Is this just a case of data mining?

Originally published August 18

There has been a lot of discussion recently about how the recession signal from the yield curve might be weakened by technical issues. For example, I have emphasized that the term premium appears to have dropped dramatically relative to its historical average and that size of inversion matters much more than is widely understood, even leaving aside the issue of the term premium.

On Thursday, Mohamed El-Erian went on CNBC to claim that the signal from the yield curve was “distorted” by global factors, such as the upcoming ECB rate cuts and restart of QE.  I do not share in El-Erians take that global or supply issues are the main source of “distortion” in the yield curve.  Nor do I accept that the curve is even distorted. The global growth worry here is much more that global yields make sense than that they do not.  But El-Erian is implicitly insisting that it’s different this time, which seems right.

And this brings us to the main point of this report, which is so mundane as possibly to be a sleeper.  It is easily conceivable that the current obsession over the slope of the yield curve is just the result of data mining.   If we look under enough rocks, we are sure to find relationships that have “worked” in the past, even if the correlations involved are spurious and inclined to break down in real time.  If we data mine, it will always be different his time.  That is not disrespect of history, but Statistics 100.