Dark Pools: What Lies Beneath
Bank of America Merrill Lynch’s James Wardle takes a look at adverse selection in public dark pools.
Dark pools provide sources of non-displayed liquidity facilitating anonymous matching between counterparties helping reduce market impact costs and minimise information leakage. Marketshare of pan-European dark volume has risen dramatically over the past few years hitting a year high of 2.5% in May 2010, see Figure 1. As executed volumes in the dark continue to rise and high-frequency volumes increase, uncertainty over fill-quality is at the front of everyone’s mind — at what cost does accessing dark liquidity come?
High-frequency (HF) participants (e.g. hedge-funds, market makers, etc.) have an investment horizon that is typically much shorter than the traditional long only institution. It is the coming together of these two distinct flows that exacerbates the occurrence of adverse selection. The use of short-term alpha forecast models by HF traders and other market participants to try and opportunistically execute at temporary lows for buys and highs for sells means that the opposing counterparty may be liable to early execution at local price maxima (minima) for buys (sells) and thus becomes the victim of adverse selection - see Figure 2.
Adverse selection can also arise from gaming. This is where the presence of a large block is detected in the dark by a number of ping orders (small orders looking for size), after which the informed trader waits for a temporary adverse price spike before they send a large order to consume the liquidity found in the dark. This results in the informed trader obtaining size at a favourable price to them at the expense of the uninformed trader who again becomes the victim of adverse selection. Across large orders the cost of adverse selection can add up and seriously damage returns.
Post-trade, adverse selection can be identified in two main ways: measuring the performance of the fill to short-term price-movements, and looking for reversion patterns postfill. Short-term price movements are well-captured by using the Time-Weighted Average Mid-price (TWAM) measure; this takes the average mid-price T seconds before the fill and T seconds after the fill. Varying the TWAM time-frame helps us to identify any executions that may have occurred at temporary price spikes. A positive return of the fill price to TWAM implies we have filled at a price better than short-term price movements (positive-selection) and a negative return implies that we have filled at a worse price (negative or adverse selection).