MiFID 2: Impact of Dark Caps on Algorithmic Trading Strategies

Alistair Cree, Assistant Vice President, Analytics
Colleen Ruane, Managing Director, Analytics


Dark trading in Europe is expected to be fundamentally altered by the implementation of MiFID II/ MiFIR. The proposed cap on dark pool volumes will require all institutional investors to reevaluate how they interact with dark liquidity, but for some the impact of the changes will be greater than for others. This research examines the potential impact on a variety of algorithmic strategies. In particular we focus on the varying degrees to which different strategies participate in dark trading and how much of that volume is traded in sizes which qualify for the Large In Scale (LIS) waiver and so are exempt from the cap.

In brief we find that:

  • The portion of dark volume executed LIS increases with order size. Merging algorithmic orders to increase order size where possible may prove a useful tactic for maintaining access to dark liquidity in the event that dark trading is suspended.
  • There is significant variation in different institutions’ use of dark liquidity. Institutions that trade more heavily in the dark also make greater use of LIS executions.


As is currently understood, dark trading regulated by MiFID II/MiFIR will differ from the present regime in three significant ways. Firstly, the closure of Broker Crossing Networks (BCNs) will limit brokers’ capacity to cross client order flow internally. Secondly, dark pool trading on MTFs will be limited to execution at the midpoint only. Finally, caps will be implemented on the volume trading both on any individual venue at 4% of market volume and on the market as a whole at 8% of market volume. ESMA will calculate the dark volumes for each venue and the market as a whole on a stock-specific basis over a rolling 12 month window. In the event that ESMA determines that the cap has been breached, dark trading for that security will be suspended for the next 6 months, either in the specific pool or for all dark pools. In this paper we focus on the potential effects of such a suspension of all dark trading for a security.

Estimating present dark volumes is inherently difficult given the fragmented nature of European markets and the absence of a consolidated tape. Most estimates place dark volumes in the range of 6 – 10% of total volumes but these estimates can vary significantly between markets and securities. For example, for the period from 2014-04-01 to 2015-03-31, Fidessa estimates European dark volumes at 6.1% of total for Europe as a whole but at 9.5% for the UK and 7.0% for Sweden.


Under MiFID, orders submitted to venues above certain sizes may be exempted from MiFID’s pre-trade transparency requirements under the Large In Scale waiver (LIS). One important subtlety of the proposed dark pool cap is that LIS executions will be exempted from both the calculation of the dark pool cap and any subsequent suspension of dark trading. Whether or not a venue order is large enough to qualify for LIS depends both on the order’s size and the average daily turnover of the security (ADT). Table 1 shows the current LIS thresholds set out by MiFID.


As trading in sizes above LIS provides a route for participants to continue to execute in dark pools once the cap has been exceeded it can be expected that the use of this waiver will significantly increase under MiFID II/MiFIR. It follows that algorithms which already make greater use of LIS executions will have to make smaller adjustments to their strategies than those which trade less volume at LIS sizes. ESMA released draft regulatory technical and implementing standards for MiFID II/MiFIR on 2015-09-28 containing revised tables for determining whether an order qualifies for large in scale. For the specific universe of data used in this analysis, using the new LIS thresholds results in only minor changes to the statistics provided below.


The universe of data for this analysis is limited to European algorithmic executions occurring between 2014-04-01 and 2015-03-31. Additionally we restrict the data to executions in securities with a market capitalization above $5 billion U.S. dollars. All buy-side participants represented in the data traded with a variety of brokers. In total the dataset included over 9 million executions ($120 billion U.S. dollars) spread across 65 separate venues through 24 brokerages.

Executions were classified as LIS based upon whether their size in Euros exceeded the thresholds shown in Table 1. The ADT values were sourced from ESMA’s table of shares admitted to trading on EU regulated markets. Where necessary the conversion from the local currency of a security to Euros was also sourced from this table. It should be noted that as our analysis uses executions rather than venue orders our estimates regarding the value of LIS trades can be considered conservative.

Throughout this analysis we make use of a system for classifying algorithmic strategies based upon their objectives and execution styles:

  • Scheduled – includes TWAP, VWAP and participation-based strategies.
  • Dark – denotes a liquidity-seeking strategy concentrating on
    dark pools.
  • IS – is shorthand for Implementation Shortfall, a family of strategies based upon cost and risk minimization.
  • Liquidity Seeking – refers to opportunistic strategies that prioritize the sourcing of liquidity.
  • Other – includes a range of alternative strategies such as Close and DMA strategies.


The term “order” has a wide variety of meanings within the broad subjects of trading and market structure. To avoid confusion, from this point, we will exclusively use “order” to describe the parent orders submitted to algorithms and not the smaller slices submitted by algorithms to venues. When discussing the fills obtained back from venues we will refer to either “executions” or “fills”.


The extent to which different algorithms engage in trading in dark pools is determined both by the objectives of the algorithm and the characteristics of the order it is required to execute. Larger orders provide both greater opportunity and incentive (in the form of reduced impact) for making heavier use of dark liquidity. Additionally the more flexibly an algorithm can vary its participation the greater use it can make of the opportunistic liquidity available in dark pools. Similar constraints affect algorithms’ use of LIS sizes.

In Figure 1 we have classified algorithmic executions into one of three types based on the size and venue type:

  • Dark ex LIS – executions in dark pools at sizes below the
    LIS threshold.
  • Dark LIS – dark executions above the LIS threshold
  • Lit – executions in lit venues.

Figure 1 shows the breakdown of the value traded between these three execution types. As expected Dark strategies make the heaviest usage of dark venues while Scheduled strategies only execute 22% of their value in dark pools. Dark strategies also make the most use of LIS sizes when trading in the dark; 18% of dark executions by value were above the LIS minimum size. For Scheduled strategies this number was much lower; only 2% of dark executions by value qualified for LIS. Both the IS and Liquidity Seeking strategies made similar use of dark liquidity (41% and 43% of traded
value respectively).


By grouping executions according to the size of the order entered into the strategy we can see that this difference between strategies’ use of LIS is not a result of differing order sizes. Figure 2 shows the value traded in lit markets as percentage of total traded value while Figure 3 shows the value traded in LIS sizes as a percentage of dark traded value. The results are grouped by strategy and the size of the order relative to the large in scale threshold. For example the group ‘A) 0 -50%’ indicates orders that were less than half the size of the LIS minimum size.



Looking at the data in this way it is apparent that the difference in LIS usage between strategies is present for all order sizes where LIS trading is possible. This suggests that the lower LIS usage by Liquidity Seeking algorithms is less a result of differences in the orders submitted to these strategies and more likely due to their different execution styles. We can therefore speculate that in the event of a suspension of dark trading by ESMA that Liquidity Seeking strategies’ ability to access dark liquidity will be more significantly disrupted than IS strategies.

Figure 3 shows a clear relationship between larger order sizes and increased use of LIS executions. Clearly for orders smaller than the LIS minimum size any LIS trading is impossible but after passing this threshold the value traded LIS steadily increases with order size for all strategies. For orders greater than 1000% of the LIS minimum size entered into Dark strategies 38% of the dark traded value was executed LIS.

How many orders are actually entered into algorithms at sizes below the LIS minimum size? In Figure 4 it can be seen that 23% of value executed by Dark strategies was submitted in sizes smaller than the LIS threshold. For IS strategies this figure was 31%. This further highlights the impact of a suspension of dark trading. In the event of a suspension it may be advisable for traders to take steps to increase the size of orders entered into algorithms in order to maintain some access to dark liquidity. It is also worth noting that interaction with LIS liquidity may vary significantly based upon any individual broker’s implementation of these strategies.



Figure 5 shows the usage of lit liquidity split by country and order size for France, Germany, Great Britain and Sweden. Figure 6 shows the same breakdown for LIS liquidity. As before, the value of lit trading is shown as a percentage of total traded value, the value of LIS trading is shown as a percentage of dark traded value and the order size displayed is relative to the LIS minimum size threshold. In line with expectations these statistics show significantly less dark trading in France and Germany than in Great Britain. For Great Britain and Sweden there is a clear relationship between the size of the order and the use of both dark liquidity and LIS execution sizes, with larger orders participating more heavily in the dark and making greater use of LIS sizes. In France and Germany both of these trends are less evident.



Dark trading in Sweden is far more likely to take place at sizes which exceed the LIS threshold. For Swedish orders larger than 1000% of the LIS minimum size, 45% of the value traded on dark venues did so at LIS sizes. This suggests that algorithmic dark trading in Sweden may be significantly less affected by a suspension of dark trading. Specifically, the limitations on access to dark liquidity will be smaller than in France, Germany or Great Britain, with 45% of dark trading for large orders able to continue without any changes to the logic of the algorithmic strategies.

To what extent are these results due to differences in trading styles and strategy usage between different countries? Figure 7 shows the breakdown of strategy usage for each of the four countries. Relative to France and Germany, trading in Great Britain and Sweden shows greater use of Dark, IS and Liquidity Seeking Algorithms. This goes some way to explaining the lower usage of dark liquidity in France and Germany. It can also be seen that the higher use of LIS sizes in Sweden cannot be easily explained by different preferences in strategy selection. Sweden has lower usage of the Dark and IS strategies that execute most heavily in LIS sizes than Great Britain, but still has considerably higher use of LIS than Great Britain.



Finally we examine the variation between how institutions use LIS execution sizes. In Figures 8 and 9 we have classified institutions according to where they fall in the ranking of contributors to the dataset by the percentage of their trading executed in lit venues. “Upper Half” refers to institutions that traded the most on lit venues and “Lower Half” those which traded the least on lit venues. Each group contributed similar amounts of traded value to the dataset so this is not simply a division of institutions according to size but rather by trading style.

The differences between the Upper and Lower Halves of the distribution of lit venue usage are large. On average institutions in the Upper Half executed 84% of their traded value on lit venues, while institutions in the Lower Half only executed 57% on lit venues. For both classes of institutions the fraction of dark value traded LIS increased with order size while the fraction of dark executions of total trading remained constant. While the results confirm the expectation that institutions which trade more heavily in the dark will be more greatly affected by the cap it also shows that this effect will be somewhat mitigated by their increased use of LIS sizes. Ultimately these results underline that there are significant differences in the ways and extent to which different institutions use dark liquidity and the degree of exposure to a suspension of dark trading will vary depending on the institution in question.




Trading on dark venues is a significant component of the logic of most algorithmic strategies. In the event of a suspension of dark pool trading, under the proposed MiFID II/MiFIR regulations, these strategies will face varying degrees of disruption particularly at smaller order sizes. For each institution the effect of the new regulations will depend significantly on their style of trading and their exposure to different markets. The question of what effect the suspension of dark trading will have on transaction costs is a complex one, beyond the scope of this paper, but is an obvious next step in the effort to quantify the potential impact of these regulations.


ID.Higgins and J.P. Urrutia, 2014, Living In The Post-MiFID II World,
IIFidessa Fragulator,
http:\, http:\, http:\
IIIOfficial Journal of the European Union, 2006, Commission Regulation (EC) No 1287/2006, Annex II Table 2.