Professor Larry Harris of the University of Southern California, author of the seminal text “Trading and Exchanges: Market Microstructure for Practitioners”, has many times proposed that exchanges randomize orders on arrival. Professor Harris’ purpose is to reduce the perceived negative effects of High Frequency Traders. His December 27, 2012 article in the Financial Times summarized his idea.
To my knowledge, none of the equities exchanges to whom Professor Harris directed his thoughts have taken up Professor Harris’ idea.
Two forex platforms, ParFX and EBS, are implementing similar algorithms. Their aim is to lessen or eliminate the advantage of faster automated traders. This is especially important in EBS’ case as a key value proposition over single bank platforms is that an EBS user can ‘make’ or post, not just ‘take’ or remove prices. If a user trades on a single bank platform, he can only take prices, i.e. hit the bid or lift the offer, and must therefore pay the bid-offer spread. By joining the bid or offer on an FX ECN such as EBS, that is to say making prices and leaving passive orders at the bid or offer level, a user assumes risk that the market might move away but, if the market moves towards him, he may trade without paying the full or any spread.
The problem, which ParFX hopes to avoid and EBS hopes to change, is that when the market moves away from a maker, fast automated traders can take prices faster than their makers can remove them. In other words, if a maker becomes slightly off-market there is a high risk of adverse selection of passive orders or, to use the vernacular, ‘getting picked off’. Traders have therefore become reluctant to commit their order book to ECNs. By randomizing orders, ParFX and EBS hope to mitigate this issue as there will no longer be as great a speed advantage.
In principle, randomization of orders sounds very plausible and the direct consequence will be to reduce adverse selection. However, with any microstructure change there are always unintended consequences. I can’t forecast them all, but I see at least two possibilities.
Firstly, automated traders generate a large proportion of the volumes that EBS and ParFX do wish to receive. However, automated traders favor deterministic matching algorithms. For them, a key value proposition over single bank platforms, in addition to the ability to make prices, is predictable matching algorithms. It is true that EBS’ architecture of three matching engines on different continents has never been wholly predictable as you never know if you order is being matched locally or in a distant region with associated transit latencies. However, randomizing orders will make the algorithm even less predictable and several automated traders have told me that this concerns them and may make cause them to move flow to other ECNs or favor single bank platforms.
Secondly, it may boost single bank platform share indirectly. There are traders whose algorithms are careful not to take prices from bank platforms and EBS at the same time. Doing so would take the liquidity from EBS that the bank platform may be relying on to cover. This is another form of adverse selection – this time of the bank. If users do not have to be so concerned about hitting EBS and a bank at the same time, then their algorithms will do so where they would previously restricted trading to ECNs. It is true that it could boost EBS and ParFX for the same reason, but I suspect the issue was more a reason for users to limit their trading on banks, to preserve important bank relationships, rather than a reason to limit trading on ECNs.
A better way?
A more direct solution to the problem of adverse selection of passive orders would be to improve means for traders to cancel orders quickly. This might include more sophisticated stop-loss order types or other order types. If you know you can remove your order before you are picked off, you would be more confident of committing your order book to an ECN.
Phil Weisberg, global head of FX at Thomson Reuters, said at a recent conference that he has always thought of randomization of orders “as a bug, not a feature”. The jury is out on which it is; we will learn from actual practice.