No, The Fed Is Keeping Rates Low

As is now well known, Ben Bernanke has just launched a blog (welcome!) at the Brookings Institution. His first analysis post was "Why are interest rates so low?". He offers an explanation using textbook economic theory. As one might guess, I disagree with him about the framework used. However, my argument here is that even if I grant that his framework is (roughly) correct, he is still not able to draw the conclusions that he does.

Not Total Disagreement

Within the article, he disputes one of the sillier ideas that floats around on the internet.

A similarly confused criticism often heard is that the Fed is somehow distorting financial markets and investment decisions by keeping interest rates “artificially low.” Contrary to what sometimes seems to be alleged, the Fed cannot somehow withdraw and leave interest rates to be determined by “the markets.” The Fed’s actions determine the money supply [BR - questionable.] and thus short-term interest rates; it has no choice but to set the short-term interest rate somewhere. So where should that be?

Other than the part about the money supply that I noted, this is a succinct version of what I discussed in my detail within my article "Yes, Interest Rates Are Artificial." Correspondingly, I think his point here is brilliant. My concern about the quip regarding the money supply partly raises a somewhat theoretical argument*, but I believe that his statement does not reflect how monetary operations are implemented. I may tie this in with my series of posts on government operations in a later article..

Long-Term Equilibrium Rates

The key point of his argument is within this paragraph.

If you asked the person in the street, “Why are interest rates so low?”, he or she would likely answer that the Fed is keeping them low. That’s true only in a very narrow sense. The Fed does, of course, set the benchmark nominal short-term interest rate. The Fed’s policies are also the primary determinant of inflation and inflation expectations over the longer term, and inflation trends affect interest rates, as the figure above shows. But what matters most for the economy is the real, or inflation-adjusted, interest rate (the market, or nominal, interest rate minus the inflation rate). The real interest rate is most relevant for capital investment decisions, for example. The Fed’s ability to affect real rates of return, especially longer-term real rates, is transitory and limited. Except in the short run, real interest rates are determined by a wide range of economic factors, including prospects for economic growth—not by the Fed.

My complaint is that the "the person on the street" (or even "the person on Wall Street") is true in a wide sense, not a narrow sense.

The reaction from some post-Keynesian bloggers was fairly critical, such as Ramanan's response in "Disappointing Start, Mr. Bernanke".  But as @neilwilson pointed out to me, the best pre-rebuttal was summarised at interfluidity in "Michal Kalecki on the Great Moderation" (and Kalecki himself was writing in 1943).

However, I want to follow the editorial line taken by "circuit" in "Ben Bernanke and the natural rate of interest."

Now, I realize that the equilibrium real rate is unobservable and varies through time, which means it's subject to uncertainty. However, we could say the same thing about the concept of potential output, yet few would deny it is a useful concept.
In fact, most people are aware of the concept of "output gap", the difference between potential output and actual output. The corollary concept for the real interest rate is the "interest rate gap", the deviation of the actual policy rate from the real equilibrium rate.

I am willing to grant that the "equilibrium real rate" is an unobservable model variable. My argument is that Ben Bernanke still cannot draw the conclusions he does even if this is true.

Theoretical Frameworks Versus Mathematical Models

Dr. Bernanke is basically writing in the same way he would address the students that he used to teach, and following standard mainstream logic. I would paraphrase his argument in this way (although I am extending his logic in what I believe is a reasonable fashion).

• We assume an economic framework in which the real rate of interest has a strong effect on behaviour.
• This framework is converted to a simplified mathematical (typically using log-linearisation), and that simplified model is fit to real world data.
• The fit is statistically acceptable (although it would not be able to predict things like the financial crisis, as that crisis was allegedly the result of a non-forecastable "productivity shock"). Therefore, we can argue that the model is "correct", statistically speaking.
• Since the rate of inflation was "acceptable" to policymakers during the post-1982 era, the policy interest rate was near the model-predicted "equilibrium" level (which changes over time). This holds because if had been held elsewhere, the model predicts that there would have been a large deviation of the inflation rate (higher or lower).
• Therefore, policymakers were following this equilibrium, and so it was not a conscious choice to bring about low real interest rates.

Missing Ingredient - Model Error

The mainstream economic modelling framework implicitly assumes there are two sources of error (or risk) within macroeconomic modelling.

1. The parameters of the fitted model are estimated incorrectly. (These are the "structural parameters" of DSGE models.)
2. The state of the system (such as, what is the output gap in March 2015?) is estimated incorrectly.

They use was appears to be highly sophisticated statistical analysis (typically Bayes' Theorem based approaches) to show that these estimation errors are acceptable. This is exactly the type of logic used by Optimal Control theorists in the 1960s. As I discussed previously, optimal control was abandoned by engineers after it killed a test pilot. From the point of view of a control systems theorist (which I am), mainstream economists are trapped in a 1960s time warp.

What is missed is the third type of risk - model risk. 

We do not know the exact model of any engineering system. For example, in aerospace, the airframe bends, and we cannot hope to capture the higher harmonics of the oscillations. But if we want sophisticated engineering systems, we cannot give up on modelling and collapse into a funk. Instead, we need to adopt approaches that explicitly incorporate model errors, which is an approach named "robust control".**

Once we take model risk into account, things get interesting.

Assume that we have a nice mainstream model, which fits the data acceptably. As I argued earlier, the realised policy rate would have been close to the equilibrium rate.

Then imagine that there is a perturbed version of the model, which has the same dynamics as this original model, under the assumption that the "fixed parameters" are the same as the original, and the policy rate is very close to the "equilibrium real rate" in the first model. (We can infer that the "equilibrium rate" exists within the second model, and is the same as the first.)

However, the difference is that for this perturbed model, the effect of the real policy rate is less than the original model if we move away from equilibrium. This might be the result of inertia in entities' economic behaviour (for example, the utility function penalises changes changes in production levels versus their levels in the previous accounting period). In other words, this model has a lower sensitivity to the real rate than the baseline model.

This new model could be fit the data in the same way, and generate a similar quality of fit. (It may be that the "best fit parameters" are slightly different, to take into account the small deviations of the realised policy rate from the model-predicted equilibrium.)

Taking this new model into account, we can no longer draw the conclusion that Ben Bernanke does. If the economy is not as sensitive to the policy rate as he assumes, it could easily be that the equilibrium nominal policy rate in the United States right now is 2%, not the 0% that he is taken as given. The Fed may have been deliberately keeping the real rate of interest several hundred basis points too low, but was not enough to have an appreciable effect on inflation. In other words, this was a policy choice by the Fed.

Since these two models are essentially indistinguishable unless the Fed deliberately engineers a huge deviation from the model-predicted real rate, no amount of statistical obfuscation can tell us which is correct. It turns into a judgement call. In my opinion, the fact that the estimated "equilibrium real rate" swings around so much is a sign that real rates are not that powerful a driver for the economy. But obviously, opinions differ.

Concluding Remarks

It is nice to see Ben Bernanke set up a blog, as he should be able to clearly enunciate the mainstream view. But we need to delve beyond the high-level academic generalities to see what the important distinctions between the mainstream and heterodox economic view are.

Footnotes:

* I am not going to dredge up the word "exogenous" here.

** There has been some attempts to introduce robust control into economics, such as in the book "Robustness" by Hansen and Sargent. However, they adopted the worst possible way of looking at robust control, which was a formulation relying upon game theory. Within this approach, model error disappears, it's just that you are contending with "malicious disturbances". This is mathematically equivalent to what I view as the proper way of looking at robust control only in the case of linear systems. My feeling is that the economists love game theory and adopted that methodology as a result, and completely lost sight of the issue of model risk.

(c) Brian Romanchuk 2015