First-in-first-out (FIFO) markets process incoming messages in the same temporal order they arrive to the matching engine. Thus, reaction time to market events is one the most important factors in the competition for scarce resources. Following the introduction of electronic trading, many firms sought to decrease their reaction time to market events by automating their strategies (i.e., algorithmic trading) and by improving their technology (e.g., optimize their trading models, establish faster network links, upgrade their software and hardware stacks). Over time, this evolved into a technological arms race between several firms competing for better placement at the order book111Orders who occupy the top of the order book have better access to liquidity .. The average reaction time has drastically reduced, initially from seconds to milliseconds, and soon after down to microseconds and nanoseconds.
For any FIFO market race to be fair, participants must be treated equally and
compete only on the basis of their respective reaction times.
In this work, we argue that unconditionally equal treatment of the participants
in FIFO markets is becoming increasingly difficult to achieve.
This is primarily due to FIFO’s implementation and infrastructural requirements
that are hard to consistently meet in practice. For example, exchanges
(whose participants may have nanosecond-scale reaction times) must ensure that their
infrastructure does not introduce any delays that could alter the
relative arrival times of competing orders [25, 28, 31].
However, networking equipment (e.g., network switches) is rarely certified to guarantee
equal latency on all ports at a nanosecond scale .
To better understand the deployment complexities of FIFO order-matching,
we study several real-life incidents where exchanges failed to maintain
a level playing field for all market participants.
Most of these incidents were due to unintentional, subtle infrastructural inefficiencies
but, in certain cases, participants intentionally further exacerbated
the timing discrepancies through technical222In this work, we use the term “technical” to refer to knowledge, machines, or methods used in science and industry. This is not to be confused with “technical analysis” that refers to the analysis of statistical trends.
Based on the insights from our survey, we then argue that unbounded
temporal fairness is an elusive goal and introduce a bounded extension.
This more practical version of fairness could improve the microstructural
transparency of financial exchanges as well as allow participants to determine if the
degree of temporal fairness provided by an exchange is acceptable given their strategies.
Overall, this paper makes the following contributions:
Defines the implementation and infrastructure requirements that must be fulfilled by every exchange using an unbounded FIFO order-matching policy.
Surveys several “unfairness” incidents from financial exchanges and investigates their root causes.
Introduces a bounded version of fairness that improves the market transparency towards participants and could serve as a reference point for regulators.
In this section, we introduce some fundamental concepts of electronic markets and outline the operation of modern exchanges.
2.1 Market Structure
Figure 1 illustrates the different actors and components of an electronic market and their interactions. Market participants () submit their orders to the exchange either through brokers or directly to the exchange’s gateways. Brokers are used by participants who trade low volumes, while those with higher volumes usually prefer to place their computers in the exchange’s premises (i.e., colocation) and connect directly to the exchange (i.e., direct market access). In both cases, the exchange receives the incoming orders through its gateways which filter out invalid orders and forward the rest to be matched. To balance the load and reduce latency under heavy traffic conditions, exchanges tend to deploy several order gateways very close to the matching engine .
Once an order is received, it is placed in the order book where the matching algorithm ranks, pairs and fills it with those of other participants. 333A more thorough treatment of the limit order book is provided by Gould et al. . Once a match is found, the order is closed and removed from the book, if the entire open quantity of the order has been filled. Otherwise, the book’s record is updated to include only the remaining (unfilled) quantity. Participants remain up to date on their pending orders and the state of the order book through the market’s data feeds which are routed through the exchange’s update servers. Depending on their feed subscription type, participants may receive only periodic snapshots of the order book or near-realtime updates on the trading activity.
2.2 Order-Matching Policies
Order-matching policies specify the manner by which a financial exchange must process (e.g., rank, match, hide) the order messages it receives from market participants. Currently, one of the most commonly used policies is FIFO (also known as first-come-first-serve matching) . Under this policy, the matching engine processes order messages in the same temporal order in which the messages were received. Consequently, in markets with scarce resources, participants who react fast to market events have higher allocation chances compared to participants with slower reaction times.
In modern exchanges, this speed-related competition has three different manifestations depending on the roles of the competing parties i.e., maker vs. maker, taker vs. taker and maker vs. taker444In finance literature, “market makers” provide liquidity to the market (i.e., post resting orders that populate the limit order book), while “takers” remove liquidity by placing orders that are ‘marketable’ (so immediately match with makers’ orders causing those maker orders to be removed from the book). These roles are not fixed and participants transition from maker to taker depending on the side of the trade they take.. In the first case, makers compete with other makers for the best position in the queue at each price level in the limit order book. An order that is earlier in the queue at a price level is much more likely to get filled than one positioned later in the queue [13, 35]. Similarly, takers compete with other takers for favorably-priced bids and offers of scarce resources . The first taker order to the market has access to the full offered volume, while subsequent ones have a lower chance of getting filled, as the volume decreases. A maker who is trying to cancel their “stale” bid or offer may also compete with taker(s) who are trying to lift that bid or offer. This practice is called “sniping” and involves fast takers buying assets from or selling assets to slower makers at ‘stale’ bid and offer prices [4, 13, 21, 27]. A more thorough analysis of matching policies can be found in .
3 Unbounded Temporal Fairness
In economic theory, fair races for a resource in contention are won by the “fastest” participant if the order book operates in a continuous FIFO manner [4, 17]. Two market participants and are said to compete for a trading opportunity if responding to the same economic stimulus (e.g., market event), they both submit order messages that seek to capture [1, 27, 29]. We define as reaction time the time that elapses between the receipt of the stimulus by a participant and the submission of an order responding to the stimulus by . In the rest of this paper, we adopt the position of [4, 17] and assume that a market participant’s speed is a function of their (direct or indirect) investment in technology i.e., the more a participant spends the faster they are expected to be. With this in mind, we can now provide a definition of temporal fairness in financial exchanges [27, 29]:
Temporally Fair Race: Given a market event occurring at time , a race between two participants and with reaction times and is fair, if for the arrival times of their orders ( and ) it holds that: and , where is constant.
The constant represents the various transmission delays of the exchange’s infrastructure (e.g., network link latency for market updates, order transmission latency, order processing latency). A more relaxed version of this definition that seeks to bring slower and faster participants onto “equal footing” has been discussed in [27, 29, 33, 32]. We can now define when an exchange is temporally fair.
Temporally Fair Exchange: A financial exchange is considered temporally fair, if all the races between all possible pairs of its participants are always temporally fair.
4 FIFO Requirements
Intuitively, FIFO matching appears to conform with the above definition of temporal fairness by-design
(see Section 3) as orders are ranked and matched based on their relative
arrival time at the exchange’s engine. However, a FIFO market design can maintain its fairness
properties only if orders always reach the matching engine in the same relative order
they were sent. Similarly, market updates should reach all the colocated participants
simultaneously so as to ensure that no one has more time to react. More formally,
a FIFO deployment is temporally fair if it remains compliant with the following three
requirements at all times:
Consistent & Simultaneous market updates
Given a market event occurring at time and two colocated participants who receive the corresponding market update at times and respectively, it must always hold that = , for every possible pair of participants.
Preserve relative submission order
Given two colocated participants whose orders were sent at times and and arrived at the matching engine at times and , it must always hold that .
Honor price-time priority
Given an order that arrives at the matching engine at time and a competing order555The term “competing orders” refers to lit orders for the same instrument that reside on the same order book side and price level. that arrives at , if , it must always hold that will be executed before (unless is explicitly cancelled by the participant).
5 Practical Considerations
As outlined in Sections 3 and 4, temporally
fair markets require that participants compete solely on the basis of their reaction time while
every other factor of the system remains constant for all of them.
However, the practicality of these strict requirements is questionable.
We now survey various incidents from the relevant literature as well as
legal reports and show that in various cases participants experienced
favourable or unfavourable order submission latency or market update delay
due to technical inefficiencies.
Electronic exchanges comprise a plurality of hardware and software components that exhibit small non-constant variations in their processing times. For example, modern computers are optimized for maximum performance (i.e., instruction throughput) and thus do not guarantee deterministic execution times. This is due to speculative execution, cache eviction in the presence of competing processes, charge-to-read and data prefetching that contribute to variances in the process execution runtimes in a range between a few nanoseconds to a few hundred microseconds[34, 15]. Furthermore, distributed network architectures (such as those used by major financial exchanges) also introduce discrepancies in the processing times and asymmetries in the data dissemination speed. This is attributed to the various performance differences between replicated components. For instance, an order gateway may be significantly less crowded compared to the rest of the gateways, thus providing favourable submission times to its users . To address this, market operators strive to maintain their systems perfectly symmetrical and load balanced at all times with nanosecond  precision. While this can reduce the magnitude of the problem, differences cannot be fully eliminated even if the exchange uses and fine-tunes the exact same hardware and software in all of its replicated components.
Uneven Information Dissemination.
Another potential point of failure with regards to temporal fairness
are the market operations that involve information dissemination.
Such operations, if not implemented carefully, have the potential to
introduce discrepancies in the times that information is received by
different participants. For example, Tick-by-Tick data feeds
update participants after every change in the order book and
must be received by all subscribed participants simultaneously.
However, perfect synchronicity is technically challenging to achieve in practice.
In a relevant incident, the data feed servers at the National Stock Exchange
(NSE) of India were reportedly transmitting the updates in a sequential, non-randomized manner
to the subscribed participants, while the order gateways differed in capabilities
and loads . In particular, market updates were transmitted in
the same order participants logged in (on a per server basis), thus
enabling participants with knowledge of this technicality to gain an advantage by logging in early.
This resulted in unfair races as some participants were consistently receiving market updates
before everyone else (Requirement 1 in Section 4).
From a technical perspective, one-shot multicast could potentially decrease
the latency discrepancies by transmitting all the updates almost simultaneously666Multicast based on fanout-spitting or application-layer multicast overlay services suffer from delay problems similar to those of unicast ..
Unfortunately, even in this case, the inter-arrival discrepancies are reduced but not eliminated .
During the same incident, participants were also found to actively delay updates to others by unnecessarily occupying positions in the queue of update servers they were not using . Similar practices have been reported in various other exchanges and markets [14, 12, 18].
As discussed in Section 2.1, major exchanges have a modular (cf. monolithic) architecture
with several replicated components to improve their responsiveness and minimize their downtime. However, component replication
can have an impact on fairness as there may be fluctuations in the performance of the different components over time.
Participants can take advantage of such discrepancies even if they have no information about the performance of the individual
components (e.g., order gateways). In past incidents, participants were found to simultaneously broadcast copies
of their orders to all available gateways .
This practice improves the likelihood that their order will be routed through one of the faster (e.g., less loaded) gateways
thus outracing competing participants who do not replicate their orders (Requirement 2 in Section 4).
Various exchanges have taken measures to counter order replication, usually by limiting the number of simultaneous connections that are
permitted per participant [8, 9, 14, 12].
A common practice for market participants is to use private
fast network links to aggregate market data from multiple venues so as to take better
informed trading decisions. In centralized exchanges, this
practice does not violate any of the fairness requirements as the order submission
latency and the update-feed delays remain constant.
However, fast external links may have an adverse effect on the fairness of exchanges with decentralized order books.
Such exchanges maintain several geographically-distributed nodes (i.e., order books)
that use a network link to route incoming orders to the node that currently offers
the best price (Figure 2). Thus, an incoming order that gets submitted
in London may be routed to New York to be matched. However, this assumes that any
participant who observes an incoming order in London will not be able to
reach the New York book before the order is routed there.
While in most cases this is a reasonable assumption (such very low-latency links across continents are limited), the ICAP EBS venue’s [18, 11] inter-region routing link was outperformed by the Hibernia Express link, thus advantaging participants who had subscribed to it. To address the problem, the operator introduced a speedbump (i.e., a short constant delay) that alleviated the latency advantage provided by the faster network link .
Communication Protocol Violations. In networks that use store-and-forward switching, each complete transmission unit (e.g., frame, packet) is copied to the switch’s memory buffer and checked for potential errors (e.g., cyclic redundancy checks) before it gets forwarded to the next node. Therefore, the processing time of each message by the switch depends on its size as the whole unit must be received before any checks are performed. In 2015, a market participant was found to exploit this process by intentionally truncating data fields in their outgoing orders. This resulted in invalid data in some non-critical fields but gave them an advantage over other competing participants as switches would process their orders faster [6, 7].
Participants have been also found to take advantage of
the inherent properties of communication protocols. In networks,
large messages are often split into fixed-size fragments that are sent to their
destination sequentially. The recipient stores the incoming
fragments into a memory buffer and combines them to reconstruct the original message.
However, message fragmentation may result in ambiguity with regards to the
arrival times of incoming messages. This is of particular importance
in electronic exchanges as fragments of two competing orders (i.e., messages) may arrive interweaved.
To resolve this, matching engines have introduced tie-breaking policies that allow them
to determine the relative order of incoming orders in a consistent manner.
One such policy timestamps incoming orders based on the arrival time
of their first fragment (e.g., IP packet). This timestamp is then used to
determine their relative precedence in the order book.
Unfortunately, this policy is prone to optimistic messaging,
a form of technical gaming. Optimistic messaging relies on the
fact that orders are timestamped with the arrival time
of the first fragment but cannot be reconstructed
until all fragments have arrived [23, 19, 25].
A strategic participant can exploit this to establish precedence in the
book for orders they may intend to submit in the future.
For example, given an economic event that is to occur at a specific time ,
a participant preemptively sends a fragment of an order shortly before
(a few milliseconds early) to establish precedence in the order book. Then,
depending on the outcome of the event, the participant can decide to
trade on the news and transmit the rest of the order’s fragments, or drop the
incomplete order (e.g., by invalidating the network/application-layer checksum in the remaining fragments).
Depending on the variant used by the participant, the pre-event fragments may
contain the fields of the order that are not trade-specific (e.g., participant ID)
or may contain event-depend data if the event has only a few potential outcomes.
Unintended Order Interactions. Electronic markets paved the way for sophisticated order types that realize conditional interactions with the order book. However, the additional complexity introduced by those orders, made it more difficult for operators to account for all possible interaction scenarios. For example, certain order types (e.g., intermarket sweep orders, hide and light orders) could under specific conditions overtake other orders who preceded them in the execution queue (Requirement 3 in Section 4) [3, 26]. While these order types were available to everyone, their existence and operation was poorly communicated thus resulting in an unfair asymmetry between the participants who were aware of their operation and those who did not [40, 5, 3, 2]. In another occasion, the interaction between two order-types on New York Stock Exchange enabled participants to use pegging orders to detect the presence (but not the volume) of hidden orders . As a result, participants who exploited this design flaw could retrieve additional information on the current state of the order book compared to those that relied only on the market feeds. While this order interaction was a bug, it highlights how unintended corner-cases may occur in exchanges with complex microstructure.
6 Bounded temporal fairness
As discussed in Section 4, an exchange that guarantees fairness unconditionally must ensure that all updates arrive simultaneously to all its colocated participants. However, unlike price, time is continuous and thus simultaneity is unrealistic, especially considering the various hardware and software imperfections. We thus argue that unbounded fairness is an elusive goal. Instead, fairness should be considered with regards to a clearly defined reference frame. We now introduce bounded versions of the definitions given in Section 3 that provide such a reference frame.
-Bounded Temporally Fair Race: Given a market event occurring at time , a race between two participants and with reaction times and is -fair, if for the arrival times of their orders ( and ) it holds that: and , where is constant.
As before, the constant represents the various transmission delays of the exchange’s infrastructure, while quantifies the time scale at which the current infrastructure of the exchange can provide fair treatment between the participants. For example, a switch that consistently offsets one of its ports by 1ms can provide -fairness for , as the faster participant in races where will still win the race (assuming the rest of the infrastructure exhibits no jitter). For brevity, we refer to -bounded temporally fair races as -fair.
-Bounded Temporally Fair Exchange: A financial exchange is considered temporally fair with respect to bounds and , if the races between all possible pairs of its participants are
-fair with probability greater than or equal to.
In the above definition, represents the probability that a race between two orders will be -fair. We introduced
, as past works have shown that the probability distributions of the latency in distributed systems (both in financial exchanges and in other applications) exhibit long tails[24, 25]. In a practical setting, the operator can derive and by monitoring the discrepancies occurring in their system. Some techniques and measurement tools for this purpose have been discussed in .
Overall, the above definitions provide a more practical notion of temporal fairness as they relax the requirements for strict simultaneity in market updates and complete precedence-preservation in orders. Bounded fairness can improve the transparency of financial exchanges and allow participants to determine if the degree of temporal fairness provided by an exchange is acceptable given their strategies. Moreover, and could serve as a well-defined reference point for regulators and could drive a positive competition between exchanges.
Note, however, that these definitions do not make a statement about the races that take place beneath the boundary. This leaves a blind spot, where all the races with reaction-time differences less than could be consistently won by the slowest competing participant. Operators can try to decrease (and increase ) by upgrading their infrastructure but similarly participants are also getting consistently faster and more sensitive to subtle differences. We argue that non-continuous market designs at
timescales could solve this problem by uniformly distributing the victories between participants for races beneath[27, 29, 33, 32]. In these market designs, the majority of the participants will experience a FIFO market, while the small minority of ultra-fast traders will be still guaranteed at least equal treatment at all scales.
7 Conclusions & Future Directions
This work discusses the gap between economic theory and the unavoidable complexities that emerge when theoretically “fair” market models are implemented. We focus on FIFO order-matching and its infrastructural and implementation requirements that “fair” deployments must fulfill. Our survey of the literature and relevant sources suggests that technical challenges make full compliance with these strict requirements unrealistic. Instead, we propose a bounded version of fairness that takes into account the unpredictability of hardware and software systems. The bounds clearly define the degree of fairness that participants can expect from an exchange and can be tightened further as the operators refine their infrastructure. For cases that fall outside these bounds (i.e., very low latency competition), discrete-time policies can be applied to guarantee at least equal participant treatment.
We thank the attendees of the workshop for the insightful discussion and the valuable feedback. Moreover, we would like to thank David Kohan Marzagão for his input with regards to our definitions of fairness. Vasilios Mavroudis was partially supported by University College London and BinanceX through the Tradescope project.
-  Angel, J.J., McCabe, D.: Fairness in financial markets: The case of high frequency trading. Journal of Business Ethics 112(4), 585–595 (2013)
-  Australian Securities and Investments Commission: Report 331: Dark liquidity and high-frequency trading. https://download.asic.gov.au/media/1338878/info178-published-15-January-2014.pdf (2013)
-  Bodek, H.: The Problem of HFT: Collected Writings on High Frequency Trading & Stock Market Structure Reform. Decimus Capital Markets LLC, Norwalk, CT (2013)
-  Budish, E., Cramton, P., Shim, J.: The high-frequency trading arms race: Frequent batch auctions as a market design response. The Quarterly Journal of Economics 130(4), 1547–1621 (2015)
-  Buti, S., Consonni, F., Rindi, B., Wen, Y., Werner, I.M.: Sub-penny and queue-jumping. Charles A. Dice Center Working Paper (2013-18) (2015)
-  CME Group: Slides from new iLink architecture webinar (part i). https://web.archive.org/web/20141005062026/http://www.cmegroup.com/education/new-ilink-architecture-webinar.html (2014)
-  CME Group: Notice of Summary Access Denial - NYMEX-16-0600. https://www.cmegroup.com/notices/disciplinary/2016/11/NOTICE-OF-EMERGENCY-ACTION/NYMEX-16-0600-ELDORADO-TRADING-GROUP-LLC.html#pageNumber=1. (2017)
-  CME Group: CME Group Customer Forum. https://www.cmegroup.com/globex/resources/files/customer-forum-q4-2018.pdf (2018)
-  CME Group: Introducing CME Group. http://web.stanford.edu/class/cs349f/slides/CME(2019)
-  Cont, R., Kukanov, A., Stoikov, S.: The price impact of order book events. Journal of financial econometrics 12(1), 47–88 (2014)
-  Detrixhe, J., Mamudi, S.: A New Fast Lane for Traders Spurs Plan to Thwart Exploiters. https://www.bloomberg.com/news/articles/2015-08-13/a-new-fast-lane-for-traders-spurs-plan-to-thwart-exploiters (Aug~2 2015)
-  EBS Service Company Limited: EBS Dealing Rules – High Level Summary. http://image.exct.net/lib/fef6127172670c/d/1/EBS(2012)
-  Farmer, D., Skouras, S.: Review of the benefits of a continuous market vs. randomised stop auctions and of alternative priority rules (policy options 7 and 12). In: Manuscript, Foresight. Government Office for Science (2012)
-  Garnham, P.: Foreign exchange: EBS vows to remove unfair advantages. https://www.euromoney.com/article/b12kjjrny27stq/foreign-exchange-ebs-vows-to-remove-unfair-advantages (2012)
-  Godbolt, M.: Memory and Caches: How they work, and how they affect performance. https://www.youtube.com/watch?v=vDns3Um39l0 (2018)
-  Gould, M.D., Porter, M.A., Williams, S., McDonald, M., Fenn, D.J., Howison, S.D.: Limit order books. Quantitative Finance 13(11), 1709–1742 (2013)
-  Harris, L.: What to do about high-frequency trading. Financial Analysts Journal 69(2) (2013)
-  Howorka, E., Horsfall, P., Iaccheo, S., Merold, M., Steyn, S.: US Patent 2007/0124419: Distribution of Data to Multiple Recipients. https://patentimages.storage.googleapis.com/80/53/fc/ab9c809e14c402/US20070124419A1.pdf (2007)
-  Hurd, M.: The Accidental HFT Firm (Forthcoming, Negative Latency Chapter) (2019), https://meanderful.blogspot.com/2018/06/negative-latency.html, accessed Mar 28 2019
-  Janecek, K., Kabrhel, M.: Matching algorithms of international exchanges. Tech. rep., Citeseer (2007)
-  Kirilenko, A., Kyle, A.S., Samadi, M., Tuzun, T.: The flash crash: High-frequency trading in an electronic market. The Journal of Finance pp. 967–998 (2017)
-  Klöcking, J.U., Maihöfer, C., Rothermel, K.: Reducing multicast inter-receiver delay jitter-a server based approach. In: International Conference on Networking. pp. 498–507. Springer (2001)
-  Lariviere, D.A., Hosman, B.P., Peck-Walden, P.I., Studnitzer, A.L., Bonig, Z., Nagarajan, M.: Message processing protocol which mitigates optimistic messaging behavior. US Patent App. 15/851,067 (Jun~28 2018)
-  Li, J., Sharma, N.K., Ports, D.R., Gribble, S.D.: Tales of the tail: Hardware, os, and application-level sources of tail latency. In: Proceedings of the ACM Symposium on Cloud Computing. pp. 1–14. ACM (2014)
-  Lohr, A.: Open Day 2019: T7 infrastructure and latency. https://www.deutsche-boerse.com/resource/blob/1637232/da0ae611905acda0d7502260903a0835/data/Open-Day-2019_T7-Latency-Roadmap_Andreas-Lohr_final.pdf (2019)
-  Mavroudis, V.: Market manipulation as a security problem: Attacks and defenses. In: Proceedings of the 12th European Workshop on Systems Security. p.~1. ACM (2019)
-  Mavroudis, V., Melton, H.: Libra: Fair order-matching for electronic financial exchanges. In: Proceedings of the 1st ACM Conference on Advances in Financial Technologies. pp. 156–168. ACM (2019)
-  Melton, H.: Ideal latency floor. US Patent App. 14/533,543 (Nov~5 2014)
-  Melton, H.: A fairness-oriented performance metric for use on electronic trading venues. In: 2017 International Conference on Computational Science and Computational Intelligence (CSCI). pp. 1027–1030. IEEE (2017)
-  Melton, H.: Understanding and improving temporal fairness on an electronic trading venue. In: 2017 IEEE 37th International Conference on Distributed Computing Systems Workshops (ICDCSW). pp.~1–6. IEEE (2017)
-  Melton, H.: On fairness in continuous electronic markets. In: Proceedings of the International Workshop on Software Fairness, FairWare@ICSE 2018, Gothenburg, Sweden, May 29, 2018. pp. 29–31 (2018)
-  Melton, H.: On fairness in continuous electronic markets. In: 2018 IEEE/ACM International Workshop on Software Fairness (FairWare). pp. 29–31. IEEE (2018)
-  Melton, H.: The paradoxical effects of jitter on fairness in financial exchanges: Engineering implications. Working paper (2019)
-  Metamako: MetaApp 32: An adaptable platform that enables intelligence at the network edge leveraging FPGA, x86_64 server and 5ns layer 1 switching technology. https://www.xenon.com.au/wp-content/uploads/2017/06/MetaApp32_data_sheet.pdf (2017)
-  Moallemi, C., Yuan, K.: The value of queue position in a limit order book. Market Microstructure: Confronting Many Viewpoints (2014)
-  Prabhakar, B., McKeown, N., Ahuja, R.: Multicast scheduling for input-queued switches. IEEE Journal on Selected Areas in Communications 15(5), 855–866 (1997)
-  Preis, T.: Price-time priority and pro rata matching in an order book model of financial markets. In: Econophysics of Order-driven Markets, pp. 65–72. Springer (2011)
-  Securities and Exchange Commission: Securities Act of 1933 Release No. 10463. https://www.sec.gov/litigation/admin/2018/33-10463.pdf (March 2018)
-  Securities and Exchanges Board of India: Order in the matter of NSE Colocation. https://www.sebi.gov.in/enforcement/orders/apr-2019/order-in-the-matter-of-nse-colocation_42880.html (2019)
-  United States of America before the Securities and Exchange Commission: In the Matter of UBS Securities LLC, Exchange Act Release No. 74060. https://www.sec.gov/litigation/admin/2015/33-9697.pdf (2015)