May 14, 2012

Goldman Sachs on Rate Markets

May 14, 2012

Goldman Sachs on Rate Markets

Here’s a great rate piece from Goldman…

-In mid-2011 the term premium in 10-year Treasury yields fell from about +50 basis points (bp) to -50bp. Unlike in 2009 and 2010, the term premium has persisted at a low level for many months. To what extent can the large and sustained drop in the term premium be attributed to Fed policy actions?

-We have already done extensive research on the impact of QE on rates, and have consistently found that “stock” effects dominate “flow” effects. In this article, we extend this analysis using a model that directly takes into account investor expectations—not only for the announced size of purchases, but for the entire path of the Fed’s balance sheet over time.

-This approach implies a potential interaction between QE and Fed communication. In particular, because the Fed has linked the eventual normalization of its portfolio to the timing of rate hikes in its exit strategy, guidance about the funds rate can indirectly affect expectations about the balance sheet. If investors expect that the Fed will hold its assets for longer, forward guidance may in turn reduce the term premium.

-We apply this framework to simulate the cumulative impact of Fed easing on the term premium to date. We find a total effect from unconventional policy—including both QE and forward guidance—of about -80bp. About 45bp of this total results from actions over the last year (the twist and communication changes).

-For the rates market, this model implies: (1) the term premium should have a natural tendency to rise without further Fed action, because stock effects are temporary in forward-looking markets; (2) rates may be quite sensitive to changes in Fed communication because it affects both the expected funds rate path and the term premium; and (3) even without “flow” effects, the end of the twist could have a negative impact on the market if it changes expectations about the future path of the Fed’s balance sheet.

With the funds rate stuck at its effective lower bound for more than three years, the Federal Reserve has used two unconventional policy tools to reduce longer-term interest rates: 1) communication about how long the low funds rate is likely to remain in place; and 2) large-scale asset purchases or quantitative easing (QE). These tools are generally thought to affect longer-term rates in distinct ways. Official communication or “forward guidance” can affect investors’ expectations about the future path for policy rates. In contrast, large-scale asset purchases reduce the supply of securities in private sector hands and, through portfolio rebalancing effects, reduce the term premium in longer-term interest rates.

In practice, however, the effects of these policy tools may interact. For instance, there may be a “signaling channel” for asset purchases, where QE announcements contain implicit information about future policy rates . In addition, there may be an aspect of balance sheet guidance implicit in the Fed’s communication about the funds rate. We think these interactions can go a long way toward explaining the persistently low level of longer-term interest rates over the last nine months.

Lower Term Premium Explains Drop in Rates

Longer-term interest rates can be thought of as the sum of expectations of future short-term interest rates and a risk premium. The risk premium reflects the additional compensation demanded for investing in longer-term securities compared to a strategy of rolling over investments in short-term securities. For default-risk-free bonds this risk premium is typically referred to as the “term premium”, and is essentially a compensation for duration (i.e. interest rate) risk.

Thus, a 10-year Treasury security can be written as: 10yr yield=10yr Expected Avg policy rate+term premium

Although market expectations for future policy rates have also fallen, we believe the decline in longer-term rates since mid-2011 primarily reflects a large and sustained drop in the term premium. We cannot directly observe the term premium, however, and must rely on model-based estimates. As our measure of the term premium in the 10-year Treasury yield, we use the estimate produced by staff economists at the Federal Reserve Board . The Fed’s term premium estimate is derived from a traditional term structure model, augmented with survey data in order to overcome known biases in yields-only models .

As shown in Exhibit 1, this estimate of the term premium declined sharply in mid-2011, from an average of about +50 basis points (bp) in the first half of the year to an average of about -50bp since September . This change explains about 70-80% of the drop in 10-year yields over this period.

Dealing Directly with QE Expectations

To what extent can the large and sustained decline in the term premium be attributed to the Fed’s unconventional policy actions, as opposed to the European sovereign debt crisis or other factors? We have already done extensive research on the impact of QE on rates . Our work (as well as the academic literature on the subject) has consistently found that changes in the expected amount of Fed assets can reduce rates, and that these “stock” effects clearly dominate “flow” effects.

However, most of our previous work has only indirectly addressed expectations. For instance, our empirical studies have focused on QE announcement dates, which recognizes that Fed asset purchase programs affect rates before the actual bond buying begins. Still, there are a few reasons why we may now want to address expectations more directly.

First, the experience of QE1 and QE2 suggests that the impact of central bank asset purchases may be relatively short-lived—at least under certain conditions. As shown in Exhibit 1, the term premium rose sharply after both of the last two programs were announced. A few recent academic papers have also argued that the impact of unconventional policy fades quickly: Wright (2011) finds a “half-life” of only two months, and Jarrow and Li (2012) find half-lives of 4-19 months, depending on the interest rate . While “flow effects” are sometimes considered a reason for these findings, shifts in investor expectations are another (and in our view more likely) explanation.

Second, we now have more clarity about the likely sequence of the exit process from the April 2010 and June 2011 FOMC meeting minutes and from Chairman Bernanke’s February 2010 testimony to Congress. More guidance may have narrowed market views about the future path for the balance sheet.

Third, the Fed is using much more aggressive forward guidance than during previous rounds of asset purchases. In those periods, the FOMC said that rates would stay low for “an extended period”, which Chairman Bernanke defined as “at least two or three meetings” before action (at the June 2011 press conference). Today the committee has given guidance that the funds rate will remain “exceptionally low … at least through late 2014”, implying at least two years (16+ meetings) before rate hikes begin. If forward guidance can move the term premium, we are more likely to see its effects today.

In order to deal directly with investor expectations of the Fed’s balance sheet, we apply the model proposed by Chung et al. (2011) . The authors argue that the QE portfolio balance effect is related not only to the initially announced stock of purchases, but to the entire profile of the Fed’s balance sheet over time. That is, the QE effect is a function of the present discounted value (PDV) of the expected “excess” size of the Fed’s balance sheet . This can be written as:

QE Effect=?*PDV[Excess Fed Assets]

where the QE Effect is measured in basis points, and ? is a scaling factor that translates the PDV of excess Fed assets into a portfolio balance effect. We follow Chung et al (2011) and measure excess Fed assets as the ratio of the Fed’s security holdings with an initial maturity greater than one year as a percent of nominal GDP (with a small adjustment for trend growth in the size of the Fed’s balance sheet).

What Forward-Looking Markets Imply

Before estimating the impact of past Fed actions, we highlight three key implications of this model:

1. The expected balance sheet path matters. The length of time that investors expect the central bank to hold its assets has implications for the size of the portfolio balance effect. If the portfolio balance effect is a function of the PDV, then the longer the balance sheet remains elevated, the bigger the impact of QE.

Exhibit 2 demonstrates this with a hypothetical $600bn QE program announced at the start of 2013. In one example, investors are assumed to expect that reinvestment of maturing securities will last for one year after the QE operation ends, and that thereafter the balance sheet will be brought down to a normal level within three years. In the second example, investors expect two years of reinvestment and a five-year normalization process after that.

Because the PDV of the Fed’s asset holdings is much larger in the second example, the announced program has a larger initial effect on yields—about 30bp instead of 20bp. The effect of the second operation also persists longer, with a half-life of about 10 quarters instead of 6-7 quarters. The size and persistence of QE effects are thus quite sensitive to the expected path of the balance sheet.

2. Credible forward guidance can affect the expected path of the balance sheet. The second implication from the model follows directly from the first: if the expected balance sheet path matters for the size of the QE effect, then credible forward guidance can affect the term premium by changing expectations about the evolution of the balance sheet (Exhibit 3).

Note that it does not matter (at least very much) whether the FOMC provides guidance about the federal funds rate or about the balance sheet. Because Fed officials have committed to a specific sequencing for the exit process, funds rate guidance and balance sheet guidance are not independent instruments. By offering explicit forward guidance about the timing of the first rate hike, the FOMC is already giving implicit guidance for the future path for the balance sheet .

3. Stock effects can be temporary. The proposed model says that the size of the portfolio balance effect on the term premium depends on the PDV of the Fed’s excess asset holdings at any given time. Moreover, the FOMC has clearly committed to normalizing the size of balance sheet in the future. Therefore, as time passes, the PDV of the Fed’s excess asset holdings begins to decline because the end of reinvestment and the beginning of asset sales gets closer (i.e. the eventual runoff is discounted less). It is not necessary to assume “flow” effects for the impact of QE to be temporary: stock effects on the level of rates are likely temporary because of forward-looking markets.

Introducing Balance Sheet Expectations

Calculating the impact of QE on the term premium from this model requires a number of assumptions, the most important of which are the expected path for the Fed’s balance sheet and the size of the portfolio balance effect parameter (? in the equation above).

We unfortunately cannot observe investors’ expectations for the Fed’s balance sheet, but we have some observable information to work from. First, Fed officials have communicated their preferences about the sequence of the exit process at various times, so we know that they prefer to end reinvestment, then hike rates, then sell assets. Second, we know something about market expectations for the timing of funds rate increases from futures prices, and also some information about balance sheet expectations from surveys. By piecing this information together, we may be able to make reasonable assumptions about how the balance sheet was expected to evolve at certain times.

Chung et al. (2011) provide assumptions about the expected path for the balance sheet during the first three phases of the Fed’s unconventional operations—QE1, the August 2010 announcement of reinvestment of the MBS paydowns, and QE2—based on this type of information. We follow their assumptions almost exactly for these periods .

For the next three phases of Fed action—the 2013 forward guidance, the twist, and the 2014 forward guidance—we make our own assumptions about how investors expected the balance sheet to evolve. For the timing of the first rate hike, we use the FOMC’s official forward guidance, cross-checked with market expectations at the time. Next, we assume that investors expected reinvestment to end two quarters before the first rate hike. This is based on FOMC guidance and responses to special questions in several of the 2011 Blue Chip surveys. Finally, we assume that asset sales begin three quarters after the start to rate hikes, and are completed within a five-year window after the start of rate hikes. This is the most uncertain aspect of the expected balance sheet path, but we again have some guidance from the June 2011 FOMC meeting minutes and Blue Chip surveys.

We made a few additional assumptions specific to the twist. First, following Chung et al (2011), we focused on the face value of the Fed’s portfolio rather than a duration-adjusted portfolio. Explicitly accounting for duration would have been preferable, but it greatly complicates the analysis. We therefore assumed that the twist was equivalent to a $600bn expansion of the balance sheet, just like QE2 . Second, we assumed a slower pace of natural run-off of the balance sheet after the end of reinvestment than before the twist. Because the Fed has sold many short-dated securities, natural runoff should be much lower. Finally, we assumed some “excess” assets from the twist remain on the Fed’s balance sheet even after five years. This is because Fed officials have committed to reducing the size of the balance sheet and to removing agency securities, but they have not committed to a specific maturity mix for their Treasury portfolio—so some of the long-duration Treasuries may linger.

Exhibit 4 shows our estimates of the Fed’s longer-term security holdings as a share of GDP, in excess of “normal” levels. The lines should be interpreted as our best guess of the path of the Fed’s balance sheet expected by investors at the time of the announcement of the asset purchase program or forward guidance (the shape of the expected balance sheet path likely changes between announcements).

Finally, we calibrated the portfolio balance effect parameter ? such that QE2 reduced the term premium by 15bp, consistent with the results from our empirical work . This gives a value for ? of -26.3. This estimate for the impact of QE is at the lower end of the range from empirical studies and lower than the value used in Chung et al. (2011). However, one reason to use a fairly conservative estimate is that not all earlier studies controlled for the “signaling channel” of QE, and we are only interested in the portfolio balance effect on the term premium.

Powerful Punch from Guidance and Twist

With assumptions about the expected path of the balance sheet and an estimate for the portfolio balance effect parameter (?), we can calculate the implied QE effect—the impact on the 10-year yield term premium—from each operation. This is shown in Exhibit 5, which graphs the incremental impact of each announcement on the term premium.

A few points stand out. First, QE1 had the largest initial impact on rates, which is not surprising as it was much bigger than the other operations.

Second, the announcements of the 2013 and 2014 forward guidance statements had an impact on the term premium beyond their impact on expected future policy rates. The reason is that the announcements changed the expected path for the balance sheet and therefore increased its PDV. The same goes for the August 2010 decision to begin reinvesting MBS paydowns: this did not increase the size of the balance sheet, but it increased its PDV.

Third, the model implies that the twist had a larger portfolio balance effect than QE2. Although the programs removed the same amount of duration from the market, the PDV of the twist was much bigger because it followed the 2013 forward guidance. Thus, investors likely expected the assets to remain on the Fed’s balance sheet for much longer. A few particular aspects of the twist likely helped too (e.g. a lower run-off speed because of short-term asset sales).

Survey data support the hypothesis that investors expected a relatively short holding period for the Fed’s “excess” assets until recently. For example, just prior to the QE2 announcement, a CNBC survey found that market participants expected the Fed’s balance sheet to increase by an average of about $475bn by August 2011. However, the mean response then expected a decline of $45bn between August and November of 2011. Similarly, in the April 2011 Blue Chip Financial Survey, 74% of respondents said they expected the Fed’s balance sheet to begin shrinking before the end of that year. In the May survey, only 5% of survey respondents said they expected reinvestment to continue for more than a year.

We do not have any recent survey data on reinvestment expectations. However, it is clear that market expectations for the timing of the first rate hike have moved much further out. If we assume that reinvestment would end a few quarters before the first rate hike, current market pricing would imply an end to reinvestment in early 2014 or possibly very late 2013. Thus, our guess would be that today the majority of investors currently would expect reinvestment to continue for at least one year—in contrast to the 5% that did in May 2011.

Large Cumulative QE Effect

Exhibit 6 summarizes our estimate the cumulative impact of the Fed’s unconventional policy measures on the term premium, taking into account both the initial announcement effects and the tendency for the effects to decay over time. The estimates imply a total impact on the term premium of around -80bp. About 45bp of this total is accounted for by the last three operations: the twist and the communication changes in August and January.

Given the large number of assumptions in our estimates there is clearly a range of uncertainty around these figures. For example, if we were to use a higher portfolio balance effect parameter (?)—such as the one used in Chung et al. (2011)—the cumulative impact of Fed easing would rise to about 100bp. In contrast, if we assume that the Fed will remove all of the “excess” assets from the twist within five years, the QE effect would shrink to about 70bp. Using a duration-adjusted balance sheet measure may also lead to smaller estimates (because this would capture the natural aging of securities while the Fed holds them).

If this framework for understanding the impact of QE is correct, we see three main implications for the rates market. First, in the absence of any new Fed operations, the term premium should have a natural tendency to rise because QE portfolio balance effects are temporary. By our estimates, the portfolio balance effect on the 10-year yield term premium should shrink (become less negative) by about 20bp by the end of 2013.

Second, rates may be very sensitive to changes in communication—which could mean the bar for pulling forward the guidance is high. We have argued that the Fed’s communication about policy not only affects the expected path of policy rates, but also the term premium through expectations about the balance sheet. Therefore, any changes in the forward guidance could have an outsized impact on the yield curve.

Third, even if “flow” effects are negligible in and of themselves, the end of the twist could affect rates if it changes investor expectations about the future path of the balance sheet—that is, the flow of purchases could act as a signal about the future stock.

Finally, we commonly encounter the view that market expectations of additional QE could be high, given the persistently low level of rates. However, our estimate of the cumulative effect of past easing is quite large, suggesting this is a plausible alternative explanation.