The latest Spotlight Review from the FICC Market Standards Board got me thinking about the waning ethos of “consenting adults” in markets, as well as about a piece of industry infrastructure that could be very useful in monitoring activity in FX markets especially.
The Review from FMSB looked at the use of machine learning in market surveillance. It describes the issues, as our report lays out, but what struck me was the section that discussed market participants use of quote or transaction activity. I found it interesting that the paper stepped into spoofing territory by noting that there was a risk of market manipulation by using quote or transaction data in one market or benchmark to influence the price of other correlated markets.
This provides an excellent example of how the correlated nature of markets can raise this risk and, at face value, I see nothing wrong in frowning upon this practice, especially if someone is, for example, bidding a commodity market up to sell AUD. They are a creating false market, it could be argued, and therefore should be stopped.
But what about when the markets to which they are posting interest are firm liquidity, i.e. no last look?
Another example used by the paper noted how placing a bid and/or offer on a primary venue and then trading elsewhere was also a conduct risk. Let’s be clear, it can be – but should it be in terms of the FX market?
I ask the question not because I think it is good that people can play games, but rather because the primary FX venues support firm liquidity and have no last look, which means you can try bidding up, for example, the Aussie, but there is a real chance you could get clobbered in the meantime. The use of minimum quote lifespans provides additional protection.
The problem with repressing such practice remains the core issue of “intention to deal”. It can be argued that placing a price on a firm central limit order book, with an MQL especially, very much represents an “intention to deal”. The FMSB paper observes how this risk is heightened during times of relative illiquidity in markets, I would humbly suggest the problem is more about periods of low trading activity – the less trading activity, the less likely someone is to get hit on their price, hence they may be tempted to play around with the markets.
This quandary has evolved because of the use of machines in pricing and trading FX markets – quite simply they have to believe the data they are fed and that makes them more vulnerable to people playing games than perhaps a human is. Don’t get me wrong, both can be misled, but it seems to me that the machines have lower thresholds of credibility – a human will question a price outside the market, a computer, if it is able, would just hit it. The complication is so many are pricing markets, so therefore their first act is not to hit the outlier, but to take it into account for their own price streams.
Whether this should be something to spend countless man hours and technology resources on combatting is not the straightforward question it seems, however, not least because of the aforementioned “intention to deal” challenge. Just as we should have responsibility and accountability for wrong decision, so too should we assume the same mantle regarding how we formulate a price. The FMSB paper does not overtly state that the actions described herein are wrong, but there is a strong hint they are seen that way. Should we, however, protect people who are claiming to be liquidity providers but who are unwilling to pay out for premium data or to build an effective and robust data framework?
The Review itself notes how data users should not “be overly reliant on quotes on any single primary venue especially when it contributes a small percentage of market volumes in that asset class”, but in FX that is not as easy as it sounds. I accept that some of those data users are going to be price engines, and if the coders choose to focus in on too narrow a subset of data sources, it is their fault, but as far as FX is concerned there is a limited subset of firm liquidity CLOBs or ECNs. Clearly the publicly expressed desire of LMAX Exchange to get deeper into the data business, as well as Cboe FX launching a firm CLOB, indicate that at least two firms feel the depth and breadth of data sources is insufficient.
Reviews like this from FMSB are valuable in that they plainly lay out the issues and, like the FX Global Code, remove the “we didn’t know” excuse from market participants’ lexicon. These issues do need to be highlighted, but I also feel that we need to be talking about the complexities underpinning them. In this case, it can be credibly argued that posting a price on a firm liquidity venue, where it can be hit at any time by anybody, cannot be construed as attempted manipulation.
There are going to be occasions, however, where it could be just such a deliberate attempt and it is there that the industry challenge lay and the need for infrastructure arises. What is needed is a central repository of trades on platforms that is available only to regulators and only needs to be accessed when genuine claims of misconduct are made. It does not need to be driven by commercials, avoiding even million, let alone billion, dollar fines should make the investment decision easy enough, therefore it can be a dormant facility – one that is only woken up when required by a designated authority or authorities. I would also, I should point out, include any client-to-client matching mechanisms run by banks or other firms.
I accept this is only a reactive response to misconduct rather than proactive, but the very existence of such a data store, with accurate timestamps (as also recommended in the FMSB paper), would surely give potential miscreants pause for thought. It would also provide regulators with the ability to access and analyse a broader subset of the market and rebuild order books and, importantly, last looked books. That way if we have people deliberately trying to spoof markets they will be found out – as an added bonus any flash events can also be analysed in more depth.
More pertinently such an infrastructure would be a valuable resource for surveillance staff, and as the FMSB paper makes clear – under the right circumstances, the advance of machine learning makes studies of such conduct much easier and quicker. We just need the data in one place and permit its use on a “need to” basis only.