The new rule-of-thumb in town pertains to the impact of Fed purchases on U.S. long term interest rates: every $100 billion purchased would reduce yield by 0.03 percentage points. This is a flow analysis that differs from the stock view of Bernanke, who believes it is not the changes but the outstanding of bonds held by the central bank that puts a lid on yields.
My view is that the flows of purchases should have an impact on yield (not the outstanding in the Fed balance sheet) as the gross supply of bonds is not static but dependent on flow factors: redemptions and the fiscal deficit. This is the reason why I was a strong proponents of tapering: a lower fiscal deficit meant a lower supply of bonds (the redemption risk being concentrated mainly in 2016) and hence a limited impact on bonds yields (all things being equal). The Fed chose to focus more on the negative fiscal shock on the U.S. economy.
Even though I am a strong proponent of the flow approach (confirmed in the chart above: public debt and real rates are correlated unless the Fed buys bonds), I doubt that there could be such a thing as a simplistic rule of thumb for the impact of quantitative easing (QE) on yields.
- All QEs were not created equal: QE1 and Operation Twist for instance have only a few features in common;
- Some QEs came along with different forms of forward guidance, while others did not; and
- The cyclical position of the U.S. economy differed significantly from one QE to the other.
A rule of thumb has to be statistically verified and to show some stability over time. I therefore resorted to econometrics to test its validity.
Here comes the first hurdle: it is very easy to model U.S. Treasury yields since 2008 without introducing the Fed balance sheet as an explanatory variable.
The chart below shows the output of a model that tracks 91% of the total variance of U.S. rates (monthly frequency). The standard error of regression remains high yet close to 20 basis points. Compare this to the 0.03 rule of thumb number to see how fragile this number is.
The model is based on traditional factors: Fed funds, slope of the Eurodollar curve, households' inflation expectations and the stock market. The S&P500 has a negative sign when the 1-month return is factored in but a positive one for longer run returns. This can be easily explained: in the short run there might be some episodes of risk aversion when stocks are down and rates are up. But in the medium run stock returns and yields move in the same direction. In both cases though the stock market is statistically significant.
I added the Fed balance sheet and introduced the absolute change as an input to be able to test the -0.03 elasticity. The 12-month change in the Fed balance sheet is statistically significant. Unfortunately, I found that a rise in the Fed balance sheet would lead to an increase of 10-year U.S. yields!
According to this model, every $100 billion purchased would increase yield by 0.04 percentage points. Interestingly enough, stability tests (CUSUM of square tests) show that the coefficient is stable throughout our sample (2008-today). In addition, an estimate based on a sample that does not encompass the tapering period gives an estimate of +0.035 basis points.
This is a simple reminder that rules of thumb not based on economic or financial identities are unreliable. I would opt out of a scenario based on such a rule (which would be, for instance, that the Fed tapers in December and that it would buy around 220 of securities until then: the rule of thumb would call for a extra decline of roughly 6/7 basis points of U.S. Treasury yields...).
U.S. Treasuries (NYSEARCA:TLO) still have some room to perform until the Fed tapers in December. U.S. Treasury yields may indeed fall down to 2.5/2.6%. But this will not be a matter of Fed purchases but rather a reflection of the extreme dovishness of the U.S. Central Bank.