30 Jan 2013
Comparing Nordic Dark Pools
Short Article(8 pages)
Nordic Dark Pools
In 2012 Nasdaq OMX launched a new trading venue, 'Nordic@Mid'.
Nordic@Mid offers mid-price matches on a subset of the more liquid Nordic stocks
that are listed on the main Nasdaq OMX exchange. The matching price is based on
the mid price of the main Nasdaq OMX exchange at the exact time of each Nordic@Mid
Nordic@Mid competes with other existing MTF mid-price matching dark pools and broker
crossing networks such as CHI-Delta, Turquoise Dark, BATS Dark and UBS MTF. These
have offered primary market mid price matches on the more liquid Nordic stocks for
One feature distinguishing Nordic@Mid is that the matching engine used to determine
the mid-price matches is co-located within the main Nasdaq OMX exchange and so obtains
near zero latency on the market data needed to calculate mid prices. The matching
engines of most other dark pools/BCNs offering trading on Nordic Stocks are based
in London and so suffer an IT/geographic latency of at least 10 ms when sourcing
the market data needed to determine price matches. Because of these latencies, there
is the potential for the fastest HFT market participants to see price updates an
instant before the updated prices are reflected in the London based pools.
This short article looks at how the execution quality and characteristics differ
between executions that occur on Nordic@Mid and other competing venues.
Trade Data / Market Data Used in This Study
We base our analysis on all dark mid-point trades executed in November 2012 on four
dark venues: BATS Dark (BATD), Chi-Delta (CHID), Turquoise Dark (TRQD) and Nordic@Mid.
As CHID, TRQD and BATD offer trading on all European stocks, we have restricted
our analysis for those venues to only trades in Nordic stocks (as defined by ISINs
starting with DK, FI, SE).
The mid-price matched trade data we have used for all four venues is taken directly
from the exchange's own market data feeds. We have used the millisecond timestamps
provided by the exchanges. The market data we use to determine primary BBO and European
wide EBBO (primary venue plus CHIX, BATS, TRQX and BURG) lit prices at the time
of each trade is based on lit market order book data. This is sourced directly from
exchange feeds using the millisecond timestamps as disseminated by each exchange.
Comparing Summary Level Statistics
The table below shows a summary of all Nordic trades on the four mid-point matching
dark pool venues for November 2012. Results are shown for
- Exact Time / PBBO (effective price based on Primary best bid / offer)
- Exact Time / EBBO (effective price based on European best bid / offer)
Execution quality statistics based on PBBO/EBBO prices
One can see the following from the table above:
- CHID and BATD have significantly more volume/value traded than the other venues.
- The average trade size of Danish stocks is consistently higher across all four venues.
- Overall, average trade size is largest on Nordic@Mid (€9321), being about 50% larger
than average trade sizes on the other three venues.
- The average on book bid-offer spread at the time of trade is also much larger on
Nordic@Mid (18.4 BPS versus around 11 BPS on the other three venues). This indicates
that the instruments traded on Nordic@Mid are not so heavily weighted towards the
most liquid / low spread stocks.
- Taking the primary market best bid and offer at the time of the trade and comparing
this to the matched trade price, we find that there are no outliers for Nordic@Mid
trades. By contrast the other London based MTFs match trades outside apparent primary
BBO around 2% of the time.
- In terms of Europe-wide EBBO prices, Nordic@Mid shows fewer price matches outside
EBBO (1% versus about 4-5% for the other MTF dark pools).
The last result is slightly surprising. Due to its geographic and co-location advantage,
as expected, Nordic@Mid has no outliers versus primary mid price. However, when
looking at outliers versus EBBO, the EBBO price is composed of prices from both
Stockholm (Nasdaq OMX, BURG) and also London (CHIX, TRQX, BATE) and hence the geographic
advantage is eliminated. The fact that there are fewer EBBO outliers on Nordic@Mid
would seem to imply that the Nordic@Mid trades are not occurring as frequently at
times when there are differences between Stockholm based lit touch prices and London
based lit touch prices (more on this later).
Another way of analysing trades occurring outside of EBBO is to consider, in the
case of outliers, which venues are presenting an apparent arbitrage opportunity.
Looking at the charts below, we see that for Swedish outliers on CHID, it is mainly
the primary venue and CHIX itself that offer lit prices that are apparently better
than the mid-point matched dark price.
For the case of EBBO outliers occurring on Nordic@Mid, almost all outliers occur
due to lit liquidity on the London based MTFs (a very small number of outliers occur
due to better lit prices on Burgundy).
EBBO Outlier Venue Analysis CHID EBBO
Outlier Venue Analysis Nordic@Mid
One of the most likely explanations that Nordic@Mid has fewer price anomalies than
the London based dark pools is the geographical latency between Stockholm and London.
We can quantify this latency in the following manner:
- We take a dark mid-point matched trade from one of the London MTFs
- We compare the traded price versus the primary mid-price at the same millisecond
as disseminated by Nasdaq OMX
- If the prices are different, i.e. the MTF trade price does not match the Nasdaq
OMX mid-price, we play the Nasdaq OMX order book forwards or backwards until we
find the closest point in time where the MTF trade price is equal to the primary
If we look at a frequency histogram of how much we need to wind the primary order
books backwards in time to get the exact price match, we see a spike at the average
Latency distribution CHID Latency distribution
Latency distribution TRQD Latency distribution
What the results show is that the London based dark pool MTFs appear to suffer a
latency of between 12-14 ms in obtaining market data from the primary Nordic market.
Nordic@Mid does not have any latency as expected.
Price Movements / Impact
One of the advantages of trading on a mid-point matching dark pool is that, due
to the nature of the matching price, it is extremely difficult for other market
participants to infer the aggressive side (Buy / Sell) of a trade. Because of this,
one might expect trades in mid-point matching dark pools to have less market impact
than equivalent trades on lit venues.
From the public trade records that our data set for this study consists of, it is
not possible to infer the side of the aggressor of the trade. However, an indirect
way of measuring the impact of the dark pool trades is to look at how often the
primary market mid price moves at various time intervals after the dark trade.
The chart above shows the percentage of times that the primary mid price changes
at various time intervals (1 sec, 5 sec, 15 sec etc) after a dark trade for each
of the four dark pool venues in our study. The first thing to note is that CHID,
BATD and TRQD exhibit extremely similar behaviours. Nordic@Mid on the other hand
is markedly different; mid prices move significantly less, often following a trade
on this market.
Another way of measuring how correlated trading on dark pools is with lit markets
is to look at the mid price a short time before a dark pool trade and see whether
the trade is correlated with a prior price movement on the primary market. The table
below shows the percentage of time that a dark trade was preceded by a primary mid-point
price move on each of the four venues. Again, trades on Nordic@Mid show far less
correlation with price movements on lit venues.
- Of the four dark pools in this study, Chi Delta and BATS Dark currently have the
highest number of trades / value traded for Nordic stocks.
- Average trade sizes are about 50% larger on Nordic@Mid when compared to the other
3 dark pools.
- From looking at the average on-book spread at the time of each trade, Nordic@Mid
trades have higher average spreads suggesting that the instruments traded are less
heavily weighted towards very liquid, low spread stocks.
- Using exact timestamps, trades on the 3 London based dark pools are outside of the
primary best bid and offer price approximately 2% of the time. The most likely explanation
for this is the observed 12-14 ms geographical latency between Stockholm and London.
- When considering trade prices versus EBBO, all venues have some outliers. That is,
it would theoretically be possible to execute an arbitrage between the matched dark
pool trade and a lit venue. However, Nordic@Mid exhibited a lower rate of EBBO outliers
(1%) compared to London based dark pools (4-5%).
- Trading on Nordic@Mid appears to be less correlated with lit price market movements
than trading on London dark pools.
In conclusion, the above results would indicate that certainly for a Stockholm based
trading organisation, executions on Nordic@Mid will be most consistent with the
primary market data and also the venue appears to have slightly larger average trade
sizes and less correlation with lit market movements. For a London based trader,
the first of these advantages is reduced as geographical latencies will affect not
just the market data that one sees but also the time taken to execute.