LORD COPPER: Knight Capital and the perils of algo-trading

The big financial scandal this month is slightly different from the normal business of rigging markets and uncontrollably large positions.

The big financial scandal this month is slightly different from the normal business of rigging markets and uncontrollably large positions.

Instead of crass overtrading, blatant lying about rates or blowing the company’s capital on eurobonds, this time we have a bank of quiet boxes with blinking lights running amok and threatening to destroy their creators.

Knight Capital – a specialist marketmaker on the NYSE – suffered a catastrophic glitch in its algorithmic trading program that cost it about $440 million in about 20 minutes on August 1.

The first thing we have to acknowledge is that in today’s terms, $440m is peanuts. JP Morgan’s London Whale and MF Group’s Jon Corzine easily put that amount in the shade, as have numerous other oh-so-clever traders of recent times.

But this situation is a bit different. Like the “flash crash” of May 2010, there isn’t an individual, or a group of individuals, at whom we can point the finger.

This time it was the machines that caused the problems. And in that, it was completely predictable. Go back to 2010 and you’ll find my own comments about unfettered algo trading, as well as those of people far more expert than me.

It is not quite like Robert Harris’s The Fear Index (although what a great advert for the book!) because we are not talking about a self-developing artificial intelligence gradually learning to program its own trades. The problem arises in a different way.

The algo traders are, as far as I can see, trying to achieve one of two things, which may frequently overlap. They are either aiming to get on to a market trend, or they are looking to exploit an arbitrage (or series of arbitrages).

Fetish for co-location
To achieve either of these requires speed – hence the fetish for co-location – since without it, in one case the exploitable arbitrage may disappear and in the other the trend may have run away before you can grab it.

The clever bit of the programming is therefore concerned with identifying opportunities, rather than the more prosaic business of executing the strategy.

To do that, they use a panoply of electronic trading techniques and tools – pinging, fishing, iceberg-seeking, sniffing and all the rest – to discover what other traders are trying to do.

This results in huge volumes of orders swilling around the market with potentially serious consequences.

Normally this is not an issue, because markets are mostly two-sided. However, the risk is on those occasions when the market is one-sided, because then the trend-following algorithms will rush into any gap they find and push prices way beyond reason.

And because they are trend-following, they will continue to do so, with potentially disastrous consequences. The flash crash and the Knight Capital issues were contained (apart from the consequences for Knight itself), but should we be betting that this will always be the case?

That is a practical issue – the potential for the market to be swamped and subsequently physically struggle to manage the weight of trades. But what about what one could call the philosophical element?

Driven by robots
We applaud the use of robots in manufacturing industry. They eliminate some of the drudgery of production-line work and offer significant advantages in terms of cost and consistency. It is also almost impossible now to envisage a world without computers, electronic communication and GPS aircraft and ship tracking, developments that we now take for granted.

So should we regard the computerisation of financial services – or more particularly, of financial market trading – as somehow different? The answer is that we probably should. There is an inherent qualitative difference, because the risk in financial market trading is not restricted to the participants.

If the only risk of loss sat with the traders and/or market makers themselves, then clearly there should be no special pleading. But it doesn’t. We all have a vested interest in the integrity of financial markets, because the security of our entire system depends on their stability. That’s why we have regulators for investment products and firms, with a brief to keep the markets stable and transparent.

Once the programs have been launched, algorithmic computers trading with no human hand on the tiller do not satisfy that brief.

Circuit-breakers are a step in the right direction, but the wholesale handing over of markets to pre-programmed black boxes should be causing the regulators sleepless nights.

As an aside, the LME has so far escaped this problem. Best not to be complacent about it, though.

Lord Copper

What to read next
Fastmarkets invited feedback from the industry on the pricing methodology for cobalt hydroxide, min 30% Co, inferred, China, $lb, via an open consultation process between May 4 and June 1, 2023. This consultation was done as part of our published annual methodology review process.
Fastmarkets will discontinue its consumer buying price assessments for machine shop turnings in the Cleveland and Pittsburgh markets effective Tuesday June 6.
Fastmarkets has decided to proceed with the launch of a new European low carbon ferro-chrome price covering material with lower chrome content.
Fastmarkets invites feedback on a proposal to increase the publication frequency of non-exchange-deliverable equivalent-grade (EQ) copper cathode premium, cif Shanghai, from once every two weeks to once every week.
The outlook for North American steel scrap prices has headed further into bearish territory ahead of June’s trade, with prices for all grades expected to fall again after a round of across-the-board decreases in May
Fastmarkets is inviting feedback on a change of publishing time for our ferro-chrome price in the Chinese domestic market as well as ferro-chrome import prices in Japan and South Korea, to 5-6pm Shanghai time from 2-3pm London time.
We use cookies to provide a personalized site experience.
By continuing to use & browse the site you agree to our Privacy Policy.