By Phil Mackintosh, Chief Economist, and Satchit Sagade, Head of Quant Research, Economic and Statistical Research – Europe, Nasdaq
In some of our recent posts, we’ve talked about how trades move between venues at close to the speed of light. We have also highlighted that speed is different for fiber and microwave transmissions. Data also suggests that some venues tend to set prices while other venues mostly peg to those prices. In fact, the evidence suggested that primary listing venues contribute much more to price discovery than other trading-only venues, quotes matter more than trades, and dark markets contribute almost nothing to price discovery – even when they trade – as they are pegged to price-setting venues’ prices.
This is especially important in a region as big as Europe. The time it takes for price updates to travel around such a large region means pegged prices will update a (relatively) long time after the actual prices have changed. That gives trades with microwave connections an even bigger advantage.
It’s even more important to European regulators if they’re still considering using the consolidated tape as a pre-trade price to peg prices and/or to make routing decisions.
Stale pricing found to impact dark pool executions
Given all this, it should come as no surprise that a recent study looking at U.K. dark pools found “a substantial amount of stale trading [in dark pools] occurs.” This finding has been replicated extensively across Europe, with research from Euronext, SIX Swiss and Deutsche Börse finding similar results.
You might say, “So what?” After all, a different study in the U.S. found that even though price dislocations could be seen on the U.S. consolidated tape, when the researcher accounted for the time that it would take to send trades back to the venues with dislocated prices, the “latency arbitrage” opportunities were mostly gone.
Dark pools peg to primary prices
Dark pools in Europe don’t (by definition) publish prices pre-trade. They usually aim to match buy and sell orders at the midpoint of the primary best bid and offer (PBBO). This makes use of the so-called reference price waiver.
Keeping orders in dark pools without pre-trade transparency (public quotes) helps a buyer “hide” their demand, reducing price impact from signal-driven traders.
However, we also know that hidden orders have higher opportunity costs, via lower probability of execution. That’s why dark pools are more attractive to larger, patient investors.
Fast traders are (almost) always right
What’s a little different about the U.K. study, though, is the research found that arbitrageurs frequently take dark liquidity when stale prices exist. The U.K. study finds that the arbitrageurs are on the winning side of the execution between 96 and 99 percent of the time (see Chart 1).
Importantly, because the fast trader always seems to be on the “informed” side of the spread, the research concluded that these stale prices were “imposing large costs on passive dark pool participants.”
Chart 1: Participation in stale trades: Winners and losers
Tracking latency arbitrage in dark pools: An example
Let’s look at an example of how this works for a hypothetical stock listed in Stockholm and trading in Nasdaq as well as in a London-based dark pool.
1. At time = 0 milliseconds (No activity, all markets in sync)
The market is in a steady state with the PBBO midpoint of €11, which is reflected in the Nasdaq data center located in Stockholm as well as in the London venue. A natural sell order is sitting in the London dark pool.
Chart 2a: At t = 0ms (market in steady state with a price of €11)
2. At time = 1 milliseconds (price discovery in Stockholm)
Price discovery happens in Stockholm when a buyer sends an order to lift all shares on the offer (PBO). That leads to a change in the PBBO midpoint to say €12. Information about the trade as well as the new PBBO midpoint starts its journey to London both via microwave and fiber connection; however, as we know, microwave is faster, so after 1ms, that update is closer to London than the Fiber message.
Note that at this time, only traders in Stockholm know that the trade has occurred, and the PBBO midpoint has changed.
Chart 2b: At t = 1ms, only traders located in Stockholm know the market is now higher
3. At time = 9 milliseconds (Arbitrageur trades arrive in London before updated midpoints)
The information arrives in London via the faster microwave connection where an arbitrageur who, knowing prices have gone up in the primary, seeks to buy all dark liquidity at the stale old (lower) midpoint price, resulting in a loss of €1 for the natural seller.
Chart 2c: At t = 9ms, traders using microwave can sweep London dark venues, hoping to buy at the old lower midpoint
4. At time = 10 milliseconds (Back in sync)
Fiber connections tell dark pools that the new mid is €12, and the midpoint peg reference price is adjusted. Both Stockholm and London prices are back in sync.
Chart 2d: At t=10ms, London dark pool updates the PBBO midpoint after receiving the PBBO midpoint update on fiber
In less than the blink of an eye (literally), lit prices updated to €12, and then dark pool trades occurred at the old €11 midpoint.
The fast arbitrageur can generate an instantaneous profit by selling at the potentially higher new bid while buying in the dark pool at the old midpoint. Chart 3, which is from the London study, illustrates this for a U.K. stock where a buyer-initiated dark trade prints at the older midpoint (£30.235), allowing the arbitrageur to sell at the new bid (£30.25) and make an instantaneous profit (£0.15).
Chart 3: A view of prices and fills across time shows dark prints at old prices, below the new bid (taken from Fig.6 in Aquilina, Foley, O’Neill and Ruf, 2023)
We find the same seems to happen in Nordic markets
We replicated some of the analyses from the U.K. study in Nasdaq Nordic markets. Specifically, we looked at companies in the OMXS30, OMXC25 and OMXH25 indices for the period from October 2022 to April 2024.
First, we know that around 60% of all EBBO updates for Nasdaq Nordic listings originate on Nasdaq (see Chart 4). That indicates that prices are mostly set in our Stockholm data center, which is around 10ms by fiber from the majority of U.K.-based dark venues.
Chart 4: EBBO setting frequency
We compared six dark pools to Nasdaq’s Nordic-at-Mid. Together, they account for an average turnover of €4.7 billion per month and almost 100% of executions in dark pools using the reference price waiver.
We find that:
- 30% of all dark pool trades, excluding Nordic-at-Mid, were executed at stale prices.
- €17.1 billion or 23% of total turnover in dark pools is executed at stale prices during our sample period.
Interestingly, there are large differences across the different dark pools, with the highest (lowest) instance of stale pricing being 50% (21%) based on trade count and 33% (8%) based on euro turnover (Chart 5).
Chart 5: Frequency of stale pricing
We also compute the latency of stale executions across the different dark pools. Latency here is measured as the difference between the trade time stamp in the dark pool and the time stamp of the true PBBO midpoint at the time of the dark trade. We observe median latency across dark pools ranges from around 9ms to 15ms. However, because larger dark pools have lower latencies, the market-wide median latency is closer to the low end, at 9ms. This maps closely with the observed fiber transmission speeds between Stockholm and London, indicating dark pools use fiber for price updates.
Costs of stale Nordic pricing could be over €3 million per year
We estimate the cost of stale pricing as the difference between the actual PBBO midpoint and the price in the dark pool for all stale transactions. The cost per dark pool is driven by each pool’s overall liquidity and its proportion of stale trades (see Chart 6).
We calculate that in 2023 a total of €3.3 million is lost for just for the 80 Nasdaq Nordic listings in the period we analyze.
Chart 6: Euro cost of stale executions
In relative terms, this cost adds up to 2.9 basis points (bps) across all dark pools. To put this in context, the average quoted spread on Nasdaq during our sample was 5.6bps. Furthermore, this cost of stale execution ranges from 2.5bps to 3.6bps across the dark pools again showing that there are large differences across them (See Chart 7).
Chart 7: Cost of stale executions in bps
Not all dark pools are the same
All the above statistics exclude Nasdaq’s Nordic-at-Mid. That’s because Nordic-at-Mid is different.
Being integrated into the matching engine for the primary lit book, it always trades at the current PBBO midpoint, experiencing zero cases of stale pricing.
Furthermore, in the case of Nordic-at-Mid, the use of size, as opposed to time, priority helps investors trade much larger quantities than competitor dark pools. In our sample, Nasdaq’s Nordic-at-Mid has the second-largest median execution size (€6,853) across all dark pools and the largest median execution size across the five largest dark pools (see Chart 8).
Chart 8: Trade size distribution across dark pools active in Nasdaq Nordic listings
Importantly, this means natural investors using Nasdaq’s Nordic-at-Mid eliminate the risk of trading at stale PBBO prices and simultaneously increase the chance their midpoint fills are executed against other natural investors crossing the spread.
That’s because zero latency means no chance for faster traders to intercept midpoint liquidity at the expense of natural orders. That, in turn, improves the liquidity available, with price improvement, for spread crossing orders on the primary.
The availability of an atomic dark-lit sweep functionality allows aggressive orders to ping the Nordic-at-Mid before hitting the far touch again with no risk of interception by faster traders. In our sample, a dark fill originating from such sweep orders, on average, results in 2.3bps instantaneous half-spread savings plus a further 3.8bps savings in price impact.
Nordic-at-Mid liquidity is growing
That makes Nordic at mid better for all investors — those looking to hide at the midpoint as well as those needing to cross the spread for liquidity. Not surprisingly, Nasdaq’s Nordic-at-Mid market share has also been increasing. Rising from 6.6% of all dark midpoints in October 2022 to 24.0% in April 2024 (Chart 9).
Chart 9: Dark RFPT market share
This again proves the consolidated tape can’t be pre-trade
In a future with a European consolidated tape, it may be tempting to use the EBBO as a benchmark to match trades in real-time. However, pegging to a consolidated tape will almost certainly make the stale pricing problem even worse than we see above.
That’s because a consolidated tape will increase delays in transmitting the “latest” quotes around the market. Not only will the tape need to wait for the last venue to update, but the physical location of the consolidated tape provider will also add another leg of transmission time to anyone “pegging” to the EBBO prices.
The BIS study seems to prove that it will create winners and losers, both across individual national markets and by trading sophistication. In short, the distance to the center of price discovery will still matter more than the distance to the consolidated tape; it’s just physics.
Pegging isn’t price setting
The other important thing to note here is that while our focus has been looking at stale pricing in dark pools, the broader argument likely also applies to other forms of pegged trading, including in some systematic internalizers who also do not contribute to price discovery and rely on prices derived from lit markets.
If it’s important to share the EBBO with all traders, it’s also important that the venues pegging to EBBO pay their fair share. Said another way: market makers setting the spread deserve to capture that spread, too.
Failure to price the EBBO correctly will result in artificial fragmentation like the U.S. has, where pegged venues can free ride on the quotes of price setters, earning trade revenues while reducing spread capture for the very traders who aresetting those prices.
And that won’t just increase latency costs; it could easily add to search costs and harm price discovery at the same time.
