Tick Pilot Update. 2

Transcription

1 AUTHOR Colleen Ruane Managing Director Analytical Products and Research Phil Pearson Director ITG Algorithms CONTACT APAC CAN EMEA US Tick Update INTRODUCTION The bell has rung, the ball has dropped and 217 is finally here. As 216 closed with a surprising rally spurred by small-cap stocks following the presidential election, it is timely to take a closer look at the biggest ongoing market structure experiment in recent memory, the Tick Size Program. Beginning with a one-month phase-in during October, the pilot has now been running for three months of the proposed 24-month cycle. As a reminder, the two primary goals for the pilot are to help facilitate IPOs and research activity for smallcap companies and to reduce trading costs and volatility for small-cap stocks. It is far too early to evaluate the first objective, but with two-plus months of data, trading patterns have begun to emerge. For a refresher on the tick pilot s rules and test groups, click here. 1 Rather than focus on marketwide statistics gathered from public data, we wanted to answer the most important question of the tick pilot: How has it affected institutional trading costs? This is the true test. Despite not being the initial objective of the JOBS Act, transaction costs will be a driving factor in the final determination of success or failure of the pilot program. Is it easier for investors to trade previously hard-to-trade stocks? As an institutional broker and broker-neutral trading analytics provider with a unique and sizable Global Peer database we are well positioned to address this question. Many industry pundits predicted that contrary to the pilot s goals, trading costs would actually increase with the mandate for wider spreads. Since the pilot began, the subjective consensus has been that it has made stocks moderately more expensive to trade. A cottage industry has grown out of giving basic marketwide statistics on the pilot. Many have noted that spreads are larger, NBBO sizes are huge, and the use of dark pools and inverted venues has increased significantly. 2 While meaningful, these results are ancillary to the question of trading costs. In our effort to calculate tangible results from the early data, we chose to compare November and December 216 trading by each pilot test group against the control group. 3 Additionally, we contrast against August and September 216 as the pre-pilot period. We look to our broker-neutral Global Peer database maintained by ITG Analytics for information on trading costs and trends in algorithm behavior, and we highlight a few trends specific to ITG s electronic brokerage products. To summarize, we find that for tick pilot stocks: Despite increased NBBO sizes, trading costs are higher, primarily due to the increase in spread sizes. Brokers are increasing the use of dark pools and inverted exchanges in an effort to capture spread. Parent order Implementation shortfall (IS) costs are 5% higher than pre-pilot costs compared with the control group. IS costs for arrival-price algorithms are up more than 4 basis points. Interval VWAP slippage for VWAP algorithms is higher by 2 basis points While the data from October 216 is interesting, the October phase-in meant inconsistent sizes in the control and test groups through the end of the month. Notably, almost all of the Test Group 3 stocks were part of the control group for the majority of October; thus, general market patterns can be mistaken for causality.

2 2 MARKET STATISTICS Recent ITG research has shown that spreads are almost twice as large since the pilot began. 4 Top-of-book depth, however, has increased even more by about 6x. These offsetting factors give some credence to the theory that liquidity is becoming more abundant, but at a price. It should be noted that the 6x increase in quoted size is not a fair comparison. Much of that liquidity would be found a few cents through the spread in a penny-spread regime. It is nearly impossible to quantify this because not every market participant places displayed orders outside the NBBO. Interestingly, both spread size and quoted depth have continued to increase consistently since the middle of October as market participants have begun adapting how they approach these stocks. Despite the smallcap rally and increased volume in November, spread sizes still increased. It seems that this is potentially rewarding market makers and other liquidity providers. This is a doubleedged sword. Another subtle, yet significant, theme is a significant increase in the proportion of trading on inverted exchanges. This seems to be in direct response to the increased queue sizes. Market makers and institutional traders alike have directed more trading to the inverted exchanges to compete for passive executions. Said differently, traders are essentially reducing their effective (realized) spreads to a measure less than the actual minimum increment of $.5, stepping in front of a large queue. INVERTED EXCHANGES AND DARK POOL MARKET SHARE 45% Dark up average of 1% 4% 35% 3% 25% 2% 15% 1% 5% Inverted up average of 93% Down 23% because of trade-at rule Midpoint up average of 5% Sharp increase (96%) in midpoint dark, which is permitted % Inverted Inverted Dark Dark Dark Mid Dark Mid Also seen in the Inverted Exchanges and Dark Pool Market Share chart, total dark pool volume has also increased for test Group 1 and Group 2, but has decreased for Group 3. The latter is because of the trade-at restriction constraining dark trading on the bid/offer. When considering midpoint dark trading exclusively, G3 has seen an increase of more than 1%, while the other test groups are up a smaller amount. 4

3 3 BLOCK MARKET SHARE 5% % One development, largely considered positive, is the increase in blocks as a percentage of trading volume. For simplicity s sake, we are considering a block to be any trade of more than 5, shares. 5 It should be noted that the control group also saw an increase so this isn t as stark as it seems, but G3 specifically had a 24% increase in block trade market share (closer to 15% when adjusted for the growth in the control group). Due to the restrictions on non-midpoint dark trading, it seems some brokers have employed use of the block exemption for G3, accounting for this increase. TRADING COST AND ALGORITHM PERFORMANCE An increase in spread cost for the pilot groups is a trivial result, as the program mandated an increase in spreads for many of the stocks. The question is whether enough additional liquidity showed up to mitigate the impact of the wider spreads and reduce costs (or at least keep them in line) for investors. For a first look at the real impact of the pilot, we ve evaluated broker-neutral transaction costs from two perspectives: the parent order and the broker placement. Parent order information is sourced from ITG Analytics Global Peer database. The sample relies on preliminary universes for Q3 and Q4 data, but includes trading from more than 1 buy-side institutions for both the pre-pilot and pilot periods. Implementation shortfall (IS) cost is measured from the point of release to the trading desk. For the purpose of this analysis, the pre-pilot period covers August and September 216, while the pilot period includes November and December 216. October 216 is excluded from the analysis due to the staggered phase-in schedule. 6 While we expect more robust patterns in transaction cost to emerge over time and with more data, initial results seem intuitive. The Peer Average IS Cost chart shows average IS costs (equal-weighted across observations) for tick pilot stocks based on parent orders from our Global Peer database. 7 IS costs for the pilot period are higher across all stocks (including the control), but each test group individually saw a larger increase than the control. Average cost for the test groups combined is approximately 5% higher than the increase in control group cost. 5 Block market share is total shares executed in blocks (greater than 5, shares) during continuous trading divided by total shares traded in each symbol. 6 A quick note on methodology: We typically rely on value-weighted cost calculations when reporting peer statistics, but valueweighted numbers can be skewed by larger orders. With only two months worth of data, we do not have sufficient information to comment on larger orders, and have thus chosen to use an equally weighted average to mitigate their (noisy) influence on the sample averages. 7 Peer costs are calculated as of 1/9/17.

4 4 PEER AVERAGE IS COST Costs up 91% for pilot vs. 43% for control -5-1 (11.29) (9.5) (8.85) (1.52) (16.15) (18.6) (17.66) (17.95) It is interesting to note how consistent the higher costs were in the test groups. There did not appear to be too much difference between the three test groups from the parent order level. We also looked at the impact the pilot is having on broker performance at the strategy level, focusing primarily on algorithms with explicit performance benchmarks. All orders included in this analysis are client-directed placements sent to an algorithmic destination; they include all algo classifications as well as smart order routers. 8 The sample includes information from more than 7 buy-side institutions trading with at least three brokers over the course of the analysis period. Altogether, 39 brokers are represented, along with 39 distinct algos. For data quality purposes, we omitted all orders without sufficient order-level information. 9 We have categorized algos based on the strategy s objective and most appropriate benchmark. For the purposes of this discussion, we will focus mainly on VWAP algos as well as arrival-based algorithms, including implementation shortfall and liquidity-seeking or opportunistic strategies. As above, the pre-pilot period covers August and September 216, while the pilot period includes November and December 216. VWAP algos are among the easiest to measure as the performance and strategies are very similar across brokers. As seen in the VWAP Algo Performance chart, there was a substantial increase in VWAP slippage for the test groups. Interval VWAP costs increased for all three test groups, by a total of about 13% versus an 11% decrease for the control group, a difference of about 2 basis points. Test Group 1 was the biggest contributor. The numbers are a little higher for VWAP slippage than we are used to seeing, partially due to the equal-weighted averaging, which we used to maintain consistency. (Equal-weighted seems to have a wider distribution for VWAP than a value-weighted method.). Small VWAP orders seem to have a larger variance in performance than algos with other benchmarks. 8 Broker-neutral placement-level data is sourced from ITG s FIX Network, as of 1/9/17. 9 This includes order missing: last market information, orders with extreme implementation shortfall costs, and orders for which we could not calculate relevant metrics, including spreads, participation rates or benchmark information. Orders with passive limits and instantaneous executions were also excluded from the analysis.

5 5 Unexpectedly, the breakdown of passive/mid/ aggressive trading did not appear to change much for this group. We attribute almost all of the increase in cost to wider spreads, as can be seen in the Trade Weighted Spread for VWAP Orders chart. Many broker VWAP algos rely on passive and midpoint orders earning the spread to compete against the benchmark. With wider spreads, the midpoint and aggressive orders appear to be more costly, causing unfavorable performance. VWAP ALGO PERFORMANCE TRADE WEIGHTED SPREAD FOR VWAP ORDERS The other primary area of interest was arrivalbased algorithms, including implementation shortfall, opportunistic and liquidity-seeking strategies. Average costs for arrival-based strategies, shown in the Arrival-Based Algo IS Cost chart, increased during the pilot period by an average of 4, although the control group also saw an increase. Average cost across test groups increased by 114% over pre-pilot costs, while the control group costs increased 35% over pre-pilot costs. G1 again saw the biggest increase but the differences were moderate. Determining the cause of the increase in cost is a little more difficult for two reasons. First, compared with scheduled algos (VWAP, etc.), arrival-based strategies vary significantly from broker to broker with dissimilar strategies employed for different purposes. Second, market conditions and outliers have a bigger influence on IS cost. Nevertheless, passive trading decreased in arrival-based algos by about 22%, which helped to account for some of the performance difference. Again, the larger realized spreads were also a factor. Despite the increased liquidity available for aggressive trading, it still appears that the larger spreads have a greater impact on trading costs.

6 6 ARRIVAL-BASED ALGO IS COST TRADE WEIGHTED SPREAD FOR ARRIVAL-BASED ORDERS TRADING BEHAVIOR We see some interesting trends across all broker algos in pilot stocks. In addition to the increased inverted venue and dark usage that we highlighted earlier, trading strategies appear to have been systemically altered. One of the major differences we noticed was a movement away from passive trading to more midpoint (and aggressive) trading. Across all algo categories, passive trading has been down during the pilot period. 1 Passive trading was lower in all test groups, with the largest decrease, 3%, seen in G3. Algorithmic midpoint trading is up in all test groups as well as the control. Compared with the control group, G1 and G2 are essentially flat compared with the prepilot period. Conversely, G3 saw a big uptick (23%) in midpoint trading, likely a result of the structural change that trade-at mandates. With substantial constraints on dark pool trading at the bid or offer, all dark trading shifted to midpoint as a way to improve costs and find natural liquidity. Furthermore, G3 algo trading saw the biggest decrease in passive fill rate, so broker algos had to rely more on mid trading to avoid having to pay the spread. 1 For passive and midpoint trading, we did not include DMA or smart order router destinations.

7 7 ALGORITHM PASSIVE AND MID USAGE 5% 45% 4% 35% 3% 25% 2% 15% 1% 5% % C G1/G2 G3 Passive Passive Mid Mid Another major change we noticed was in how clients used algos. Strategy selection did not change much, but the average order duration did. The numbers were fairly consistently higher across most algo strategies, but arrival-price algorithms had the most notable increase more than double the pre-pilot duration. Orders are taking longer to finish, perhaps because brokers are having more difficulty finding liquidity. The control group also saw an increase in duration, so this could have something to do with market conditions, but it appears to be exacerbated by the tick pilot. AVERAGE ALGO ORDER DURATION (IN MINUTES) C G1/G2 G3

8 8 ITG BROKERAGE TRENDS As a broker, we have noticed many of the same trends as the broader market. Specifically, we have seen an uptick in our use of inverted venues and increased trading costs for tick pilot names. All ITG algos use a proprietary order routing system called SLimit for passive trading. SLimit uses historical and real-time data on average queue length and consumption rate to optimally allocate shares among markets. Recognizing a volume shift to the inverted and flat-priced exchanges in tick pilot names, we saw a substantial increase in market share for the inverted venues. G1 and G2 increased by almost 25% but G3 increased much more, by 64.4%, versus the pre-pilot period. SLIMIT INCREASE IN INVERTED VENUES FOR TICK PILOT STOCKS 7% 6% 64.4% 5% 4% 3% 2% 23.2% 24.3% 1% % 7.1% Despite the uptick in inverted usage, SLimit s participation rate (as measured by executions divided by same-sided passive fills) decreased 18%-23% across the test groups, belying the difficulty in competing for passive executions in pilot stocks. As further proof of that, we look at ITG s VWAP algo in the VWAP Trading Within the Spread chart. This shows the breakdown of passive, midpoint and aggressive trading in the various tick pilot groupings. Traditionally, ITG s VWAP has focused on maximizing passive executions to minimize slippage versus the VWAP benchmark, executing more than 6% passively. As previously noted, passive trading has decreased for many algorithms across the Street and ITG is not immune to this. Passive trading during the pilot has become far more competitive, making it more difficult to earn the full spread. What is interesting here is that almost all of the shares lost on the passive side have moved to midpoint. Rather than earning the whole spread, we have been forced to pay half of the spread. However, and most important, very little trading has moved to the opposite (aggressive) side.

9 9 VWAP TRADING WITHIN THE SPREAD 1% 8% 6% 4% 2% % C C G1 G1 G2 G2 G3 G3 pre pilot pilot pre pilot pilot pre pilot pilot pre pilot pilot Passive Mid Active CONCLUSIONS The hype of the presidential election and the Trump market rally that followed took much of the attention off the tick pilot. In the context of massive political upheaval and a major infrastructure project in draining the swamp, a regulatory experiment for smallcap stocks may seem relatively insignificant. Furthermore, the pilot program has run relatively smoothly, with no major issues to date despite a large technology lift across the industry. However, early data suggests quite clearly that trading costs are higher across the board. With larger spreads, we speculate market makers have been the beneficiary of higher institutional costs. All things considered, the tick pilot has had very minimal impact outside of the cost increases. Market pundits had speculated there might be an increase in volatility and a decrease in trading volume, but neither has seen a material adjustment. The lack of change in intraday volatility is significant considering how much stress was put on the market in the unpredictable days following the election. Before the pilot began, we speculated that closing auction sizes may be larger due to higher trading costs throughout the day, but that has not come to fruition; they are largely unchanged. Volume profiles have also remained consistent. The only real difference in trading appears to be within the spread: which exchanges are being used, the amount of dark pool activity, how to avoid crossing the spread excessively. Almost all of the increase in trading cost is directly attributable to the increase in spread sizes. By the measure of improving markets for institutions, the pilot appears to be unsuccessful but, again, it depends how that is quantified. Nonetheless, it is the early days of this experiment and we are going to continue evaluating the outcome to determine the pilot s true costs. 217 Investment Technology Group, Inc. All rights reserved. Not to be reproduced or retransmitted without permission These materials are for informational purposes only, and are not intended to be used for trading or investment purposes or as an offer to sell or the solicitation of an offer to buy any security or financial product. The information contained herein has been taken from trade and statistical services and other sources we deem reliable but we do not represent that such information is accurate or complete and it should not be relied upon as such. No guarantee or warranty is made as to the reasonableness of the assumptions or the accuracy of the models or market data used by ITG or the actual results that may be achieved. These materials do not provide any form of advice (investment, tax or legal). ITG Inc. is not a registered investment adviser and does not provide investment advice or recommendations to buy or sell securities, to hire any investment adviser or to pursue any investment or trading strategy. The positions taken in this document reflect the judgment of the individual author(s) and are not necessarily those of ITG. Global Peer and SLimit are trademarks of the Investment Technology Group, Inc. companies. Broker-dealer products and services are offered by: in the U.S., ITG Inc., member FINRA, SIPC; in Canada, ITG Canada Corp., member Canadian Investor Protection Fund ( CIPF ) and Investment Industry Regulatory Organization of Canada ( IIROC ); in Europe, Investment Technology Group Limited, registered in Ireland No ( ITGL ) (the registered office of ITGL is Block A, Georges Quay, Dublin 2, Ireland). ITGL is authorized and regulated by the Central Bank of Ireland; in Asia, ITG Hong Kong Limited (SFC License No. AHD81), ITG Singapore Pte Limited (CMS License No ), and ITG Australia Limited (AFS License No ). All of the above entities are subsidiaries of Investment Technology Group, Inc. MATCHNow is a product offering of TriAct Canada Marketplace LP ( TriAct ), member CIPF and IIROC. TriAct is a wholly owned subsidiary of ITG Canada Corp.