One of the brilliant rhetorical devices deployed by Michael Lewis in his public interviews about Flash Boys is referring to some HFT practices as “legal front running”. By inserting the “legal”, he takes off the table any accusations of lawbreaking. Nevertheless, defenders of HFT seem to cling to legalistic definitions of “front running” in making their arguments that HFT algos do not front-run. As an example, let me take on Rishi K. Narang, founding principal of T2AM, who denounced the study authored by Elaine Wah and myself in a recent commentary published by CNBC:
Some allege that HFTs front-run other players’ orders because HFTs have access to direct feeds from various exchanges, while “regular” investors generally rely on the SIP-provided national best bid-offer (NBBO), which aggregates all the various exchanges’ order books, but which also arrives with considerable latency (due not to conspiracy, but to outdated technology). As a recent University of Michigan report claims, “[b]y anticipating future NBBO, an HFT algorithm can capitalize on cross-market disparities before they are reflected in the public price quote, in effect jumping ahead of incoming orders[…]”. This is blatantly false.
Narang’s argument is that HFTs cannot front-run because they are not specifically trading ahead of one of their own customers in order to then trade with that customer.
When an order is placed, it takes some time to be reflected in the NBBO. But that order is already in the market before the HFT can see it, even on the direct feed, by definition. HFTs never know what a customer’s order is before it’s in the market. HFTs have no customers.
The distinction between whether the order is from an HFT’s own customer or not would be relevant to a legal allegation of front-running, but that is not in question here (and I certainly would not be qualified to speak on such legal points). The other part of the argument is that the order is already in the market, so there is no front-running. However, a lot hinges here on what we mean by “already in the market”.
The HFT tactic we investigated in our paper is latency arbitrage (what Michael Lewis calls “slow market arbitrage”), in which the HFT profits on a price disparity between fragmented markets, arising due to latency in the NBBO. Suppose an order to sell exists in exchange A at $100.00, but the outdated NBBO ASK quote is $100.02. If an order to buy at $100.01 comes into exchange B (and stays there based on the outdated NBBO), the HFT can arbitrage between the two markets, buying in A and selling in B, pocketing $0.01.
So is Narang arguing that this cannot happen, or that it does not amount to “in effect jumping ahead of incoming orders”? There is in fact much evidence that it can and does happen, so we are probably arguing about semantics of “in effect jumping ahead”. Ultimately what we call it does not matter very much, but I don’t see how one can deny that the HFT is exploiting its information advantage to take the sell order on exchange A before the incoming order at B sees it. The statement that the order at B is “already in the market” is true in a sense, but somewhat misleading. “The market”, as we know, is not a centralized entity, but rather fragmented over a plethora of exchanges and non-exchange trading venues.
Back to Narang:
What’s actually happening behind the scenes may be frustratingly complicated, but it’s not immoral, unethical, harmful or illegal. Nor does it cost the non-HFT anything.
As noted, we are not arguing about legality. Nor did our paper suggest anything about morality or ethics. We did focus on harm, however, and found that indeed latency arbitrage is harmful to market efficiency. Not only does it cost the non-HFT (it’s quite obvious that some transfer of surplus much be at play in latency arbitrage), we found that the non-HFT costs exceed the HFT profits. In other words, it’s a negative-sum game. That is not even counting the costs to the HFTs themselves for achieving their speed advantage.