades’ (Kaplanis interview). Of course, the growing number of
arbitrage traders in investment banks and hedge funds did not sit down together
in a room to identify good arbitrage opportunities. Rather, ‘the arbitrage
philosophy... had been disseminated, well disseminated by August 98; it was
there in quite a few hedge funds, it was there in quite a few firms. So Salomon
[and LTCM] lost their uniqueness in doing these things. There were many, many
others that could do them’. There was some communication: if you talk[ed] to another
arb trader in the street they’d say, “Oh yes, I have this as well, I have that
as well” ’ (Kaplanis interview). But even had there not been communication,
many traders would still have identified the same opportunities. ‘And what
happened by September ’98 is that there was a bunch of arb trades that became
consensus. People knew that the UK swap spreads was a good trade, people knew
that US swap spreads was a good trade’ (Kaplanis interview). No other market
participant had the same portfolio as LTCM—many arbitrageurs were restricted
to particular portions of the spectrum of arbitrage trades—but, collectively,
much of LTCMs portfolio of positions was also being held by others.
The initial effect of
imitation was probably to LTCM’s benefit. If others are also buying an
underpriced asset and short selling an overpriced one, the effect will be to
cause prices to converge more rapidly. However, imitation also meant that when
existing trades had been liquidated profitably, replacing them was more
difficult:
Author:
Did you find that, as the years went by with LTCM—’94, ’95, ’96, ’97 and so
on—did you find... that the opportunities were drying up a bit?
Rosenfeld: Yes, big.
In the
summer of 1998, imitation switched to become a disastrously negative factor
because of two decisions, neither of which had anything directly to do with
LTCM. In 1997, Salomon Brothers was taken over by the Travelers Corporation,
whose famously risk-averse chair, Sandy Weill, was building the world’s largest
financial conglomerate, Citigroup (Booth 1998). According to Kaplanis,
Salomon’s US arbitrage desk had not been consistently successful since the
departure of Meriwether and his group, and in the first half of 1998 it was
loss-making. Though Kaplanis, promoted to head of global arbitrage for Salomon,
advised against it, the decision was taken to liquidate the US arbitrage desk’s
portfolio as quickly as possible, and responsibility for the liquidation was
passed to Salomon’s US customer desk. Since the latter was ‘not accountable for
the losses generated as a result of the liquidation, the speed of the latter
was faster than would otherwise have been the case’. This caused losses not
just to Travelers/Citicorp but also to all of those who had similar positions:
‘not only did we lose money as the positions went against us as we were selling
them, but all the other funds that also had these consensus trades also
started losing money’ (Kaplanis interview).
If the
liquidation of Salomon’s arbitrage positions was a background factor in the
problems of the summer of 1998, the immediate cause of the 1998 crisis was
Russia’s August 17 default on its ruble-denominated debt. That Russia was in
economic trouble was no surprise: what was shocking was that it (unlike
previous debtor governments) should default on debt denominated in domestic
currency. ‘I was expecting them [the Russian government] to just print money’
to meet their ruble obligations, says Kaplanis, and he was not alone in this
expectation. Initially, the default seemed to be an event of only modest
significance for firms, such as LTCM, that had little exposure to Russia or
similar ‘emerging markets’: on August 17, the Dow Jones rose nearly 150 points
(Lowenstein 2000: 144). In the days that followed, however, it became
increasingly clear that the default had triggered what Kaplanis calls an
‘avalanche’. The default was combined with a devaluation of the ruble and a
month’s ban on Russian banks complying with forward contracts in foreign
exchange (Dunbar 2000: 200-1). Since western investors used these contracts to
hedge against the declining value of the ruble, widespread losses were
incurred. LTCM’s losses in the Russian market were limited, but other arbitrageurs
carrying losses began liquidating positions elsewhere to meet the demands of
their counterparties. A hedge fund called High-Risk Opportunities, which had a
large position in ruble-denominated bonds, was forced into bankruptcy, owing
large sums to Bankers Trust, Credit Suisse, and the investment bank Lehman
Brothers. Rumours began to circulate that Lehman itself faced bankruptcy. For
weeks, Lehman ‘went bankrupt every Friday’ according to the rumour mill. Though
the bank survived, its stock price suffered badly.
In a
situation in which the failure of a major investment bank was conceivable,
there was indeed a flight-to-quality. Though there are exceptions, convergence
and relative-value arbitrage typically involves holding the less liquid of a
pair of similar assets. In August and September 1998 the prices of illiquid
assets fell sharply and those of liquid ones rose, causing losses to
convergence and relative-value arbitrageurs. LTCM had known perfectly well that
a flight- to-quality could happen and that this would be its consequence.
Indeed, it was of the very essence of convergence and relative-value arbitrage
that spreads could widen—prices could move against the arbitrageur—before a
trade finally converged. For that reason, LTCM had required investors to leave
their capital in the fund for a minimum of three years: it was this restriction
that made the fund Long-Term Capital Management. If spreads widened,
however, it was assumed that arbitrage capital would move in to exploit them,
and in so doing restrict the widening (Rosenfeld interview). Indeed, once
spreads had become wide enough, the actions of ordinary investors were expected
to reduce them.
The
configuration of the markets by August 1998, however, was that the widening of
spreads was self-feeding rather than self-limiting. As arbitrageurs began to
incur losses, they almost all seem to have reacted by seeking to reduce their
positions, and in so doing they intensified the price pressure that had caused
them to make the reductions. In some cases, senior management simply became
‘queasy’ (Rosenfeld interview) at the losses that were being incurred, and
unwilling to incur the risk of further, possibly larger, losses before trades
turned profitable. In the United Kingdom, for example, Salomon, LTCM, a large
British clearing bank, and others had all taken positions in the expectation
of a narrowing of sterling swap spreads. As those spreads widened, the senior
management of the clearing bank decided to exit. Such a decision by management
might even be anticipated by the traders: ‘you know that if your manager sees
that you’re down $10 million... the likelihood that he will ask you to get out
of this position is very high. It’s not a formal stop-loss but...it’s there’ (Kaplanis
interview).
Another
factor may paradoxically have been modern risk management practices,
particularly the ‘value-at-risk’ method of measuring and managing the exposure
of a portfolio of assets to losses. This statistical technique allows senior
management to control the risks incurred by trading desks by allocating them a
risk limit, while avoiding detailed supervision of their trading. When a desk
reaches its value-at-risk limit, it must start to liquidate its positions. Says
one trader: ‘a proportion of the investment bank[s] out there... are managed by
accountants, not smart people, and the accountants have said, “well, you’ve hit
your risk limit. Close the position” ’ (Wenman interview). An international
change in banking supervision practices increased the significance of
value-at-risk. Banks are required to set aside capital to meet the various
risks they face, and in 1996 they began to be allowed to use value- at-risk
models to calculate the set-aside required in respect to fluctuations in the market
value of their portfolios (Basle Committee on Banking Supervision 1996). The
change was attractive to banks because it reduced capital requirements, but it
had the consequence that as market prices move against a bank and become more
volatile, it has either to raise more capital to preserve its trading
positions, a slow and often unwelcome process, or to try to liquidate those
positions.
The
consequences for LTCM of these processes went beyond losses on individual
trades. ‘[A]s people were forced to sell, that drove the prices even further
down. Market makers quickly became overwhelmed, where the dealers, who would
[normally] be willing to buy or sell those positions were simply unwilling to
do it, and they either said, “Just go away. I’m not answering my phone” or set
their prices at ridiculous levels’ (Shaw interview). The simple fact that the
crisis occurred in August, the financial markets’ main holiday month and thus
typically the worst time to try to sell large positions, may have exacerbated the
effects on prices. Crucially, correlations between the different components of
LTCM’s portfolio leapt upward from their typical level of 0.1 or less to around
0.7 (Leahy interview). Suddenly, a whole range of positions— hedged, and with
little or nothing in common at the level of economic fundamentals—started to
incur losses virtually across the board. LTCMs losses were stunning in their
size and rapidity: in August 1998, it lost 44% of its capital. However, though
massive, and far greater than had seemed plausible on the basis of LTCM’s risk
model, this loss was not in itself catastrophic. LTCM was, it seemed, a long
way from being bankrupt, and indeed, the widening of spreads meant that the
arbitrage positions it held had become more attractive. Spreads could be
expected to fall—indeed, they have subsequently fallen—and as they did LTCM’s losses
could be recouped and profits made.
This
would happen, however, only if LTCM survived to make those profits. At this
point a social process of a different kind intervened: in effect, a run on the
bank. ‘If I had lived through the Depression’, says Meriwether: ‘I would have
been in a better position to understand events in September 1998’ (Meriwether
interview). Unlike investment banks, which report their results quarterly, LTCM
and other hedge funds report monthly. On September 2, Meriwether faxed LTCM’s
investors its estimate of the August loss. His fax, intended to be private to
LTCM’s investors, became public almost instantly: ‘Five minutes after we sent
out first letter...to our handful of shareholders, it was on the Internet’
(Merton interview). In an already febrile atmosphere, news of LTCM’s losses fed
fears of the fund’s imminent collapse. These fears had two effects. First, they
had an immediate effect on the prices of assets LTCM was known or believed to
hold in large quantities. Such assets became impossible to sell at anything
other than distressed prices. Beliefs about LTCM’s portfolio were sometimes far
from accurate: after the crisis LTCM was approached with an offer to buy six
times the position it actually held in Danish mortgage- backed securities
(Meriwether interview). Nevertheless, presumptions about its positions were
accurate enough to worsen its situation considerably.
The
second effect upon LTCM of fears of its collapse was even more direct. Its
relationship to its counterparties typically was governed by ‘twoway
mark-to-market’: as market prices moved in favour of LTCM or its counterparty,
solid collateral, such as government bonds, flowed from one to the other. In
normal times, in which market prices were reasonably unequivocal, it was an
eminently sensible way of controlling risk by minimizing the consequences of
default. In September 1998, however, the markets within which LTCM operated had
become illiquid. There was ‘terror’ that LTCM was going to liquidate, says
Meriwether (interview). The loss caused to a counterparty if that happened
could be mitigated by it getting as much collateral as possible from LTCM
before liquidation, and this could be achieved by ‘marking against’ LTCM: by
choosing, out of the wide spectrum of plausible market prices, a price
unfavourable to LTCM, indeed predicated upon the latter’s failure (Merton
interview; Meriwether interview). LTCM had the contractual right to dispute
unfavourable marks: in its index options contracts, for example, such a dispute
would have been arbitrated by getting price quotations from three dealers not
directly involved. These dealers, however, would also be anticipating LTCM’s
failure, so disputing marks would not have helped greatly. The outflows of
capital resulting from unfavourable marks were particularly damaging in LTCM’s
index option positions, where they cost the fund around $1 billion, nearly half
of the September losses that pushed it to the brink of bankruptcy (Rosenfeld
interview).
LTCM kept its
counterparties and the Federal Reserve informed of the continuing deterioration
of its financial position. On September 20, 1998, staff from the Federal
Reserve Bank of New York and Assistant Secretary of the Treasury Gary Gensler
met with LTCM. By then, it was clear that without outside intervention
bankruptcy was inevitable. In the words of William J. McDonough, President of
the Federal Reserve Bank of New York:
Had Long-Term Capital been
suddenly put into default, its counterparties would have immediately ‘closed
out’ their positions... [I]f many firms had rushed to close out hundreds of
billions of dollars in transactions simultaneously... there was a likelihood
that a number of credit and interest rate markets would experience extreme
price moves and possibly cease to function for a period of one or more days and
maybe longer (McDonough 1998: 1051-2).
If ‘the
failure of LTCM triggered the seizing up of markets’, said Alan Greenspan, it
‘could have potentially impaired the economies of many nations, including our
own’ (Greenspan 1998: 1046).
McDonough brokered a
meeting of LTCM’s largest counterparties, which concluded that a
recapitalization of LTCM would be less damaging to them than a fire sale of its
assets. Fourteen banks contributed a total of $3.6 billion, in return becoming
owners of 90% of the fund. LTCM’s investors and partners were not bailed out:
they were left with only $400 million, a mere tenth of what their holdings were
worth not long previously. The recapitalization did not immediately end the
crisis: many feared that the consortium that now owned LTCM might still decide
on an abrupt liquidation. On October 15, 1998, however, the Federal Reserve
cut interest rates without waiting for its regular scheduled meeting, and the
emergency cut began to restore confidence. It also gradually became clear that
the consortium was intent on an orderly, not a sudden, liquidation of LTCM’s
portfolio, which was achieved by December 1999.
What,
then, might a sociology of arbitrage consist in, and how does the case of LTCM
bear upon it? Three key points emerge. First, arbitrage has a ‘Granovetterian’
sociology (Granovetter 1985, 1990): it is an activity conducted not by anonymous,
atomistic economic agents, but by people who are often personally known to each
other. Second, included in the possible forms of interaction amongst these
people is imitation, and this has particularly dangerous consequences (as in
the more general economic sociology models of White and Fligstein). Third, for
this and other reasons the capacity of arbitrage to insulate ‘the economic’
from ‘the social’ is limited: indeed, the interweaving of the economic and the
social is too intimate to be captured even by notions of imperfect insulation.
Interviewee
David Wenman’s use of the phrase ‘arbitrage community’ is not happenstance:
arbitrageurs often know each other and are affected by each other. ‘Community’
does not imply harmony. For example, one interviewee at LTCM suggested that it
had generated resentment amongst Wall Street investment banks (for instance by
pressing hard to reduce ‘haircuts’) and that others ‘were, I think, jealous of
the money we made’. Resentment and jealousy, however, are indicative that those
involved were not atomistic individuals, but mutually aware and mutually
susceptible. Positive forms of this awareness and susceptibility were also
evident: I was struck, especially during the process of getting interviewees’
permission for quotation, how exercised they often were not to give offence to
each other.
These
issues of mutual susceptibility are not matters incidental to the real business
of arbitrage, because that real business depends upon mundane forms of social
interaction with personally known others. To perform its arbitrages, the
Salomon/LTCM group had to borrow money (via what participants call ‘repo’, in
which the borrowed money is used to buy securities that are pledged as
collateral for the loan) and also had to borrow bonds (for short sale). Others
of its trades, for example the Royal Dutch/Shell arbitrage, were implemented by
arranging ‘total return swaps’ with banks. All these were wholly legitimate
activities, but getting the best possible repo, bond borrowing and swap terms
was critical to the profitability of arbitrage exploiting small price
discrepancies. It could be done better amongst personally-known people, rather
than by anonymous commercial interaction. In the 1970s and 1980s, for example,
‘repo...wasn’t done by the top people at the firm: it was almost like a clerk’s
job’, and Rosenfeld and his Salomon and future LTCM colleagues ‘always spent a
lot of time with those guys and that was very important to us’ (Rosenfeld
interview).
The
emphasis in commentary on LTCM on its use of mathematical models has diverted
attention from the extent to which its arbitrage activities (and also those of
its predecessor group at Salomon) rested upon a Granovetterian, institutional
understanding of the embedded nature of markets. Meriwether’s reputation as a
trader in the US bond market rested less on mathematical sophistication than on
his understanding of matters like who held which bonds and why. ‘Mathematics
was helpful’, he says, but understanding the institutional structure of the
bond market was ‘more important’ (Meriwether interview).
As
Salomon’s arbitrage activities began to expand overseas, Meriwether realized
that it would not be enough simply to send Americans, however sophisticated
mathematically, into overseas markets. ‘Knowing the culture was more important
than just quantitative knowledge’, he says. Typically, Salomon would seek to
recruit people brought up overseas, train them in New York, and then send them
back to the markets in the countries in which they were raised. The head of
Salomon’s trading activities in Japan, the legendarily-successful Shigeru
Miyojin is an instance. Someone who did not know Japanese would be at a
disadvantage, and in Japan (as elsewhere) the price discrepancies that were of
interest to arbitrage would typically be ‘driven by the tax and regulatory
framework’. An outsider would often find that framework hard to comprehend in
sufficient depth (Meriwether interview).
The
Granovetterian sociology of market embedding is thus evident in the normal
practice of arbitrage. In the case of LTCM, however, that embedding took a very
specific form, imitation, and this is the second aspect of the sociology of
arbitrage that needs emphasizing. The underlying general point is well-known to
economic sociology, and has been emphasized, for example, by White (1981, 2001)
and Fligstein (1996, 2001). Firms do not choose courses of action in isolation:
they monitor each other, and make inferences about the uncertain situation they
face by noting the success or failure of others’ strategies. When this leads to
diversity—to firms selecting different strategies and coming to occupy
different niches—a stable market structure can result. But if firms imitate,
each choosing the same strategy, disastrous crowding (White 2001: 139-44) can
occur. That is what took place in global arbitrage in the 1990s.
The
effects of imitation run deep: it can, for example, affect the statistical
distributions of price changes, causing distributions to become dangerously
‘fat-tailed’ (i.e. the probability of extreme events is far higher than implied
by standard normal or log-normal distributional assumptions). That imitation
can affect statistical distributions in this way was shown in theoretical work
by Lux and Marchesi (1999); the case of LTCM appears to show it happening in
practice. The unraveling of the imitative superportfolio caused ‘fat tailed’
price changes far beyond those anticipated on standard models.5
Imitation
led to extreme price movements and to disaster because of a third feature of
the sociology of arbitrage: the possibility of ‘arbitrage flight’, the risk
that arbitrage positions that, if held for long enough, have to be profitable
may nevertheless have to be abandoned.6 (LTCM’s arbitrage positions were eventually profitable: the
consortium that recapitalized the fund not only recouped its investment but
made a modest profit on it, and would have made a larger profit had its goal
not been to liquidate LTCM’s positions in an orderly but rapid fashion.) This
possibility was expressed to me, separately, by two partners in LTCM who used
the same analogy. Suppose they had been vouchsafed a little peek into the
future: that they knew, with absolute certainty, that at a particular point in
time the stock price of company X would be zero (these conversations took place
during the dot.com bubble). Could they, they asked me, make money with
certainty from this knowledge? Their question was rhetorical: they knew the
answer to be no. Of course, they could sell the stock short (see the glossary
in Table 3.1). If they could hold their position until the stock price
became zero, they could indeed profit handsomely. But a rise in price in the
interim could still exhaust their capital and thus force them to liquidate at a
loss.
The consequence
of this third feature of arbitrage, when conjoined with the second feature
(imitation),7 is that arbitrage’s capacity to ‘insulate’ the
economic from the social is limited. This constitutes, for example, a limit on
the performativity of economics: under some circumstances, arbitrage may be
unable to eliminate what economic theory regards as pricing discrepancies.
Ultimately, the metaphor of ‘insulation’, the Parsonian view of the economy as
a differentiated subsystem, is itself inadequate. The financial markets are not
an imperfectly insulated sphere of economic rationality, but a sphere in which
the economic and the social interweave seamlessly. In respect to arbitrage,
the key risks may be social risks from patterns of interaction within the financial
markets, rather than shocks from the real economy or from events outside the
markets. That, at least, is what seems to be suggested by the contrast between
August 17, 1998 (the Russian default, a relatively minor economic event,
triggered a disastrous unravelling of an imitative superportfolio) and
September 11, 2001 (a dramatic external shock that failed to trigger dangerous
internal social processes).8
The
interweaving of the economic and the social is not simply a matter of
analytical interest. It affects the technical practices of risk management,
because imitation of the kind evident in 1998 can undermine the protection
flowing from the basic precept of such management: diversification. The most
important way in which LTCM’s successor, JWM Partners, has altered its
predecessor’s risk model to take account of the lessons of 1998 is that all the
fund’s positions, however well diversified geographically and unrelated in
asset type, are now assumed to have correlations of 1.0 ‘to the worst event’ (Meriwether
interview). In an extreme crisis, it is assumed that diversification may fail
completely: all the fund’s positions may move in lock-step and adversely, even
those positions where the fund holds assets that should rise in relative value
in a crisis.
One way
of expressing the forms currently taken by the inextricable interweaving of
the economic and the social is via Knorr Cetina and Bruegger’s notion of global
microstructure. The financial markets are now global in their reach, but
interaction within them still takes the form of ‘patterns of relatedness and
coordination that are microsocial in character and that assemble and link
global domains’ (Knorr Cetina and Bruegger 2002: 907). In a sense, it was
globalization that undid LTCM: ‘Maybe the error of LongTerm was that of not
realizing that the world is becoming more and more global over time’, says
Myron Scholes (interview). Of course, no one was more aware than LTCM’s
principals of globalization as a general process (they had surfed globalization’s wave, so to
speak), but what caught them unawares were the consequences of the global
microstructure created by imitative arbitrage. What happened in August and
September 1998 was not simply that international markets fell in concert (that
would have had little effect on LTCM), but that very particular phenomena,
which at the level of economic fundamentals were quite unrelated, suddenly
started to move in close to lock-step: swap spreads, the precise shape of yield
curves, the behaviour of equity pairs such as Royal Dutch/Shell, and so on.
The ‘nature of the world had changed’, says Meriwether, ‘and we hadn’t
recognised it’. LTCM’s wide diversification, both internationally and across
asset classes, which he had thought kept aggregate risk at acceptably modest
levels, failed to do so, because of the effects of a global microstructure.
Since September 1998, this
particular microstructure has dissipated as arbitrage capital has withdrawn
from the markets. The failure of the shock of September 11, 2001, to ramify and
amplify through the markets is testimony to the way in which market linkages
driven by imitative arbitrage have been very much weaker subsequently. LTCM’s
successor fund, JWM Partners, was active then too, but its capital base was
smaller and its leverage levels lower, so its arbitrage positions were
considerably smaller (Silverman and Chaffin 2000). The amount of capital
devoted to convergence and relative value arbitrage by other market
participants such as investment banks was also much smaller (interviewees
estimate possibly only a tenth as large in total). There was thus no
significant superportfolio in 2001. September 11 sparked another
flight-to-quality, but there was no equivalent crisis. While LTCM had been
devastated in 1998, JWM Partners’ broadly similar, but much smaller, portfolio
emerged unscathed from September 2001: the partnership’s returns in that month
were ‘basically flat’. Of course, the linkages manifest in 1998 may well
return, albeit most likely in different forms. But that, indeed, may precisely
be the point. Globalization is not a once-and-for-all event, not a unidirectional
process, not something that can be stopped, but a composite of a myriad
microstructures, often contradictory, waxing, and waning.
Partners
in and employees of LTCM:
Haghani, Victor, Gerard Gennotte, Fabio
Bassi, and Gustavo Lao, London, February 11, 2000.
Leahy,
Richard F., Greenwich, Conn., October 31, 2000.
Meriwether,
John W., Greenwich, Conn., November 14, 2000.
Merton,
Robert C., Cambridge, Mass., November 2, 1999.
Rosenfeld,
Eric, Rye, NY, October 30, 2000.
Scholes,
Myron S., San Francisco, June 15, 2000.
This
article also draws on a wider set of interviews (numbering 60 in total)
conducted by the author with finance theorists and market practitioners, of
which those drawn on most directly here are:
Kaplanis,
Costas, London, February 11, 2000.
Shaw,
David E., New York, November 13, 2000.
Wenman,
David, London, June 22, 2001.
Not all interviewees were
prepared to be identified, and some quotations and interview material are
therefore anonymous.
1. See Barry and Slater (2002)
and the subsequent papers in the May 2002 issue of Economy and Society.
2. Strictly, the fund was the
investment vehicle (Long-Term Capital Portfolio) that LTCM managed, but to
avoid complication I shall refer to both as LTCM.
3. Figures for total returns
are calculated from the data in Perold (1999: A19); the figures for returns net
of fees are taken from Perold (1999: A2).
4. See Scholes (2000) for an
interpretation of the crisis in terms of the ‘liquidity premium’.
5. The dollar swap spread, for
example, has a daily volatility (standard deviation) of around 0.8 basis
points. Perhaps the single most dramatic event in the crisis of August and
September 1998 was the widening of the dollar swap spread in half a day (the
morning of Friday, August 21, five days after the Russian default) of 19 basis
points (Perold 1999: C2): a 35a event. Of course, nothing can safely be inferred from
a single event plucked from amongst many, but it is worth noting that the
aggregate movement in price of LTCMs positions in August 1998 (a 44% loss) was
a —14a event in terms of the 3.2% historical monthly volatility of the fund’s
portfolio and a —10.5a event on its risk model’s 4.2% monthly volatility.
Either is wildly unlikely on standard distributional assumptions.
6. This feature has been
modelled by behavioral finance specialist Andrei Shleifer (Shleifer and Vishny
1997; Shleifer 2000). Shleifer’s work is prescient: the Shleifer and Vishny model
captures well one key aspect of 1998, the arbitrage flight that occurs when
those who invest capital in arbitrageurs withdraw it prematurely in response to
adverse price movements. But in another respect even Shleifer preserves the
Parsonian boundary around the ‘economic’. The Shleifer-Vishny model’s
arbitrageurs are not influenced by each other, and each has perfect individual
knowledge of the true value of the asset they trade. As we have seen, however,
a key dynamic leading to the crisis of 1998 was imitation amongst arbitrageurs.
The resultant correlation of prices that were otherwise essentially unrelated
economically— the second key aspect of 1998—is not captured by the
Shleifer-Vishny model’s single asset market and non-imitative arbitrageurs.
7. Were it not for the risk of
imitation-induced correlation, the dangers posed by arbitrage flight could be
reduced greatly by holding a large portfolio of diverse arbitrage positions.
8. 2002 saw sharp falls in
global stock markets, but these were not the direct effect of September 11.
After recovering quickly from the initial shock of September 11,
stock markets continued to
rise for several months before succumbing to the effects of events such as the
Enron and WorldCom scandals.
沒有留言:
張貼留言