I had a short chat with Nassim Taleb this morning about his new paper with Charles Tapiero, entitled “Too Big to Fail, Hidden Risks, and the Fallacy of Large Institutions”.
There’s a great deal of mathematics in the paper, which is full of equations and greek letters, but the gist of it is explained in pretty plain English:
Societe Generale lost close to $7 Billions dollars, around $6 Billions of which came mostly from the liquidations costs of the (hidden) positions of Jerome Kerviel, a rogue trader, in amounts around $65 Billions (mostly in equity indices). The liquidation caused the collapse of world markets by close to 12%. The losses of $7 Billion did not arise from the risks but from the loss aversion and the fact that costs rise disproportionately to the size of the bank…
Consider the following two idealized situations.
Situation 1: there are 10 banks with a possible rogue trader hiding 6.5 billions, and probability p for such an event for every bank over one year. The liquidation costs for $6.5 billion are negligible. There are expected to be 10 p such events but with total costs of no major consequence.
Situation 2: One large bank 10 times the size, similar to the more efficient Société Génerale, with the same probability p, a larger hidden position of $65 billion. It is expected that there will be p such events, but with $6.5 losses per event. Total expected losses are p $6.5 per time unit – lumpier but deeper and with a worse expectation.
In other words, small mistakes we can live with. Large mistakes we can’t, because when a mistake the size of Kerviel’s is unwound, the costs are enormous — not only to SocGen, which lost upwards of $6 billion, but also to all shareholders globally, who saw the value of their holdings marked down by trillions of dollars thanks to the effects of SocGen’s enormous and chaotic forced unwind.
The lessons here are broader, and apply to the practice of M&A more generally: when industries consolidate, there might well be economies of scale — but at the same time tail risks increase. What happens when a massive amount of technology outsourcing is consolidated in Bangalore, or computer-chip manufacture is consolidated in Taiwan? Efficiency rises — but so does the risk that one disastrous event could have massive systemic consequences.
The solution for banks is relatively simple: just put a cap on their size. (I’ve been suggesting $300 billion.) What’s the solution for other industries, which also naturally tend to consolidate and cluster? I’m not sure, but in an increasingly interconnected and just-in-time world, the risks are greater than ever.
Update: A couple of good comments from dsquared; the first points out that the paper ignores problems of correlation, which is true. But as Rick Bookstaber is more than happy to point out (and Taleb is no fan of Bookstaber), correlations tend to pop up in the most unlikely places, and in general they just make everything more dangerous — not only the systemic risk of lots of small players failing at once, but also the systemic risk associated with one large failure. So add in correlations to Taleb’s model and I think it only becomes scarier.
Dsquared then asks whether we really want to move to a world of “Fewer SocGens, More Barings”. My memories of the systemic implications of the Barings collapse are hazy (maybe John Gapper or even Nick Denton can help out here), but I think the answer might well be yes: while the Barings collapse was bad for Barings, it didn’t have the kind of negative externalities that we saw with Kerviel. But I might be wrong on that front.