The Gateway to Algorithmic and Automated Trading

Liquidity sensing and smart order routing

Published in Automated Trader Magazine Issue 05 April 2007

Regulation in both the US and Europe is having a significant impact on smart order routing in relation to best execution and venue selection. An important part of the routing process revolves around the sensing of hidden liquidity.

AT talks to five major sellside providers about their take on the technology and the challenges for liquidity sensing and smart order routing


  • Richard Balarkas, Managing Director and Head of AES Sales at Credit Suisse
  • Toby Bayliss, European Head of Self-Directed Sales, Citi
  • Brad Hunt, Managing Director Equities Electronic Trading, Goldman Sachs
  • Chris Jackson, Director, Execution Sales, Merrill Lynch International
  • Owain Self, Executive Director, European Algorithmic Trading, UBS
  • Peter Sheridan, Head of European Algorithmic Trading, Goldman Sachs
  • Jarrod Yuster, Head of Global Portfolio and Automated Trading, Merrill Lynch

If smart order routing is to be truly smart, is there an implicit requirement Post RegNMS/MiFID for the sellside provider to be connected to every possible execution venue?

Balarkas: While regulation such as MiFID certainly raises the bar in this respect, it doesn't make this a specific requirement. However, there is still the need to remain competitive, so one therefore cannot afford to ignore any venue that has reasonable liquidity.

Bayliss: I think the pressure to connect will come more from competitive considerations than regulation. Ultimately, the decision whether or not to connect to individual pools of liquidity will be determined by their current or forecast volumes. The proliferation of venues is also likely to lead to partnerships being created between venues sharing liquidity information.

Hunt: The answer in Europe is not necessarily. MiFID states that asset managers as well as broker dealers (investment firms generally) must have a "best execution" policy. If in that policy they state that they will access each and every available venue, then that is what they are obliged to do. MiFID asks investment firms to consider criteria that they would use in order to make a routing decision. Within those criteria, price is obviously the dominant factor, but speed and certainty of execution also have to be considered. Regarding certainty of execution, smart routers will need to have a history of the likelihood of being hit and the order of magnitude of the liquidity available at the venue.

Finally, cost is another significant consideration in Europe, given the relatively inefficient clearing infrastructure and potentially high fixed costs associated with dealing on alternative venues.

Self: I think one has to connect to liquidity venues if they are seen as being beneficial to providing best execution. There may be some venues that appear to exhibit sufficient liquidity, but because of the nature and activity of the other participants on those venues connecting might not actually improve the execution quality for our clients. Therefore there is a need to monitor the various execution venues on an ongoing basis and only connect to those that offer the "right" sort of liquidity.

Yuster: For RegNMS, brokers have to be directly connected to all eleven Self-Regulatory Organizations (SROs) or allow those SROs to re-route orders on their behalf (which is less than ideal in terms of best execution, for reasons of latency and potential order concentration at a single venue). However, this is still incomplete connectivity, because it omits ECNs and dark pools.

In order to achieve best execution, a broker needs to be connected to all these venues and be processing all the market data from all their books. In this last respect, in order to optimise performance, we believe it is important to process the full book of data (not just top of book) directly from source and not via a third party provider.

In Europe, the regulatory environment is moving towards a principles-based interpretation of best execution. It's our understanding that the buyside will have to prove that they have the infrastructure and processes to achieve best execution. Over the long term, it will become a standard requirement that smart routers access all available public venues, private venues will continue to be subject to individual client/ broker agreements.

Required to connect everywhere?

Are "compliance snapshots" (the capture of contemporaneous trading and quote information to substantiate order routing decisions) likely to prove a significant hurdle for the sellside?

Balarkas: Such snapshots aren't a particular obstacle to larger sellside participants, who will be collecting that data anyway, but nor are they of much assistance in guaranteeing you got the best result. The problem with the concept of a compliance snapshot is that it is only applicable to a market/situation where one is supposed to be chasing the best price, regardless of all other circumstances.

In practice this is not ideal because one often finds that what appears to be the best price isn't actually available if one tries to trade against it. A relatively common problem in this respect is the ghost price that appears briefly but disappears by the time an order arrives. Over time one becomes aware of the venues and situations where this is likely to happen. Therefore truly advanced best execution actually involves deploying that expertise so as not to route orders there and miss real opportunities, even if the price displayed currently appears to be the best.

Bayliss: We don't think this will represent a problem. Apart from the fact that larger sellside firms will be collecting this data anyway, existing market data vendors have already started to make it clear that they will be stepping into this space. We expect product offerings from these vendors to include the ability to replicate the theoretical combined order book and incorporate all information available to the trader at the time of execution (for example whether or not a quote is protected).

Hunt: For some smaller sell side firms this could potentially prove a rather expensive endeavour. Therefore I suspect that some of them will try to resolve this by using vendor-based solutions.

In our case we don't see this as a problem, as we have reconfigured our systems so that they are scanning additional data feeds for alternative liquidity resources. We also use time-stamping to ensure that the order management systems are synchronised and that we have a comprehensive audit trail. In addition, the post trade reports we send to all customers can include this granular level of detail if they require it.

Jackson: Smart routing technology, full book market data as well as trade and quote databases are all expensive to build, maintain and require significant expertise. Therefore probably only the top ten or fifteen brokers will be able to do this and amortise the cost over global businesses across their order flow, which will likely make compliance snapshots a challenge for smaller sellside participants.

It seems likely that a consolidated tape of real time publicly displayed quotes will be a standard requirement in the market. However, the question whether historical and trade reporting information will be available in a consolidated form is less certain, the Merrill Lynch supported project Boat is a key initiative to address this.

Self: This could potentially be a problem for smaller sellside participants that are only connected to one or two venues and don't have the capability of aggregating all the market data in one place. The challenge then is how to obtain a sufficiently comprehensive view of the market. If third-party data vendors can provide this, then all well and good, but if they cannot then the challenge is significant.

From our perspective we don't see this as a problem, as we are collecting all this data anyway. However, there is a need to make an evaluation whenever a new trading venue emerges. As a result, we might well be collecting compliance snapshot data for a particular venue before we were actually trading live on it. I think it is prudent to be collecting data from a market even before you have an actual trading link to it, as this obviously helps in assessing the quality of the venue for possible order execution.

Compliance snapshots - an overwhelming task?

Are the search costs associated with smart order routing now so small as to have negligible effect on price priority calculations?

Balarkas: The cost of searching is now pretty minimal. The only substantive cost associated with searching (as opposed to simply trading on the first available venue) is the opportunity cost. There is also obviously a certain amount of latency involved in looking at more than one venue, but if your systems are fast enough that is very small.

There has been a fair amount of academic research in this area, which has been largely inconclusive. However, from a practical perspective we don't regard search costs as being problematic.

Bayliss: Search costs are not insignificant and so any disparity between venues will continue to be an input into price priority calculations. However, while the initial set up costs for connecting to a trading venue are considerable, if the order flow to that venue is substantial then the connection costs will be quickly amortised over time.

Self: It is expensive to connect to multiple venues, but the need is there. Particularly in the US where there are a lot of crossing networks it is very expensive, but because of regulatory pressure it is just one of the costs of trading. If there are trading venues that offer sufficient liquidity and they improve best execution quality, then you have to connect.

You obviously have to factor in the individual costs of connecting to a venue (which may vary widely) when calculating search costs. This could be a self-fulfilling prophecy in that the costs of connecting to and trading on a venue could have an impact over the longer term on the liquidity available there.

Yuster: The "search cost" associated with smart order routing may be viewed as the latency involved in re-routing part of an order from one destination to another as well as the loss of priority in the queue on an individual order book when doing so. In many cases, as liquidity fragments across more venues, these factors are outweighed by the opportunity of achieving a better price and/or to exhaust hidden reserves on another venue.

However, certain stocks that trade at fewer price points during the day will have longer queue times, and for these stocks retaining an order's position in the order book is essential. In these cases, the cost of removing an order from one venue and moving to another may be much greater.

If a trading venue is consistently failing to show liquidity, should it still be scanned as part of the smart order routing process?

Balarkas: If one is already connected to a venue, then there is nothing to lose by scanning it. If it is very inactive, it might not be the venue one pings first or places one's own passive orders. However, once connected, there is little or nothing to be lost in trying it out. There may be a point at which one disconnects from such a venue in order to save unnecessary costs, but the liquidity would have to be pretty dire to justify that step.

Toby Bayliss: "...the competition among trading venues is serving to keep the costs of connection down..."

Bayliss: If it is felt that the trading venue has in certain cases the potential to still be a valuable source of liquidity, then it will continue to be scanned as part of the order routing process. If liquidity falls to such an extent that costs of accessing the venue do not offset the perceived price benefits, then scanning would cease. Similarly, if accessing a very illiquid trading venue reveals too much information about your particular trading strategy, it should be omitted from the scanning process.

Particularly in the US, the competition among trading venues is serving to keep the costs of connection down, so in general they are not disproportionately high in relation to the expected return from executing trades on those venues.

The position In Europe is currently slightly different. For example, I don't believe a single stock trader would see much value in being connected to a Continental exchange to trade a major UK security. Such a stock may have an alternative listing on the Continent, but the liquidity is likely to be minimal in comparison to the LSE.

Hunt: Possibly, but it is important to note that there is a difference between smart routing and algorithmic trading - their objectives are different. Algos are the liquidity seekers and proactively scan all venues including internal. Algos can be effective at that because they have knowledge of the parent order, trading objective and benchmark.

So, you could say smart routers optimise the routing of child orders to ensure the best child price is obtained across venues, and algos aim to achieve the best execution result for the whole order size, over the life of the order, by seeking out hidden liquidity. Both look at previous history to develop a sense as to where you are likely to be filled, but the reality is that smart routers are not in fact scanning, as algorithms do, but rather receiving information. This information comes from two distinct sources: public venues that provide quote feeds which we consolidate and non-displayed venues, such as liquidity that is internal to a broker dealer.

Self: I think a twofold approach is appropriate here. One is historical - obviously you do historical analysis as to where liquidity is available in terms of regional exchanges and other liquidity venues. However, while trading your systems also need to keep track of where you are finding liquidity on a particular day. So if you aren't finding liquidity on a certain venue on a particular day then you might start trying other venues first.

Nevertheless, there is an opportunity cost to consider when deciding venue priority. If you ping a venue in considerable size, then that obviously ties up a lot of your order, so a balance of historical and real time liquidity profiles should ideally be considered when making the priority and weighting decision.

However, if you have a connection to a liquidity venue anyway, there isn't any reason not to test it for liquidity, as there isn't an additional cost for pinging it.

Yuster: The historical liquidity available on a particular destination is an important factor in a smart routing decision, and at the start of a trade, the weighting given to each venue will be based on the historical liquidity available. Prioritisation of trading venue is then revised dynamically based on available liquidity.

Additionally, we believe that clients should have the option to decide for themselves by setting their own individual rules as to which venues our smart order router polls when searching for liquidity. Some clients want to limit what they publish on certain less discreet venues - often only taking liquidity from rather than posting liquidity to that platform. In addition, clients have the option to prioritise venue polling on the basis of the real-time hit rate achieved for a particular name on that venue.

Pulling the plug on illiquid venues

Brad Hunt: "MiFID asks investment firms to consider criteria that they would use in order to make a routing decision."

Do you feel that smart order routing systems are now so effective in consolidating the various trading venues that they already replicate/exceed the claimed liquidity benefits of a centralised limit order book?

Balarkas: A common view amongst economists and one that I share is that exchanges should be natural monopolies. By that I mean that there is no reason why a single venue cannot provide all the functionality and benefits of reduced search costs, liquidity, and pre/post trade transparency. If such a venue existed it would exist alone, as there would be no reason for competing venues to exist. However this assumption depends upon the natural monopoly not extending into monopoly pricing, which is a common concern.

I think that the process of aggregating trading venues does actually replicate many of the single exchange benefits outlined above. This creation of a virtual single market probably matches the liquidity benefits of a centralised book, but I am not sure it exceeds them.

On the other hand, the fragmented model adds value in terms of maintaining competition and therefore bearing down on costs. There are other benefits as well - for example, if one venue is prepared to price at a finer tick size than others, then that represents a major advantage.

Bayliss: The most efficient, fair and effective market is based around a centralised limit order book. This however assumes that the centralised order book has all the required functionality, order types, speed and also operates at minimal cost. Unfortunately the monopoly effect has lead to centralised trading venues failing in one or more of these areas. The efficiency of consolidated order books ensures that the basic functionality of a centralised exchange is matched.

In addition, there are potential additional benefits derived from liquidity hidden from a traditional exchange. For example, the ability to execute large blocks of stock by accessing dark pools of liquidity and trading with less impact will be a significant trading benefit. Other advantages, such as being able to trade within the spread and the availability of additional order types, will offset the setup costs of smart order routing systems. The overall result is that smart order routing across multiple venues represents an overall net gain versus a centralised limit order book.

Hunt: Particularly in the US, where liquidity fragments across a number of different ECNs, I think smart routers have been very effective in consolidating that liquidity. They are effectively showing participants a virtual centralised limit order book. However, the liquidity that makes up that virtual book is still residing in different systems, so there is no single location for price and time priority executions to occur. So, while they are effective at consolidating liquidity across market centres and venues, they will never replace one central limit order book.

This of course prompts the question of whether a single central limit order book in which all the liquidity resides is really the end objective. At this point, we would say no. One of the things that multiple trading venues and smart order routing have achieved is a reduction in overall trading costs by creating more competition between exchanges and tightening spreads.

Jackson: A single centralised limit order book effectively represents a monopoly on trading with all the associated implications of monopoly pricing due to lack of competition. Therefore, in terms of competitive pricing, you could argue that the smart order routing across multiple venues offers lower costs due to competition.

By offering unprecedented access to additional, undisplayed inventory - both hidden reserves and dark pool liquidity - smart order routers can also be said to increase available liquidity beyond the visible exchange order book. In addition, it should be noted that smart order routers do more than just aggregate liquidity, they also provide clients with a suite of tactical strategies to satisfy a variety of trading objectives.

Self: I think that for those who have effective smart order routing then this is certainly the case. Entities that are supplying best execution services simply have to have this kind of technology available. If they use this in the correct way, then it does indeed largely replicate a centralised limit order book.

The one thing it doesn't do is take account of the unsophisticated market participant. For example, if you happen to be sitting upon the best bid in a particular liquidity venue and an unsophisticated participant hits a bid lower than yours somewhere else, then that still affects your execution deleteriously. However, regulation is effectively compelling trading sophistication, so over time even this caveat should disappear.

Chris Jackson: "A single centralised limit order book effectively represents a monopoly on trading..."

Is smart order routing that omits the sensing of hidden liquidity intrinsically dumb?

Balarkas: Yes!

Bayliss: The short answer is of course yes. The idea of routinely neglecting to search hidden pools contradicts the most basic concepts of transaction cost analysis. Executing with hidden liquidity can enable significant volume to be transacted with minimal impact.

However, if you are trading an order that represents a very small percentage of the daily volume in a liquid name, targeting hidden liquidity may not be necessary to facilitate best execution. Under these circumstances, if you are happy to get done just at the touch, then of course you don't need to search all the venues for liquidity anyway. There is not a lot of point pinging all the exchanges and trying to find out what is going on in all dark pools if all you are trying to do is a thousand Vodafone shares.

Hunt: A smart order router that only accesses displayed liquidity on public venues is not accessing all the liquidity available in the market place at that given point in time. Therefore, if you think access to all available liquidity is important to achieving your investment or trading objectives, then you would have to consider that as a factor. We have discovered that accessing both displayed and non-displayed liquidity is vital for our smart router. This is because the non-displayed liquidity component is significant in improving execution performance and accessing greater size than is available at the touch price, thereby reducing market impact.

I think it is too early to tell how much difference, in percentage terms, accessing hidden liquidity actually makes in terms of performance. It really depends on the size of the order. However, what we are seeing in Europe is spread compression and the dispersion of liquidity across the order book. So while touch spreads are tightening, the weighted spread cost for a larger order is not necessarily narrowing and that is a result of existing liquidity being dispersed across a larger number of tiers. Therefore, while it is too early to say for sure, evidence suggests that there are significant benefits of non-displayed liquidity in terms of size improvement at a given price, which intuitively leads to a lower market impact cost.

Self: The need for sensing hidden liquidity can differ depending on the situation. For smaller orders where enough liquidity is available on the touch, sensing hidden orders may have little material benefit. However, for other situations where your need for liquidity is greater, it would be a mistake to omit the sensing of hidden orders. As I said earlier, if you are already connected to a venue the amount it costs to send an order to test for hidden liquidity is small. It might cost fractionally more in certain venues in that you might have to leave the order there a little bit longer because of the cyclical nature of other participants' trading.

Yuster: If you are pinging venues for liquidity and not using the results obtained from that process as an input to future trading behaviour, then yes that is less than smart.

Smart order routers require extensive databases of statistical information in order to seek hidden liquidity. A smart router must have knowledge of the available order types and parameters at a destination - for example minimum size constraints, support for iceberg order types - as well as historical statistical information per venue. For example, an important determinant in making routing decisions is the typical duration required to rest in a dark venue in order to maximize fill rates.

Owain Self: " of the most effective techniques is using intelligent learning to optimise your next trade."


In general terms, which liquidity sensing techniques do you feel are most likely to be effective?

Balarkas: Historical experience is a significant input to this. Over time as one continually operates on the same venues one appreciates where it is easiest to execute trades against hidden orders. However, while this represents solid empirical evidence, like any historical data it isn't infallible for predicting future activity. There is always the possibility that hidden liquidity will move to another venue for a myriad of reasons.

On the other hand, if one is trading against more hidden liquidity on a particular venue, then that in is itself an incentive for other participants to place more such hidden liquidity on there. Their desired outcome is for their hidden orders to be found and traded against and moving to that venue will obviously facilitate this.

Bayliss: The ability to "ping" multiple sources or dark liquidity allows dark pools to be probed to detect interest from other market participants. This is achieved by simultaneously sending off minimum size orders to each dark pool to investigate if any appetite exists. This then enables liquidity to be directed to the venue with the highest probability of execution.

The "networking" effect is also important; remembering and acting upon successful destinations from previous orders can improve the hit rate of which trading venues to target.

The use of a "Dark Book" at the order routing/EMS stage allows orders to be routed where it is already known that liquidity exists. It also enables users to specify total order size without revealing full details to the market or third party dark pool.

Self: Trying multiple sources and using immediate or cancel orders are very simple ways of accomplishing this. In general terms, one of the most effective techniques is using intelligent learning to optimise your next trade.

The key is how you marry this real time learning with historical experience. Your starting point will always be history, but that alone is not enough. If you get that marriage right then it will provide a good insight not only into where liquidity might arise, but also when and who from.

Sheridan: In general terms the key objective with most trading strategies is to reduce information leakage. We have found that algorithms are a very effective tool to seek both displayed and non-displayed liquidity in a stealthy manner.

A major objective for the investment community is to reduce the trading costs associated with small and mid-cap securities. There are a number of different ways of doing that, including:

  • An effective feedback loop: When trying to access non-displayed liquidity you need to be smart in the way that you ping it. This includes trying to achieve a mid-price execution to see if there are any discretionary order types in the market. That needs to be supported by a good feedback loop, so that as you are pinging you utilise the fill rates on those pings going forward to adjust your trading behaviour to best effect.

  • Real and Unreal liquidity events: There are still occasions when market participants will advertise size in the order book with the intent of bringing a buyer or seller into the market. In these circumstances it is important not to look at the displayed liquidity and assume that it is a liquidity event. A far more important consideration is whether it is a real liquidity event and specifically whether or not it is one that makes showing one's own hand worthwhile.

Assessing the characteristics of certain liquidity requires a historical understanding of what the order book looks like in that particular security - for example, hat is the average bid offer size? Is the event occurring now genuinely significant? Is it significant in that the current bid/offer spread is tighter than normal? Just because somebody is offering 100,000 shares doesn't necessarily mean it is worth trading. If you execute, you are showing your hand and potentially marking up the balance of your order.

Yuster: We tend to use a mixture of historical and real time data to help locate liquidity. Early in a trading session, we tend to rely more on historical statistics per symbol and per venue. However, as the session develops, you are able to see if real time liquidity distribution differs from the historical profile and accordingly, dynamically adjust your routing calculations.

Richard Balarkas: "...Money managers have different views on the risk that is represented by showing their order..."

Is the cost of developing liquidity sensing algorithms only justifiable in the case of more volatile stocks with a higher exposure risk?

Balarkas: No I don't believe that this is the case. The simple fact is that there are many reasons why people feel they can achieve alpha but this is not obvious just from the nature of the stock. Therefore money managers have different views on the risk that is represented by showing their order and the effect that will have on the alpha they are trying to capture. I don't believe that this is mirrored within the qualities of the stock itself, such as size and volatility.

Bayliss: There are many techniques for sourcing liquidity that are relevant for both liquid and illiquid names; as such, the skills and knowledge are directly transferable. The risks of trading liquid names are less prevalent as there is a reduced risk of predatory gaming. However executing large size in liquid names still raises impact issues which trading with dark pools can reduce.

Jackson: No - you need liquidity sensing algorithms across the whole universe of stocks for which you offer smart routing. Volatility is only part of the story, liquidity is obviously also important - and both these characteristics of a stock can change hourly. Therefore there are frequent occasions where finding liquidity outside the current touch on even low beta utility names has significant client benefit. Furthermore, the costs of this technology are mostly fixed so extending the universe of names covered is very straightforward.

Self: There are a number of factors which contribute to the overall cost of trading - spread, volatility, liquidity etc. Any order that is large relative to ADV or in a stock which has a wide spread will naturally benefit from finding liquidity and therefore is a good candidate for these algorithms. Additionally where the trader has high alpha content in their flow, in any type of stock, the benefit from picking up non-displayed liquidity and thereby reducing their duration of execution will be immensely beneficial.

Sheridan: No. We think there are benefits to finding liquidity in any stock. If you look at the US, whenever an order is executed in even a large cap name in the order book, information is released to the market. There are many different strategies that react to that information that potentially have a price impact on the stock, even if it is a large cap. So accessing liquidity in a stealthy manner is vitally important, irrespective of the market capitalisation or the turnover in the particular name you are trading.

I think there is definitely a requirement to have a liquidity seeking component on all security types, not just on those that are volatile. Ultimately, liquidity sensing becomes a tool that augments the existing benchmarked algorithmic strategies. Participants are not just looking to mimic the benchmark; they are looking for some out-performance, which is where I think these strategies can be very useful for an active manager.

Peter Sheridan: "This time of day effect is more exaggerated amoung small and mid-cap securities than large caps..."

Are liquidity sensing algorithms already starting to make the use of hidden limit orders redundant?

Balarkas: No not at all. When I place a hidden limit order I want counterparties to be able to find me but without moving the price ahead of trading (pre-trade risk). In practical terms there is simply no way one can accurately model the depth of an iceberg order. (Though one can make a rough estimate based upon the amount that is being shown).

Bayliss: Hidden limit orders are necessary when liquidity is not available at the required level and you wish to ensure you capture liquidity at that level or better, without signalling your intentions to the market.

Therefore I would disagree and say that there remains a very strong case for using hidden limit orders, regardless of the capabilities of liquidity sensing algorithms. In practical terms I cannot see any way in which it is anywhere near possible to recreate the entire order book. Unless you interact with all the flow how can you actually know what is there?

Self: I don't believe so - in fact liquidity sensing algorithms are actually making those hidden limit orders more effective. If the sensing algorithm finds your hidden limit order then you trade, which is what you want.

The exception is when a counterparty is continually taking your liquidity in small size to see if you are really there and then using that information against you in a predatory fashion. This is why I alluded earlier to the importance before connecting to a venue of checking whether the activity of other existing participants is actually prejudicial to your trading.

Sheridan: No, I don't think they are making them completely redundant. There are tools that can look for hidden liquidity but there is no guarantee they are finding it all. They can look at the average bid/offer size, the average trade size/count and other data to make estimates. However, these are just informed estimates as to what lies behind the visible order, so there is still some value to hidden limit

orders. Therefore I don't think liquidity sensing algorithms will ever make hidden orders completely redundant.

Yuster: No, because there will always be limits on how accurately you can reconstruct the order book. In the first instance, you have to send some flow to a venue to gauge what is there. In the second, historical data is only of value in predicting what may be in an order book if it is statistically significant and has a high probability of recurring in the future. Hidden limit orders will always be an important way of participating in the market without exposing the order.

Do you believe that the effectiveness of liquidity sensing is strongly correlated with the time of day? If so, does this imply that other types of algorithm should be automatically substituted at certain times of day?

Jarrod Yuster: "...there will always be limits on how accurately you can reconstruct the order book."

Balarkas: An effective sensing algorithm will work consistently throughout the day; therefore I cannot see any reason why one would not wish to use it because the market was illiquid. Typically more than 75% of all money managers' orders have negative trend (price moves adversely away from the desired entry point). On that basis one needs to get trades done faster than pure participation with visible liquidity will allow, so there will always be a need to search for and exploit hidden liquidity.

Bayliss: Every dark pool has its own rules on how hidden orders are matched, some are continuous, while others operate on an auction basis at set times of the day. The discontinuity of the trading in dark pools naturally causes their effectiveness to be correlated with times of day. Information on liquidity and auction times are therefore an intrinsic part of the smart order routing process.

However, despite all these factors, it is interesting to observe from practical experience that if a robust liquidity sensing algorithm is being used, no particular time of day seems to yield better or worse results.

Jackson: There is certainly a strong correlation between liquidity patterns and time of day. You can clearly see that the liquidity, volatility, trade frequency and intensity characteristics of certain markets are very different at points over the day. It is important that the historical statistics driving an algorithm are not just daily averages, but specific to the time of day.

Liquidity seeking algorithms should be able to model liquidity patterns based on historical statistics but then adjust real-time to changing market conditions. Our smart routing algorithms did just that during a recent market outage. When liquidity on the main market declined, the algorithm compensated by increasing the proportion of the order worked and executed on alternative venues.

Self: Depending on the liquidity venue (and particularly on the larger dark pools with passive institutional flow) you will definitely see a correlation between the number of hidden limit orders and the less volatile points of the day. However, each venue tends to have its own characteristics, which is why (particularly in the US) we will try certain venues at certain times during the day, as we know there is likely to be more liquidity available.

Timing hidden liquidity

Sheridan: I think the answer to the first part of the question is yes. From a European perspective, there is a very significant "time of day" effect in European markets. If you look at the LSE, the bid/offer spread of even large names during the price discovery period is four or five times wider than the average spread for the whole day. Liquidity (in terms of displayed liquidity) is very limited during that period and the trading cost implication of that is why some clients have historically tended to avoid trading at this time of day.

This time of day effect is even more exaggerated among small and mid-cap securities than large caps that have a continuous and stable order book throughout the day. Volume in small and mid-cap securities is very often skewed towards the latter part of the day when more investors are coming into the market.

Generally speaking it is always important to understand the trading patterns of an individual stock or market, and time of day is a significant part of that. Therefore algorithms need to be tuned to take advantage of these characteristics. However, I don't think it is necessarily a case of using different algorithms. I suppose the question clients should ask themselves instead is whether the algorithms they are using take into account the time of day effect and, if so, how?