Who’s Zoomin’ Who?
The notion of what rates “should be” from my last blog post stirred a few (well…) comments. On one side there was the – valid – objection that there is no equilibrium rate (r*) the same way there is no one rate in the real economy. On the other side was the expected “even lower rates and more unconventional monetary policy (UMP) measures will make the asset price bubble even worse” that I, believe it or not, have quite a lot of sympathy for.
However, while I have some sympathy for that belief, the standard economic framework is quite dismissive of it, as monetary policy is held to have, at best, only temporary effects on real variables.
In today’s blog, I thought we could try to shed some light on this dichotomy, both before and after the global financial crisis, and see if there have been any changes between the two periods, which were both ripe with accusations of monetary policy causing asset price bubbles.
For simplicity rather than for completeness, I chose to interpret the “bubble theory” as a monetary-policy-induced fall in 5y5y forward real rates (and 10y real rates). This is quite opposed to my last blog post, where I assumed that the ever-lower policy rates were a reflection of lower r*.
 To be fair, the notion of money neutrality is often being discussed even among ‘mainstream’ academics, so I see no shame in continuing to hold sympathy for “the bubble of everything” theory. For a recent example, c.f. this)
Macrobond moment: As the statistical analysis adds complexity to the application by default, it is largely underused by Macrobond users. I, however, am a great fan, as straight-forward hypothesis-testing is so much easier to do in in Macrobond in comparison to other, dedicated, statistical tools. Here, I have chosen to use the Impulse-Response Function (IRF) or “dynamic multipliers” to explore both causality and persistence of policy (FFR) and natural (r*) interest rate shocks in a standard VAR setting where I have “controlled for” economic activity and inflation.
The Pre-crisis Sample (January 1999 to July 2007)
As can be seen, the fall in long-term real rates long preceded the global financial crisis, and discussions of monetary policy causing asset bubbles (and, in turn, even the financial crisis itself) were almost as heated then as they are now. Between 1999 and 2005, real (long) rates halved from around 4% to around 2% and, as long rates are often used as a discount factor for other asset prices, it is no wonder that many felt these developments were the cause of the boom in asset prices.
To get an idea of what caused the fall in long real yields, I have chosen to estimate a VAR-model with the conference board’s coincident index, the FRB Dallas trimmed mean PCE, the Federal Funds Rate (FFR) and the 5y real rate in 5y time (5y5y, which is also the chosen measure of ‘r*’) – in that order.
This ordering assures that in an impulse-response simulation, monetary policy does not instantaneously affect either economic activity or inflation (albeit, the opposite might hold true). The reason for putting FFR before 5y5y last is analogous, as it implies that shocks/innovations to r* (5y5y) does not impact monetary policy directly but instead attributes all within-period correlations to FFR. By doing so, we make sure that we – if anything – maximize the explanatory power of FFR on the expense of r*.
First, let’s have a look at 5y5y response to a shock/innovation in FFR:
What this means is that a one-standard deviation (positive) shock to FFR has a very small, negative, and only intermediate effect on r*. The long-run effect on r* from an FFR change approaches zero. Hence, it would seem that the decrease in r* during the period preceding the GFC was not the result of an overly accommodative monetary policy stance.
How about the other way around then? – If a shock to r* (5y5y) has a more pronounced and lasting effect on FFR, it would be a strong indication that it has rather been the low r* that has depressed policy rates.
Admittedly, the effect on FFR from a positive shock to 5y5y (r*) is quite small, and even negative to begin with (which actually makes sense). That said, in time, the effects obviously become positively correlated and seem to be much more permanent, lingering well into the third year and beyond. – In other words, it is much easier to corroborate a story where the causality runs from a less benign outlook on long-term growth and interest rates to short-term – policy – rates, than the other way around.
The Post-crisis Sample (Aug 2007 – Now)
As the policy rate, FFR, has been held stable for more than 5 years (half the sample period), and many UMP-measures have been undertaken during this sample period, caution on the statistical results are indeed warranted. Hence, I have tried a number of different specifications (e.g., different shadow rates) and it is of course possible that yet another specification or approach would be more successful. Either way, you’re welcome to play around with different variables when downloading the Macrobond document.
Somewhat surprisingly, the results are quite stable regardless of what policy rate measure is used. And gladly, the results still hold when adding and combining different dummy-variables for QE-periods and obvious inefficiencies in TIPS-pricing (such as around the first negative yielding TIPS auction).
The response of r* (5y5y) to an innovation in FFR is admittedly somewhat different in the post-GFC sample period than in the pre-crisis period. Initially, there is a small positive correlation, which swings into a small negative effect that gradually dissipates as the IRF approaches zero. All in all, the conclusion here is similar to the first sample period. FFR holds very little sway over r, virtually non-existent in the long run.
What about the opposite relationship, then? – Surely, all the different UMP measures (QE:s etc) have changed the quite clear and persistent relationship between FFR and r*?
Nope, I’m glad to say. – If anything, the response of FFR to innovations in r* (5y5y) is somewhat more pronounced after the GFC and, also, lingers on.
Discussion and Conclusion
Even though the results are reasonably stable to specifications and also make sense over the full sample period, the exact shape and size of the more permanent impact from r* to FFR is somewhat sensitive to the sample length. Smaller and shorter (especially more recent) samples can, e.g., provoke a longer-run impact that is closer to zero. That said, longer samples are definitely to prefer, and here dummies actually make a difference, again stabilizing the results in line with what we’ve discussed above.
Thus, and all in all, our results are conducive to the view that it is indeed r* (FFR) that has lead policy rates (FFR) lower, and not the other way around. Hence, the standard economic model – with money being long-run neutral is verified as we find no or very little evidence of overly expansionary monetary policy having any strong or lasting effects on r*.
In all honesty, I doubt this is the last piece of the puzzle. I cannot even categorically deny any possibility of bubbles being built by a historically accommodative monetary policy. But I do prefer these rather straight-forward exercises to more obscure model builds that seem hellbent on proving one thing or the other. To that end, the work above has at least informed me that I should place the bar high when faced with the not all-too uncommon opinion that lofty asset prices are all the central banks’ doing. – Because, if it’s not market participants doing it then, probably: It’s the economy, stupid!