When Michael Lewis published the book Flash Boys, a heated and highly-mediatic debate started on whether high frequency traders (HFT) were front running other investors by arriving faster at the available quotes - a form of latency arbitrage. Even a new exchange, the Investors Exchange (IEX) - led by Brad Katsuyama, was founded on the basis of this idea, i.e. under the premises that for markets to be fair such latency advantage should be eliminated. On the other side of the debate, many rose up to advocate in favor of HFT and the exchanges.
What is Michael and Brad's thesis?
Firstly, some context:
- In the US, the BBO that is publicly disclosed is an aggregate of the BBO of all US exchanges (apparently 13, when Flash Boys was written), and is called the National BBO or NBBO;
- Additionally, in the US, whenever a trading order is submitted to one exchange, if it is larger than the volume available at the NBBO in such exchange, it will be redirected to other exchanges so that it is fullly traded at the NBBO, i.e. at the best price possible across all US exchanges;
- Whenever one order is submitted, it takes time (miliseconds) to go from the trader's server to the exchange's server, and such time is as higher as longer it is the distance between the two points;
- HFT trading firms pay (huge sums of money) for co-location, i.e., to have their servers placed the closest as possible from the exchange's server, in order to minimize latency.
Michael and Brad argue that whenever one large market order (or limit order priced at the NBBO) arrives at one exchange, HFT in co-location can react and immediately send orders to the other exchanges and arrive there miliseconds before the original order. By the time the original order gets to the other exchanges, the NBBO has already changed by one or a few cents - as HFT removed all orders at the original NBBO (by cancelling them or buying against them), making the traditional investor to trade at less favorable prices against the HFT.
EXAMPLE:
To counter that, IEX implemented a speed bump - a mechanism to slowdown HFT. In particular, IEX delays by 350 microseconds the sending of execution confirmation messages – meaning that when the HFT learn about an execution, it is already too late for them to send orders to the other exchanges.
What do the critics of this thesis have said?
Some argue that Michael and Brad's claims are simply false, a pile of nonsense that make part of a deceiving marketing strategy focused on raising business to IEX through the demonization of HFT-friendly exchanges.
Others argue that, even if Michael and Brad's thesis is correct, the net impact of HFT in trading costs is very advantageous. I.e., the few cents lost to HFT latency arbitrageurs are insignificant compared to the benefits that HFT have brought over the recent years: rise in liquidity, sharp decrease in bid/ask spreads and increased market making activity.
Many say that speed advantages have always existed in financial markets. For example, before computerized trading, traders who paid to be at the exchange's trading floor were able to react much faster to information than other traders. Similarly, today, investors that are willing to pay will have acess to real-time market data, while other investors can only see what is happening in the market with a (15 or 20 minutes) delay. Additionally, even if a traditional investor has access to real-time market data, it will never be as fast to react to a earnings or macro indicator release as an HFT. These and many other speed advantages have always existed and have never been made illegal.
Others say that the idea of "an HFT buying ahead of a traditional investor and selling him back at a higher price" is not feasible in a relatively liquid market. For such to be possible, there cannot be any order standing at the second possible level of prices: if the traditional investor sends an order to buy at 10,54$, the HFT can indeed buy all the volume available at that price, but then, if it places an order to sell at 10,55$, such order will go to the end of the line and thus will likely not be traded - as orders are usually fulfilled on a first come first served basis.
Another view is that what is happening is not front-running, but a legitimate form of rapidly-adjusting market making: when high frequency market makers see an increase in the buy interest in one security at one exchange, they rapidly adjust their quotes for such security at all exchanges – note that this view is compatible with the one exposed in the previous paragraph.
Finally, one can also argue that, even if it is true, it is not really front running, because the HFT do not know whether the orders will be directed to the other exchanges, they are just guessing. It may well be an easy guess – i.e. with a high likelihood of getting it right – but it is still just guessing.
So... what is the truth after all?
Well, it hard to say... Only by doing a deep on-site investigation complemented with frank insights from top HFT experts would allow us to give a proper answer. However... It seems that at least part of Michael and Brad's thesis was true, as several US and European exchanges implemented or intend to implement speed bumps – though mostly on specific products or small venues... Here goes a list of such venues (source: WSJ), ordered by implementation date (yes, you are seeing correctly... IEX was not the first...):
- (2013) ParFX;
- (2013) EBS Market;
- (2013) IEX;
- (2015) Aequitas NEO-N;
- (2015) TSX Alpha;
- (2016) Refinitiv Matching;
- (2017) Eurex, on some products;
- (2017) NYSE American (NYSE MKT);
- (2018) Nasdaq's Midpoint Extended Life Orders;
- (2018) IntelligentCross;
- (2019) Moscow Exchange's USD:RUB FX market;
- (2020E) London Metal Exchange, on gold and silver futures;
- (2020E) ICE's US Futures market;
- (2020E) CBOE EDGA.
Note that these speed bumps, even though are all intended to slowdown HFT, vary in kind and in goal. For example, while some speed bumps slowdown every investor, others are asymmetric (or discriminatory) and slowdown only some type of investors - usually liquidity providers are given full speed. Additionally, there are some speed bumps that have a constant length, while others have a variable random length.
Comments
Post a Comment