Two months ago, as we were carefully reading the latest Goldman explanation [11]of how the firm had completely missed something Zero Hedge predicted back in January, namely the record warm winter's impact on skewing seasonal adjustments for payroll data (which has since validated our day 1 of 2012 predication that 2012 will be a carbon-copy replica of 2011, and which has made the comedy value of another Goldman masterpiece, that of Jim O'Neill's idiotic "2012: Not a Repeat of 2010 or 2011" [12]soar through the roof) we stumbled upon something we knew was about to get much, much more airplay: Goldman's quiet and out of place admission that what matters for a country's central bank is the flow of its purchases, not the stock (another massive economic misconception we have been trying to debunk since the beginning). Recall these words [11]: "...we have found some evidence that at the very long end of the yield curve, where Operation Twist is concentrated, it may be not just the stock of securities held by the Fed but also the ongoing flow of purchases that matters for yields..." This is how we summarized this observation two months ago (pardon the all caps): "UNLESS THE FED IS ACTIVELY ENGAGING IN MONETIZATION AT EVERY GIVEN MOMENT, THE IMPACT FROM EASING DIMINISHES PROGRESSIVELY, ULTIMATELY APPROACHING ZERO AND SUBSEQUENTLY BECOMING NEGATIVE!"
All caps aside, what this means is simple: if it is indeed flow that matters (and it is), then Fed intervention can never stop, period. If the stock of a central banks' assets is irrelevant, the Fed can have $1 on the left side of the balance sheet or $1 quadrillion: it does not matter - if the market expects the Fed to stop buying assets tomorrow, then the crash is as good as here. That has precisely been the biggest flaw with the Fed-accepted stock model, per which Bernanke can buy up a few trillion in MBS and the stock market will be flat as a frozen lake. Alas, this is increasingly becoming obvious is not the case. Hence flow.
Which is why today, two months later, and a week before Bernanke will almost certainly announce the NEW QE, we were not surprised at all to see that Goldman has actually made the case for flow in the form a of a white paper titled "Flow Effects at the Ultra-Long end of the Curve."
For monetary theory purists this is equivalent to Martin Luther walking up to the front door of the Marriner Eccles building and nailing his 95 theses: we have now entered the era of the monetary reformation, which incidentally as more and more classical economists follow suit, will throw all of Keynesian and neo-classical economics into a tailspin where virtually every core assumption will have to be reevaluated.
Congratulations economists: in their pursuit of another record year of bonuses at any cost, Goldman just sacrificed your precious voodoo. Because where Goldman goes, everyone else promptly follows.
From Goldman Sachs:
Flow Effects at the Ultra-Long End of the Curve (Shan/Stehn)
- With the scheduled end of the Fed's twist approaching, market participants are debating the extent to which the end of the Fed's purchases will affect the yield curve. The "stock view" – which Fed officials and we have generally subscribed to – suggests that markets tend to price in the Fed's purchases at announcement and then show little responsiveness to the subsequent flow (and end) of purchases. The "flow view," however, would suggest that yields increase when the twist concludes.
- Using a simple model of the Treasury yield curve, we revisit this issue in today's daily. Our estimates suggest that the flow effect is negligible for short and intermediate maturities (of less than 20 years) but statistically significant at the ultra-long end of the curve (with maturities of 20+ years). Although the uncertainty is significant, these estimates suggest that – all else equal – the end of the twist will have negligible effects on the short and intermediate part of the curve, but might push up yields at the ultra-long end of the curve by around 5 basis points.
With the scheduled end of the Fed's twist approaching, market participants are debating the extent to which the end of the Fed's purchases will affect the yield curve. Economic theory suggests that we need to distinguish between the effects of the announced stock of Fed purchases and the flow of actual purchases. In forward-looking and liquid markets, bond yields should primarily depend on the announced stock of purchases. Therefore, markets should price in the size of the purchase program at announcement and show little response to the subsequent flow of purchases. This means that when the flow of Fed purchases is discontinued—but the size and duration of the Fed's balance sheet is unchanged—there should be little effect on yields. Empirical evidence has generally reinforced this prediction. Our own work, for example, has confirmed that stock effects dominate flow effects. (See Sven Jari Stehn, "Stocks vs. Flows Revisited: End of QE2 Unlikely To Have Significant Effect on Bond Yields," US Daily, April 13, 2011.)
Although the "stock view" appears to be a good description of the effects of Fed purchases at the short and intermediate maturities, flows might be more important at the ultra-long end of the Treasury curve. Intuitively, this would fit with the observation of investment habitat – how purchases of 20-30 year bonds are mostly conducted by more heterogeneous investors that are less sensitive to changes in demand and supply in the Treasury market. Consistent with this view, we found tentative evidence for flow effects at the ultra-long end of the curve in earlier work (see US Daily cited above). However, the number of observations was very small and so the estimates were very imprecise.
With more data on hand and the end of the twist in sight, we revisit the issue of flow effects from Fed asset purchases at the long end of the curve in today's daily.
Following our previous work, we focus not just on one particular point on the yield curve at a time but also explore how the Fed’s purchase program has affected the entire yield curve. Doing so allows us to better separate the effects of economic factors (which affect the entire yield curve) from the Fed purchases (which differ across the yield curve). Making use of the relative movement of yields at different maturities provides more information and should thus provide better identification. (This disaggregated approach is motivated by Stefania D’Amico and Thomas King, “Flow and Stock Effects of Large-Scale Treasury Purchases,” Finance and Economics Discussion Series, Federal Reserve Board, 2010-52.)
Specifically, we construct our model in a number of steps.
First, we use the New York Fed’s Treasury yield curve estimates, which provide coupon-equivalent par yields for maturities between one year and 30 years.
Second, we construct a dataset of daily flows of Treasury purchases from the New York Fed’s website and allocate these into different “maturity buckets.” For example, we match purchases of bonds that have remaining maturity of between 9.5 and 10.5 years with the 10-year bond yield discussed above. Our sample period – which runs from March 2009 through April 2012 – can be divided into three phases: QE1 (March 2009 through October 2010), QE2 (November 2010 through August 2011), and the twist (since September 2011). The distribution of the purchases is shown in Exhibit 1 below.
Third, we control for the announced stock of Treasury purchases. Given our focus on testing for flow effects and the difficulty of identifying the announcement effect at individual maturities, we use a very flexible approach to capture the effect of the stock of purchases on yields. (Specifically, we use an intercept dummy and a linear trend for each maturity bucket in each QE phase.) The advantage of this approach is that we do not have to make a priori assumptions on the magnitude of the stock effect and doing so should raise the bar for finding flow effects. The drawback, of course, is that our model focuses solely on generating a flow estimate and cannot provide an estimate for the magnitude of the stock effect.
Finally, we combine the data on yields and flows with our stock dummy variables to construct a panel model for these thirty maturity buckets with daily data since March 2009. To take into account variations in duration and/or liquidity factors across maturity buckets, we allow the constant in our model to vary across maturities (that is, we include so-called maturity “fixed effects”). And to disentangle the effect of the Fed’s purchases at the different maturity buckets from economic factors that affect yields across the maturity spectrum, we allow the whole yield curve to shift over time (that is, we include so-called “time fixed effects”). In a nutshell, we estimate the following panel regression:
yield = ?*flow + ?*stock+ fixed effects
where the flow variable captures the purchases, the stock variable is given by the dummies described above and the fixed effects represent maturity and time fixed effects as discussed above. If there is a flow effect, then we should find a negative ? in this regression. To explore whether the flow effect differs at different parts of the curve, we allow ? to vary across different maturity buckets. In our baseline specification, we split the yield curve into seven segments (namely, 1-2 years, 3-4, 5-7, 8-10, 11-14, 15-20, and 21-30 years).
Our results are summarized in Exhibit 2 below. (For the full set of details, see Table 1 in the appendix). Our estimates reveal statistically insignificant ? coefficients at the short and intermediate maturities (up to 20 years), but negative and statistically significant estimates of ? at the ultra-long end of the curve (with 21-30 years maturity). In other words, our estimates confirm previous findings that the flow effect is negligible for short and intermediate maturities but significant at the ultra-long end of the curve. In terms of magnitudes, our results suggest that a $1bn purchase at the ultra-long end of the curve (all other things equal) lowers the yield by 3.3bp at that part of the curve.
We performed a few robustness checks. First, we split the regressions for the 1-10 maturities and 11-30 maturities in two regressions to address the concern that the daily time effects are not appropriate when grouping all 30 maturities together in one regression. The results are qualitatively similar: the flow effect at the very long end of the curve remains significant but the size of the effect drops from 3.3 to 2.3bp per $1bn (see Table 1 in the appendix). Second, we omitted the 1-2 years and 1-3 years maturities since the yields of these maturities have effectively been pinned down by the Fed's guidance language. Again the results are qualitatively unchanged (not shown).
In a final step, we can look at the implication of our estimates for the yield curve should the twist end in June as currently scheduled. As discussed, our model suggests that – all other things equal – we should not expect to see significant effects on the short and intermediate parts of the yield curve. But – again all other things equal – our model would point to an increase in yields at the ultra-long end of the curve. A simple approach to gauging the implied magnitude of this effect is to look at the average monthly purchases at this part of the curve over the last few months and ask by how much yields would move if this dried up. The Fed has purchased around $13bn per month at the ultra-long end of the curve (21-30 years) since the start of the twist. Taking into account that no purchases were actually made at the 21-23 year maturity (see Exhibit 1 above), this comes to an average of around $1.8bn per month in each of the maturity buckets where purchases actually took place. Applying our estimates of the flow effect – the 2.3bp to 3.3bp range per $1bn of purchases – this would imply that an increase in yields at the ultra-long end of the curve of around 5bp.
While we have reasonable confidence in the qualitative conclusion of our analysis – that flows tend to perturb yields only at the ultra-long end of the curve – it is important to keep in mind a number of caveats when interpreting the quantitative implications of our model. First, disentangling the influence of stocks, flows and other variables on the yield curve is a very difficult exercise and the uncertainty is therefore considerable. Second, our estimated magnitudes are only about as large as the average daily volatility of yields over the last three years (around 6bp at the 30-year maturity). Finally, it is important to note that our model's estimate only refers to the effect of the end of the flow of purchases on yields and does not take into account other factors that might influence yields at the same time. For example, the end of the twist could have a significant effect on the yield curve beyond the analysis presented here if Fed officials deliver something different from what the market is expecting for the June meeting