Contents Crib Sheet Physics says the SIPs can t compete How slow is the SIP? The SIP is 99.9% identical to direct feeds SIP speed doesn t affect most trades For questions or further information on this report, please contact Phil Mackintosh at pmackint@kcg.com or 201.963.2724 March 2016 The Need for Speed IV: How Important is the SIP? by Phil Mackintosh and Ka Wo Chen Crib Sheet We talked in our Speed III report about why the SIP will always be slower, regardless of what investments we make to speed it up. This is because SIP messages have farther to travel, which takes more time. But just how much slower, and how different, is the SIP? Typically, the SIP is less than 0.8ms slower than direct feeds. Data show that the SIP and direct feeds are identical almost all of the time, indicating that quote flicker is actually rare. Latency really only occurs when quotes change, otherwise the SIP and direct feeds are identical. For institutional investors, a slow SIP matters less than you d think. While passive posts are most likely at the NBBO regardless of SIP speed or usage, taking liquidity relies more on fast routing. The bigger problem is the value (and cost) of direct feeds which is ultimately passed on to investors. 1
Physics says the SIPs can t compete In our recent report we discussed how data travels around the different venues in the market. We showed that: All data is delayed Even at the speed of light, transmitting data and trades takes time, resulting in latency between venues. This means all data is delayed even direct feeds and fills. Speed depends on processing too The SIP also has more processing to do as it needs to compile the NBBO and compute reference prices like LULD triggers as well as record TRF trades. Direct feeds need processing too but if speed is important, you can just subscribe to minimal price-change data. Conversely, if your trading strategy requires full depth of book by venue, fills, and individual cancels and amendments, you may prefer to receive more information than the SIP provides. A centralized SIP won t stop race conditions It s not just about the distance your data travels or the amount of data you consume. The type of hardware and software you use, or even the skill of your programmers affects how quickly you can process data. Slower or centralized data does not stop race conditions. 2
Comparing our direct feeds to the SIP Because of these facts and the fact that RegNMS prescribes the SIP as the official NBBO, there are concerns about how slow the SIP is, and how different direct feeds are from SIP prices. In this report, we compare timestamps on synthetic NBBOs from one of our trade desks to timestamps from SIP NBBOs arriving at the same data center. The synthetic NBBOs are compiled from direct feeds from nine exchanges. We compute: Latency: SIP message arrival timestamps minus direct feeds message Match Ratio: The percentage of time the SIP NBBO and synthetic arrival timestamps, on average, per stock. NBBO are in agreement. Direct-feed BBO isn t simple One of the problems with direct feeds is the fact that updates are sent on a change-by-change basis, as we discussed in our Speed III report. This means the receiver of direct feeds needs to create a synthetic NBBO. If just one of those updates is dropped or even arrives out of order, the BBO that is calculated may be wrong. As a consequence, some of the differences between the SIP and direct feeds might also be caused by errors in synthetic BBO calculation. What timestamps are important? Note that we ignore SIP timestamps in favor of our own data arrival timestamps as we can t act on new information until we actually see it ourselves. But because we are using arrival timestamps: The location of our databases in relation to each of the SIPs also affects results, and will differ from direct feed studies based elsewhere, and will be impacted by what venue trades and where the primary listing is (more on this in our Speed III report). The transmission time from the SIP, not just the difference in processing times, is added to the latency we observe. 3
How slow is the SIP? There are a lot of conspiracy theories about SIP performance and the benefits of direct feeds. What do the statistics really show? The SIP isn t that different from direct feeds Latency of the SIP NBBO only occurs when the market ticks. At all other times the SIP NBBO and the direct feed (synthetic) NBBO are the same ( Exhibit 3). Quote changes are typically event driven, for example: A trade takes all the passive liquidity, widening the spread or setting a new level bid or offer. A trader steps inside and narrows a 2-cent (or wider) spread. Passive liquidity completely fades at a price level, perhaps triggered by futures re-valuations, book depth signaling, or trades occurring at other venues. 4
High quote-to-trade doesn t mean inefficient prices The NBBO fading in the absence of (lit) trades is a common complaint about our electronic markets. It is a phenomenon often measured using a quote-totrade ratio, which in turn is sometimes the basis of proposed transaction taxes. The data show that most stocks have relatively low quote-to-trade ratios. However, a high quote-to-trade ratio often reflects a very efficient and competitive market. In general, it reflects how sensitive bids and offers are to valuation, and therefore how finely tuned price discovery has become. An example of this can seen in ETFs (the light green circles in Exhibit 4 show that ETFs often have much higher quote-to-trade counts, especially illiquid ETFs). 5
Ironically, despite high quote-to-trade ratios, ETFs are one of the biggest beneficiaries of our automated and electronic markets. Automation has brought cost efficiencies and certainty to hedging. As a result, many ETFs trade with spreads well inside the underlying spread, often less than 1bp wide. The value of the ETF, however, can be very accurately calculated from underlying securities prices (inav). Consequently, the ETF can (and should) reprice to the correct price without a trade taking place. High quote-to-trade ratios can also be seen in the SEC s MIDAS data. In addition to ETFs, exchanges with low market share also see high quote-totrade ratios. Yet market makers maintain fair quotes at those venues even though trades rarely occur. The SIP is approximately 780µs slower If we look at just the times when the SIP differs from direct feeds, and compute how long the difference lasts, we see that: Latency varies for each stock and each trade. This is caused by a Average latency across all stocks is 780µs. This is consistent with a Latency seems to increase as trade activity increases, which number of things from the venue that trades happen on, to the primary listing of the stock, to the network utilization at the time of the trade. recent study that estimated the SIP was 500µs slower than direct feeds, although that study was using exchange dissemination timestamps, which are different from our data arrival computations. makes sense as the SIP has more processing to do (Exhibit 5). 6
The SIP is 99.9% identical to direct feeds For all the industry focus on SIP latency, including millions of dollars being spent to speed up the SIP, how much does it really matter? What we find might surprise you. How much of the day are SIPs different? If we multiply the average latency for each ticker by the number of quote changes that ticker has each day, we have an estimate for how long the SIP is likely to be different to the direct feeds. We estimate that the SIP is identical to the direct feeds almost all the time. In fact: Almost 5,000 NMS stocks match at least 99.99% of the day (direct 85% of stocks have SIP prices that match at least 99.97% of the day 97% of stocks have NBBOs that match at least 99.90% of the day feeds differed for less than 3 seconds each day). (direct feeds differed for 7 seconds each day). (direct feeds differ 23 seconds each day). 7
This would indicate that widespread quote flickering, often cited in the media, is just not happening. It also means that posting or processing orders using the SIP results in no difference in price almost all the time. Trading a lot makes direct feeds more different More liquid stocks trade more often, and usually through more price levels, as they tend to have higher prices. We also see in Exhibit 5 that their latency tends to be higher. These factors should combine to make the direct feeds more important as liquidity increases. Not surprisingly, we see that large cap stocks dominate the results below the 99.98% matched level (Exhibit 6). Looking at each stock separately ( Exhibit 7), we see a fairly linear relationship between trade activity and matched time. 8
SIP speed doesn t affect most trades Don t get us wrong, being fast is important for all traders, no matter their time horizon. But we ve found this requires use of direct feeds and constant investment in trading hardware and software. Posting faster and smarter is important In our report, Speed I: Speed is Important for Spread Capture, we found that fast routing was important to get good queue priority for passive orders. Taking liquidity on the NBBO is easy if you re fast In our next report, Speed II: Beating Fade is Easy if You're Fast, we showed that taking liquidity and avoiding fade was easy (Exhibit 8), but this requires direct feeds and dedicated links to exchanges. 9
The SIP can never be as fast as Direct Feeds In our last report, Speed III: Physics and Trigonometry, we showed that even direct feeds are delayed, but because SIP messages have further to travel, and updates sometimes come from remote exchanges, the SIP will always be slower, and sometimes out of sync. The SIP is less different than you think Finally, in this report, we show that quotes don t flicker nearly as much as people seem to think, and that the SIP is the same as direct feeds almost all of the time. This makes the SIP a perfectly good tool for your Bloomberg or TCA especially if you use mark-outs of more than 1ms. But speeding up the SIP will not increase its accuracy. Solve for data, not the SIP Direct feeds are vital to knowing where the market was most recently and to taking liquidity. As we said before, a better data solution would be for aggregators to colocate at each exchange, and compete based on pricing with different levels of speed to suit different traders all of which could still meet the SEC definition of NBBO (currently up to 1 second of latency) for best-ex computations. A larger problem is the value of direct feed data. Exchanges are one of the few areas of finance seeing increasing fees despite decreasing market share. 10
Philosophically, it s all our data. If the NBBO is going to be protected, then perhaps the data to compute it should be free. This would also shift the incentive to exchanges from selling data to competing on transactional efficiency. The change might help reduce fragmentation as well as lower transaction costs. Disclaimer This material is not research and is not intended as such. It has been produced by personnel employed within the Trading Strategy Department and/or by Sales and Trading personnel of KCG Americas LLC ( KCGA ). This material is intended only for clients of KCG Holdings, Inc. and its subsidiaries (collectively KCG ). It is not intended for further dissemination in its present form or any other and may not be further disseminated beyond the initial recipient without express consent from KCG. This material is not intended to constitute an offer, or solicitation of an offer, to purchase or sell any security or financial instruments or to participate in any investment strategy. The securities and strategies referenced in this material are not intended as recommendations and may not be suitable for you. KCGA believes that the information contained herein is accurate but does not warrant that it is accurate, complete, or up to date. The views expressed herein are based on assumptions. Any of those assumptions may be incorrect. KCG does not guarantee the performance or success of any opinion or idea expressed in the information. KCG assumes no liability to anyone as a result of the use of this material. This material provides the observations and views of the persons who prepared the material. These observations and views may be different from, or inconsistent with, the views of KCG or other Departments or persons within KCG. The observations and views expressed in this material may change at any time without notice. KCGA makes markets and trades for its own account and may hold positions in any of the securities of the issuers referred to herein or in related financial instruments. The persons who prepared this material may also advise our trading desks. In the US, products and services are offered by KCGA, member FINRA/SIPC. In Europe, products and services are offered by KCG Europe Limited ( KCGEL ) which is authorized and regulated by the Financial Conduct Authority. KCGA and KCGEL are affiliates and are indirect subsidiaries of KCG. For additional information about KCG (NYSE: Euronext: KCG) please visit www.kcg.com 2016 KCG Holdings, Inc. All rights reserved. 11