A look at term premia

Why are bond yields so low?   On the face of it, that's an easy question to answer.  While you may not agree with the market pricing, you can nevertheless point to low inflation, massive central bank purchases of global bonds, the advent of negative rates, weak global growth, etc. as a reason why yields are so low.

But how satisfying is that answer?  Does the 1.71% that one currently receives from lending to the US government accurately reflect the market's belief in the likely trajectory of short term interest rates over the next decade?   Does the 0.10% that one gets from owning Bunds?  Yes, it's true that short rates in the US, Europe, and elsewhere are currently very low at the moment, and the precedent of Japan suggests the possibility that they could stay that way ad infinitum.  Yet short rate futures strips, particularly in the US, do price an eventuality of higher rates, and the FOMC dot plot (ha ha ha) also suggests likewise.  Surely the risk is of a more abrupt return to normalcy for the yield curve?

This is one of the reasons that academics and bond market analysts like to think of a term premium, the added yield beyond short rate expectations that investors require to own a bond for a duration beyond the immediate future.  The term premium essentially reflects the risk that your embedded short rate forecast is incorrect and the compensation that you require for that risk.

There are a number of econometric models of the term premium, some of which yield markedly different results.  A few years ago some New York Fed economists published their own model (known as ACM after the authors), which is an "affine class of term structure model which characterize yields as linear functions of a set of pricing factors."  While Macro Man isn't really competent to judge the relative merits of this model versus some of the others, it does confer the signal benefit of being downloadable on the NY Fed website and on Bloomberg, which allows the masses to play with it to our heart's content.

What's interesting to observe is that term premia across the curve are currently negative; in other words, bond yields are actually lower than those that would directly reflect the market's expectations for the trajectory of short-term interest rates.  To an extent, that is a satisfying result, insofar as it likely reflects the biases of a number of a number of market observers who don't happen to own several hundred billion dollars' worth of bonds.

Why should the term premium be negative?   Look no further than the March FOMC (and indeed much of the last 18 months); if you think that there's a risk that the Fed will continue to jelly on moving rates, it's natural that this should be reflected in the price.  Interstingly, if we look at the level of yields adjusted for this term premium (call it the "natural" level of yields), we can see them rise sharply across the curve with the announcement of the taper in December 2013.

Macro Man also had a look at the 2-10 term premium curve to get a sense of how it is impacting the nominal yield curve on an historical basis.    What do you know....the difference between the 2y and the 10y term premia is as small as it's ever been (reflecting a substantial flattening influence on the curve.)   The implication would appear to be that the balance of risk is heavily skewed towards a re-steepening, particularly given the volatility of the term premium curve.

Or is it?  If one were prepared to take the term premium as an independent explanatory variable, that forecast would resonate a lot.  However, given that these term premia estimates are generated by running regressions on the curve, Macro Man suspects that they are descriptive rather than prescriptive.   In other words, he reckons that bond yields don't rise and fall because of shifts in the level of the term premium, but rather the term premium rises and falls along with changes in bond yields and the shape of the curve.

Ask a CTA if he's piling into TY because he's adjusted his view on the term premium, and he's likely to respond "01001100 01001111 01001100".