Body
Four prominent economists (David Greenlaw, James Hamilton, Ethan Harris, and Kenneth West, or GHHW) attracted much attention earlier this year when they argued that the consensus of previous studies overstates the effects of quantitative easing (QE) on long-term interest rates. However, a careful reading of their paper and the associated data suggests that their conclusion is also overstated. It is true that some studies of the initial round of QE, or QE1, may have found effects on bond yields that were larger than would be expected in noncrisis circumstances. But the evidence suggests that moderately lower estimates, such as those employed by staff at the Federal Reserve in recent years, are reliable measures of the effect of QE in normal times.
GHHW conclude that (1) the lasting effect of QE1 was perhaps half as large as some prominent estimates, and (2) the effects of the remaining rounds of QE were either transient or negligible. Their paper relies on the "event study" methodology for measuring the effects of QE. An event study measures the effect of a policy on an economic variable by adding up the movements in the variable in narrow windows around news about the policy. Critical assumptions of event studies are that (1) the policy in question was not expected prior to the news events, (2) the news events are the only times the market changed its expectations about the policy, (3) those expectation changes occurred entirely within the event windows, and (4) nothing else happened during the event windows to affect the economic variable of interest.
Arguably, the assumptions are not unreasonable for major news events concerning QE1, but they are clearly violated for subsequent rounds of QE. Prior to the first QE1 announcement, market expectations of future Fed purchases of long-term bonds were probably quite low.1 During QE1, especially during the early months, Fed announcements were probably the main source of information about future purchases. Under these circumstances, adding up interest rate movements around QE1 announcements is a reasonable approach to measuring the total effect of QE1 on interest rates.
In my 2011 paper with Matthew Raskin, Julie Remache, and Brian Sack (GRRS), we summed the movements in the yield on 10-year Treasury securities on eight days with Federal Reserve announcements that contained news about QE1 to get a total decline of 91 basis points.2 We also expanded the event set to include all Federal Open Market Committee (FOMC) statements and minutes over the life of the program (November 25, 2008 through March 31, 2010), which implied a total decline of 55 basis points. The smaller total in the larger event set reflects the fact that yields tended to rise on the additional dates.
GHHW conducted a similar calculation, adding all speeches by the Fed chair that touched on monetary policy to all FOMC statements and minutes, and found a total decline of 17 basis points over the life of the program. However, through communication with GHHW, I discovered that this estimate omits two important news events: the initial announcement of QE1 purchases on November 25, 2008 and a speech by Chair Ben Bernanke on the mortgage market on December 4, 2008.3 Adding yield movements on these two dates to the GHHW total implies a cumulative decline of 49 basis points. GHHW compute another total based on days in which Reuters reported that Fed policy was a significant factor behind changes in bond yields. The total decline in the 10-year yield on these Reuters days over the life of QE1 was 48 basis points.
It is likely that QE has an especially large effect during periods of financial stress; this may explain the particularly large effects on the 10-year yield during the initial QE1 announcements, which were the focus of the baseline event sets in GRRS and some other studies. Some of these effects likely would have dissipated as financial stress eased. There is no compelling reason that the dissipation of the large initial effects would occur only on Fed news days, but it is interesting to note that the cumulative effects mentioned above that include event dates beyond the initial months of QE1 (around 50 basis points) are quite close to the estimated effects of QE1 obtained from a very different methodology discussed below. In any event, a reversal of the large declines in bond yields after several months in no way diminishes the stimulative effects of QE1. A better interpretation is that QE1 shored up business and investor confidence, preventing a descent into deflation as happened earlier in Japan and allowing bond yields to rebound somewhat.
For QE programs after QE1, the event study methodology provides little useful information. Bond yields declined substantially between the end of QE2 and the beginning of the Maturity Extension Program, reflecting the surprisingly weak economic recovery, which raised expectations of additional QE bond purchases. Most of these declines occurred on days with no Fed news events, so event studies do not attribute them to QE. Yield movements on Fed event days had little cumulative effect until well into QE3, reflecting the fact that Fed policy was close to what the market had expected. In other words, after the establishment of the QE precedent in QE1, event studies tell us how the Fed deviated from market expectations of QE purchases, not how such purchases affected bond yields.
The taper tantrum of 2013 shows that surprise changes in expected QE purchases do have significant effects on bond yields in times of normal market conditions. GHHW calculate that the 10-year yield rose nearly 40 basis points on Fed event days between March and September 2013, probably reflecting a reduction in the expected total bond purchases under QE3.4
GRRS also pursued an alternative methodology for estimating the effect of QE on bond yields, using a time-series regression of the 10-year yield or 10-year term premium on various factors including the net supply of long-term bonds by the government. The regression was based on the period 1985–2008, and thus did not include many times of significant financial stress or, indeed, any of the QE programs. However, QE purchases reduce the supply of long-term bonds, and their effect on yields can be calculated using the estimated supply coefficient from the regression. We found that bond purchases of the size of QE1 would have been expected to reduce the 10-year yield between 40 and 80 basis points, with a central tendency around 50 to 60 basis points.5 One of the GHHW authors (Hamilton) in a separate paper with a different coauthor also found a similar estimate using a somewhat different regression approach. Staff at the Federal Reserve also use a regression-based estimate in this range when they model the macroeconomic effects of QE. They estimate that all of the Fed's QE programs combined had a peak effect on bond yields in late 2013, reducing the 10-year yield by about 120 basis points.
Although smaller than the initial event-study estimates from QE1, the regression-based estimates support a substantial effect of QE on bond yields that is not limited to times of financial stress.
Author's Note: I thank Chris Collins for gathering data and information on speeches.
Notes
1. GHHW point out that the November 10, 2008 issue of Blue Chip Economic Indicators found that 54 percent of economists surveyed expected the Fed would engage in some quantitative easing soon. But the survey question explicitly referred to the case of Japan, in which such purchases were of relatively short-term government bonds that had little macroeconomic effect, so the response may overstate the market's expectation of the type of QE that the Fed eventually adopted.
2. A basis point is one-hundredth of a percentage point.
3. The speech discussed mortgage rates and availability and mentioned the Fed's QE purchases aimed at improving conditions in the mortgage market. The only speech in the GRRS event dates occurred on December 1 (it is also in the GHHW dates); it was the first time a Fed official raised the possibility of buying long-term Treasury bonds.
4. Applying the Fed staff's benchmark estimate (discussed below) a yield increase of this magnitude suggests that Fed communications reduced market expectations of the length of the QE3 program by 12 months at $85 billion per month.
5. My 2016 survey paper found broadly comparable effects in other countries.