The Monetary Base and the Bond Yield Channel of Monetary Transmission in the Age of ZIRP

A Mark Sadowski post

In this post we are going to add US bond yields to the baseline VAR which I developed in these three posts (here, here & here).

In particular, we are going to add the yield of 10-Year Treasury Constant Maturity Securities (GS10) and yield of Moody’s Seasoned Aaa Corporate Bonds (AAA).

The first thing I want to do is to demonstrate that the monetary base Granger causes bond yields during the period from December 2008 through May 2015. Here is a graph of the natural log of SBASENS and of GS10 measured in percent.

Sadowski GC5_1

And here is a graph of the natural log of SBASENS and of AAA measured in percent

Sadowski GC5_2

The following analysis is performed using a technique developed by Toda and Yamamoto (1995).

Using the Augmented Dickey-Fuller (ADF) and Kwiatkowski-Phillips-Schmidt-Shin (KPSS) tests I find that the order of integration is one for all three series. I set up two two-equation VARs in the log levels of the data including an intercept for each equation.

Most information criteria suggest a maximum lag length of two for the VAR that includes 10-year Treasury Bond yields as a variable. An LM test suggests that there is no problem with serial correlation at this lag length. The AR roots table suggests that the VAR is dynamically stable, and the Johansen’s Trace Test and Maximum Eigenvalue Test both indicate that the two series are not cointegrated at this lag length.

Most information criteria suggest a maximum lag length of five for the VAR that includes Aaa Corporate Bond yields as a variable. The LM test suggests that there is a problem with serial correlation, but this problem disappears when the lag length is increased to six. The AR roots table suggests that the VAR is dynamically stable at this lag length, and Johansen’s Trace Test and Maximum Eigenvalue Test both indicate that the two series are not cointegrated at this lag length.

Then I re-estimated the two level VARs with one extra lag of each variable in each equation. But rather than declare the lag interval for the two endogenous variables to be from 1 to 3 and 1 to 7 respectively, I left the intervals at 1 to 2 and 1 to 6 and declared the lag of each variable to be exogenous variables. Here are the Granger causality test results.

Sadowski GC5_3

Thus, the results are as follows:

  • I fail to reject the null that 10-year Treasury Bond yields does not Granger cause the monetary base, but I reject the null that the monetary base does not Granger cause 10-year Treasury Bond yields at the 5% significance level.
  • I fail to reject the null that Aaa Corporate Bond yields does not Granger cause the monetary base, but I reject the null that the monetary base does not Granger cause Aaa Corporate Bond yields at the 1% significance level.

In other words there is strong evidence that the monetary base Granger causes bond yields, but not the other way around.

Since the monetary base Granger causes bond yields they should probably be added to our baseline VAR model. This is because, under these circumstances, we might expect shocks to the monetary base in the VAR model to lead to statistically significant changes in bond yields.

With 10-year Treasury Bond yields added to the baseline VAR model, most information criteria suggest a maximum lag length of three. However, an LM test suggests that there is problem with serial correlation at this lag length. Increasing the lag length to four eliminates this problem. An AR roots table shows the VAR to be dynamically stable.

With Aaa Corporate Bond yields added to the baseline VAR model, most information criteria suggest a maximum lag length of five. An LM test suggests that there is no problem with serial correlation at this lag length. An AR roots table shows the VAR to be dynamically stable.

The Johansen’s Trace Test and Maximum Eigenvalue Test both indicate that there exists one cointegrating equation at this lag length in both VARs. But this is expected, since we already have evidence that the monetary base is cointegrated with industrial production. This matter is addressed in greater detail in the three posts where the baseline VAR is developed.

I am using a recursive identification strategy (Cholesky decomposition), which is the dominant practice in the empirical literature on the transmission of monetary policy shocks. Such a strategy means that the order of the variables affects the results. For the four-variable VARs I am arranging the output level first, the price level second, the monetary policy instrument third, and the financial variable last in the vector. This ordering assumes that the Federal Open Market Committee (FOMC) sees the current output level and price level when it sets the policy instrument, and that the output level and price level respond to a policy shock with one lag, but that financial markets respond to a policy shock with no lag.

As before, the response standard errors I will show are analytic, since Monte Carlo standard errors change each time an Impulse Response Function (IRF) is generated. Here are the responses to the monetary base and bond yields in the four-variable VARs.

Sadowski GC5_4

Sadowski GC5_5

The instantaneous response of bond yields to a positive shock to the monetary base is positive, but statistically insignificant in both cases. This is followed by a statistically significant positive response in the third and fourth month. Furthermore a positive shock to 10-year Treasury Bond yields in month one leads to a statistically significant positive response in the level of industrial production in months three and four, and a positive shock to Aaa Corporate Bond yields in month one leads to a statistically significant positive response in the level of industrial production from months three through five.

The IRFs show that a positive 2.1% shock to the monetary base in month one leads to a peak increase in 10-year Treasury Bond yields of 0.13 percentage points in month four. In turn, a positive 0.17 percentage point shock to 10-year Treasury Bond yields in month one leads to a peak increase in industrial production of 0.19% in month ten.

The IRFs also show that a positive 2.2% shock to the monetary base in month one leads to a peak increase in Aaa Corporate Bond yields of 0.097 percentage points in month four. In turn, a positive 0.13 percentage point shock to Aaa Corporate Bond yields in month one leads to a peak increase in industrial production of 0.21% in month thirteen.

So Quantitative Easing (QE) raises 10-year Treasury Bond and Aaa Corporate Bond yields?!?

Yes, in fact Michael Darda has repeatedly shown that long-term Treasury yields have generally risen under QE programs in figures such as the following.

Sadowski GC5_6

Confusingly, this runs counter to the stated objectives of the Federal Reserve’s Large Scale Asset Purchase Program (LSAP).

If the Federal Reserve purchases 10-year Treasury Bonds in large quantities that should drive the price of 10-year Treasury Bonds up and their yields down, right? Wrong!

That would be true if households or firms did that. But when the Federal Reserve increases the monetary base to purchase 10-year Treasury Bonds it increases expected Nominal GDP (NGDP), or expected aggregate demand (AD), and higher expected AD means higher inflation expectations, ceteris paribus.

Why might an increase in inflation expectations lead to an increase in bond yields? Well one reason would be that an increase in inflation expectations lowers the expected return for bonds causing the demand for bonds to decline and their demand curves to shift to the left.

This is almost certainly true in the case of 10-year Treasury Bonds, since it is unlikely that their supply would increase endogenously to an increase in expected NGDP. Rather, as NGDP increases it is more likely that the quantity supplied of 10-year Treasury Bonds will decrease, as Federal tax revenues increase, and Federal spending on social insurance, such as unemployment compensation, decreases, resulting in a decrease in the Federal deficit. Thus whatever increased demand that the Federal Reserve created for 10-year Treasury Bonds through its LSAPs was almost certainly more than counterbalanced by decreased demand by households and firms due to increased inflation expectations.

Moreover, there is another way that increased inflation expectations may lead to an increase in bond yields. For a given interest rate, when inflation expectations increases, the expected real borrowing cost falls, hence the quantity of corporate bonds supplied increases at any given bond price and interest rate. Thus an increase in inflation expectations causes the supply of corporate bonds to increase and the supply curve to shift to the right.

Furthermore, an increase in NGDP expectations may lead to an increase in the expected profitability of physical investment opportunities. The more profitable equipment and structure investments that a corporation expects it can make, the more willing it will be to issue bonds in order to finance those investments. Thus an increase in expected NGDP may cause the supply of corporate bonds to increase and the supply curve to shift to the right, resulting in a decrease in the price of bonds and an increase in their yields.

And, pointedly, increased nominal spending on equipment and structures probably means increased real output.

Next time I shall add domestic stock market indices to the baseline VAR. I suspect that the results will be less controversial than the results on long term interest rates probably are, but who knows?

7 thoughts on “The Monetary Base and the Bond Yield Channel of Monetary Transmission in the Age of ZIRP

    • Thomas,
      Thank you for the paper. First of all I want to be careful to distinguish the bond market from the concept of “safe assets” as defined by Gary Gorton. There is considerable overlap but, for example, safe assets excludes all nonfinancial sector bonds. Moreover, safe assets includes short term securities such as commercial paper and T-Bills as well as checking and savings deposits and money market funds. I shall demonstrate the results of including deposits in the baseline VAR in a later post.

      However, the results of your paper are broadly consistent with what I have empirically demonstrated here, namely that the Federal Reserve’s LSAP have increased bond yields. Theoretically this must be occurring through some combination of reduced demand for, and increased supply of, bonds.

      With respect to the original Japanese QE, Honda et al estimated a VAR almost identical to the one I have estimated here and showed the original Japanese QE increased the yields of Japanese government bonds by statistically significant amounts (Figure 3).

      http://www2.econ.osaka-u.ac.jp/library/global/dp/0708.pdf

      Also, with respect to the term premium, I replicated the work of Gagnon et al.

      http://www.ijcb.org/journal/ijcb11q1a1.pdf

      over the time period from December 2008 through August 2013 and found results that suggest that LSAP *increases* the term premium. This is of course the complete opposite of what Gagnon et al.’s results were over a pre-ZIRP time period, namely from January 1985 through June 2008. But this is one of my biggest complaints about most studies of the effects of QE: they don’t actually estimate the effects of QE *during* QE.

      In any case my results are in full agreement with your paper’s results concerning the term premium.

      Perhaps I shall bring them up to date and do a post on it.

  1. Thanks for sharing this very thorough work, Mark. Have you thought of adding any restrictions to the model for certain lag lengths for certain variables (restricted VAR, or SVAR)? Also, besides the Feds various announcements, there were several discrete events in this time period (e.g. budget battle of August 2011 and Dec 2012, and the “fiscal cliff” which was also followed by permanent tax rates (Jan 2013) that had large impacts on certain data. Also, I have modeled some of these data for a while (also in EViews) and have found in some cases, a term spread (10/2yr or 10/GS3M) and a risk spread (baa-aaa) have been helpful in a forecasting framework (mainly with BVARS and more traditional structural econometric models). The GS10/GS3M is typically I(1) and (baa-aaa) I(0), of course, depending on the time periods.. The Effective fed funds rate wasn’t much of a market based rate in this period so hence GS3M. When I add an equity market variable (endogenous) into the model its always interesting to see the IRF and eventual market predictions! So I await your results on that!

    • gofx,
      1) Yes, SVAR is on my to-do list. However, right now I’m trying to show the results with a minimum of restrictions, as well as with a model that is as utterly conventional as possible. Hence the moniker “Simple Baseline VAR”.
      2) Eventually I plan on doing a post with spending and tax variables added to the baseline VAR to show the effects of fiscal policy, and to demonstrate how it interacts with monetary policy. So, stay tuned.
      3) Actually I haven’t gotten around to adding term or risk spreads to the baseline VAR yet. Thanks for the head up!
      4) Believe it or not, right now I’m doing this all with a creaky old version of EViews 5. I installed EViews 8 on another computer earlier this year, and consequently qualify for a free update to EViews 9. But for now I’m using EViews 5 simply because that’s what I’m most familiar with. And, as you probably know, Bayesian VAR (BVAR) was only added to EViews starting with EViews 8. So I have no experience with BVAR (yet).

      One reason for doing these posts is in fact to to get feedback on the VAR model, which I’ve been getting not only in the comment threads but also via email. So thank you very much for your comment!

      • Mark, Thanks for your reply. Yes your modeling strategy makes sense, especially with EV5! The nice thing about EV9 is that they added that ARDL option. So you can run motifs of your VAR (as an Equation object) and see how it kicks out different lag lengths for the “dependent variable” and the “indiependent” variables (unlike the unrestricted VAR) (i.e. variations such as gs10 = f(lindpro, lnsbasesns,lpcepi). The default model selection criteria in ARDL is AIC.

  2. Sorry but Mark Sadowski’s long series at Nunes’s blog (which is aptly named “little stories”) seems like rubbish. Sadowski plays with his Mathematica toolkit when he says: “The first thing I want to do is to demonstrate that the monetary base Granger causes bank credit during the period from December 2008 through May 2015.” — keyword: GRANGER CAUSES, which is prone to data mining (e.g., world GDP was found once to be highly correlated to the price of butter in Bangladesh). Hence Sadowski states in this post: “In other words there is evidence that the monetary base Granger causes bank credit, but not the other way around.” (from a mere sample size of 73, from 2009 to 2014, hardly an authoritative sample).

    Proof that Sadowski is engaged in data mining is this passage *MY COMMENTS IN CAPS*: “The response of the Dow Jones Industrial Average to a positive shock to the monetary base is significantly positive from month one through month three. The instantaneous response of the S&P 500 Index is positive but statistically insignificant. FUNNY–SO DJIA SHOWS A RESPONSE TO THE MONETARY BASE BUT THE BROADER S&P500 DOES NOT??? HOW IS THIS NOT DATA MINING?

    More fundamentally, the root question of WHY and HOW the Fed decides to increase the monetary base is not addressed by Sadowski; specifically, does the Fed increase in response to member banks, or does it ‘force’ an increase onto the member banks (the former being the money neutrality hypothesis, the latter being the monetarists hypothesis)? What Sadowski could be measuring is simply a routine business expansion. To wit, Sadowski could be measuring this: leading Fed member banks see a pickup in demand for credit from their Fortune 100 customers, borrow money from the Fed, and lend this money to the Fortune 100; Fortune 100 customers use the money to lend to their suppliers, which through the well-known ‘bank multiplier’ effect ends up in a ‘Main Street’ bank credit expansion several months later. It’s well known that Fortune 100 companies are in the vanguard in any business expansion. So Sadowski is doing nothing more than statistically measuring a routine business expansion. This is not proof that the Fed causes the expansion.

    • Ray Lopez,
      1) The software package in question is EViews, not Mathematica.
      2) If by “data mining” you mean the “process of discovering patterns in large data sets” then I agree.

      https://en.wikipedia.org/wiki/Data_mining

      Granger causality tests are very useful precisely for this purpose. That’s why they are usually considered a necessary precursor to the construction of well specified Vector Auto-Regression (VAR) models.
      3) Thirty observations is the usual minimum to be considered a large sample, although more is usually preferred.
      4) It’s not uncommon for two very similar series to be statistically significant at slightly different time periods.
      5) Since late November 2008 the FOMC has been increasing the monetary base in ad hoc amounts commonly referred to as Quantitative Easing (QE). If these QEs were being done simply to satisfy the “demand for credit” from the Federal Reserve’s “Fortune 100 customers” because of a “routine business expansion” don’t you think there would have been some acknowledgement of this fact by the Federal Reserve, or at least some speculation to that effect by the news media? How is it that I always first get wind of these bizarre theories from you?

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s