Over on Slashdot [1], there was a long debate about the NASDAQ market deciding to cancel a dud series of orders that sent shock waves through the US trading markets, based on this NYT article .
The core problem underlying this event is that the trade - and all trades - are not real. That is, they are promises to make good on the orders, not trades in and of themselves. It's like imagining that everyone in the system, including the system itself, works with credit cards with really high limits on them.
This leads to two problems. Firstly, the trader might max out his creit card, and not make good on a promise. That's called settlement risk [2].
Secondly, any other party might decide to stick their wick in and futz with the order, because their own credit card gets maxed out. In this case, the other parties included
- the computer program that sent the repeating series of unreal orders,
- NASDAQ which cancelled these unreal orders after other traders had made yet further unreal orders, and
- Archipeligo which re-opened without alerting traders to the potential unwinding of the unreal state.
That's called operational risk. It's what happens when the systems make mistakes on you, and leave you with the mess to clean up [3].
Mistakes happen [4]. As time goes on, routine mistakes get caught and exotic ones slip through. Then, as they are discovered, these exotics are turned into routines by new patches. So, over time, more and more complexity marches together with more and more exotic mistakes.
Within any system, mistakes will be caught, or not, as the case may be. This mistake was not caught before it sent the shock waves through the system. Systems-wise, it's fairly clear that most mistakes will be caught, and an occasional one will get through and cause havoc. This will cause a rethink and repatching, which will catch that very mistake next time.
Until the next exotic mistake slips through, which by definition is more exotic, and more damaging [5].
As each ensuing trade is based, one after the other, on the original promise, the result is that the financial system works both as a breeding ground for bigger and more exotic mistakes, and as an amplifier of the exotic mistake. As is intiutively seen by the slashdotters (but glossed over by the financial participants) the original order is now lost in the confusion of responses by various agents trying to establish the "right" thing to do.
Say hello to systemic risk!
This is the risk that a shock cascades through the system, causing the collapse of entities unrelated to the original cause, and thus further propogating and amplifying the original shock.
Systemic risk in today's financial system rests primarily on the core premise that the whole system is based on promises to pay. It's as if every participant has a credit card, and too many of them are maxed out.
There is an easy solution to this. It's called real time gross settlement, and it means making every trade real. This isn't to eliminate credit, it instead amounts to taking away the credit cards from the systems and institutions in the center. That is, NASDAQ, Archipeligo, etc, who are not traders, but are part of the system.
Which would then make them into sinks of systemic risk, rather than amplifiers of it. Mistakes would be smothered one step away from the source, rather than cause waves of systemic failures through the global system.
But, don't hold your breath waiting for this solution. Keen students of financial cryptography will know that it is both a real solution, and a cost effective one. In fact, too cost-effective, as it will reduce the opportunities to make money by those with lots of credit cards.
iang
[1]
Slashdot discussion
[2] definition: "Settlement risk is the risk that a settlement in a transfer system does not take place as expected. Generally, this happens because one party defaults on its clearing obligations to one or more counterparties." The BIS calls this credit risk.
[3] It's also related to Herstatt Risk, after a German bank was suspended leaving counterparties sandwiched between such one-sided trades (see URL in [2]), and to legal risk, as some contracts got breached, but others got enforced, arbitrarily.
[4] vadelais reports on slashdot:
"Mistakes have been made in market trading before by other companies.
In May last year, London's FTSE 100 index dropped by more than 2%, after a trader typed 300m, instead of 30m, while selling a parcel of shares.
In 1998 a Salomon Brothers trader mistakenly sold 850m-worth of French government bonds by LEANING ON HIS KEYBOARD.
And at the end of 2001, shares in Exodus, a bankrupt internet firm, jumped by 59,000% when a trader accidentally bid $100 for its shares, at a time when its value was 17 cents."
[5] Lawrence_Bird on slashdot suggests an advantageous and more exotic mistake:
"To this day, I do not believe the order was entered in error. I believe a hedge fund or other firm had a need to buy, and buy large. What better way to get filled then to trigger a large move in a closely correlated market (Dow futures) where you know the majority of the trades will be cancelled, and just sit on the bid in the other market (CME) as people panic sell. And do not rule out collusion."
Posted by iang at December 7, 2003 12:30 PM | TrackBack