Dark pools have attracted significant attention over the last few years – going so far as to make an unlikely leap from the business press to popular bookshelves.

One technical concern has been a gradual increase in reports among industry participants about the reliability of reference prices in dark pools. The concern being that investors could be persistently trading at stale, out-of-date prices.

For policy makers and industry participants alike, this would (if true) raise questions for a number of significant reasons.

One is technological. The amount of time it takes for data to flow from point A to point B, or ‘latency’, is critical in today’s trading world. Dark pools must obtain prices from other markets (see explainer below), a process that is measured in microseconds in the UK if it’s performing well, and milliseconds if it’s not. How much and how often that latency occurs, matters to the market and prompts investigation.

Another is related to fairness. Best execution considerations are heavily dependent on reliable prices. A lack of reliability in reference prices could contribute to the perception of a lack of ‘fairness’ in modern markets, harming market liquidity and depth if investors were to stop trusting and participating in dark pools.

Essentially, this means the central question of ‘how reliable reference prices are in dark pools’ is not an esoteric one. It’s one of general economic significance.

To help answer it, we combined order book data from the major UK exchanges, with matching engine timestamps and participant identifiers. This allowed us to carry out a comprehensive analysis of reference prices in the UK. The first analysis of its kind.

We focussed on what, in our view, are the two most important aspects of reference prices - namely how often dark trades occur at stale prices and how often they are the best available prices in the market. We also assessed what classes of investors benefit and lose when these cases materialise.

Stale reference prices are essentially instances in which transactions take place at prices that are not reflective of the latest available in the market. The City’s equivalent to a traveller purchasing a plane ticket at an ‘old’ price as live airfares change. A process that can, of course, create ‘winners’ as well as ‘losers’.

The topline is that we found about 3.5% of all trades in dark pools referenced a ‘stale’ price – a figure that incorporates highs of up to 11.5% on peak days in one of the dark pools for which we have data (which we also found had the highest overall prevalence of stale prices). It is worth noting that these figures are based on a pretty conservative estimate of what constitutes a ‘stale price’. We put it at prices from two milliseconds ago or more.

When a trade takes place at a stale price, there is plainly a party that benefits from the mispricing and a party that loses (depending on which side of the trade the party is). The distribution of these costs is not accidental: high frequency traders (HFTs) are on the winning side of the trade in essentially all of cases in which they trade at stale prices. In fact, we estimate HFTs ‘win’ in around 96% of their stale price trades in dark pools.

Moreover, we found that the percentage of trades happening at stale prices is increasing over time, but we find this is related to increases in message traffic and volatility over the sample period.

In some cases, these prices are stale for a period of time which, by modern standards, is quite long – coming in at over 11 milliseconds. Figures that suggest some dark pools experience latencies far in excess of the one millisecond or under baseline. Or to put it another way, far in excess of the fastest achievable time beyond the speed of light).

But it’s notable that the overall costs (or benefit) associated with stale prices are not, from the economic perspective, significant – standing at approximately £4.2m per year. For comparison, the average daily order book equity value traded on the London Stock Exchange (LSE) is £4.9bn.

In our analysis of the choice of reference price, we found that a reasonably small percentage of trades reference a price that is not the best available (between 0.6% and 1.2% depending on definitions) and that the costs associated with this are low. We find this is less prevalent among more sophisticated participants, venue operators and HFTs.

Overall, we find that the costs associated with inferior reference prices are small, and do not outweigh the useful service dark pools provide to market users by providing price improvement and reducing price impact.

We do find evidence, however, that suggests dark pools sometimes experience significant delays in accessing data (a point picked up in the FCA’s recent thematic review on dark pools) from other venues, and it is only the most sophisticated participants that systematically benefit from these delays.

Dark pools and reference prices – fast explainer

What’s in a name? It’s an important question in modern finance, which has a knack for describing its activities in ways that pique public interest.

There is no better example of this than in the naming of ‘dark pools’. Call them ‘trading venues with no pre-trade transparency’ and they suddenly sound sober and of narrow interest. But the reality is dark pools play an important and distinct role in today’s financial world.

So while ‘lit’ venues, like the LSE, display orders to other market participants, in dark pools this is not the case. So participants in dark pools are unaware of the quantity of shares available to trade and at what price, a feature of the LSE and others.

This means dark pools have an advantage in allowing investors to submit an order, without revealing their intentions to the market as a whole and, potentially moving prices against them. The linked disadvantage is that participants do not know whether or not anyone is willing to match their order, creating a degree of uncertainty.

This lack of pre-trade transparency also means that most dark pools have to rely on ‘non-dark’ venues to determine execution prices. In practice, this requires running data feeds from those venues in order to set prices in the dark pool. Two forms of delay, or ‘latency’, in obtaining these prices exist: processing and transmission latency.

Processing latency exists because of the time it takes the hardware and software to calculate and disseminate the data. Transmission latency exists whenever the dark pool is physically located away from the lit venues because it takes time for the information to travel from one to the other.

Substantial levels of latency can cause problems for the reliability of prices, causing different outcomes for participants of varying abilities to observe and control it. This can undermine a perception of fairness and market integrity in these markets.

This article was co-authored by Sean Foley, Assistant Professor at the University of Sydney, and Thomas Ruf, Assistant Professor at the University of New South Wales. Our thanks go to both.