Saturday, July 11, 2009

Cointegration Requires Sound Economics

Because (Multi)cointegration could coincide with spurious regression, the relationships needs economic logic to hold validity. Blind number crunching then could lead to poorly designed relative value arbitrage strategies.

Cointegration/Multicointegration defined

From Wikipedia

Cointegration is an econometric property of time series variables. If two or more series are themselves non-stationary, but a linear combination of them is stationary, then the series are said to be cointegrated.

Example: 2 individual stock prices display varying probability distributions over time. When you subtract their daily returns, that difference in returns however does exhibit the same distribution at all periods, therefore becoming more “predictable”.


Multicointegration extends the cointegration technique beyond two variables, and occasionally to variables integrated at different orders.

Issues with spurious correlation/regression

The non-Stationarity of individual security returns then makes random data set cointegration test unreliable. This partly explains how some incompetent banking wizards had picked tested numbers to make them “work”, backing unreliable risk assessments on credit instruments like CDOs, MBSs, and etc.

Back to fundamental economics

Number crunching alone is not enough, while those who acknowledged the crazy leverage and bad debt-packaging realized the oncoming crisis in 2007. When it comes down to it, understanding underlying economic relationships still rules them all.

0 Reflections: