Event studies are one method that has been used to try to assess the potential effects on markets of nonstandard monetary policy measures such as QE2. The Federal Reserve Bank of St. Louis recently hosted a conference whose objective was to evaluate evidence on the effects of these policies. Here I relate remarks I made at the conference on some of the challenges from trying to use event studies to answer this question.
Event studies look at a narrow window of time around which a significant policy initiative was announced to see how markets responded at the time. The hope is that, over the short period studied, the policy announcement itself is the most important news item to which markets were responding.
For example, a paper by Joseph Gagnon, Matthew Raskin, Julie Remache and Brian Sack presented at the conference identified 8 key days on which major details of the Fed’s initial large-scale asset purchase (sometimes referred to as “QE1”) were communicated to the public. These included:
- Nov 25, 2008: LSAP announced
- Dec 1, 2008: Bernanke in a speech described LSAP as involving “substantial quantities”
- Dec 16, 2008: FOMC said it “stands ready to expand its purchases of agency debt and mortgage-backed securities as conditions warrant”
The graph below shows the yield on 10-year U.S. Treasury securities during November and December of 2008, with the particular days mentioned above marked by vertical bars. This yield fell 170 basis points over these two months. Most of us would agree that the primary cause of this decline was not QE1, but instead news of a rapidly weakening economy, which would have been a reason for falling yields even if the Fed had done nothing. The problem plaguing any effort to measure the effects of the policy is the fact that QE1 was itself also a response to that same news. How much of the decline in yields was due to news of a weakening economy, and how much was due to LSAP?
10-year Treasury yields, Nov 3, 2008 to Dec 31, 2008. Vertical lines at Nov 24, 25, 28, Dec 1, 15, 16. (2011).
The idea behind the event-study methodology is to focus on the particular 3 days highlighted above, the hope being that the primary news on these days was the actions of the Fed rather than the deteriorating economy. It turns out that 61 basis points, or more than a third of the total decline over these two months, occurred on these 3 days alone. That observation suggests that the Fed’s actions may have had a significant effect of their own.
But it’s also interesting to look at the behavior of other variables over this period, such as the graph of nominal oil prices below. These also were coming down dramatically at the time, falling 30% over the same two months, and again presumably in response to incoming news of a worsening economy. And it turns out that 16% of this decline again occurred on the same 3 days highlighted above.
Daily closing prices of West Texas Intermediate, Nov 3, 2008 to Dec 31, 2008. Vertical lines at Nov 24, 25, 28, Dec 1, 15, 16. (2011).
Now, one could tell a story for why the oil price might in fact have been responding to the LSAP announcements on those particular days. For example, traders could have figured, “gosh, the Fed wouldn’t do this unless they were really scared about what’s happening to the economy. Things must be even worse than I thought!” But, to the extent this was the case, such a story may also be part of an effect incorrectly attributed to LSAP on yields themselves. Another plausible interpretation is that there was in fact some other bad news about the economy arriving at the same time as the LSAP announcements.
This is just another illustration of the principle that correlation does not establish causality. Notwithstanding, correlations are not irrelevant. If LSAP did have an effect on yields, one should see a correlation like that in the top diagram. On the other hand, if LSAP also had an effect, as some claim, of raising commodity prices, one would not expect to see a correlation like that in the lower diagram, and thus the second correlation is harder to reconcile with that hypothesis. The observed evidence is what you’d expect to see if you believed that LSAP lowered yields, and not what you’d expect to see if you believed that LSAP raised commodity prices. Correlations are in my opinion always worth looking at, even though they’re always subject to a variety of interpretations.
Fortunately, there are a number of other candidate event-days besides the ones I listed above. Papers by Gagnon, et. al., Krishnamurthy and Vissing-Jorgensen, , and others have used a number of other key announcement dates associated with QE1 and QE2. Joyce et. al. looked at LSAP announcements in the U.K., while Eric Swanson examined the effects of “Operation Twist” in 1961. Yet other studies, including Greenwood and Vayanos, D’Amico and King, and my own research with Cynthia Wu, documented longer time-series relations between the term structure of interest rates and the supplies of bonds of different maturities. Any one of these investigations, taken by itself, is perhaps not completely convincing. But the overall conclusion from looking at this question using a broad range of different data sets and methodologies is that large-scale asset purchases do seem to have a modest potential to influence the economy.
Guest Posted by James Hamilton