New FCA Research Paper Concludes That HFT’s Do Anticipate Order Flow


A new paper from the UK’s Financial Conduct Authority (FCA) titled “Are High Frequency Traders Anticipating The Order Flow?” has been making the rounds and causing quite a controversy.  The amount of tweets and activity coming from the pro-HFT crowd made us very suspicious of this paper and we were pretty sure that we were not going to like it.  But a funny thing happened during our review of the paper, we read past the headlines and found that the paper is actually an excellent piece which relies on unique data to draw some expected conclusions.  Here is our take on the paper:

The Questions

The authors try to answer two questions:

–  Do HFT’s exploit their small (milliseconds) latency advantages to anticipate orders arriving in very quick succession at different trading venues from other market participants?

–  Can HFTs anticipate the order flow over longer timeframes (seconds or tens of seconds)?

The Data

The authors state: “We use a novel dataset with full order-book data on 120 stocks traded on lit venues in the UK for the year 2013.”   While the data is only from lit venues, it is robust data that we normally don’t see in HFT studies:

The data include information on the date, time (at the millisecond level), direction, price, quantity and information on the order type. Furthermore, our data have a major advantage compared to most other datasets used for research purposes: we can observe the identity of the member of the trading venues behind the event, i.e. the order book is not anonymised.”

When reviewing an academic paper, we first always look for the source of the papers data.  This paper gets an A+ for excellent data.

The Results

The authors state:

“We do not find evidence that the first behaviour is occurring systematically: there is no evidence in our sample that HFTs can “see the true market” and trade in front of other participants at a millisecond frequency.”

While HFT proponents from around the globe have already used the above statement to justify their existence, it’s important to note a few facts:

1) The UK stock market structure is not the same as the US market.  The authors clearly state that “the regulatory set-up makes it more difficult to predict where orders will be routed, compared to the US market.”  The authors are referring to the fact that there is no “order protection” or “trade through” rule in the UK.

2) The authors only looked at “lit venues”.  The problem of latency arbitrage in the US is partially driven by stale pricing in dark pools so we were not surprised they couldn’t find many millisecond issues.

While HFT proponents were quick to point out the millisecond conclusion, most of them stopped short of mentioning another conclusion from the paper:

“We do find patterns consistent with HFTs being able to anticipate the order flow over longer time periods (seconds and tens of seconds). This is particularly true for those market participants that adopt “pure” non-HFT strategies (i.e. those participants that do not use low-latency technology).”

Graphically, this is how this anticipatory behavior looks:


The authors are referring to broker algorithms that use predictable behavior like VWAP algos: 

“Furthermore, when looking at specific non-HFTs (something that market participants themselves cannot do as the orders they see are anonymous), we find evidence that a number of them send some of their orders in a somewhat predictable way, by sending them a few milliseconds apart on a regular basis.” 

Both conclusions of the paper make sense to us:  

1) Since the authors are only looking at lit pools and not dark, they are not able to identify millisecond latency arbitrage that happens when a dark pool prices off the slow consolidated quote generated by the SIP (of course, this would not occur anyway in the UK since they don’t have a SIP or Consolidated Tape).  

2) We have often opined about the predictability of broker algorithms and routers due to various factors like cost, ownership of dark pools, reciprocal relationships, etc.  Based on unique data which includes customer identification, the authors have confirmed that HFT’s can spot and take advantage of poorly designed algorithms and routers that leave a predictable path in their wake.

Overall, we believe this is an excellent paper that is a must read for all market structure observers. We are even going to go so far as to recommend that it be added to the next version of the RT Leuchtkafer Market Structure Bibliography.

A Notable Reference

We were happy to see that we were referenced in the paper alongside Michael Lewis.  The authors noted:

“A specific allegation made by Arnuk and Saluzzi (2009), and popularised in the bestselling book Flash Boys by Michael Lewis, is that HFTs prey on other market participants and only intermediate trades that would take place anyway. These authors claim that by exploiting their speed advantage, HFTs can predict when orders are going to arrive at different trading venues and trade in advance of slower traders; in effect, they view them as a tax on trading paid by other participants.”