Dark Pools “Uniquely Susceptible to Latency Arbitrage: Paper

A new Working Paper published by the Bank for International Settlements looks at the behaviour of high frequency traders (HFTs) in dark pools and says the mechanism is “uniquely susceptible” to latency arbitrage because they rely upon outside sources for the reference pricing and any trades are pegged to the stale reference price.

The study argues that the process by which changes in reference prices create an ‘old’ current price and a ‘new’ price once the update is propogated to the dark venue, creates a situation that does not exist on lit venues. It observes that while prices may vary between lit venues, some uncertainty remains whether a quote update in one market will be followed by convergence in another or a reversal in the original market. “By contrast, in situations where the dark pool reference price (and its pegged dark limit orders) are stale, fast traders aware of the new price can engage in near risk-free latency arbitrage by trading at these now stale prices at the expense of other market participants,” the authors state.

The study finds that “conservatively”, 4% of dark trading in a UK-based sample occurs at stale reference prices, which translates into an average cost of 2.4 bp, and an annual toll of GBP 4.2 million.

The authors note that by construction, not much is known about the identity of traders in dark pools, however they add their results provide some insights into who is active in dark pools and who provides and consumes liquidity. “We show that it is almost exclusively high frequency trading firms that are on the benefiting side of dark pool executions at stale prices (between 96 and 99% of the time),” they write. “Furthermore, stale trading does not happen at random; rather, in the vast majority of cases (83%), it is driven by aggressive dark orders from HFTs.

“This is even more notable given another of our findings: HFTs as a group almost never provide marketable liquidity in the dark (although they do post non-marketable limit orders in some stocks),” they continue. “These data points, taken together, imply very strategic order submissions by HFTs that are mostly liquidity consuming, including those instances where they take advantage of stale reference prices.”

The paper looks at design changes to help combat the issue, namely speed bumps and batch auctions. It notes that as long as order entries are held for a certain length of time, as long as the reference price updates quicker than that hold time, then latency arbitrageurs are thwarted.

Similarly, frequent batch auctions (often sub-second) also mitigate the speed advantage by allowing slower traders to check the last auction results before submitting a new order. A third method could also be to automatically cancel any stale orders after a certain length of time.

The paper concludes by noting that as the automation and speed of markets has increased, so has the importance of latency in markets, and in particular, latency in reference prices. It observes that two event studies in the paper show that removing the ability for aggressive participants to race the market data feed by making the timing of dark executions unpredictable (through random uncross mechanisms and batch auctions) is effective in reducing the cost of liquidity provision.

“Our evidence has significant policy implications,” the authors conclude. “Reference price determination is important to the fairness of dark pool pricing. Regulators should focus on the adoption of alternative market designs like random uncross features and batch auctions. Indeed, it may be the desire to avoid these ’sharks in the dark’ which has led to the endogenous adoption of speed bumps and batch auctions by an increasing number of exchange venues around the world.”