Measuring the Consequences of the Zero Lower Bound Constraint

In a period of deleveraging such as the U.S. has been going through, it is possible for the natural rate of interest to become negative. Since cash is always an option for earning at least a yield of zero, no asset should ever pay less than zero. This lower bound of zero on nominal interest rates can put a constraint on the ability of the economy to self-correct or the Fed to provide stimulus in such a situation.

The Fed still has some tools to try to reduce longer-term yields, namely large-scale asset purchases and signaling the Fed’s future intentions. A new research paper by Federal Reserve Bank of San Francisco President John Williams and Senior Research Advisor Eric Swanson proposes a creative new approach to measuring when and to what extent the zero lower bound is a relevant constraint on interest rates of any maturity.

Market watchers are familiar with the observation that bond yields often are moved by economic news. For example, when the Bureau of Labor Statistics released a better-than-expected employment report two weeks ago, the yield on 10-year bonds jumped 10 basis points. In their new paper, Swanson and Williams measured the average response over 1990-2000 in the n-period bond yield to a dozen different economic news releases such as nonfarm payrolls, new home sales, retail sales, and the CPI. They summarized that historical response in the form of a regression:

swanson_eq1.gif

Here Δitn denotes the change on day t in the yield on an n-period bond, an is the regression intercept, Xt is a (12 x 1) vector containing the difference between the announced and market-expected value for a dozen different macroeconomic variables (with most of the elements zero except on the day of a news release for that variable), bn is a (1 x 12) vector of estimated OLS regression coefficients, and etn is the regression error. They then looked at how the response of that yield to these same news announcements over a 1-year interval around some day τ in the more recent period differed from the 1990-2000 typical value, as summarized by another regression

where the values of the regression intercept γτn and slope δτn are estimated just using data in a one-year interval around τ and bn is the value of the coefficient vector as estimated by the first regression over 1990-2000. Thus a value of δτn = 1 would indicate that the n-period yield was responding to news around date τ about the same way it did in the 1990s. A value of δτn greater than one means the interest rate was even more sensitive to news in the later period, and a value less than one means the interest rate was less sensitive to news than it had been historically.

The following graph shows Swanson and Williams’ estimated value for δτn for n the 3-month U.S. Treasury bill, plotted as a function of τ. As the T-bill rate fell below 1% in 2003, the value of δτn became statistically significantly less than one (indicated on the graph by a yellow region). The estimated sensitivity continued to fall and became statistically indistinguishable from zero (indicated by pink shading) in 2004– in other words, the T-bill rate was basically not responding at all to news at this time. Since the beginning of last year, we’ve been in a new episode in which the T-bill rate is unresponsive to news. The interpretation is it is not just the overnight rate, but the 3-month rate as well, that everyone sees as stuck near zero. Nothing that happens in the economy seems to change that perception.

Time-varying sensitivity of 3-month T-bill rate to 12 different news releases. Dotted lines denote 95% confidence intervals. A value of δτ = 1 corresponds to normal Treasury sensitivity to news; δτ = 0 to complete insensitivity. Yellow shaded regions denote δτ significantly less than 1; red shaded regions denote δτ significantly less than 1 and not significantly different from 0. Source: Swanson and Williams (2012).

By contrast, the 1-year Treasury yield continued to have a statistically significant response to news throughout 2003-2006, although it became a milder response than exhibited historically. On the other hand, since the middle of last year the 1-year rate doesn’t really seem to move in response to economic news, just like the 3-month rate.

Source: Swanson and Williams (2012).

The two-year rate continues to exhibit a measurable response to daily news, although the size of the response is only about half of what would historically be expected.

Source: Swanson and Williams (2012).

Interestingly, the 5-year yield seems to respond to news pretty much the way it always did.

Source: Swanson and Williams (2012).

University of Chicago Professor Cynthia Wu and I developed a model that could explain this behavior in a research paper that just appeared in the Journal of Money, Credit and Banking. We proposed that investors know that the overnight rate is currently stuck at zero, but see some possibility of escaping from the zero lower bound at some point in the future. We developed a formula to predict the response of the n-period yield to news as a function of that assessed probability. Longer-term bonds respond to news much more than shorter-term bonds at the zero lower bound because investors anticipate that the overnight rate will be nonzero over more of the life of the longer-term bond. The Swanson-Williams approach suggests an interesting way to come up with an empirical estimate of how the market’s perceived longevity of the zero-lower-bound regime has changed over time.

In particular, Swanson and Williams found that the 2-year yield continues to have at least some response to basic news releases, which could signal that investors aren’t fully persuaded by the FOMC’s declared intention to maintain “exceptionally low levels for the federal funds rate at least through late 2014.”

The 2-year yield didn’t budge when the strong January employment report was released, though it’s been creeping up a few basis points in the two weeks since then.

Source: FRED.

This post originally appeared at Econbrowser and is posted with permission.