I showed in the first End of QE2 paper that there will be a 5% GDP shortfall in demand for Treasuries when the Fed stops its quantitative easing policy. According to many long term models, a 1% increase in the US budget deficit leads to a 15/20 basis point hike in UST 10year yields. The table below suggests this, and recaps the main drivers and their influence on 10year yields before QE2.
Even though Fed purchases did not focus solely on the 10year tenor, a rough calculation would suggest a 75/100 bp risk on US long term interest rates (this is when the Fed stops buying not starts selling – which would accentuate the trends).
There were no clearcut ex ante estimates of the potential impact (direction and extent) of quantitative policies. The ex post assessments have not been straightforward either. In the second issue of the End of QE2 Series, I propose two different estimates of the impact of QE on UST 10year yields. The results are somewhat consistent with many other estimates but are at odds with the risk of a sharp increase in UST 10year yield in the second half of the year.
1. Traditional Yield Model
The model is based on factors that are considered longend drivers of the Treasury yield curve. Longterm interest rates reflect:
expected inflation: the current level of core inflation drives the expectations of investors;
the inflation risk : the uncertainty surrounding the level of inflation during the life span of the bond. It constitutes the inflation risk premium. I model it by using the dispersion of 5year inflation expectations of economists using the Survey of Professional forecasters;
the current level of economic activity: I use a news/surprise index to gauge the growth momentum (more growth means tightening and higher risk appetite);
the uncertainty on the future conduct of monetary policy. I use the implied volatility of 1year10year swaptions (highly correlated to VIX).
I add a dummy variable for Lehman (I don’t use a short term of fed fund rate in the equation as the combination of growth momentum and core inflation is nothing more than the Taylor rule (that gives the desired level of repo).
The fit of the model is quite good. Even for the last period (first 3 months of 2011), the fit remains within the +2/2 standard error of regression. This means that traditional factors such as inflation and growth momentum can explain most of the past and current levels of 10year UST yields.
The residual of our equation has been drifting downward since the implementation of QE2. The second chart shows the estimate of this residual using the net issuance of bonds (excluding foreign investors and fed purchases). The constant term is around 25 b, which suggests that QE2, on average, brought a negative premium of 25 bp on US 10year interest rates– an impact much lower than what is widely accepted.
Click to enlarge
(Click charts to enlarge)
This model is does not rely on either fed funds or fiscal deficits. It shows that one can explain particularly well the orientation of long term rates over the last month with the growth momentum and inflation expectations. If the oil shock proves temporary and the ongoing negative momentum on US activity continues to drift downward, there is no clear cut evidence that long term rates will rise sharply if the Fed stops QE2.
2. Assessing the supply impact
Estimating the impact of net issuance and public deficit on interest rates is complicated by the fact that GDP, UST 10year yield and public deficits are all procyclical. In other words, any slowdown in economic activity brings lower interest rates and higher net issuance (see left chart below).
To assess the supply impact of higher deficits on the long end of the curve I:

adjust the deficit from the economic cycle. Technically, I take the residual from a simple estimate of net issuance using quarterly change in GDP;

consider that the impact of the net issuance on the risk premium embedded in the long end of the yield curves does not come from the current level of structural deficit but its change (worsening or improving structural imbalances).
Drawing upon this, I estimated the UST 10year yield using the current levels of VIX and Fed Funds and the change in the structural net issuances. The model fits well up to the second half of 2010 (left chart below). The residual plunges down to ‑80b, which can be considered the impact of Fed purchases on long term interest rates.
Click to enlarge
Conclusion: There is much uncertainty surrounding the estimates of the impact of QE2 on interest rates. A number of factors can influence them: inflation risk, risk appetite, growth momentum, fiscal stance. Models based on pure macro factors rule out any strong impact. More traditional models suggest an impact of 50/75 basis points.
In addition, event studies show that rates reacted before, not after the implementation of QE1 and QE2. Remember that QE1 aimed specifically at reducing rates (refinancing) whereas asset prices and inflation expectations were the target of QE2. Therefore, QE2 has been a success, as it managed to prop up risky asset prices and inflation expectations (the economic impact is less straightforward). The fact that 10year yields have not reacted yet to the widely anticipated end of QE2 may be surprising as it suggests that the commitment to maintain low rates for a protracted period of time is low and that there may be a demand shortfall ahead.
The gridlock in Washington, and the decision by S&P to give a negative outlook on US public debt did not harm the UST 10year yield for more than a few hours. Both Fed demand and credit risk have been shunned so far.
There may be another explanation in the current low level of rates. The post QE2 announcement rise in longterm rates may have already integrated the endofQE2 premium. The upward momentum of yield may have been propped up by the combination of stronger growth data and bullish activity in risky asset markets. Since the earning season disappoints and the growth momentum have reversed, the risk of a strong upward reversal of long term US rates may remain contained, even after June.
Click to enlarge