Value at risk, and what it does not see

Graduate · enrolled learners

This lesson opens with Sumitomo and Yasuo Hamanaka, 1996.

What happened

Yasuo Hamanaka was a copper trader at Sumitomo who, over years, built enormous positions in the copper market, at times large enough to influence the world price, in what amounted to an attempt to corner it, and much of this activity was unauthorized and hidden from the firm. When it finally unraveled in 1996, the losses came to about 2.6 billion dollars. A conventional risk measure, looking at the normal day-to-day fluctuations of the positions a firm knows about, would have seen little unusual, because it was never shown the true scale of the concealed book and because its whole framework assumes the trader is a small participant in a large market rather than the participant whose own size is holding the price up. When a position is that large, the risk is not the ordinary wiggle of the price; it is what happens to the price when the position that has been supporting it is finally unwound, and a value-at-risk number computed on normal fluctuations sees none of that. The measure was not wrong on its own terms; it was answering a question that had nothing to do with the real danger.

The decision point

Value at risk and similar measures estimate how much the positions a firm knows about could lose under normal market fluctuations, which means they are blind to two things that matter enormously: positions they are not shown, and the danger of a position so large that the trader is no longer a price-taker but is holding the price up themselves. Sumitomo is the case: Hamanaka's concealed, market-cornering copper book lost about 2.6 billion dollars, and a conventional risk measure saw little wrong because it was never shown the true position and because its framework assumes a small participant in a deep market, not the participant whose own size is the market. This is the central limit of a risk number: it is only as good as what it is allowed to see, so it answers precisely the question it is given, the normal fluctuation of the disclosed book, and says nothing about a hidden position or about the price impact of unwinding a dominant one. A risk system that trusts its value-at-risk number as the whole of the danger is trusting that nothing is concealed and that the firm is always small relative to the market, two assumptions that fail in exactly the situations that produce catastrophic losses. So the discipline is to treat a risk measure as a bounded tool rather than a verdict, to ask what it is not being shown, whether any position is large enough that the firm's own size distorts the price, and how a book would behave not in a normal day but in the unwind, because Sumitomo shows that the most dangerous position is often the one the risk number cannot see, either because it is hidden or because it has become the market the number assumes it is only a small part of.

Recorded loss

$2,600,000,000

Orientation and Year One are open: anyone can read them without an account. From Year Two onward the lessons are for enrolled learners, because progress through the later years only means anything if it is tracked against a record.

It is free. We do not sell the list and there is nothing to buy at the end of it.