Mechanics, Fading, and Performance

Mechanics, Fading, and Performance

Mechanics, Fading, and Performance The performance of conditional orders has been a growing point of discussion as more Alternative Trading Systems now support the order type and nearly every algorithmic provider uses it within their offering. Gaming and other forms of information leakage are the primary concerns. These concerns stem from the very function of conditionality: two or more potential contras are notified of an opportunity to trade but have the choice to not trade. While there are legitimate reasons why an invitation on a conditional order may not be accepted (aka firmed up), the reason is not disclosed to the ATS or to the counterparty. In the following report we explain the mechanics of the conditional order and analyze the performance of two scenarios. First, we look at the unfilled and filled conditional orders across several strategies within Liquidnet s two Alternative Trading Systems. Second, we compare the performance of orders resting only at Liquidnet with orders resting in multiple pools (through our Liquidnet Dark aggregation strategy 1 ). The statistics cover Q1 2018. The key findings are as follows: Interactions with at least one manual counterparty were 1.5x times larger than interactions between two conditional algo orders. We believe this is largely due to the loss of optionality resulting from committing shares to any algorithmic strategy. Conditional orders represented only at Liquidnet outperformed conditional orders within a dark aggregation strategy by 75% when measured against the impact adjusted arrival price. However, the dark aggregation strategy had nearly twice the fill rate. External algorithmic strategies 2 responded to a firm-up request nearly 97.7% of the time when the contra was a manual trader; compared to 84.8% when the contra was another algo. The mid-price movement one second after one party fades on a conditional interaction is equally distributed across three categories: Favorable, Neutral and Negative. The mid-price movement after five minutes is nearly equally distributed across Favorable and Negative. So what did we learn? While there will always be a degree of uncertainty regarding the intentions of conditional behavior, conditionals did achieve their goal of increasing trade size and fill rate with no measurable negative performance on failures to firm-up. Representation at multiple pools (leveraging conditionals) had nearly twice the fill rate of a single pool. However, single pool representation had better performance than multiple pool representation against Impact Adjusted Arrival Price and PWP 20. The act of fading within the Liquidnet ATSs did not lead to consistent negative performance to the jilted counterparty. However, this may or may not be the case within other pools or algo providers. Traders should ask other ATSs and algo providers for detailed performance statistics on conditional behavior. 1 Liquidnet Dark is a dark aggregator algorithm that routes conditional orders into the Liquidnet ATSs, BIDS, ITG Posit, and BlockCross, among other pools. 2 External algorithms include conditional orders sent from Liquidity Partners (brokers who submit orders into Liquidnet H2O) and buy-side clients who send orders directly into one or both Liquidnet ATSs. 2

The old school block trading process probably sounds absurd to anyone who joined the industry after the advent of electronic trading. And yet, nearly two decades later there are still manual block desks putting together trades. It would be easy to use clichés to explain the survival of the high touch block trade. (It s still a relationship business. Muscle memory. Paying bills.) In the age of automation though, the traditional Block Crossing Network manual workflow can seem equally anachronistic. Why do either of these processes still exist? It comes down to information and optionality. Traditional block desks still exist because they have information that helps bring buyers and sellers together that cannot be read in from an asset manager s order pad. There is information (such as a deep understanding of a Portfolio Manager s historical decision-making process) that can help unlock latent liquidity but not many Block Crossing Networks have access to it 3. There are also trader preferences, special situations, and risk factors that poorly map into electronic solutions. However, the velocity of change and innovation coupled with liquidity changes will continue to redefine the role of the traditional sales trader. Block Crossing Networks have automated the process of matching real-time opportunities to trader appetites. Optionality is key because it allows traders to maximize their interest in a specific name while attempting to source liquidity elsewhere. Immediacy is the enemy of size. The risks associated with block trading are well known and directly related to information and optionality. You can t source liquidity without giving up information. The optionality that you receive with a match is what allows counterparties to fall down. There is no silver bullet for that dilemma. It requires a combination of accountability, empathy, self-awareness, and trust. Attempts to create mechanisms that constrain optionality have failed, not because of poor implementation, but because they tried to defy basic laws of liquidity. Most electronic (aka ATS) blocks are executed at venues that have direct buy side OMS/EMS integrations. In Q1 2018, five Alternative Trading Systems (BIDS, ITG POSIT, Liquidnet Negotiated, Liquidnet H2O, and Blockcross) accounted for 63% of ATS blocks 4. Luminex, which also has direct integrations, was the 10th largest ATS block venue over that same period. The willingness to allow any of these venues to establish a blotter sync integration with an OMS or EMS depends on the non-committed status of these orders. Obviously, there is also a great degree of trust in allowing these venues to access this sensitive information. The trust is operational and behavioral: block venues need to police its participants to create safe, curated, and consistent optionality. 3 In the spirit of limiting the overt commercial nature of this piece, we will refrain from talking about Targeted Invitations and other components of our Liquidnet Virtual High Touch Platform. 4 Does not include Tradeweb Dealerweb which although technically an ATS, is a broker-only RFQ platform. 3

Exhibit One: ATS Block Market Share Source: FINRA ATS Block Data 4

Participants The composition of the block crossing networks has evolved along with the industry. At first, manual traders dominated, with most interactions taking place between two humans. As the use of algorithms and other automated strategies (primarily dark aggregation) increased, block venues came out with functionality so manual traders could automate interactions in the pool. As the buy side began to automate some of its routing decisions, the block crossing networks supported conditional orders, which help smart order routers source block liquidity while minimizing opportunity cost. In some ways the conditional order has similar trade-offs to the traditional high touch indication ; the conditional order gives the automated strategy the ability to rest at a venue while seeking liquidity elsewhere. However, in practice the use of the conditional order is very different from the traditional manual interaction. In the manual interaction, it is not necessary for either party to commit shares to a venue to see a trading opportunity. With a resting manual block order, the full parent quantity is available for matching without being committed, the natural result is a bigger print. Using a dark aggregation strategy requires that the shares are committed to the algo provider. This is a barrier to maximizing the size of the print 5. Below are the average trade sizes across the various types of interactions for the first three months of 2018. The largest print sizes occur when both parties are manual. When one party is manual, the trade size is smaller but still block-sized. When both parties are automated 6 but leveraging conditional orders, the average trade is larger than standard market-wide trade sizes but relatively small compared to a manual interaction at a block crossing network or an OTC block trade. As a point of comparison, we are also including the average block size for OTC prints off-exchange block trades that did not occur in an ATS. This category could include traditional block desks, capital commitment scenarios and other forms of market making. INTERACTION TYPES Q1 2018 Liquidnet ATS Buy Side Manual Negotiation 41,250 Manual vs Buy Side Algo 31,982 Manual vs Broker Algo 21,611 All Buy Side Algo 19,319 All Broker Algo 13,594 OTC/Upstairs 7 OTC Block Trades 37,345 All US Exchanges Average Trade Size 202 5 It could be that the cause and effect are reversed. Perhaps traders use algos precisely because they do not want to trade a very large block, but are happy to get a large print within the use of the algo strategy. This strikes me as odd behavior but could be true. 6 Automated includes conditional orders and algorithmic orders. 7 This includes all non-ats off-exchange block trades. 5

The ATS workflow for a conditional order is simple. The FIX message for a conditional order is like any other new order message with an additional tag. The ATS cannot execute against the conditional order because it is not a committed order. When the ATS receives an order that could execute against the conditional, the ATS notifies the conditional sender. After sending a firm-up request, the ATS waits some specified period for the conditional order sender to respond. If the conditional sender responds positively and market conditions are normal and both orders are still executable, the trade occurs. If the conditional sender does not respond, it is logged as a failure-to-firm event. This is commonly referred to as a timeout. The notification of a pending trade is commonly referred to as a firm-up request. This bit of jargon is used because the ATS can only execute against an executable, or firm order. Technically, the firm-up request is an unsolicited cancel message. The message usually includes the minimum size necessary for the firm order to execute. The most basic approach to the routing of the initial orders is called spraying. Spraying is when multiple conditional orders are routed at the same time (or at least with no intentional delay). A more nuanced and common approach uses logic to sequence the order of the routes, which is often called staggering. When receiving firm-up requests, the most basic and common approach is self-explanatory: first in, first out (FIFO). A more nuanced approach is that after a firm-up request is received, the router waits a very short period (<1sec) to see if it receives any additional firm-up requests and then determines which ATS will receive the firm order. This is often called batching. Each approach has its pros and cons. On the one hand, spraying takes full advantage of the conditional order ecosystem: there is no risk in over-executing so the router attempts to find as much liquidity as fast as possible. Below is an example of spraying: There are two main logical points within the conditional routing process: a) initial routing of the conditional orders, and b) responding to a firm-up request. How a router handles these decisions can impact the performance of each order as well as the broader statistics around the effectiveness of conditionals. TIME ACTION 11:00:00 am Trader sends order to Router A to buy 100k shares 11:00:05 am Router A sends 100k conditional order to BIDS ATS 11:00:05 am Router A sends 100k conditional order to ITG ATS 11:00:05 am Router A sends 100k conditional order to Liquidnet ATS 6

In less than a second, there are 400,000 shares of conditional liquidity being represented across the four venues. In the instance when there is unique firm liquidity being represented within a single ATS, spraying will access this liquidity as fast as possible. Let s continue the example: TIME 11:00:06 am ACTION BIDS ATS sends firm-up request with min quantity of 25k shares Multiple Firm-Up Requests In the instances when the contra liquidity is also being represented in multiple pools, the flow becomes a bit messier. While the algo sprayed the conditional orders out at the same time (with some minimal and unintentional difference in the actual routing time), it is unlikely that the ATSs will either receive or respond at the same time, but the result is the same: the algo receives a flurry of firm-up requests. Let s start over: TIME ACTION 11:00:07 am Router receives firm-up request for 25k from BIDS 11:00:00 am Trader A sends order to Router A to buy 100k shares 11:00:08 am Router cancels and replaces conditional orders at other venues to 75k shares 11:00:05 am Router A sends 100k conditional order to BIDS ATS 11:00:08 am Router sends firm order for 25k shares to BIDS 11:00:05 am Router A sends 100k conditional order to ITG ATS 11:00:09 am BIDS ATS receives firm order and executes trade for 25k shares 11:00:05 am Router A sends 100k conditional order to Liquidnet ATS 11:00:10 am BIDS ATS sends fill message for 25k shares At the time, there is no matching liquidity. Then a few seconds later, Router B does the exact same thing. It s important to note that most routers treat a firm-up request like a fill, decrementing the shares resting in the other pools by the quantity of the firm order being routed. TIME 11:01:00 am 11:01:05 am 11:01:05 am 11:01:05 am ACTION Trader B sends order to Router B to sell 100k shares Router B sends 100k conditional order to BIDS ATS Router B sends 100k conditional order to ITG POSIT ATS Router B sends 100k conditional order to Liquidnet H2O ATS However, because of unintentional timing issues, the ATSs responses are not received by Router A and Router B at the same time. If Router A and Router B both process firm-up requests via FIFO, it could take several attempts (and dozens of messages) on behalf of the routers to finally find each other at one of the three ATSs. While no one in this scenario is acting in bad faith, it creates unnecessary complexity. It is possible that a bad actor could attempt to take advantage of this complexity by claiming its failures to firm-up are procedural when in fact they are strategic. 7

The Concern If one of the key components of building a block trade is the optionality that enables longer resting time, then one must accept that there are legitimate reasons to not trade. However, one of the fundamental concerns regarding trading optionality is understanding the impact of when a participant declines an opportunity to trade. In the trading world, most people call this fading. A fade is an inherently frustrating experience. One of the participants took an action to try to complete a trade and the counterparty didn t respond. The frustration is a cost to the reputation of the pool, and so most venues monitor, penalize and/or police this behavior. Beyond the frustration, we also want to understand the cost of a fade. For example, it s possible that even though the fade was a frustrating experience, the stock moved in favor of the jilted party. But more importantly, we need to see if a fade consistently favors the party declining to trade. To describe the situation, we will refer to the side that firms-up as Show and the party who doesn t respond as Fade. There are several ways to analyze the situation: How much does a fade cost to Show? Is Show consistently losing performance on a fade? Did Fade make the right decision to not trade? When is a Fade not a fade? Within the two Liquidnet ATSs, there are interactions between manual traders, and algorithmic strategies and a mix of the two. The analysis below focuses on the interactions with an algorithm on at least one side. However, for a Member s own indications, we can expand the analysis to include all scenarios. First, there are five different outcomes that result from a firm-up request. The most common and obvious category is when all parties respond which accounts for 75% of all events. The other categories describe the situations where one or more parties fail to respond. We try to determine when a non-response is due to competing request for liquidity (Trade Away) or when both parties decline to trade. There are also some edge cases, such as when both parties respond but the minimum quantity is not met. % CONDITIONAL FIRM-UP RATES BY COUNTERPARTY CATEGORIES ALL MANUAL VS ALGO (39%) ALGO VS ALGO (61%) All Parties Respond 75.0% 97.7% 60.2% Only One Party Responds 15.5% 1.8% 24.6% Neither Party Responds 4.3% 0% 7.2% Trade Away 5.2%% 0% 4.2% Edge Cases 2.7%% 0.4% 1.2% Total 100.0% 100.0% 100.0% 8

During Q1 2018, approximately 39% of firm-up events occurred between a manual trader and an algo, with the remaining 61% events occurring between two algos. What stands out, of course, is that even though an algo does not know the contra type, there is a stark difference in the firm-up rates. When an algo receives a firm-up request due to a manual contra, it responds nearly every time. When two algos receives a firm-up request against each other, both parties respond 60.2% of the time. The most common explanation for the discrepancy in firm-up rates is that both algo orders were represented in multiple pools. When the orders become eligible to trade, race conditions occur among the multiple pools. The race conditions could include fills on firm orders and simultaneous firm-up requests on conditional orders. Using a relatively broad measure (an execution between 50%-115% of firm-up quantity within three minutes of firm-up request) a Trade Away accounted for 2.5% of all events. However, this number may be low since we are not including other activity that would prevent an algo from responding to a firm-up request until it receives additional fills and/or an out from the other trading venues. For example, an algo could be receiving a stream of smaller fills from an exchange and so it is unable to respond until it processes those fills and cancels the order at the exchange. In other instances, neither party responded to the firm-up requests. This could be due to a number of reasons such as a quote changes that render both orders non-marketable, competing requests from other pools, etc. This accounts for 4.3% of algo-to-algo interactions. The remaining outcome when only one party responds accounted for 24.6% of algo-to-algo interactions. This is one of the more concerning situations and while any high quality, well-monitored algo provider will have legitimate reasons for not responding, all trading venues worthy of institutional order flow need to monitor and measure these events. 9

There are many ways to measure the performance of conditional orders. Comparative algo provider firm-up rates can identify potential concerning behavior and suggestions for improved experience. The Q1 2018 top 25 conditional algo providers in shares executed (which account for two-thirds of all firm-up requests, had an aggregate firm-up rate of 80%.) Below we show the firm-up rates for each of these top 25 participants, in descending order (left to right) of number of firm-up requests. The firm-up rate fell within a range of 92% and 60% for the top 25 participants, with a wider range among smaller firms. Smaller participants tend to have more jagged results. Source: Liquidnet 10

While firm-up rate is a good starting point for monitoring behavior, it is not a panacea. We also want to measure performance, particularly in the outcome where only one party responds. The table below shows if a change in the mid-point was a negative or positive for the Fade. Neutral means the mid-point did not move. The equal distribution of the data, and the similar distribution between the two outcomes, suggests that there is no systematic information leakage immediately following the failure to respond. However, you want to make sure that there isn t any one participant systematically winning when failing to respond. Among the top 25 participants, 15 had net positive instances and 10 had net negative instances. Even among the net positive responses, none were suspiciously lopsided the distribution seems appropriately random. MID-POINT +1 SECOND FROM FIRM-UP REQUEST NEGATIVE NEUTRAL POSITIVE Trade Away 31% 40% 29% One Party Responds 29% 43% 29% AVG PRICE +5 MIN FROM FIRM-UP REQUEST NEGATIVE NEUTRAL POSITIVE Trade Away 57% 3% 40% One Party Responds 45% 3% 52% Everywhere, All the Time? Another legitimate concern regarding conditionals is that it leads to orders being over-represented in the market. For example, if there is some amount of information leakage that occurs when resting in any order book, and conditionals facilitate the shares being represented simultaneously in more than one venue, it should lead to more market impact. To see if this is true within the Liquidnet ecosystem, we measured the performance of two different types of orders during Q1 2018. The first are interactions when manual traders told us to automatically firm-up against a trading opportunity (known as Automate Negotiate ). These orders can only be created when there is a trading opportunity and the shares can only be represented in our pool. Therefore, this is a good example of a block order only being represented in a single pool. The second are Liquidnet Dark orders Liquidnet Dark is a dark aggregator algorithm that routes conditional orders into the Liquidnet ATSs, BIDS, ITG Posit, and BlockCross, among other pools. The analysis only includes Liquidnet Dark orders that were Dark Only, meaning the algo only used dark pools, as opposed to non-displayed Exchange order types. Across both interactions, we narrowed the analysis to executions greater than 5,000 shares so that we were comparing similar situations. However, a few notable differences should be pointed out. First, the average spread of the names only represented in one pool (Automate Negotiate) were wider than those that are put into Liquidnet Dark. In addition, the average trade size of orders that were only resting in one pool were larger. This feels intuitive and aligns with the data we showed earlier. The most significant trade-off between the two strategies is fill rate and the performance versus a Participation Weighted Price (PWP) of 20%. It is intuitive that the fill rate for accessing a single pool would be lower than accessing multiple pools. Orders only represented at Liquidnet (called Automate Negotiate) had a fill rate of 12.2%, while Liquidnet Dark orders had nearly double the fill rate at 27.4%. On the other hand, Automate Negotiate beat PWP 20 by 4.37 basis points, compared to Liquidnet Dark of -.60 basis point. It also makes sense that when you find a large block, you re going to outperform a benchmark based on a longer time horizon. 11

ORDER AVERAGE SPREAD CENTS PER SHARE BASIS POINTS TRADE SIZE FILL RATE VS ARRIVAL VS IMPACT ADJUSTED ARRIVAL PWP 20 Liquidnet Dark (Aggregation Strategy) Automate Negotiate (represented only in Liquidnet ) 4.78 9.93 16,218 27.4%.28 6.50 -.60 6.20 13.11 36,595 12.2% -.02 11.53 4.37 Based on the data set we looked at, conditionals help bridge the gap between a manual block crossing interaction and traditional algorithmic routing practices. However, the diminished optionality when committing shares to an algorithm leads to smaller trade sizes. When a trader can exercise patience, representation in a single dark pool will lead to better performance. However, when a trader needs to complete, a dark aggregation strategy should achieve a higher benchmark and still beat PWP 20 and Impact Adjusted Arrival Price benchmarks. The increased support and usage of conditionals among dark pools and dark aggregators has led to lower firm-up rates. Part of this is a necessary side effect of conditionals, however, in our opinion the use of conditionals should be limited to venues that are focused on block activity. We believe that traders should be aware of the conditional venues that each of its algo providers uses. In addition, traders should insist that other ATSs and algo providers measure and disclose statistics surrounding conditionals such as firm-up rates and performance on fades. While our study shows that fades were an unfortunate experience and not a statistical penalty during Q1 2018, this may not be true across all venues and could change over time. The use of conditionals should be predicated on sourcing block liquidity; not a blunt tool for solving all opportunity costs. Finally, the best protection against the real threat of gaming the firm-up process is to measure the performance of failures to respond to firm-up requests. This should become standard. Traders should also understand how each of their algo providers (and other conditional providers) handle the initial routing of the conditional order and the response to firm-up requests. While there are arguments for spraying versus staggering, it is our opinion that more sophisticated and experienced conditional routers are using some combination of staggering and batching. The use of conditionals should be a data-driven process, same as any venue-based routing decision. The use of conditionals has a rightful place in the execution ecosystem. Conditionals enable a flexibility within a trader s workflow that can help maximize block liquidity. However, conditionals are a poor substitute for traditional electronic block negotiation. When, where, and how to use conditionals should be a point of discussion, not an assumption. 12

Adam Sussman is Global Head of Market Structure. EXECUTION & QUANTITATIVE SERVICES (EQS) Liquidnet s EQS team aims to maximize your execution quality and productivity with solutions that utilize a full suite of algorithmic, analytic, and quantitative products and services. EQS also provides specialized institutional expertise and insight designed to help you achieve your execution goals and minimize your order s footprint in the market. We solve challenges, tailor solutions, help you seize on trading opportunities, and can execute on your behalf leveraging our suite of execution tools and array of liquidity sources. Through it all, we ensure that you are always in control of your execution experience. For more information and insight from Liquidnet s Execution & Quantitative Services (EQS) team, contact us: EQS-US@liquidnet.com +1-646-674-2274 Liquidnet.com 13

2018 Liquidnet Holdings, Inc. and its subsidiaries. Liquidnet Europe Limited is authorized and regulated by the Financial Conduct Authority in the UK, is a member of the London Stock Exchange and a remote member of the Warsaw Stock Exchange and SIX Swiss Exchange and is licensed by the Financial Sector Conduct Authority in South Africa. Liquidnet Canada Inc. Is a member of the Investment Industry Regulatory Organization of Canada and a member of the Canadian Investor Protection Fund. Liquidnet Asia Limited is regulated by the Hong Kong Securities and Futures Commission for Type 1 and Type 7 regulated activities and is regulated by the Monetary Authority of Singapore as a Recognized Market Operator. Liquidnet Japan Inc. is regulated by the Financial Services Agency of Japan and is a member of JSDA/JIPF. Liquidnet Australia Pty Ltd. is registered with the Australian Securities and Investment Commission as an Australian Financial Services Licensee, AFSL number 312525, and is registered with the New Zealand Financial Markets Authority as a Financial Service Provider, FSP number FSP3781. 6/18