Well Monday’s column put the cat amongst the pigeons – as writing about SNB-Day and re-papering trades always does. I am indebted as always, to those of you who shared and re-lived your experiences, the only pity was we were not in the usual location for the swapping of war stories!
The exception, it is said, often proves the rule, but it is clear to me that even five years on, people have very entrenched views of what is, and is not, acceptable practice when it comes to re-papering trades. On one hand there is what might be termed the “old school” (who, I should add, are anything but long in the tooth, this merely reflects their view) who believe that if someone is dumb enough to sell EUR/CHF at 0.1 then they should pay the consequences. Alternately, there are those who believe that every trade should be re-papered to a rate established by a third party, in the case of SNB-Day, EBS.
Interestingly, very few people sit in the middle and, if I may be so bold as to categorise people, my anecdotal evidence suggests that the more willing you are to take on risk, the more likely you are to sit in the former camp. Regular readers will know that I tend to sit in that camp as well, mainly because I think algo systems should have adequate checks on them to ensure that silly rates are simply not hit. It is the lazy approach to business that basically revolves around building or deploying a sloppy piece of technology without checks and balances and if anything goes wrong, rely upon someone else “doing the right thing”.
Why should we protect people who don’t take enough care to put hard limits on their algos? I have previously suggested that a protection against flash crashes would be central bank volatility bands that sit, perhaps, five or 10 big figures either side of mid and fluctuate accordingly. The pace of movement in rates would be the trigger to hold the band in place and thus provide support for, and maintain a, stable market.
In reality, this would be less necessary if market participants using algos had simple barriers per currency pair. In the case of EUR/CHF, we all knew that there was a huge build up of positions and that the market would, if 1.20 broke, fall sharply. If an algo was coded, on a daily basis maybe, not to trade below 1.00 then we would not have had the problems we had. It means that a human would actually have to OK the algo to trade or hit the sell button themselves. If they do so, then it’s all on them.
This would also slow the market down during flash events which could help alleviate matters, if for no other reason that it would give sensible traders – people who are willing to take risk – time to step in and bid/offer accordingly and bring some stability to proceedings.
I was chatting about this to someone earlier this week after Monday’s column and they immediately countered with ‘what about the Cable flash crash?’ Indeed, what about it? Cable dropped 12 big figures and rebounded seven or eight. It was at a rubbish time of day liquidity-wise, therefore it was what it was. We were in the midst of the Brexit drama remember, so a sharp drop in the value of the pound was always a threat, so again, stick a 10 big figure limit on the algo and while people would have been selling at 1.1600, that is not that far off the actual low of 1.1490 – people would be able to do their business as suited them. My view only really comes into play when things get silly (and on Brexit night, the higher risk could have meant a price limit 25 big figures away for example).
I come back to my central point about allowing people to get away with a weak control environment – there are, at times, obvious errors when the trade before and the trade after are dozens of big figures away. Fine, it’s obvious there has been an error so re-paper. It is equally obvious to me, however, that the vast majority of those wanting to re-paper on SNB-Day (and other occasions) want redress for losing their heads or having poor technology solutions.
In 2016 I reviewed a paper by eCo Financial Technologies which discussed the use of “last mile” checks in markets to ensure these type of events didn’t happen by accident. Looking at the response to Monday’s column I would argue this type of check is needed more than ever and that providers should be encouraged to provide it and users should be urged to check whether it exists (and go elsewhere if it doesn’t).
People are always going to lose their heads, the problem is the reliance upon technology stops short of actually protecting them when they do. Equally, people are rarely honest enough to acknowledge the probability of someone either panicking, or perhaps not having the skill set to deal with fraught market conditions. We should acknowledge that likelihood and what seems to me to be a fairly simple and manageable tweak to execution algorithms could help solve at least part of the problem.