A presentation at The 2015 Copenhagen Business School Symposium on High-Frequency Trading. Robert Almgren, President and Head of Research at Quantitative Brokers (New York)
2. Main points
1.Bond trading is big, electronic,
and of very active interest
2.Algorithmic execution is important
3.Techniques for achieving good execution
2
3. Outline
1. Rates trading is becoming electronic
futures first, then cash
2. Business of algo execution
new in futures and fixed income
3. How to get good results
understanding of market microstructure
4. How do we improve toward that result?
use a market simulator
5. How do we build and use a simulator?
3
4. 4
501.5 Treasury
201.2 Agency MBS
27.3 Corporate
9.5 Muni
3.4 Non-agency MBS
1.5 ABS
4,6 Agency
748.9 Total
193 Total
US average daily traded volume in $Billion
2015 Jan-Aug
Fixed Income Equity
Source: SIFMA
Fixed Income markets are bigger than equities
8. 8TPMG White Paper,April 9, 2015
of both market makers and providers of algorithmic trading solutions to investors. As a result,
cutting-edge trading technology increasingly resides not only with dealers but also with their
clients who deploy algorithmic trading strategies to minimize trading cost. In the inter-dealer
market, market making in on-the-run Treasuries now appears to be dominated by dedicated
trading firms and dealers with the know-how to develop the cutting-edge algorithms required
to compete.
Several important initiatives to accelerate the automation of trading in Treasury markets mirror
key innovations from the foreign exchange (FX) market.13
Recent trends with FX antecedents
include the creation of new liquidity pools in the form of single-dealer platforms streaming
executable quotes, the emergence of aggregators that provide a single interface with streaming
quotes from multiple underlying markets, and the increased “internalization” of orders by
banks.
The evolution of automated trading in the Treasury securities market is also likely to be
influenced by innovation in the Treasury futures market and the interest rate swaps market, in
which execution is becoming increasingly automated. The Treasury futures and interest rate
11
For example, while Treasury securities are exempted from important aspects of the Volker Rule, leverage ratios
make holding Treasuries as expensive as corporate bonds. As a result, dealers are incentivized to shift balance
sheet to assets with higher yield and/or wider bid-ask spreads.
12
Source: Greenwich Associates 2014 North American Fixed Income Study.
13
The over-the-counter spot FX market is in many ways similar to the cash Treasury market. Liquidity in both
markets is concentrated in a few venues, with trading in the Treasury market focused in a few on-the-run
securities, and trading in the spot FX market concentrated in a few major currency pairs (though arguably across
more trading venues). In both markets, the vast majority of trading occurs with or between market makers, with
end-investors almost never trading bilaterally and intermediation facilitating flows between buyers and sellers.
TMPG CONSULTATIVE PAPER
swap markets are closely linked to cash Treasury securities markets by active spread trading
between the markets.14
BOX 1: TMPG Review of October 15, 2014
On October 15, 2014, U.S. Treasury securities experienced record-high trading volumes and
significant intraday volatility. The yield on the 10-year Treasury note traded in the fourth-largest
intraday range since the 2008 financial crisis, and saw a 15-basis-point round trip during a short 15-
minute window. This price action was outsized relative to the fundamental economic news of that
9. 9
High-frequency trading now accounts for roughly
40% to 50% of the daily volume in Treasuries,
compared to a "negligible amount" just five years ago,
according to research firm Tabb Group.
The enormous size of the Treasuries market makes it
appealing to firms that make money from huge
volumes of trades.
As with stocks, the fear is that flash crashes could
cause sharp plunges in Treasury prices.That would
lead to soaring yields, from all-time lows below 1.5%,
to as high as 4%.
Some of the biggest players in high-frequency trading,
DRW Holdings, Sun Trading,Virtu Financial, Hudson
River Trading, Jump Trading and GSA Capital Partners,
have all been expanding their reach to Treasuries,
according to numerous trading sources.
The Treasury Department wasn't able to provide
recent breakdowns on trading in government debt.
"It's been less profitable for high-frequency firms to
trade equities, so these firms are looking at other
asset classes," said Justin Schack, a managing director
at the institutional brokerage Rosenblatt Securities.
"Treasuries are one of the most liquid markets in the
world, so it's very fertile ground for high-frequency
market making."
That liquidity has attracted hedge funds that use
computer driven formulas to capture profits from
tiny price movements.
12. 12
Oct 15, 2014, NY time
Treasury "flash crash"?
Produced by QB from CME and cash market data
09:33 09:40
RetailSales
08:00 08:20 08:40 09:00 09:20 09:40 10:00 10:20 10:40 11:00
Scaledprice
5-year note futures
10-year note futures
30-year bond futures
Ultra bond futures
5-year note cash
7-year note cash
10-year note cash
30-year bond cash
08:30: Retail sales dropped more
than forecast in September on a
broad pullback in spending that
indicates American consumers
provided less of a boost for the
economy in the third quarter.
Victoria Stillwell, Bloomberg
15. 15
U.S. Department of the Treasury
Board of Governors of the Federal Reserve System
Federal Reserve Bank of New York
U.S. Securities and Exchange Commission
U.S. Commodity Futures Trading Commission
Joint Staff Report:
The U.S. Treasury Market
on October 15, 2014
T
RY
1789
July 13, 2015
U.S. Department of the Treasury
Board of Governors of the Federal Reserve System
Federal Reserve Bank of New York
U.S. Securities and Exchange Commission
U.S. Commodity Futures Trading Commission
Joint Staff Report:
The U.S. Treasury Market
on October 15, 2014
THEDEPART
M
ENT OF THE
TREASURY
1789
U.S. Department of the Treasury
Board of Governors of the Federal Reserve System
Federal Reserve Bank of New York
U.S. Securities and Exchange Commission
U.S. Commodity Futures Trading Commission
July 13, 2015
The Evolving Structure of the U.S.Treasury Market
Oct 20-21, NY Fed
The U.S.Treasury market is the deepest and most liquid government securities market in the world. However, the events
of October 15 of last year, when yields experienced an unusually high level of volatility and rapid round-trip in prices
without a clear cause, highlight the need to better understand the factors that impact the liquidity of the Treasury market,
especially during stressed market conditions. Because of the Treasury market’s unique role in the global economy, its
liquidity and functioning have implications for the cost of financing the government, the market’s role as a risk-free
benchmark for pricing financial instruments, the costs borne by investors transacting in the Treasury market, and the
implementation of monetary policy.
The U.S. Department of the Treasury, Board of Governors of the Federal Reserve System, Federal Reserve Bank of New
York, Securities and Exchange Commission, and Commodity Futures Trading Commission invite you to a conference to
explore the key factors underlying the evolution of the Treasury market's structure and liquidity. The conference will
consist of a variety of panels on subjects ranging from the October 15th Joint Staff Report, automated and algorithmic
trading, market making, liquidity, end investor perspectives on market structure, operational risks, repo markets, academic
and practitioner perspectives on potential improvements to market structure, and regulatory requirements applicable to
the government securities market. It will include prominent industry and academic experts and will feature remarks by
several authorities in the official sector.
report
conference
17. 17
Dealers claim they are victims of predatory
behaviour in US Treasury markets dominated
by high-frequency traders
The interdealer US Treasury markets have long
been a sanctuary for primary dealers – a safe place
to lay off the risks of their client-facing activities
among themselves. But an influx of high-frequency
traders has changed all that. Non-banks armed
with speedy algorithms and superior technology
are now the dominant players on the main
interbank platforms, Icap's BrokerTec and Nasdaq-
owned eSpeed. Dealers claim they are being
pushed out of what has always been one of their
core markets.
"BrokerTec and eSpeed have become markets
where the fastest player wins and everyone else
comes last," says a senior rates trader at a primary
dealer in New York. "The banks are just not set up
to compete in that scenario. We've been crowded
out of those markets, and we're not actively
making markets there anymore."
19. 19
100. The bottom line is that, even after controlling for movements in secondary market
yields for previously issued Treasuries, almost every Treasury maturity exhibits a consistent
pattern of negative post-auction yield movements (i.e., in Defendants’ favor), as seen by the
consistency of the below-zero purple bars in the following chart.
Case 1:15-cv-06782 Document 1 Filed 08/26/15 Page 36 of 85
100. The bottom line is that, even after controlling for movements in secondary m
yields for previously issued Treasuries, almost every Treasury maturity exhibits a consistent
pattern of negative post-auction yield movements (i.e., in Defendants’ favor), as seen by the
consistency of the below-zero purple bars in the following chart.
Case 1:15-cv-06782 Document 1 Filed 08/26/15 Page 36 of 85
UNITED STATES DISTRICT COURT
SOUTHERN DISTRICT OF NEW YORK
AND BAKERS AND TEAMSTERS
FUND, CLEVELAND BAKERS
MSTERS HEALTH AND
E FUND, AND MASTERINVEST
ANLAGE GMBH, on behalf of
and all others similarly situated,
Plaintiffs,
NOVA SCOTIA, NEW YORK
BMO CAPITAL MARKETS
NP PARIBAS SECURITIES
ARCLAYS CAPITAL INC.;
FITZGERALD & CO.;
UP GLOBAL MARKETS INC.;
Z MARKETS LLC;
YWIDE SECURITIES CORP.;
UISSE SECURITIES (USA) LLC;
APITAL MARKETS AMERICA
UTSCHE BANK SECURITIES
LDMAN, SACHS & CO.; HSBC
ES (USA) INC.; JEFFERIES LLC;
GAN SECURITIES LLC;
LYNCH, PIERCE, FENNER &
CORPORATED; MERRILL
OVERNMENT SECURITIES
UHO SECURITIES USA INC.;
Case No.
CLASS ACTION COMPLAINT
JURY TRIAL DEMANDED
Case 1:15-cv-06782 Document 1 Filed 08/26/15 Page 1 of 85
UNITED STATES DISTRICT COURT
SOUTHERN DISTRICT OF NEW YORK
CLEVELAND BAKERS AND TEAMSTERS
PENSION FUND, CLEVELAND BAKERS
AND TEAMSTERS HEALTH AND
WELFARE FUND, AND MASTERINVEST
KAPITALANLAGE GMBH, on behalf of
themselves and all others similarly situated,
Plaintiffs,
vs.
BANK OF NOVA SCOTIA, NEW YORK
AGENCY; BMO CAPITAL MARKETS
CORP.; BNP PARIBAS SECURITIES
CORP.; BARCLAYS CAPITAL INC.;
CANTOR FITZGERALD & CO.;
CITIGROUP GLOBAL MARKETS INC.;
COMMERZ MARKETS LLC;
COUNTRYWIDE SECURITIES CORP.;
CREDIT SUISSE SECURITIES (USA) LLC;
DAIWA CAPITAL MARKETS AMERICA
INC.; DEUTSCHE BANK SECURITIES
INC.; GOLDMAN, SACHS & CO.; HSBC
SECURITIES (USA) INC.; JEFFERIES LLC;
J.P. MORGAN SECURITIES LLC;
MERRILL LYNCH, PIERCE, FENNER &
SMITH INCORPORATED; MERRILL
LYNCH GOVERNMENT SECURITIES
INC.; MIZUHO SECURITIES USA INC.;
MORGAN STANLEY & CO. LLC;
NOMURA SECURITIES INTERNATIONAL,
INC.; RBC CAPITAL MARKETS, LLC; RBS
Case No.
CLASS ACTION COMPLAINT
JURY TRIAL DEMANDED
UNITED STATES DISTRICT COURT
SOUTHERN DISTRICT OF NEW YORK
CLEVELAND BAKERS AND TEAMSTERS
PENSION FUND, CLEVELAND BAKERS
AND TEAMSTERS HEALTH AND
WELFARE FUND, AND MASTERINVEST
KAPITALANLAGE GMBH, on behalf of
themselves and all others similarly situated,
Plaintiffs,
vs.
BANK OF NOVA SCOTIA, NEW YORK
AGENCY; BMO CAPITAL MARKETS
CORP.; BNP PARIBAS SECURITIES
CORP.; BARCLAYS CAPITAL INC.;
CANTOR FITZGERALD & CO.;
CITIGROUP GLOBAL MARKETS INC.;
COMMERZ MARKETS LLC;
COUNTRYWIDE SECURITIES CORP.;
CREDIT SUISSE SECURITIES (USA) LLC;
DAIWA CAPITAL MARKETS AMERICA
INC.; DEUTSCHE BANK SECURITIES
INC.; GOLDMAN, SACHS & CO.; HSBC
SECURITIES (USA) INC.; JEFFERIES LLC;
J.P. MORGAN SECURITIES LLC;
MERRILL LYNCH, PIERCE, FENNER &
SMITH INCORPORATED; MERRILL
LYNCH GOVERNMENT SECURITIES
INC.; MIZUHO SECURITIES USA INC.;
MORGAN STANLEY & CO. LLC;
NOMURA SECURITIES INTERNATIONAL,
INC.; RBC CAPITAL MARKETS, LLC; RBS
Case No.
CLASS ACTION COMPLAINT
JURY TRIAL DEMANDED
...
when the Department of Justice’s investigation was announced and Defendants curtailed their
improper conduct.
8. To take one example (many others are described in the body of this Complaint),
Plaintiffs examined what are known as “reissued” Treasuries. In this scenario, the United States
Treasury sells a security at auction, and then later sells the exact same security (i.e., a Treasury
Security with the same principal amount and the same maturity date) in a later auction.4
By the
time of the later auction, the Treasury Securities that were sold in the earlier auction are already
trading in the secondary market. Accordingly, it is possible to compare the yield/price of the
identical security at the same point in time at the auction in which Defendants had an “insider”
position. If the yields at the auction were repeatedly higher than the yields in the secondary
market, it would demonstrate that Defendants were artificially increasing the auction yields
(correspondingly suppressing prices). Conversely, if both the auction and the secondary markets
were truly competitive, the yield for the security would be substantially similar in both markets,
or at least vary in a random fashion.
9. The data shows that the yields for these identical securities indeed repeatedly
diverged as between the auction and secondary markets, almost always in the direction of a
higher yield (lower price) in the auction relative to the lower yield (higher price) in the secondary
market. Across all tenors (i.e., lengths of time to maturity) of Treasuries, the yields of reissued
Treasuries in the primary market were inflated in 69% of the auctions, by 0.91 basis points, a
clearly significant disparity. This repeated bias cannot be explained as a result of random
4
Reissued Treasuries also have the same coupon rate as the previously-issued security.
For reissued securities, price is thus the variable that is determined at the auction, and which was
directly manipulated by Defendants.
chance; instead, the only plausible explanation is that Defendants coordinated artificially to
influence the results of the auctions in the primary market.
10. This is just one example of anomalous economic data. Indeed, these and the other
data analyses discussed herein confirm that Defendants took full advantage of their privileged
Case 1:15-cv-06782 Document 1 Filed 08/26/15 Page 9 of 85
21. Quantitative Brokers
Algorithmic execution and cost measurement
No prop trading or market making
Interest rate products, starting with futures
Equities, FX already well served
Live on 5 futures exchanges (all products
Cash Treasuries: eSpeed, BrokerTec, dealer pools
basis trading futures vs cash
Good execution depends on
microstructure expertise
21
22. 22
20,000 lots20,000 lots
EIAPetroleumStatusReport
BML 1000 lots
98.795
98.800
98.805
98.810
98.815
98.820
98.825
09:22 09:24 09:26 09:28 09:30 09:32 09:34
CDT on Wed 14 May 2014
BUY 129 GEH6 BOLT
09:23:31
Doneat09:32:52
Exec 98.806
VWAP 98.812
Strike 98.8075
Sweep 98.8100
GEH6 Midpoint fills
Passive fills
Cumulative exec
Market trades
Limit orders
Cumulative VWAP
Cointegration
Microprice
Bid-ask
Exec = 98.806 Cost to strike = -0.31 tick = -$3.92 per lot
Our limit orders
Cointegration signal
(indicating down move)
Benchmark =
Arrival price
Order book
(direct + implied)Midpoint liquidity
Our fills
buy sell
Produced by QB from CME and internal data
Chicago time
31. Pro rata order matching
31
1 2 3 4 5 6 7Best bid price
Incoming volume divided
among all resting orders
at best price
Incoming market sell order
Versions of pro rata matching
are common in short-term
interest rate products,
because of large tick size
and low volatility.
32. 32
1. Orders placed during the “pre-opening” or at the indicative opening price (IOP) will be matched on
a price and time priority basis. Note that implied orders are not taken into consideration, as
they are only active during the continuous trading session.
2. Priority is assigned to an order that betters the market, i.e. a new buy order at 36 betters a 35 bid.
Only one order per side of the market (buy side and sell side) can have this TOP order priority.
There will be situations where a TOP order doesn’t exist for one or both sides of the market (for
example, an order betters the market, but is then canceled). There will never be a situation that
results in two orders on the same side of the market having TOP order status.
3. Only outright orders can be TOP orders, however the TOP orders of underlying orders that are
creating implied orders will be taken into consideration during the matching process so as not to
violate the TOP order rule in any market.
4. TOP orders are matched first, regardless of size.
5. After a TOP order is filled, Pro Rata Allocation is applied to the remainder of the resting orders at
the applicable price levels until the incoming order is filled.
6. The Pro Rata algorithm allocates fills based upon each resting order’s percentage representation
of total volume at a given price level. For example, an order that makes up 30% of the total
volume resting at a price will receive approximately 30% of all executions that occur at that price.
Approximate fill percentages may occur because allocated decimal quantities are always rounded
down (i.e. a 10-lot order that receives an allocation of 7.89- lots will be rounded down to 7-lots).
7. The Pro Rata algorithm will only allocate to resting orders that will receive 2 or more contracts.
8. After percentage allocation, all remaining contracts not previously allocated due to rounding
considerations are allocated to the remaining orders on a FIFO basis.
! Outright orders will have priority over implied orders and will be allocated the remaining
quantity according to their timestamps.
! Implied orders will be then allocated by maturity, with the earliest expiration receiving the allocation
before the later expiring contracts. If spread contracts have the same expiration (i.e., CONTRACT A-
CONTRACT B and CONTRACT A-CONTRACT C), then the quantity will be allocated to the earliest
maturing contracts making up that spread (i.e., the CONTRACT A-CONTRACT B will get the allocation
before the CONTRACT A-CONTRACT C because the CONTRACT B expires before the CONTRACT
C).
Example: Pro Rata Allocation Matching
CME Eurodollar order matching
33. Mixed time priority / pro rata
33
g algorithm
ucts, LIFFE takes
fj =
(V Pj 1)k
(V Pj)k
Vk
.
g is obtained for k = 1, and FIFO in the limit k ! 1.
polate between the two. LIFFE sets k = 2 for Euribor,
d Euroswiss (effective May 29, 2013, [LIFFE, 2013]).
d this algorithm, note that by the Mean Value Theorem
fj =
Vj
V
g(xj)
Pj = volume preceding order j
Robert Almgren and Eugene Krel May 31, 2013 2
0 1
0
LIFFE
q
k = 1 (pro rata)
2q
k = 2 (Euribor)
4q
k = 4 (Short Sterling)
0 1
0
1
CME
p q
(1−p) q
Figure 1: Mixed allocation algorithms, for small market order sizes; q denotes the
LIFFE
2-year Treasury
& some Treasury calendar spreads
Market order size as fraction of resting limit orders
k=1: pro rata
k=∞: time priority
38. 38
$/lot
BEST EXECUTION FOR FIXED INCOME AND FUTURES
Page 1
May 5, 2014
Algorithm Performance Comparison
Algorithmic futures trading has recently become very popular among the world’s largest
commodity trading advisors (CTAs), global macro hedge funds, pension funds and asset
managers. Buy side traders use algorithms to efficiently, anonymously, and cost effectively
execute their futures trades across global markets. Above all, these traders value consistently
good performance in line with benchmarks and expectations. In this environment, it has
become utterly essential for traders to analyze and compare alternative sell side algorithm
providers in order to select the best broker to achieve investment goals.
Quantitative Brokers (QB) is a fully independent, agency-only broker that services fixed-
income, equity index, and energy futures traders with a suite of algorithms designed to
minimize market impact on outright, spread, and multi-leg inter-commodity transactions. QB
offers existing and prospective clients free access to both real-time and historical simulators
for back testing and algorithm performance comparison analysis. Beyond simulations, QB
strongly encourages its current and prospective clients to simultaneously route similar
actual orders through alternative algorithms to conduct the most realistic performance
comparison analysis.
Revolution Capital Management (RCM) is a CTA based in Broomfield, Colorado. RCM
directs trading for approximately $680 million in its Mosaic Institutional Program, a diver-
sified program that trades over 50 markets with an average holding period of just over
3 days. Revolution’s research approach is statistically rigorous and can be summarized
as short-term pattern recognition. RCM traders executed similarly sized orders using the
arrival price algorithms at Quantitative Brokers and at three of the world’s largest bulge
bracket banks. RCM was gracious enough to share the results of this comparative analysis.
This analysis covers over 1.56MM futures contracts traded in Revolution’s Mosaic Institu-
tional Program from April 10, 2013 to April 17, 2014. Markets included are the E-mini S&P
500 and NASDAQ, Eurex Bund and Bobl, LIFFE Long Gilt, US 5-, 10-, and 30-year Treasury
$8.5MM in 21 months: 88 bp annual improvement in return
April 2013 through Jan 2015
1.4MM orders 1.2MM orders
39. −2 −1 0 1 2
0.00
0.25
0.50
0.75
1.00
Take
Post
Cumulativevolumefraction
Slippage to arrival price in ticks
CCP:mean=0.53
QB:mean=0.37
Thu 15 Aug 2013 to Fri 31 Jan 2014
All Eurodollar outrights
39
client
Produced by QB from internal data
$2MM savings in 6 months
40. 1 2 5 10 20 50 100 200 500 1000 2000 5000 10000
-2
0
2
4
Parent order size in lots
Slippagecosttomidpointasfractionofminimumpriceincrement
Take
Post
ZN -- Fri 04 Apr 2014 to Thu 02 Oct 2014
Size is not most important variable for rates
40
10-year Treasury futures
Produced by QB from CME and internal data
For rates products,
order size is not
most important factor
in slippage
41. Execution interval VWAP - AP in units of min px incr
Averageexecutionprice-AP(bid-askmidpoint)inunitsofminpxincr
Fri 04 Apr 2014 to Wed 01 Oct 2014
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
ZB 0.47
ZB
=
-0.02
+
0.69
x
ZF 0.63
ZF
=
-0.00
+
0.87
x
ZN 0.30
ZN
=
-0.04
+
0.64
x
ZT 0.13
ZT
=
0.03
+
0.70
x
T-Bond (589 ZB)
5-year (618 ZF)
10-year (632 ZN)
2-year (120 ZT)
0
20
40
60
80
100
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Thousandsoflotsexecuted
BOLT CME IR
0 20 40 60 80 100 120
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
Thousands of lots executed
ZT
ZN
ZF
ZB
41Produced by QB from CME and internal data
For rates, slippage is largely
controlled by ability
to forecast price motion
(and passive fills).
Price change during execution
Slippage
to
Arrival
Price
42. 1 2 5 10 20 50 100 200 500 1000 2000
-10
-5
0
5
10
15
20
Parent order size in lots
Slippagecosttomidpointasfractionofminimumpriceincrement
Take
Post
NG -- Fri 04 Apr 2014 to Thu 02 Oct 2014
Size matters more for non-rates
42
Natural gas futures
Produced by QB from CME and internal data
For non-rates products,
slippage depends on
order size (market impact)
43. Impact cost model
43
1 5 10 50 100 500 1000
-1
0
1
2
3
Executed size in lots
Slippageasfractionofminpxincr
mean = 0.423
50 100
0.250 0.295
ES from Thu 02 Jan 2014 to Thu 02 Oct 2014SP500 futures
Produced by QB from CME and internal data
44. What does performance depend on?
Passive fills
Short term price signals
Reliable performance in different mkt conds
44
45. Futures vs cash
45
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
19:00 21:00 23:00 01:00 03:00 05:00 07:00 09:00 11:00 13:00 15:00 17:00 19:00
real time basis
10 min exponential moving average
CT30 intraday basis on 2015.01.29
EST
18
19
20
21
22
23
24
25
26
27
28
29
30
19:00 21:00 23:00 01:00 03:00 05:00 07:00 09:00 11:00 13:00 15:00 17:00 19:00
real time basis
10 min exponential moving average
Cheapest−to−deliver intraday basis on 2015.01.29
EST
Basis = CF x Bond - Futures
Strongly mean-reverting: price signal
Produced by QB from CME and cash market data
46. How to develop and improve algos?
1. Pure theory and quant modeling
2. Experiment with real client orders
3. Simulator
46
47. 47
“S
kate to where the puck
is going to be, not where
it has been,” Wayne
Gretzky once told an interviewer.
As the Great One described it, what
cuts certain players a level above isn’t
native instinct alone, so much as end-
less practice seeing the ice and, frankly,
the hard work of getting to where a
scoring opportunity will be, before it
reveals itself.
Gretzky’s advice is one of Robert
Almgren’s favorite lines—but not
because the co-founder of Quantitative
Brokers (QB) is a hockey fan. Instead,
he says a similar idea applies to the
business of algorithmic futures execu-
tion: the more you see, the more you
test, the more instinctual an algo-
based strategy can become. Changing
expectations at systematic hedge funds
and commodity trading advisors
(CTAs) have put increasing emphasis
on an active and dogged approach to
execution, with recognition that in
correlated markets like interest rates,
the extra step is crucial, even if it is also
more expensive.
Shops both big and small now argue
it’s worth it, including AHL—Man
Group’s managed futures fund—and
Revolution Capital Management, a
Broomfield, Colo.-based CTA. And
much of the value, they say, derives
from what comes before any trades are
even made.
A potent mixture of in-house, futures
commission merchant, and boutique
brokerage-provided algorithms now
play a part in commodity trading
advisors’ and managed futures funds’
trading activities. Tim Bourgaize
Murray examines why a new cadre of
simulation tools is helping to organize—
and perhaps re-mold—these buy-side
specialists’ order flow.
April 2014 waterstechnology.com
development of our own execution
algos has allowed traders to move up the
value chain and focus on more complex
markets and strategies. Currently, we
rithms designed to handle multi-legged
Indeed, complex strategies like
legging are where the two streams of
QB’s information—post-trade total
ker takes on projects like constructing a
es the target
price on a synthetic instrument using
SALIENT POINTS
• Managed futures specialists are increasingly
taking advantage of boutique agency brokers’
algorithms, citing their ability to be opportunistic
and adjust to markets’ behavior, as well as
faster speed to implementation and greater
alpha realized through price slippage.
• Rates futures, particularly, are ripe for these
applications given their correlation and the char-
acteristics of the complexes within which they’re
traded, and are well-serviced by Quantitative
Brokers (QB), among other independent shops.
Hedge fund AHL and CTA Revolution Capital
Management are among QB’s users for rates.
• Another value-added feature at smaller shops
like QB is their simulation environments, which
mimic the matching engine logic of relevant
futures exchange venues and can test new
adjustments to algorithms with real-time market
data before putting the algos into production.
• Sources expect a greater variety of such
brokers to crop up in coming years, while sell-
side futures commission merchants (FCMs),
sensing greater competition, are also expected
to mature their offerings and continue bundling
futures algos with other execution and clearing
services.
trading technologies for financial-market professionals
ALGO CHASE
Managed Futures’
Waters Technology
Tim Murray
April 2014
The algorithmic “palette” for futures
is diversifying for two reasons. The
rst is that the style of trading is in
transition, somewhat belatedly, just as
oading more of
ing algo development—built in the
ers—many of them, including QB
and Pragma, run by former quants like
Almgren—put a premium on quality
rather than price. Sure, the uptake
“We do perform reviews on all
algos internally using our own
simulator, and are always keen
to compare these results with
those of the provider. If they
cannot provide a simulator, it
takes a lot longer to see if we
believe their story.” Murray
Steel, AHL
48. Computational fluid dynamics
48
Examples of CFD applications
CFD simulations by L¨ohner et al.
Examples of CFD applications
Smoke plume from an oil fire in Baghdad CFD simulation by Patnaik et al.
Introduction to Computational Fluid Dynamics
Instructor: Dmitri Kuzmin
Institute of Applied Mathematics
University of Dortmund
kuzmin@math.uni-dortmund.de
http://www.featflow.de
Fluid (gas and liquid) flows are governed by partial differential equations which
represent conservation laws for the mass, momentum, and energy.
Computational Fluid Dynamics (CFD) is the art of replacing such PDE systems
49. Computational fluid dynamics
Complete simulation is impossible
Discretize to capture key features:
• Conservation of mass, momentum, etc
• Positivity of density, etc
• Vortex dynamics
• Chemical reactions
• 2-D, 3-D, axisymmetric, etc
• Nonlocal effects (incompressible flow)
49
50. Computational market simulation
Complete simulation is impossible
(Human reaction is very complicated)
Key features to include:
queue position and match algorithms
price movement
Features to neglect for simplicity:
market impact
(Literature on agent-based markets)
50
51. Market simulator
Tool for developing and testing execution
algorithms for interest rate products.
Capture essential features of main markets:
• matching algorithms and passive fill probabilities
• short term pricing signals
Will have limitations -- useful anyway
Does not embody model of market impact
The one most natural way to build a simulator
51
52. What did not work
52
CME Test
Environment
Algo being tested
Dummy algos
Dummy algos
Dummy algos
ZI
ZI
ZI
ZI (Santa Fe
zero-intelligence
market algorithm)
Random order
sizes and times
No real market data:
no price signals
54. Criteria for simulator
• If no algo orders, reproduce market data
• If no market data, reproduce match engine
• Challenge: combine market data with orders
54
56. How to use simulator
Historical
rerun scenarios for algo improvement
backtests for potential clients
Real-time
clients can connect to “test-drive” algos
Algorithm development
test new signals on historical orders
multi-market legging trades
Real-time splitting for testing
compare simulator executions with real
56
57. Signal development
1. Propose idea
plausibility tests
2. Statistical tests on historical market data
nonzero correlation with future price movement
3. Rerun actual orders executed
show improvement in slippage
57
65. Summary
Fixed income trading is becoming electronic
Need full range of algo execution tools:
Transaction Cost Analysis (TCA) reporting
Market microstructure analysis
Algorithm optimization
Market simulator
65
66. Disclaimer
This document contains examples of hypothetical performance. Hypothetical performance results have many inherent
limitations, some of which are described below. No representation is being made that any account will or is likely to
achieve profits or losses similar to those shown. In fact, there are frequently sharp differences between hypothetical
performance results and the actual results subsequently achieved by any particular trading program.
One of the limitations of hypothetical performance results is that they are generally prepared with the benefit of hindsight.
In addition, hypothetical trading does not involve financial risk, and no hypothetical trading record can completely account
for the impact of financial risk in actual trading. For example, the ability to withstand losses or to adhere to a particular
trading program in spite of trading losses are material points which can also adversely affect actual trading results. There
are numerous other factors related to the markets in general or to the implementation of any specific trading program
which cannot be fully accounted for in the preparation of hypothetical performance results and all of which can adversely
affect actual trading results.
The reader is advised that futures are speculative products and the risk of loss can be substantial. Futures spreads are not
necessarily less risky than short or long futures positions. Consequently, only risk capital should be used to trade futures.
The information contained herein is based on sources that we believe to be reliable, but we do not represent that it is
accurate or complete. Nothing contained herein should be considered as an offer to sell or a solicitation of an offer to buy
any financial instruments discussed herein. All references to prices and yields are subject to change without notice. Past
performance/profits are not necessarily indicative of future results. Any opinions expressed herein are solely those of the
author. As such, they may differ in material respects from those of, or expressed or published by or on behalf of,
Quantitative Brokers or its officers, directors, employees or affiliates. Quantitative Brokers, LLC, 2010.
66