By Jeff Bacidore, Founder and President, The Bacidore Group
A recent paper by researchers at Dimensional Fund Advisors (DFA) and Harvard Business School document how execution quality varies across inverted and non-inverted venues. Specifically, the authors look at trade performance statistics like effective spread, realized spread, and market impact.
The authors pay particular attention to the realized spread inclusive of exchange fees. The motivation is that seeing differences in gross performance across venues is not sufficient to justify routing since fees vary across venues as well. Rather, to make comparisons “apples to apples”, one must compare the performance net of fees/rebates.
The DFA/Harvard team documents an extremely interesting result: the net costs of inverted and non-inverted venues are “statistically indistinguishable” from one another. In other words, while there are differences in gross performance across venues, the fee differences effectively offset those performance differences. For a trading desk that pays its own fees – either directly via a “cost plus” arrangement or implicitly in its gross commissions – such a finding suggests that the routing decision doesn’t matter. Based on this, the authors conclude that the optimal routing strategy is to simply route orders pro rata to the different venues.
This finding mirrors one of the key results of a prior academic study by Robert Battalio, Shane Corwin, and Bob Jennings (BCJ), which was discussed in a prior post. Like the DFA/Harvard study, BCJ finds evidence that the realized spreads inclusive of fees are similar across venues. But BCJ note that realized cost of filled limit orders provide an incomplete measure of trading performance, as it ignores potential differences in fill rates across the venues. And fill rates are critical when measuring “all in” costs.
For example, consider a trader working an order that must be completed. Any quantity that cannot be filled via non-marketable limit orders must eventually be priced more aggressively to ensure completion, e.g., by converting unfilled orders to marketable orders at the “end time”. If use of one venue type makes it more likely that an order will be left unfilled and ultimately converted to a marketable order, then a trader is no longer indifferent between the two venue types. Instead, if the inverted venues have higher fill rates, traders would prefer the inverted venue since they not only provide the same net costs for their non-marketable fills, but they also have higher fill rates. And these higher fills rates make it less likely that the trader will eventually have to rely on spread-paying marketable orders to complete any unfilled orders. BCJ show that, once “clean up” costs are considered, the routing decision does matter.
And one other finer point on these studies. While the results suggest that routing doesn’t matter conditional on filling, the indifference may be due to the fact that the orders in their studies may already have been “optimally” routed in order to maximize performance. Inverted venues, for example, may be used only when the performance gains justify the cost. Consequently, while one may find that the average costs across the two venue types are similar, this does not necessarily mean that the marginal costs are identical at all points in time. In fact, they almost certainly aren’t, as most smart router developers would attest to.
Nevertheless, the broader message of the DFA/Harvard study is an important one, namely that the market has adapted to differences in fee structure so much so that the realized performance across venues appears to be statistically indistinguishable on average from the buyside perspective.
 The effective spread is defined as the premium/discount paid relative to the prevailing quotemidpoint. The realized spread is defined the premium/discount relative to a future quote midpoint. Market impact is defined as the change in quote midpoint following the trade.