It is beyond doubt that the past crisis-laden few years have not been good for the concept of statistical correlation when it comes to financial markets.
Simply stated, what was supposed to be correlated in a certain way turned out to be correlated in a completely different fashion. This applied particularly forcefully to mortgage securities, naturally: while those convoluted collateralised debt obligations directly responsible for the mayhem typically assumed that the trashy US housing-based assets that made them up were slightly correlated (ie the probability of their defaulting concurrently was characterised as supremely small), in the end they did all go down in sync (ie the true default correlation turned out to be closer to one in practice than in theory).
Such turn of events caused huge losses for those who had chosen to believe the original, complacency-building estimate. CDO investors lose their shirts as soon as a number of the underlying securities go underwater; the models said that such losses could not take place, given the assumed low correlation. Clearly, the models were wrong.
We have seen correlations play nasty games in finance before. Usually, this involves assuming different asset families (subprime credit, equities, foreign exchange) will not move in tandem, only to later witness how they all sink together. It happened during the Asian and Long-Term Capital Management crises in 1997-1998 and it happened during the latest meltdown. Punters who thought themselves “well diversified” were left contemplating how everything they owned was being lambasted in unison. Putting faith in ex-ante correlations has had unsavoury effects for many.
One could then be tempted to conclude that we should stop paying heed to the notion of correlation in the markets. Co-dependence between assets is too indecipherable and undependable a concept; what seemed to work in the past is rendered untrue soon after, what was naïvely categorised as causational is revealed as spurious or fortuitous. Perhaps there’s no such thing as real, dependable correlations in the chaotic and unruly market universe.
Or is there? It has recently been revealed that uber-famous mega-successful hedge fund Renaissance Technologies does in fact seem to make money by trading correlations. Ren Tech (consistent 30 per cent-plus annual returns for the past two decades) would dig deep into data sets, detect co-movements among certain financial variables, accept some of those co-dependencies as trustworthy, and trade on their future repeat. If this story is accurate, and given the fund’s miraculous performance, there is no alternative but to conclude that hidden in the tumultuousness of market activity there are in fact reliable correlations to be found.
As important as that discovery would be the way Ren Tech apparently goes about it. Not by religiously abiding by arrogant mathematical predictions and theoretical precepts, but by humbly analysing and interpreting what the market does. Instead of acting based on preconceptions as to what should work, Ren Tech would non-judgmentally accept what market activity tells it. If two assets appear to be cosily related then this must be accepted as a fact of life, regardless of whether you find it unintuitive or your mathematical model deems it impossible. Ren Tech´s correlations discovery process seems to be based on the notion that the market is right.
While the message from Ren Tech would be that actual correlations can be discovered and profitably juiced, some may still argue that plenty of evidence suggests that correlations may still be nothing but a treacherous farce. Look, again, at those sub-prime CDOs: traders, quants, rating agents, regulators, and investors all believed default co-dependence to be minuscule and yet the opposite happened. Should anyone trust the notion of correlation at all after all this? Maybe yes. Those disappointing results came from models that conveniently churned out the outputs most conducive to business taking place: for the CDO machine to run profitably, it was essential that default correlation be categorised as tiny. There were too many interests invested in that notion. Even if many players knew that correlation estimate to be a total fluke, they could not afford to yell the truth. The wrongful estimates were making them too much money.
Outfits like Ren Tech have a much bigger incentive to be truthful. They don’t pay their bills with the fees generated by selling purportedly risk-lite products to Norwegian pension funds. They pay their bills through successful punts. If their correlation estimates are wrong, they’ll blow up. Thus, perhaps the conclusion is that whether correlations figures should be trusted depends on who’s doing the talking.
Pablo Triana is the author of Lecturing Birds On Flying: Can Mathematical Theories Destroy The Financial Markets? (Wiley, 2009)
Get alerts on Financials when a new story is published