Вы находитесь на странице: 1из 59

ESLSCA

Market Liquidity: Measures, Models


and Order Execution
by
Olivier Milla
A paper submitted in partial fulllment to
obtain the Master 2 - Market Finance Diploma
in the
P ole Trading
December 2010
ESLSCA
Abstract
Pole Trading
by Olivier Milla
This paper attempts to present several topics surrounding the concept of liquidity on
the nancial markets. Liquidity is dened as the ability in time or price to trade assets.
Macro-economical liquidity and market liquidity are tied through money supply and
ination. As markets evolved over the past centuries, banks took a preeminent place in
the creation of money, and more recently through leverage in investment banking. As
such, the 2007 subprime crisis can be seen as a liquidity shortage that hit the nancial
markets, starting with credit and real estate.
At a smaller scale, liquidity on a single market is closely tied to the balance sheet of
the actors and the timing of the orders. This conclusion can also be extended to the
micro-structure of the nancial markets. The May 6, 2010 ash crash exhibits the
importance of timing and behavior to create liquidity on a nancial market.
Market participants usually read liquidity through rst or second order measures such
as volumes, open interests, bid/ask spreads, etc. While this lecture is the same for
everyone, each participant does not benet from the same access to liquidity due to
legal, operational and technological costs.
Models exist to integrate these costs along with the costs of slippage into mark to
market models. The Almgren-Chriss model presented in this paper is one of them. We
also expose how it can be used to optimize the execution of orders to reduce slippage.
Acknowledgements
I would like to thank my project advisor, Naji Freiha.
Special thanks go to Adrien Mouillon and David Arnaud for their conversation that lead
to parts of this report.
ii
Contents
Abstract i
Acknowledgements ii
List of Figures v
Symbols vi
1 Markets And Flows 3
1.1 What Is A Market, Anyway? . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 A little Bit Of History . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Banks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Credit Intermediation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 The Monetary Policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 Controlling Ination . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.2 Controlling The Money Supply . . . . . . . . . . . . . . . . . . . . 10
1.4 Evaluating Market Liquidity . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4.1 Denition Of The Measure . . . . . . . . . . . . . . . . . . . . . . 12
1.4.2 Subjectivity Of The Measure . . . . . . . . . . . . . . . . . . . . . 12
1.4.3 Non-Market Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5 Market Liquidity Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5.1 Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5.2 Market Width and Market Depth . . . . . . . . . . . . . . . . . . . 16
1.5.3 Market Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.5.4 Late Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Three Liquidity Crisis 20
2.1 The 2008 Fall of the Shadow Banking System . . . . . . . . . . . . . . . . 20
2.1.1 Chronicle of the Crisis . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 The November 2000 Turkish Overnight Liquidity Crisis . . . . . . . . . . 25
2.2.1 Chronicle of the Crisis . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 The May 6, 2010 Flash-Crash . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.1 Chronicle of the Crisis . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.2 Cross-market propagation . . . . . . . . . . . . . . . . . . . . . . . 31
iii
Contents iv
2.3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3 Modeling Market Liquidity 35
3.1 The Almgren-Chriss Liquidity Asset Price Model . . . . . . . . . . . . . . 35
3.1.1 Price Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.2 The Denition Of A Trading Strategy . . . . . . . . . . . . . . . . 37
3.1.3 Cost Of Trading . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4 Order Execution 40
4.1 Impact Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1.1 Linear Impact Functions . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1.2 Exponential Impact Functions . . . . . . . . . . . . . . . . . . . . 43
4.1.3 Empirical Impact Functions . . . . . . . . . . . . . . . . . . . . . . 46
4.2 The Ecient Frontier Of Optimal Execution . . . . . . . . . . . . . . . . 47
5 Conclusion 50
Bibliography 51
List of Figures
1.1 Credit Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Maturity Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Liquidity Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Year on Year CPI vs Transaction Volume of the SPX Index . . . . . . . . 9
1.5 U.K. money supply and a combined capital market price index, 1950-1972. 10
1.6 Real Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.7 OTC Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1 Shadow Bank Liabilities vs. Traditional Bank Liabilities . . . . . . . . . . 21
2.2 The 2000 Turkish Liquidity Crisis . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 May 6,2010 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 E-mini S&P 500 Futures Volume and Price . . . . . . . . . . . . . . . . . 30
2.5 E-mini S&P 500 Futures Volume Market Depth . . . . . . . . . . . . . . . 30
2.6 Market Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.7 E-mini SPY and Aggregated S&P 500 Stocks Buy-Side Market Depth . . 33
3.1 Order Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Random Walk with Order Execution . . . . . . . . . . . . . . . . . . . . . 37
4.1 Minimum Variance for Linear Impact Functions . . . . . . . . . . . . . . . 43
4.2 Minimum Expected Value for Exponential Impact Functions . . . . . . . . 45
4.3 Minimum Variance for Exponential Impact Functions . . . . . . . . . . . 45
4.4 Minimum Expected Value for Linear Impact Functions . . . . . . . . . . . 48
4.5 Minimum Variance for Linear Impact Functions . . . . . . . . . . . . . . . 49
4.6 Minimum Utility Function for Linear Impact Functions . . . . . . . . . . 49
v
Symbols
X Initial number of securities
T Time taken to sell the X securities
t
k
Discrete time k (0 . . . N)
S
k
Securitys screen price at time k

S
k
Securitys paid price at time k
x
k
Number of securities held at time k
n
k
Number of securities sold between time k and k 1
g() Permanent impact function
h() Temporary impact function
U Utility function
C Cost of trading
Volatility of returns
Drift
Time between two orders
Average rate of trading
n
k

Fixed cost of execution


Permanent impact function parameter
Temporary impact function parameter

equals
1
2

Utility function parameter


vi
Introduction
The deance witnessed between operators of the money and repo markets during the
Lehman Brothers crisis froze entire spans of the nancial markets. When bid/ask spread
widened or quotes simply vanished from the screens (as it was the case for the equity repo
market for instance), assets lost their value by lack of consideration. Mark-to-market
becoming impossible, assets were marked as null in the banks balance sheets until
eventually the opportunity to unwind these positions shows up, which is still not the case
for several asset classes (think credit derivatives). Even now, two years later, numerous
banks see their quarterly results improve as some of their assets under management
regain value with the slow revival of their respective markets.
Along with the mechanical devaluation of assets caused by the rise of risk and uncer-
tainties, the crisis shed light on a component of the nancial markets: liquidity.
The trouble is, what is liquidity? How can it be measured? How much does it contribute
to an asset valuation? How does it impact the mark-to-market of assets and the execution
of orders? Can liquidity be hedged?
This paper aims at reviewing some of the prevailing concepts to date that can help
clarifying the perception of liquidity both from practical and theoretical standpoints.
The paper will develop as follows:
Chapter 1 will review market liquidity at three dierent scales: the global macro-
economy, a local market and a markets micro-structure. Three crisis of the corre-
sponding magnitude will picture the mechanics leading to a shortage of liquidity.
At the macro-economical scale, we will start by exposing the denition of a nancial
market and the historical conditions foregoing its inception, underlying its role as
a macroeconomic engine for cash allocation. To remain close to our main topic, we
1
Chapter 1. Introduction 2
will simply aim at revealing the vital strings and stakeholders that link a markets
life to the economy and which should be closely followed to assess its liquidity.
The 2008 Fall of the Shadow Banking System will serve as an illustration.
We will then expose liquidity at the scale a of a single, national market by studying
the Turkish overnight market and its 2000 crisis, showing how information and
behavior shape the events on a nite market.
Finally, at the scale of a markets micro-structure, we will reveal how participants
add or remove liquidity, underlying the importance of technology and algorithms
in that process. The May 6, 2010 Flash-Crash will serve as an illustration.
Chapter 2 will present how market liquidity is usually read by traders and risk man-
agers. We will list both generic and market-dependent measures.
Chapter 3 will expose a simple model developed by Robert Almgren and Neil Chriss
in 2000 to add execution pricing and xed execution costs to mark-to-market.
Chapter 4 will briey use the model of Chapter 3 to show it is possible to better the
execution of large orders on a market.
Chapter 5 will conclude the study.
Chapter 1
Markets And Flows
1.1 What Is A Market, Anyway?
A market is a virtual place. Two people meeting and exchanging an object or an
information spawn a one-time market that lives for the time of the transaction. Once
they are done, the market vanishes until two people meet and eventually trade the same
underlying again.
Over the last centuries, as people started to steadily trade a variety of nancial contracts
and as exchange methods became more standardized, a certain idea of continuity and
tradition got attached to the activity of trading, spawning the idea of the nancial
markets as a constant, unmovable activity.
For instance, the history of the Amsterdam Stock Exchange is telling.
1.1.1 A little Bit Of History
The stock market was born in Holland in 1602 when the Dutch East India Company
issued the rst shares to nance its trading with Asia (dividends were delivered as
spices). As investors (both institutional and retail, already at that time) wanted to keep
informed about the company with other investors, they took the habit of meeting daily
at a certain spot in Amsterdam which soon became the Amsterdam Bourse, with an
ocial building.
3
Chapter 2. Markets And Flows 4
The concept grew all over Europe and national exchanges were rapidly created in all
the major capital cities. The number of quoted securities grew and many corporations
started to fund their activities through the capital markets. As investments throttled
and crisis occurred, contracts hedging against certain risks appeared, along with the
derivatives contracts (Futures, Options, etc.) that were introduced as contracts linking
their price to the evolution of one or more other contracts or market parameters. In 1978,
the European Option Exchange (EOE) was founded in Amsterdam, bringing complete
legitimacy and transparency to these new products.
Another milestone worth reporting is the addition of a stock market index to the place
in 1983 (the AEX)
1
. This abstraction layer allowed for a direct appreciation of the
market-as-a-whole and to more developments regarding the study of the stock market
and the economy (and also to the development of technical analysis, behavioral analysis,
etc.)
In 2000 the Amsterdam Stock Exchange merged with its equivalents in Brussels and
Paris to form the Euronext exchange
2
.
Beyond the technical and nancial creativity that sustained this growth, nancial mar-
kets slowly entered the daily life of the crowd. In America, daily stock prices were rst
published in the newspapers in 1795 by the New York Price Current which aimed at
listing daily goods prices in the city. In 1889, the Wall Street Journal began printing
as the rst newspaper dedicated to the stock market and related aairs
3
.
In 1915, pension (to cover old age, death, disability, etc.) became part of the social
claims. Private - and later public - funds spawned, as well as insurances. Numerous of
these institutions used the nancial markets to nance part or all of their returns.
All companies irreversibly joined the action with the institution of mark-to-market as
an accounting standard. In the late 1980s this accounting principle entered the U.S.
Generally Accepted Accounting Principles (G.A.A.P.). From this moment on, both
companies, banks, institutional and retail investors balance sheets were tied to market
values, with nancial risks in their balance sheets.
1
The Dow Jones Industrial Average - the rst stock index - was created by Charles Dow in 1884.
2
For further details about the inception of exchanges and nancial markets, [See 1].
3
Source: History Of Business Journalism, http://www.bizjournalismhistory.org/.
Chapter 2. Markets And Flows 5
However how important nancial markets became, they serve in essence the same pur-
pose as they were in their rst days: put money to work. In practical terms, that means
matching borrowers and lenders through various contracts.
Trading was also institutionalized by another important wheel that needs to be presented
to understand the ows we see on the nancial markets : banks and the most important
of them, the central banks.
1.1.2 Banks
Banks were introduced very early in the system. They rst acted as custodians (safe-
keepers of the securities), then as consultants (giving advices to clients) and represen-
tatives (taking actions in name of other parties) and nally as intermediaries (allowing
any nancial institution to reach any fund through the network of banks). This late step
completed the integration of nance in economies as it linked, compared and matched
all the reachable sources of cash and risk.
As noted by Zoltan Pozsar [2]: Relative to direct lending (that is lenders lending directly
to borrowers), credit intermediation provides savers information and risk economies
of scale by reducing the costs involved in screening and monitoring borrowers and by
facilitating investments in a more diverse loan portfolio and therefore banks rapidly
became the core users of the nancial markets.
This recycling of money by banks through the nancial markets is called Credit Inter-
mediation.
1.2 Credit Intermediation
For a bank, credit intermediation has three main dimensions that represent the dierent
natures of investments and associated risks it can carry:
Credit Transformation That is when a bank enhances the credit quality of the debt
it issues on the nancial markets. This is done via the use of equities as collaterals
and the denition of priority of claims
4
. Credit Intermediation creates a dierence
4
Priority of claims: The order following which claims against the bank will be treated if it becomes
insolvent.
Chapter 2. Markets And Flows 6
in terms of credit quality between the cash ows in and out of the bank, the
remaining being the exposition which could result in a prot or a loss.
A good example of credit transformation is the issue of any note by a depository
bank. The credit quality of the note on the market depends on its seniority and
is backed by deposits that can be of far less credit quality. Another example are
Collateralised Debt Obligation (CDO)
5
which are products available on the credit
market as having very good quality grades while they are backup by a pool of
much riskier assets.
Credit transformation throttled in the recent years and was at the root of the 2007
subprime crisis. Section 2.1 will exhibit some of the mechanisms that can lead
to a liquidity crisis on the credit market.
Figure 1.1: Credit Transformation
Maturity Transformation That is when a bank uses short-term deposits to fund long-
term loans. That operation creates a dierence in duration between the cash ows
in and out of the bank, leaving it at risk. A detailed example of a liquidity crisis
involving maturity transformation constitutes the section 2.2 (The 2000 Turkish
overnight market crisis).
Figure 1.2: Maturity Transformation
5
CDOs are a type of structured asset-backed security whose value and payments are derived from a
portfolio of xed-income underlying assets.[...] The rst CDO was issued in 1987.(Source: Wikipedia)
Chapter 2. Markets And Flows 7
Liquidity Transformation That is when a bank uses liquid deposits (say cash) to
fund illiquid assets (say any security, compared to cash). That operation creates a
dierence in liquidity between the banks cash ows in and out. The bank, again,
carries the dierence in the hope to gain from it. Section 2.3 will exhibit a subtle
liquidity crisis at the micro-structural level of a market (between S&P Futures and
the underlying Stocks/ETFs).
Figure 1.3: Liquidity Transformation
These three dimensions are the main risks that one faces when taking positions on a
nancial market. Respectively Credit Risk, Duration Risk (also known as Curve Risk)
and Liquidity Risk. It is obvious that any loan mingles these three risks. For instance
the funding of a the construction of a private company headquarters (a 20 years loan on
a BBB-rated corporation for the construction of an illiquid commercial asset) by issuing
1-year rolling B-rated commercial papers.
One bank supervises and regulates the relationships between lenders and borrowers: the
central bank
6
. The central banks mandate is - since its inception in 1913 - to set the
most important economic parameters to allow for a suitable economic expansion. As
such it provides stability by acting as a lender of last resort
7
and as a controller of the
ination levels. It also stimulates the economy when needed by regulating the money
owing in the system. All these operations are referred to as the monetary policy
8
.
6
When talking about the central bank, we are talking about the Federal Reserve System - The Fed
- everywhere in this paper, except where expressively notied.
7
A lender of last resort is a nancial institution ready to lend money when nobody else will.
8
For more details, see [3].
Chapter 2. Markets And Flows 8
1.3 The Monetary Policy
The Feds monetary policy is to amount the cost of money and credit in the U.S. economy
to achieve growth and price stability.
To achieve growth, money and credit must grow at a pace that allows economic activity
to expand at a sustainable rate without excessive price increases. If money and credit
grow too slowly, the cost of money (the interest rates) is too high and people and business
will not be able to aord it, slowing the pace of consumption. Conversely, if money and
credit increase too rapidly, ination will rise (interest rates will drop in real terms).
This is most crucial as this sets the return expectations for the investors. In other
words, it sets how repealing or attractive market returns are compared to o-markets
placements (direct consumption, real estate, etc.) and therefore dictates the liquidity
(in the sense of usage) of the nancial markets.
1.3.1 Controlling Ination
Ination is a persistent increase in the level of consumer prices or a persistent decline
in the purchasing power of money, caused by an increase in available currency and credit
beyond the proportion of available goods and services
9
.
In absolute terms, ination hurts entrepreneurship and corporate management by mak-
ing price evolution and cost evaluation dicult (and therefore adds diculty to pricing
future cash ows). It is also a very unfair process. For instance retirees, sheltered by
state money and workers, who negotiate their wage do not have the same elasticity
regarding ination.
In relative terms, even though a high ination well anticipated by the market gives
protection to both borrowers and lenders, it puts at a disadvantage all forms of money
not carrying interests (such as paper money). As such, the nancial markets provide
in essence a protection against anticipated ination by setting the terms of contracts
between borrowers and lenders.
9
The American Heritage Dictionary of the English Language, Fourth Edition, Copyright c 2000
Houghton Miin Company.
Chapter 2. Markets And Flows 9
In the United States, the ination rate for ordinary consumers is most commonly mea-
sured by the percentage rise in the Consumer Price Index (CPI)
10
reported by the
Bureau of Labor Statistics (BLS). For an evaluation of the ination of the economy as
a whole, the GDP deator
11
is more suited as it includes the prices of non-consumers
goods and excludes the prices of foreign-produced goods.
Ination is directed by the money supply through the equation of exchange:
MV = Py
where M is the money supply (see below), V is the number of times per year the average
dollar turns over in transactions for goods and services, P is the general price level and
y denotes the economys real income (as measured, e.g., by real GDP)
12
.
Figure 1.4: Year on Year CPI vs Transaction Volume of the SPX Index
10
The CPI is a statistical measure of a weighted average of prices of a specied set of goods and
services purchased by wage earners in urban areas. It is a price index which tracks the prices of a
specied set of consumer goods and services, providing a measure of ination. (Source: wordIQ.com)
11
The GDP deator is a price index measuring changes in prices of all new, domestically pro-
duced, nal goods and services in an economy.[...] A simple GDP deator formula goes like this
GDP deator in % =
Current Year GDP
Base Year GDP
100. (Source: wordIQ.com)
12
For more details about Ination, [See: 4]
Chapter 2. Markets And Flows 10
1.3.2 Controlling The Money Supply
The money supply is the amount of money in the economy, measured according to
varying methods or principles.
13
Measures
The main measures are called M1, M2 and M3. M1 is a narrow measure of moneys
function as a medium of exchange. It sums currency and checking account deposits.
M2, a broader measure also reects moneys function as a store of value as it adds some
types of saving deposits to M1. M3 is a still broader measure that covers items widely
regarded as a close substitute for money in addition to M2.
Figure 1.5: U.K. money supply and a combined capital market price index, 1950-1972.
[Source: 5]
Over the years, changes in the relationship between the money supply and the economy
have complicated the Feds decisions and the correlation between the nancial markets
and these measures.
For many years M1 was an accurate measure of the Gross Domestic Product (GDP)
until in the 1980s when banks started paying interest on checking accounts, people put
a lot of money into checking accounts. The rapid growth in M1 broke the relationship
between M1 and the GDP.
Similarly, the relationship between M2 and the economy broke down in the 1990s, when
the interest rates were very low and people pulled money out of saving accounts to
13
The American Heritage Dictionary of the English Language, Fourth Edition, Copyright c 2000
Houghton Miin Company.
Chapter 2. Markets And Flows 11
put it into nancial investments outside of banks, such as mutual funds (which are not
included in the M2 money supply measure).
As a result, the Fed tended to look more at direct economical data (as opposed to
monetary ones) such as the employment rate, prices indexes or commodities prices
14
.
Even though the relationship between nancial markets and money supply distended, it
didnt disappear and markets and money supply remain tied together. A loose monetary
policy may increase liquidity and encourage more investment by making margin loan
requirements less costly, and by enhancing the ability of dealers to nance their positions.
For instance, monetary expansion increases equity market liquidity and unexpected in-
creases (decreases) in the Federal Funds
15
rate lead to decreases (increases) in liquidity
and increases (decreases) in stock and bond volatility. This establishes a link between
macro liquidity, or money ows, and micro or transactions liquidity[7].
The Feds Tools
To control the creation of money, the Fed has three main tools at its disposal:
Reserve Requirement It is the amount of cash that a deposit bank needs to keep,
the rest being allowed to be lent to other parties. The Monetary Control Act of
1980 sets the Reserve Requirement to stay between 8% and 14%. The last reserve
requirement move happened in April 1992, when the Fed lowered it to 10% (from
12%).
Discount Window This is the interest rate at which the Federal Reserve Banks make
very short-term loans to banks. It is a national rate, xed since Frebruary, 18,
2010 to 0.75%. Since the beginning of the 1980s, the Fed uses the Federal Funds
rate as the key gure for its lending target.
Open Market Operations are the Fed buying or selling securities in the open market.
The Fed could do so to oset a seasonal need of cash from banks (Christmas
shopping season, etc.), an exceptional need of cash (the bank run of late 1999
14
For more details about Money Supply, [See: 6].
15
Federal Funds are overnight borrowings by banks to maintain their bank reserves at the Federal
Reserve. (Source: Wikipedia)
Chapter 2. Markets And Flows 12
when people worried about the possible 2000s computer bug), or to add/remove
cash from the system by growing/reducing the banks balance sheets.
Finally, the Fed looks at the value of the dollar in terms of foreign currencies as inter-
national trade and nancial activity picked up in the recent years.
In the end, both ination and money supply are reected through the nancial markets.
This supply and demand of money in turn leads transaction liquidity in the sense of
the availability of counterparties on markets. This market liquidity is visible through
dierent indicators.
1.4 Evaluating Market Liquidity
1.4.1 Denition Of The Measure
Generally, liquidity is broadly dened from two point of views
16
, price and time:
Price The degree to which an asset or security can be bought or sold in the market
without aecting the assets price.
Time The speed at which an asset (resp cash) can be converted (invested) into cash
(an asset).
It is important to see that this very dichotomy shows that the denition of liquidity
hasnt yet converged to a sole, broadly accepted and clearly stated denition - such as
the denition of volatility for instance. In other words, liquidity remains an hidden,
implicit parameter of the market with no universal appreciation tool available.
1.4.2 Subjectivity Of The Measure
In addition, market liquidity mingles market risks and non market risks, both not being
equally accessible by the operators, giving them dierent access to liquidity. For example,
you want to enter a security at 100, screen price. By actually taking your position you
will go through costs that will impact the book value of the operation (broker, taxes,
16
see http://www.investopedia.com/terms/liquidity.asp
Chapter 2. Markets And Flows 13
etc.) and which will not be the same for another given operator. Any operator sees
in fact a modied bid/ask that includes these eects. The following diagram shows in
black a screen bid/ask spread for a security and in red the eective bid/ask if an
order were to be executed.
Figure 1.6: Screen price (mark to market price) vs eective price.
While some of these costs are known and stable (broker fees, taxes, etc.), some others
remain hidden and non-predictable.
1.4.3 Non-Market Costs
Transaction costs The transaction costs are an important factor to take into account.
There are commonly three:
Exchange/Clearing Fees which are taken by the exchange for their services.
Broker/Execution Fees which are taken by the broker. They can include fees
for the usage of execution algorithm or other services.
Fiscal Fees which can be taken on a per trade basis (such as the Federal Stamp
in Switzerland).
Tax costs are a more intricate topic. They are usually thought of at the scale of
portfolios and evaluating each trades impact is not instrumental.
Operational costs are mostly seen on an annual basis and in term of throughput.
These costs can be broken down to each trade. They comprise all the investment
done to compute position and account statements (Net Asset Value, etc.).
Legal costs comprise all the costs that are needed to bring a legal stature to the com-
pany and all the legal costs induced by market activities.
Chapter 2. Markets And Flows 14
More important, these hidden costs represent about 5/6 of the overall cost
17
and give
to the dierent operators very dierent access to liquidity depending on their scale,
country, broker, technology, etc.
Considering that the overall cost is assumed to represent 150 billions dollars per year
just for the American market, which represents, according to BNPPAM
18
from 1% to
15% of an investments value (with an average of about 2%), the question of assessing
these costs becomes key for any investment institution.
The other components of liquidity are shared by all operators and depend on market
conditions. Any trader or risk manager taking a position assess liquidity following a set
of indicators.
1.5 Market Liquidity Indicators
Market liquidity can be observed through several market variables which are independent
from the users point of view. While none of them mirror entirely the state of a markets
liquidity, they together give a good picture of it at the time of the trade. Practitioners
tend to focus on a chosen set of indicators, depending on their underlying and their needs.
For instance, a commodity trader will closely follow the Open Interest (OI) evolution
while a bond trader will concentrate on the bid/ask spread. The same dierence prevails
between a long-term and a short-term trader. Here is a review of the main indicators
currently looked at:
1.5.1 Flows
Volume represents the number of trades of a given security during a period of time. It
measures the activity going on in the security, as agents allocate or reallocate their
portfolios. The more volume, the more it is possible to get in/out of a position
fast and without aecting the price of the security.
Volume follows strong seasonality (see Book Depth) that are not without aecting
prices and how the mar-to-market of a position should be read.
17
Result published by Plexus, a company specialized in the evaluation of transaction costs and quoted
by Naji Freiha and Berge [8].
18
quoted by Naji Freiha and Berge [8].
Chapter 2. Markets And Flows 15
Technical analysis has derived two other indicators to link price trends to volume
trends:
On Balance Volume (OBV) is the cumulative total volume updated say every-
day (volumes are taken at each days market close or at midnight):
OBV
today
= OBV
yesterday
+

+volume if close
today
> close
yesterday
0 if close
today
= close
yesterday
volume if close
today
< close
yesterday
This indicator underlines whether a trend in price is followed by a trend in
volume, or whether price is changing for other reasons than the availability
of the asset on the market.
Money Flow is typical price multiplied by volume.
typical price
today
=
high
today
+low
today
+close
today
3
money ow
today
= typical price
today
volume
today
This indicator attempts to evaluate roughly the cash value of a days trans-
actions.
Fund Flows They are surveys conducted by agencies such as TrimTabs
19
that track
the cash allocation into mutual and hedge funds. As these funds allocate these
ows in securities, these surveys reveal large volume and allocation patterns that
can also be linked to market prices. See for instance Using Equity ETF Flows as
a Contrary Leading Indicator[9].
Order Imbalance is dened as the notional value of buys less the notional value of
sells over a time period, divided by the total value of buys and sells over that same
period of time:
Value of buys Value of sells
Value of buys + Value of sells
. It is still an aggregate account of the transactions but it adds informations
concerning the trade initiations. A trade can be initiated as a buy or a sell and if
the amount of buys (resp. sells) outnumbers the amount of sells (resp. buys), the
available liquidity to buy or sell a security diers.
19
http://www.trimtabs.com.
Chapter 2. Markets And Flows 16
1.5.2 Market Width and Market Depth
Bid/Ask spread On OTC markets, the bid/ask spread is the dierence between the
prices at which market makers are ready to buy and sell a certain quantity (known
in advance) of a security. On listed markets, the same dierence can be seen in the
order book, between the rst oered price and the rst bid price. It is important
to note that , rst of all the prices between the bid price and the ask price are
totally illiquid. Second, that even though market makers operate in a best eort
fashion, liquidity outside this spread is unknown. Nevertheless, by continuously
publishing a spread and an order size, market makers oer a certain liquidity.
A market can also be deemed illiquid when either the bid (ask) price moves away
from a (virtual) market middle price. This eect exists for all markets and a shock
on the bid/ask spread in one market is followed by the same behavior in other
markets, as characterized for instance by [7]. This study also underlines a time
correlation between the US bond and US stock markets bid/ask spread of +28%
along with a magnitude correlation between a move of the bid/ask spread of the
10Y bonds and stock for the same issuer (when the bid/ask spread widens for the
bond, the stocks bid/ask spreads by a relative 1/4 of that magnitude.)
20
.
Bid/Ask spread follows dierent seasonality. (see Book Depth)
Figure 1.7: Bid/ask spread of some currency pairs as published by the retail broker
fxpro.com on Oct.13, 2009
Book Depth is dened as the dierence between the price of the highest and the price
of the lowest orders placed in an order book. This price is often smoothed as a
20
This also underlines the credit risk of the issuer.
Chapter 2. Markets And Flows 17
bell, centered around the current market price. It is used to evaluate the number
of people ready to trade at prices distant from the current market price. The more
people, the more orders one can hit, even if it reduces the return. As pointed out by
[7], book depth is also strongly linked to the tick size. As with the Bid/Ask spread,
book depth suers from seasonality of dierent frequencies. These seasonality are
very market-dependent.
Daily frequency following the open trading hour and the overlapping of the dif-
ferent timezones.
Quarterly due to the numerous accounting variables linked to this frequency
(renancing rate, company results publishing, dividend paying), the natural
seasons which impact most industries and the importance of Future contracts
which also follow this frequency (Euribor 3M contracts, Bund contracts, etc.).
Yearly also due to the numerous deadlines associated with that frequency (com-
panies yearly balance sheet and turnover, asset managers yearly target, etc.).
In addition, depths are lower around holidays, lower on Friday compared to the
other days of the week, higher in August and September relative to January and
slightly lower during daily lunch hours. Some markets also have specic seasonality
[see 7, Page 12]. US bond depth is relatively high in February, May and July
whereas the US stock depth is relatively low on Monday and high in March. Also
the stock depth decreases during bond market crisis and both markets depth are
correlated by +20% [7].
1.5.3 Market Size
Open Interest (OI) is used by Futures and options traders. It represents the number
of contracts alive at a certain point in time. As opposed to spot contracts, Futures
and options contracts have no xed number. Thus, open interest represents the
size of the secondary market for these derivatives and it also impacts the bid/ask
spread. Associated with volume, open interest can reveal important details about
a contract. For instance, a relatively high number of derivatives contracts in
comparison to the daily volume underlines the possibility of a squeeze at maturity.
Level II quotes The NASDAQ is a computerized system created in 1971 that facili-
tates trading and provides price quotations on more than 5,000 of the more actively
Chapter 2. Markets And Flows 18
traded over the counter stocks. It permits to see who is trading and how. The
three dierent types of players are:
Market Makers They provide liquidity as they are entailed to sell (resp buy)
when everybody is buying (resp selling). The most important market maker
is known as the Ax. The Ax of a stock is the market maker who leads the
price action of a given stock.
Electronic Communication Networks (ECN) They are computerized order
placement systems. They act as proxies for institutional or retail investors.
Wholesalers They also act as proxies but only for retail traders. They are also
known as Order Flow rms.
Commitment of traders The Commitment of traders (COT) is a report issued by
the Commodity Futures Trading Commission (CFTC) enumerating the holdings
of participants in various Futures and options markets in USA along with their
natures: Hedger, Speculator, Other. It focus on commodities (agriculture, energy,
whether,...) and is updated every Friday at 3:30 Eastern Time.
21
Some indicators are only available after an order execution:
1.5.4 Late Indicators
Slippage is the indicator that matches the closest the liquidity denition given above
but with the disadvantage of only being available after a trades execution. Slip-
page can either refer to price slippage (commonly called slippage) and to time
slippage (commonly called Execution time). Here is an example of slippage
(taken from Taleb [10]):
A fund manager needs to go long the JPY currency against the dollar. The 115
calls on the Chicago Mercantile Exchange are quoted 88/92 which means that the
middle market price in theory is at 90. He would then assume that for buying 4000
contracts, he would have to pay 92 for the rst 1000, 93 for the next 1000 and up
to 98 for the balance of 2000 as he would drive the currency itself higher thanks to
option traders hedging their deltas. His weighted average will then be 95.25 and
he will count an overall slippage of 5.25 ticks for his execution. He should make
21
Homepage of the COT: http://www.cftc.gov/marketreports/commitmentsoftraders/index.htm
Chapter 2. Markets And Flows 19
the same allowance for his unwinding the trade, provided he picks the same time of
the day. Slippage would be more important when the market gets more volatile or
when overlapping of time zones reduces the total number of market participants.
Slippage is often wrapped into order execution conventions such as VWAP
22
or
OHLC
23
.
Historical Volatility is a mixed indicator of a markets liquidity. There exists count-
less studies linking liquidity and volatility using both empirical and mathematical
terms. All of them suggest that both parameters are driven by the same factors.
None of the studies can separate them or at least, dierentiate them entirely. (See
for instance the paper by Leland which integrates liquidity as a modied volatility
parameter). In all cases, volatility is a key indicator to liquidity even though its
interpretation is intricate.
22
Volume Weighted Average Price
23
Open High Low Close
Chapter 2
Three Liquidity Crisis
This chapter serves two purposes. First, to expose some of the mechanisms of credit,
maturity and liquidity transformation presented in section 1.2. Second, to illustrate what
are liquidity crisis at the global (section 2.1), local (section 2.2) and micro-structural
scales (section 2.3). The latter serving as an introduction to the last two chapters where
micro-structural liquidity and order execution will be studied.
The 2008 crisis will expose how credit intermediation is performed by some institutions
and how their activities collapsed in a gigantic liquidity crisis.
2.1 The 2008 Fall of the Shadow Banking System
Hopefully, macro-economical liquidity collapse are not so frequent (generally one or two
per century). The most famous one is the bank run of the nineteen-thirties. While the
creation of the Fed as lender of last resort in 1913 certainly reduced the occurrence of
bank runs, one happened very recently, more subtle in its causes but as dramatic in its
eects: The fall of the shadow banking system, which began in the summer of 2007 and
peaked following the failure of Lehman Brothers.
Actors of the Shadow Banking System are nancial intermediate as presented in 1.2
with the only dierence being that they do not have access to central bank liquidity or
public sector credit guarantees.
20
Chapter 3. Three Liquidity Crisis 21
Instead, they were supposedly backed by the private sector. As dened by McCulley who
forged the term shadow banking in 2007: unregulated shadow banks fund themselves
with uninsured commercial paper, which may or may not be backstopped by liquidity
lines from real banks. Thus, the shadow banking system is particularly vulnerable to
runs - commercial paper investors refusing to re-up when their paper matures, leaving
the shadow banks with a liquidity crisis - a need to tap their back-up lines of credit with
real banks and/or to liquidate assets at re sale prices.
Indeed, they notably used supposedly AAA assets as collaterals for their operations. The
problem was that these assets were the product of a range of securitization and secured
lending techniques, that imply risks that were underevaluated such as the agency risk
or the correlation risk.
Figure 2.1: Shadow Bank Liabilities vs. Traditional Bank Liabilities in USD trillion
[Source: 2]
2.1.1 Chronicle of the Crisis
There exists many timelines describing the events of the credit crisis of 2007. The Fed
of New York has published one underlying their actionsRyan [11] and the international
Chapter 3. Three Liquidity Crisis 22
components of the crisis. Academics also have published detailed chronicles, such as [12]
or spontaneous consortium such as [13].
From 2001 to 2004, the number of mortage-backed securities
1
jumped as lending condi-
tions loosen and home prices increased following the economical growth and the zero-rate
policy of the dot-com bubble burst of 2001. Banks stued themselves with subprimes,
notably Fennie Mae and Freddie Mac. In 2006, home prices began a rapid decline be-
cause mortgage loans terms changed as interests rose and people couldnt pay up. The
excess supply of homes put pressure on home prices.
Between February 2007 and April 2007, 25 major subprime lending rms declared
bankruptcy
2
. All major banks published losses in the subprime markets including HSBC,
Bearn Sterns, Merril Lynch, J.P.Morgan Chase, etc. BNP Paribas locked two of its funds
as it couldnt value the assets in them, owing to a complete evaporation of liquidity.
During the summer of 2007, most banks amounted their exposure to the subprimes
market. In September, Northern Rock faced a big problem as liquidity was cut o and
it couldnt nance its positions. It appeared that even as a deposit bank, Northern Rock
relied mostly on the markets to nance its mortgage lending activity. A traditional bank
run followed until the Bank of England provided the bank with an emergency funding.
The rst gures of the banks exposure were revealed. Most major banks (UBS, Citi,
Merril Lynch, etc.) started to announce their exposure to the subprimes. Citigroup
losses already summed up to $40 billions within the last six months. The Fed attempted
a global, coordinated plan with the ve major central banks to oer billions of dollars in
loans to banks. The ECB auctioned $500bn to help commercial banks over the Christmas
period.
In January 2008, the nancial markets suered the largest fall since September 11, 2001
on fears that the recession might become global. The scope of the nancial crisis just
started to be grasped. The G7 amounted the possible losses due to the subprimes to
$400 bn. Series of interest cuts were started by central banks. U.S. home prices felt
most in 25 years. Foreclosures throttled. In March 2008, Bear Sterns was purchased by
J.P.Morgan Chase at $2 a share. A year earlier share prices reached $170 a share.
1
Mortgage-backed securities were created in 1983 by Salomon Brothers and First Boston. They are
tradable securities which payments come from xed-income underlying assets.
2
Ben Bernanke announced at that time that growing number of mortgage defaults will not seriously
harm the US economy.
Chapter 3. Three Liquidity Crisis 23
In September 2008, Fannie Mae and Freddie Mac were taken over by the U.S. govern-
ment. They revealed owning over $5 trillions of mortgage-backed securities. Lehman
Brothers had incurred billions of dollars in losses due to the mortgage crisis and couldnt
nd a buyer. All markets got aected. All Lehman debts, credit lines and market posi-
tions incurred losses to most of the nancial institutions worldwide. Even money markets
jumped, revealing that even short term debts could tumble. Bank failures continued.
In October 2008, Eastern Europe was hit by a currency crisis. People were doing carry
trade by borrowing in Eastern Europe currencies and investing abroad. When returns
fell, people closed their lines, selling Eastern-European currencies and facing huge debts.
This fueled the crisis in Europe, particularly in Germany.
The TARP (Troubled Assets Relief Package) was released in the U.S. The government
injected $700 billions into banks balance sheets against preferred shared. Consolidation
of the nancial sector continued. The IMF started to provide help to Eastern-European
countries. China announced a $586 billion stimulus package. Worldwide central banks
cut their interest rates. Iceland government collapsed.
A $787 billion stimulus package was released by newly elected President Obama followed
by quantitative easing from the Fed amounting to over $800 billions.
2.1.2 Conclusion
We can outline a liquidity spiral at the macro-bank-level as follows ([14]) :
Banks balance sheets deteriorate All banks see exceptional losses in their activi-
ties. What was priced as an idiosyncratic risk appears to be a systemic risk: a
whole set of lendings deteriorate. That could come from mortgage, consumption
or investment lending. In our case it came from mortgages in the U.S.
Banks de-lever, selling assets This is amplied by the fact that banks tend to have
a pro-cyclical leverage, meaning that when things are going goods (balance sheets
grow), they lend even more, and when balance sheets are smalls, banks do not
leverage as much[15]. Volumes rise considerably as agents still think it is a conjec-
tural problem and therefore the slide appears as a good buying opportunity.
Chapter 3. Three Liquidity Crisis 24
Risk management tighten, lending reduced, counterparty exposures minimized.
A beginning of deance between operators appears. Prerequisites for lending cash,
reverse-repos, etc. become more strict. Volumes are reduced to levels below the
beginning of the crisis as prices keep on sliding.
Margins increase At this point, people may think theyre safe. Some crisis can stop
at this point if the lendings at the root of the problem nd their real value or that
extra cash incomes balance the deterioration. In such case, volumes can go back
to normal, pre-crisis levels.
Liquidity vanishes This is the point where market liquidity problems become signif-
icants. Prior to the crisis, market prices made sense compared to each others as
idiosyncratic risks were mirrored by relevant spreads and where exceptional dete-
riorations could be priced in and be diluted in the entire market. In our case, the
entire mortgage-backed securities market had to be reevaluated.
At this point, most parties nd themselves caught on the same side and adverse
interest is reduced by far. Assets loose value both as their inner value is reevaluated
but also because of lack of consideration: no other party is ready to take these
assets against cash.
If assets still nd no solid bid, markets are deserted. Volumes dive, prices become
extremely volatiles. Assets are marked-to-value in the books: they are given a
value (can be zero) in the balance sheets.
Traders face funding liquidity risk as they become unable to fund their positions
and are forced to unwind. Some OTC governmental deal can appear where public
institutions take parts of these assets against cash to protect the banks balance
sheets.
It is also important to underline that a fair amount of cash available do not entail
liquidity. Decisions as to whether rescue plans by governments and central banks helped
restoring liquidity are still questioned.
We will now assess the liquidity ows at a smaller scale through another crisis: The
November 2000 Turkish overnight crisis.
Chapter 3. Three Liquidity Crisis 25
2.2 The November 2000 Turkish Overnight Liquidity Cri-
sis
Figure 2.2: The Turkish Overnight Rate from 1998 to 2001. [Source: Bloomberg]
A liquidity crisis hit Turkey on November 20, 2000 that spanned for about 15 days. At
the peak of the crisis, overnight rates reached 2000%, forcing the intervention of the
Turkish central bank, an emergency loan by the IMF and the takeover or bankruptcy of
several banks of the country.
A close study of the events shows that the rationale behind the crisis lay between macroe-
conomic liquidity and market microstructure. The relatively small size of the Turkish
overnight market ( 180 nancial institutions executing about 1000 trades per day) allows
for a close understanding of the timing of the crisis and of the liquidity problems.
2.2.1 Chronicle of the Crisis
During the 1990s, Turkey was an active emerging market with an ination rate close
to 100% and an annual real GDP growth rate of 6.8%
3
. It signed numerous agreement
3
Global Finance Magazine - http://www.gfmag.com/gdp-data-country-reports/157-turkey-gdp-
country-report.html
Chapter 3. Three Liquidity Crisis 26
with the IMF stipulating that local rates should remain oating, that the government
had to keep o the overnight market, that banks were limited to 20% of their total assets
in foreign currencies.
Many banks used o-balance sheet transactions to exceed this ratio using bonds as
collateral. At the end of 2000, yields increased, reducing the value of these collateral and
banks faced margin calls. These operators often used the overnight market as a source
of funds to cover these margin calls, which in turn pushed short term yields higher,
decreasing bonds value, generating margin calls and so fueling more needs for cash to
cover margin calls. As a result, a vicious loop between short term yields and the cash
market appeared.
In the second half of 2000, the Turkish yield curve ended up inverted and some banks who
were on the edge of bankruptcy were aggressively borrowing on the overnight market,
including the biggest actor on that market: Demirbank. In November, banks tried
unsuccessfully to dump assets, stocks as well as local T-bills (1-year maturities bonds).
At that time, market operators started to question the solvency of each others and
solvent banks reduced their exposure to institutions rumored to be in trouble. Foreign
creditors started to withdraw their credit lines, rapidly selling the local currency and
forcing the central bank to provide liquidity by buying back the local currencies (it still
didnt intervene on the overnight market though).
On November 30, the central Bank stopped providing funds to the domestic market as
it reached its Net Domestic Assets target. The crisis culminated the very next day (Dec.
1), when the overnight interest rate reached 2000%.
On December 5, the IMF announced an emergency loan to Turkey which stabilized
the countrys economy but didnt save numerous nancial institutions from bankruptcy,
starting with Demirbank which stopped all banking activities that day. On December
6, the government took over Demirbank (which eventually was sold later on to HSBC).
The total outow during the crisis was summed to USD 6 billions, eroding about 25%
of he foreign exchange reserves of the Turkish central bank.
Also note that on November 22, in the midst of the crisis, Standard & Poors upgraded
the rating of Demirbank to B+ long-term and B short-term, completely overlooking the
liquidity needs of the bank.
Chapter 3. Three Liquidity Crisis 27
2.2.2 Conclusion
The liquidity shortage cycle is the same as the one exposed in the previous section.
However this crisis is also interesting as there was no real need to reevaluate assets in
a way similar to the asset-backed securities. Banks were simply over-leveraged through
o balance sheets techniques and the overnight market, tying together the entire rate
curve of the country. Another main dierence is the scale of the market, which allows
for agents to clearly read it and assess the troubles that may face other participants.
These informations allowed for numerous agents to withdraw from the market in time.
A last scale will be considered to assess liquidity at the execution level : minutes or
hours. To underline this, we will take as an example the nancial markets crash of May
6, 2010.
2.3 The May 6, 2010 Flash-Crash
Figure 2.3: Intraday quotation of the iShares S&P 500 Index Fund on May 6, 2010.
[Source: 16]
Chapter 3. Three Liquidity Crisis 28
On May 6, 2010, the prices of many U.S.-based equity products experienced a seldom
seen price decline and recovery within a time frame of about 20 minutes. This free
fall hit both equity indexes, equity ETFs, stocks and their related derivatives. Some
equities lost up to 15% in a few minutes before recovering most of the loss. Over 20,000
trades were executed at prices 60% away (up or down) from the values they had a few
minutes before, at prices ranging from one penny to $100,000.
On September 30, 2010, the U.S. Commodity Futures Trading Commission and the
U.S. Securities & Exchange Commission jointly released a memorandum regarding the
events of that day as seen both from the exchange aggregate ows and the various market
operators, market-makers, funds, etc.[16]
The report reveals the mechanisms that lead to a reduction of 99% of the market depth
for several contracts during few minutes.
2.3.1 Chronicle of the Crisis
May 6 started as a turbulent day over the European debt crisis, leading European CDS
up, U.S. equities slightly down and Euro currency down against the USD and the JPY.
At 1:00 PM, the bearish momentum started to lead to above average automatic execution
volumes, triggering a mechanism on the New York Stock Exchange (NYSE) known as
Liquidity Replenishment Point
4
at an unusual rate.
At 2:30 PM, the VIX Index, tracking the implied volatility of the S&P 500 options, was
up 22.5 percent from the opening level, U.S. Treasuries were ying and the Dow Jones
was down about 2.5%. At that time, both the E-mini Futures contracts and the ETFs
tracking the S&P 500 Index had their book order depth reduced by respectively 55%
and 20%.
At 2:32 PM, a mutual fund initiated a sell program to sell a total of 75 000 E-mini
Futures contracts (valued at $4.1 billion) as an hedge to an existing equity position.
This sell program was operated via an automated execution algorithm programmed to
target an execution rate of 9% of the volume of the preceding minutes (sliding), without
regard to price or time.
4
Liquidity Replenishment Points are moments during which automated execution is paused and then
resumed a few second later, allowing other, non-automated participants to catch up with the ow
Chapter 3. Three Liquidity Crisis 29
This execution resulted in the largest net change in daily position of any trader in the
E-mini since the beginning of the year. The price of the E-mini went down by 3% in
just four minutes as the other participants couldnt or didnt want to provide buy-side
liquidity. At that time the market depth of the E-mini fell to 1,050 contracts ($58
million) which is less than 1% of the depth level from that morning. With virtually no
participants left, the contracts price fell an additional 1.7% in 15 seconds to its intraday
low of 1056 at 2:45:28 PM.
Between the beginning of the sell program and that point, the sell algorithm had pushed
35,000 E-mini contracts in the market. During the same time frame, reacting to the
events, all the other traders combined sold more than 80,000 contracts and bought
50,000 which represent levels respectively 15 times and 10 times bigger than any 13
minutes interval during the past three days. In other words, volumes spiked, fueling
even more the execution rate of the sell algorithm.
At that moment, a circuit breaker of the Chicago Mercantile Exchange (where the E-
mini Futures are quoted) triggered a safety pause for ve seconds, allowing buy-side
interest to increase. When trading resumed, prices stabilized and shortly after the E-
mini began to recover. The sell algorithm kept operating until 2:51 PM but prices were
quickly rising.
Only two sell programs of similar size were executed in the E-mini within the 12 months
prior to May 6. These two other programs used a combination of manual trading and
several automated execution algorithms taking price, time and volume into account to
execute 75,000 contracts in over 5 hours. However, on May 6, as the markets were
already under stress and the sell algorithm only targeted volume, the sell program took
just 20 minutes to ll the orders.
Due to the relationships between the E-mini Futures, equity ETFs and stocks, (they
share the same underlyings), what could have been a local crisis to the E-mini Futures
also became a market-wide crisis, aecting all contracts. An important parameter is the
way market participants operate depending on their strategies and how they provide
liquidity to each others. Taking a closer look to these mechanisms will for instance
elucidate why both the market depth on the buy and the sell sides fell, why the volumes
spiked when no market depth was left and how some stocks happened to be traded at
irrational prices.
Chapter 3. Three Liquidity Crisis 30
Figure 2.4: E-mini S&P 500 Future Volume and Price on May 6, 2010. [Source: 16]
Figure 2.5: E-mini S&P 500 Future Market Depth on May 6, 2010. [Source: 16]
Chapter 3. Three Liquidity Crisis 31
2.3.2 Cross-market propagation
We segregate the dierent market participants depending on their role and strategies:
Exchanges Their role is to match orders and to publish data (volumes, prices, interests,
etc.) to the other participants.
Traders They are the clients of the market. They seek exposure to market moves
through various products, using management strategies and research analytics to
bet on price directions or gain dividends from their investments.
HFTs / Arbitrageurs They primarily focus on beneting from cross-market prices
discrepancies but with limited or no exposure to subsequent price moves in those
products. They can operate on the futures, ETFs or the underlying securities.
They trade signicant volumes (they account for over 30% of a days volume) but
keep very limited exposure.
Market-makers Their activity is very similar to HFTs and Arbitrageurs except that
it is mandatory for them to keep publishing bid/ask quotes to the market. Their
business is not to try to benet from product prices dierences but to gain the
bid/ask spread while keeping as little exposure as possible.
When the sell program began to operate, the E-mini Futures became relatively cheap
compared to ETFs or the underlying securities. HFTs and arbitrageurs were naturally
the rst to react and started to buy aggressively the E-mini to sell ETFs or the securities.
They traded nearly 140,000 E-mini contracts (33% of the volume) between 2:41 PM and
2:44 PM, while maintaining an overall exposure of less than 3,000 contracts at all time.
The Sell Algorithm responded to that increase in volume by increasing the rate at which
it was feeding the orders in the market even though the previous orders were obviously
not fully absorbed by fundamental buyers. As HFTs and Arbitrageurs reached their
exposure limit they started to act as liquidity-consumers instead of liquidity-providers
to reduce their exposure. At the same time numerous HFTs also stopped operating as
their systems triggered security pauses (taking into consideration prices moves, prices
integrity, volume, risk limits, etc.).
Chapter 3. Three Liquidity Crisis 32
Figure 2.6: U.S. Equity Market Participants.
At 2:45 PM, volume spiked as HFTs and Arbitrageurs were passing each others the
contracts (hot-potato eect) as no sucient buyer appeared and the sell algorithm
accelerated. Between 2:45:13 and 2:45:27 PM, HFTs traded over 27,000 contracts (49%
of the total trading volume) while only buying only about 200 additional contracts net.
That is when buy-side market depth in the E-mini fell to 1% of its depth from that
morning level and that the E-mini dropped by an additional 1,7% to reach its intraday
low.
For stocks and ETFs market-makers, who trade by submitting non-marketable rest-
ing limit order and capturing a bid-ask spread, rapid price movements and volatility
triggered both a reduction of the number of shares oered as well as a widening of their
quotes. Some market-makers completely left the market thus publishing mandatory
stub quotes
5
. Eventually, when liquidity vanished, some of these stub quotes were hit
by market orders.
It also appeared that many of the securities experiencing the most severe price disloca-
tions on May 6 were equity-based ETFs.
5
Stub quotes are quotes at unrealistic prices. They are submitted to respect the obligation of quotes
publication but are not expected to be hit.
Chapter 3. Three Liquidity Crisis 33
There are two ways of quoting an ETF. Either by providing quotes to the market as if
it was a stock, letting the price ow with the buy and sell interests. Either by using the
underlying securities to price th ETF. While the rst category of market-makers was
not severely injured by the crisis, the latter faced pricing problems when the volatility
increased, thus widening their quotes and stopping to provide liquidity to the underlying
securities.
Figure 2.7: E-mini, SPY and Aggregated S&P 500 Stocks Buy-Side Market Depth
on May 6, 2010. [Source: 16]
The cross-market propagation is clearly visible on the gure 2.7 where the E-mini leads
the fall, followed rst by the SPY ETF
6
and later on by the underlying securities. Note
that the recovery is also lead by the E-mini. In fact market participants generally
acknowledge that the E-mini leads the ETFs and the securities and not the other way
around.
2.3.3 Conclusion
This crisis also underlines the problem of assessing liquidity. For instance, measuring
liquidity through volumes during the crisis would have proven to be completely wrong
6
The SPY is the S&P ETF with the highest volume.
Chapter 3. Three Liquidity Crisis 34
(which measure was used by the sell program). One would have had to look at bid/ask
spread for market-makers and market depth on the exchanges.
The next Chapter will present models used to assess liquidity. The following chapter
will present the principles behind execution algorithms such as the Sell Program of this
chapter.
Chapter 3
Modeling Market Liquidity
This section presents the Almgren-Chriss Liquidity Asset Price Model (ACLPM) which
tries to mirror the market behavior using an asset price model generated from a discrete
Arithmetic Brownian Motion (ABM) and xed transaction costs. It was published by
Robert Almgren and Neil Chriss in December 2000[17].
The reader may nd a more comprehensive review of the literature surrounding the topic
by reading Mitton [18, Page 134].
The section Symbols (1) at the beginning of the document lists the letters and symbols
used in this chapter to facilitate the reading.
3.1 The Almgren-Chriss Liquidity Asset Price Model
3.1.1 Price Dynamics
We assume a security price with an initial price S
0
. At time t
0
we hold XS
0
. The
security price evolves naturally according to two factors: volatility () and drift ()
following the arithmetic random walk:
S
k
= S
k1
+ +w
k
with k going from 1 to N and w
k
being a Brownian w
k
N(0, ), that is w
k
=
k

with

k
a draw from independent random variables. Note that in extremely volatile markets
35
Chapter 4. Modeling Market Liquidity 36
or over a long period, it would be important to consider the geometric instead of the
arithmetic Brownian motion. Nevertheless, for our present situation, the arithmetic
motion should be enough.
We add to this behavior our own impact on the security through two parameters: a
temporary impact, caused by the surplus of supply caused by our order, and a permanent
impact, which implies a change in the equilibrium price of the security for a time at least
equal to the liquidation time. The permanent impact is thereafter named g() with
being the average rate of trading
n
k

(in units/time) between times t


k1
and t
k
.
Figure 3.1: Example of a generated buy order execution happening every 5 ticks.
In the end, we will consider the securitys price to evolve following:
S
k
= S
k1
+ +

k
g(
n
k

) (3.1)
with = 0 at rst to imply that we have no information about the direction of future
price movements, hence not permitting to optimize the execution using the knowledge
of an existing trend.
The temporary impact represents a slippage which is resorbed before passing the next
order. It is the paid price (

S
k
) on a give order, which depends on the previous price but
which will not remain until the next order (S
k+1
). This dierence, thereafter named
h(), depends on the average rate of trading during one time interval. Hence the actual
Chapter 4. Modeling Market Liquidity 37
Figure 3.2: Example of a generated walk with a buy order execution happening every
5 ticks.
price received on the kth sale is

S
k
= S
k1
h() (3.2)
and h() represents the dierence between the quoted price and the paid price.
3.1.2 The Denition Of A Trading Strategy
Suppose we hold a block of X units of a security that we want to liquidate before time T.
We divide this interval into N periods of equal length =
T
N
. We dene subsequently
the discrete time t
k
= k with k = 0, 1, ...N.
A trading trajectory is a list of holdings x
0
, x
1
, . . . , x
N
where x
k
is the number of units
that we plan to hold at time t
k
. Our initial holding is x
0
= X and we must obtain
x
N
= 0. (We take the case of a sell program. The construction stands the same for a
buy program.)
We also dene a strategy by its trade list n
1
, n
1
, . . . , n
N
where n
k
= x
k1
x
k
, that is
the number of units sold between the times t
k1
and t
k
. The relation between x
k
and
Chapter 4. Modeling Market Liquidity 38
n
k
therefore writes:
x
k
= X
k

j=1
n
j
=
N

j=k1
n
j
with k=0,. . . ,N.
A trading strategy is a rule determining the n
k
depending on the informations available
at time t
k1
. Such a strategy can either be static or dynamic. In a static strategy, the
rule for determining each n
k
depends on the information available at time t
0
, that is
prior to execution:
t
0
n
k
while in a dynamic strategy, each n
k
depends on the information up to t
k1
:

t
0
t
1
.
.
.
t
k1

n
k
3.1.3 Cost Of Trading
What we really want is to nd the cost (C) of a complete liquidation (from X to 0)
which is the dierence between the total paid price and the quoted price observed just
before the liquidation. That is,
C = XS
0

n
k

S
k
with
N

k=1
n
k

S
k
= XS
0
....
(a)
+
N

k=1

k
x
k
. .. .
(b)

k=1
g()x
k
. .. .
(c)

k=1
n
k
h()
. .. .
(d)
with (a) the initially quoted price, (b) the eect (positive or negative) of volatility, (c)
the loss of our total position and (d) the price drop suered for the n
k
units sold during
the period t
k
. Finally, the total cost of trading is:
C =
N

k=1
(

k

N

k=1
g())x
k

N

k=1
n
k
h() (3.3)
Chapter 4. Modeling Market Liquidity 39
which is a random variable centered around
E(x) =
N

k=1
x
k
g() +
N

n=1
n
k
h() (in dollars) (3.4)
and of variance
V (x) =
2
N

k=1
x
2
k
(in dollars squared) (3.5)
When a trader must execute an order, he has to take a decision according to the level of
risk he is willing to take. A defensive trader will try to reduce V (x) as much as possible,
thus carrying no uncertainty, while a trader willing to take some risk will try to minimize
E(x) given a maximum V (x).
Chapter 4
Order Execution
The directive of the European parliament dated of September 25, 2003 species the duty
of the market agents to operate as best as possible and thus, to quantify, report, and
optimize an order price.
Today, investors rely on simple averaging methods to get a clean price of a portfolio
transaction. Such methods encompass VWAP
1
or OHCL
2
. In all cases, the hidden part
(the dirty price) of the execution lies within the market impact of the order and the
risk carried during its execution.
According to Joe Ratterman, C.E.O of BATS Exchange, Inc.
3
, the average execution
size in the displayed markets today is under 200 shares. And theres so much investment
in technology to take some rms large interest in moving either buying or selling a
security, and making sure that thats get put into the market in a way that doesnt
move the market adversely against them[19].
Best market execution is simply the optimization of bots pushing orders in the market
following the principles exposed in the previous chapter.
1
calculated as the average price weighted by the volumes during the length of the execution
2
calculated as the median price of the highest, the lowest, the opening and the closing price of a day
3
Third largest exchange in the world by volume behind the NYSE and NASDAQ.
40
Chapter 5. Order Execution 41
4.1 Impact Functions
To see practical results of section 3.1, we have to dene the two impact functions. We
will take here two important cases of g() and h(). In the rst case, the permanent and
temporary impacts are a linear function of the volume while in the second case, they
are an exponential function of the volume.
4.1.1 Linear Impact Functions
By taking g() = (with a real absolutely positive number), the selling of n units will
depreciate the securitys price by , whatever how long it takes to sell them. Similarly,
taking h() = (with a real absolutely positive number), we get a proportional
eect. The xed costs of selling can also be taken into consideration as they have a
simple yet important impact. This amount is taken as a constant per order, that we
name . Finally,
h() = sign(n
k
)
. .. .
constant
+
....
linear
Rewriting equations 3.1 and 3.2 with these new impact functions we get:
S
k
= S
0
+
k

j=1

j
(X x
k
) (4.1)

S
k
= S
k1
sign(n
k
) +

n
k
(4.2)
which leads to rewriting equations 3.3 and 3.4 as:
C =
N

k=1
(

k

n
k

)x
k

N

k=1
n
k
( sign(n
k
) +

n
k
)
=
N

k=1

k
x
k

N

k=1
n
k
x
k

N

k=1
n
k
( sign(n
k
) +

n
k
)
=
N

k=1

k
x
k

1
2
(X
2

k=1
n
2
k
)
N

k=1
n
k
( sign(n
k
) +

n
k
)
=
N

k=1

k
x
k

1
2
(X
2

k=1
n
2
k
)
N

k=1
n
k
( +

n
k
) as we are selling
Chapter 5. Order Execution 42
and
E(x) =
1
2
X
2
+
N

k=1
|n
k
| +

k=1
n
2
k
with

=
1
2

=
1
2
X
2
+ |X| +

k=1
n
2
k
as we are selling all the way
while the variance 3.5 remains unchanged.
So, given these impact functions, we can either want minimum variance by selling ev-
erything at once, or minimum impact, by selling at a constant rate.
Minimum Variance In this case we liquidate everything with a single order: n
1
=
X, n
2
= n
3
= = n
N
= 0 (implying x
1
= x
2
= = x
N
= 0), which leads to:
E = Xh(
X

) = X +
X
2

V = 0
As expected the variance is null and the expected value does not depend on as there
is a single order. On the other hand, the loss can be arbitrarily large and depends on
the xed cost and the ratio

.
Minimum Impact In this case we liquidate at a constant rate n
k
=
X
N
(implying
x
k
= (N k)
X
N
) which leads to:
E =
1
2
XTg(
X
T
)(1
1
N
) +Xh(
X
T
)
=
1
2
X
2
+X +

X
2
T
V =
1
3

2
X
2
T(1
1
N
)(1
1
2N
)
Chapter 5. Order Execution 43
Figure 4.1: Plot of V (N) for X = 10000, = 20%, = 0, 01
We can compare this expected value with the previous one. By replacing

by its
expression and T by N we get:
E =
1
2
X
2
+X +X
2
(

N


2N
)
then we add and remove
X
2

:
E =
1
2
X
2
+X +X
2
(

N


2N
)
X
2

+
X
2

and we factorize to get:


E = X +

X
2
. .. .
Expected value of minimum variance
+(1
1
N
)X
2
[
1
2

]
. .. .
a
(4.3)
It is now clear that there is a risk as the expected value can be greater/smaller than the
previous one depending on whether

2
is greater or smaller than

.
4.1.2 Exponential Impact Functions
While linear impact functions constitute an intuitive reference, studying exponential
impact functions can be also lead to valuable conclusions. Indeed, even though denitive
proper impact functions for a given market should be calibrated empirically, exponential
Chapter 5. Order Execution 44
impact functions are the simplest continuous, non-liear forms of behaviour. We take:
g() = e

(4.4)
h() = sign(n
k
) +e

(4.5)
which leads to rewriting equations 3.1 and 3.2 as:
S
k
= S
0
+
k

j=1

j

k

j=1
e

n
k

(4.6)

S
k
= S
k1
sign(n
k
) +e

n
k

(4.7)
equations 3.3 and 3.4 become:
C =
N

k=1
(

k

N

k=1
e

n
k

)x
k

N

k=1
n
k
(sign(n
k
) +e

n
k
)
=
N

k=1
(

k

N

k=1
e

n
k

)x
k

N

k=1
n
k
( +e

n
k
) as we are selling
and
E(x) =
N

k=1
e

n
k

x
k
+ |X| +
N

k=1
n
k
e

n
k

(4.8)
while the variance 3.5 remains unchanged.
It is still obvious that we can get minimum variance by selling everything at once:
Minimum Variance In this case we liquidate everything with a single order: n
1
=
X, n
2
= n
3
= = n
N
= 0 (implying x
1
= x
2
= = x
N
= 0) and which leads to:
E = Xh(
X

) = X( +e

)
V = 0
As expected the variance is null. On the other hand, the loss can be arbitrarily large.
Minimum Impact Finding the minimum of equation 4.8 is not trivial. Nevertheless,
we can note a few things. First, that without any impact ( = = 0)we have E =
X(
N1
2n
+ + 1) which represents the oor of the expected value.
Chapter 5. Order Execution 45
Now, if we suppose that we sell at a constant rate n
k
=
X
N
(implying x
k
= (N k)
X
N
)
as in the minimum variance solution for linear impact functions we get:
E =
X
2
e

X
N
(N 1) +X +Xe

X
N
= X(
N 1
2
e

X
N
+ +e

X
N
)
and V =
2
X
2

(N + 1)(2N + 1)
6
Figure 4.2: Plot of E(N) for X = 10000, = 20%, = 0, 01, = 1.10
6
, = 1.10
6
The minimum is found for N = 15.
Figure 4.3: Plot of V (N) for X = 10000, = 20%, = 0, 01, = 1.10
6
, = 1.10
6
Chapter 5. Order Execution 46
4.1.3 Empirical Impact Functions
The linear and exponential impact functions allow us to grasp the markets behavior.
In practice, dierent market may have dierent behaviors, and one market can have
impact functions with more complicated expressions, including gaps, nite liquidity, etc.
Moreover, the two impact functions above display additional problems: First, trying to
calibrate the parameter (say using a linear regression) can lead to important errors for
certain points. Second, if we were to try to calibrate in real time, we would have
to recalculate the and parameters using all the observed data, which is very time-
consuming. Therefore, we can try to build a better solution, which could be easily
exploited and which could reduce the average as well as the maximum error. The idea
is to use the linear impact functions with dierent parameters according to . Formally,
it is:
We x limits to : {0, . . . ,
max
} Now, we build a partition of M subsets
{
1
, . . . ,
M
} on this set. For each set, we have:
g(
k
) =
k
h(
k
) = sign() +
k
with k {1, . . . , M}. Finally, we can have a single expression for all possible values of
:
g() =
M

k=1

k
,

k
(4.9)
h() = sign() +
M

k=1

k
,

k
(4.10)
with

k
,
the Kronecker symbol such as:

k
,
= 1 if
k

k
,
= 0 if
k
Chapter 5. Order Execution 47
4.2 The Ecient Frontier Of Optimal Execution
A trader is entailed to support a limited risk. This risk deploys over time and is rep-
resented by the variance of its strategies V (x). The optimal strategy is therefore the
strategy that minimizes execution costs E(x) in respect to a given limit V

for V (x).
This reduces to minimizing the function
U(x) = E(x) +V (x)
where is a number specifying the level of accepted variance. Given 3.4 and 3.5, the
general expression of U(x) writes:
U(x) =
N

k=1
x
k
g() +
N

n=1
n
k
h() +
2
N

k=1
x
2
k
(4.11)
As it is convex for any 0, the minimum of U(x) is found by setting its partial
derivative to zero:
U
x
k
=

g(
k
)
k+1
g

(
k
)

+
h(
k+1
) h(
k
)

+h

(
k+1
)

k+1

k

+2
2
x
k
(4.12)
for k = 1, . . . , N. Then setting
U
x
k
= 0 leads to:

g(
k
)
k+1
g

(
k
)

+
h(
k+1
) h(
k
)

+h

(
k+1
)

k+1

k

= 2
2
x
k
(4.13)
If we use the linear impact functions of the previous section we have:
g() = g

() =
h() = sign() + h

() =
By replacing in 4.13, we get:

(x
k1
2x
k
+x
k+1
) =
2
x
k
(4.14)
This is a recurrence equation, which if solved can generate the list of x
k
and n
k
.
Chapter 5. Order Execution 48
According to [17], the solution may be written as a combination of exponentials e
t
j
where satises:
2

2
(cosh() 1) =

2

and which gives, for x


0
= X and x
n
= 0:
x
k
=
sinh((T t
k
))
sinh(T)
X
and
n
k
=
2sinh(
1
2
)
T
cosh((T (k
1
2
)))X
For small steps of , [17] gives the following expected value and variance:
E(x) =
1
2
X
2
+X +

X
2
tanh(
1
2
)(sinh(2T) + 2Tsinh())
2
2
sinh
2
(T)
V (x) =
1
2

2
X
2
sinh(T)cosh((T )) Tsinh()
sinh
2
(T)sinh()
which is not a trivial solution.
Figure 4.4: Plot of E(N) for X = 10000, = 20%, = 0, 01, = 1.10
6
, = 1.10
6
,
= 0.01
Chapter 5. Order Execution 49
Figure 4.5: Plot of V (N) for X = 10000, = 20%, = 0, 01, = 1.10
6
, = 1.10
6
,
= 0.01
Figure 4.6: Plot of U(N) for X = 10000, = 20%, = 0, 01, = 1.10
6
, = 1.10
6
,
= 0.01
Chapter 5
Conclusion
Market liquidity is a dicult parameter to evaluate as it encompasses the desire that
people have to trade along with their constraints and their behaviors. As such, most
of the measures of liquidity in fact grasp historical liquidity and there is no way to
date to predict future liquidity (as opposed to the implied volatility extracted from
options prices). Dierent proposals have been put forward to cover that problem such
as American forward contracts (contracts allowing to buy/sell securities at any point in
time at screen price, see [18]).
The growing size of the nancial markets sure helped liquidity a lot and say stock
liquidity today is miles away from what it was only a few years ago (as an example,
the NYSEs average speed of execution fell to 0.07 second in 2009 from 10.1 seconds in
2005[21]). Nevertheless, when markets freeze, assets can loose all their values and create
a systemic crash.
A good part of the bettering of liquidity is also due to technology and the arrival of high-
speed trading and dark pools. For some, like George Sauter, chief investment ocer at
Vanguard Group Inc.
1
, high frequency trading is good as it has reduced transaction
costs by 50 percent over the past 10 years[21]. For regulators, such techniques make
nancial markets more opaque (as price manipulation can be hidden more easily) and
add systemic risk as they speed up the rate of execution and can be the source of costly
mistakes and avalanches[20].
1
The biggest U.S. mutual fund manager with $1.4 trillion in assets.
50
Bibliography
[1] Niall Ferguson. The ascent of money: A nancial history of the world. Penguin
Press, 2008.
[2] Adam Ashcraft Hayley Boesky Zoltan Pozsar, Tobias Adrian. Shadow banking.
Federal Reserve Bank of New York, (Sta Reports No.458), July 2010.
[3] Steven Marlin Ed Steinberg and Jesse Chen. The story of monetary policy.
Federal Reserve Bank of New York, 2010. URL http://www.newyorkfed.org/
publications.
[4] Lawrence H. White. Ination. Library of Economics and Liberty. URL http:
//www.econlib.org/library/Enc/Inflation.html.
[5] Gordon Pepper with Michael J. Oliver. The liquidity theory of asset prices. 2007.
[6] Anna J. Schwartz. Money supply. Library of Economics and Liberty. URL http:
//www.econlib.org/library/Enc/MoneySupply.html.
[7] Asani Sarkar Tarun Chordia and Avanidhar Subrahmanyam. An empirical analysis
of stock and bond market liquidity. Federal Reserve Bank of New York Sta Reports,
no. 164, March 2003.
[8] Amaury de Ternay Naji Freiha and Julien Berge. Couts de transaction et risques
dexecution : la question de la mesure. Banquemagazine, (653), December 2003.
[9] Vincent Deluard. Using equity etf ows as a contrary leading indicator. TrimTabs
Investment Research, 2010. URL http://www.trimtabs.com/global/pdfs/ETF_
Flows_and_Market_Returns.pdf.
[10] Nassim Taleb. Dynamic hedging, managing vanilla and exotic options. John Wiley
and Sons, Inc, 1997.
51
Bibliography 52
[11] William Ryan. International responses to the crisis timeline. Federal Reserve Bank
of New York, 2010. URL http://www.ny.frb.org/research/global_economy/
IRCTimelinePublic.pdf.
[12] Mauro F. Guillen. The global economic and nancial crisis: A time-
line. The Lauder Institute, Wharton, University of Pennsylvania, 2010.
URL http://lauder.wharton.upenn.edu/pdf/Chronology%20Economic%20%
20Financial%20Crisis.pdf.
[13] Economics Of Crisis. The great contraction: Timeline of events. 2010. URL
http://www.economicsofcrisis.com/economics_of_crisis/timeline.html.
[14] Lasse Heje Pedersen. Liquidity risk and the structure of nancial crises. Presenta-
tion prepared for the International Monetary Fund and the Federal Reserve Board,
October 2008.
[15] Tobias Adrian and Hyun Song Shin. Liquidity and leverage. Federal Reserve Bank
of New York Sta Reports, (328), May 2008. URL http://www.newyorkfed.org/
research/staff_reports/sr328.pdf.
[16] U.S. Commodity Futures Trading Commission and U.S. U.S. Securities & Exchange
Commission. Findings regarding the market events of may 6, 2010. September 2010.
URL http://www.sec.gov/news/studies/2010/marketevents-report.pdf.
[17] Robert Almgren and Neil Chriss. Optimal execution of portfolio transactions. 2000.
[18] Michael David Mitton. Derivative pricing and optimal execution of portfolio trans-
actions in nitely liquid markets. Thesis, university of Oxford, Trinity 2005.
[19] Alexandra Zendrian. Get briefed: Joe ratterman interview.
Forbes, June 2009. URL http://www.forbes.com/2009/08/14/
ratterman-bats-tradebot-intelligent-investing-exchange.html.
[20] Henri Emmanuelli. Rapport de la commission denquete sur les mcanismes de
speculation aectant le fonctionnement des economies. December 2010. URL http:
//www.assemblee-nationale.fr/13/pdf/rap-enq/r3034.pdf.
[21] Kambiz Foroohar. Speed geeks. Bloomberg Markets, 19(11), November 2010.

Вам также может понравиться