Will somebody please think of the Data Generating Process!

Or: "Should Finance People be allowed out unaccompanied by a macroeconomist?".

If the central bank is doing its job right, monetary policy ought to appear to be irrelevant.

The Economist says (HT Mark Thoma) "Interest rates do not seem to affect investment as economists assume", and this means that monetary policy is irrelevant. (Let's just set aside the fact that monetary policy is not interest rate policy.) The Economist refers us to an empirical study by S.P. Kothari, Jonathan Lewellen, and Jerold B. Warner.

Read their second paragraph below (I messed up the copy-and-paste, as usual):

Take a simple macro model in which investment is a negative function of the rate of interest. The second-year ISLM model will do fine for our purposes. Then ask yourself this question:

If this model were true, under what circumstances would we expect to observe data showing a negative correlation between investment and the rate of interest?

Hmmm. Well, if there weren't any shocks, so the IS and LM curves never shifted, we wouldn't observe anything at all, except for one point. So we need to introduce some sort of shock. What sort of shock? Here are two examples:

1. Monetary policy shocks. Assume the central bank does random stupid things, that shift the LM curve up and down. Yep, then we would observe a negative correlation between investment and interest rates. But notice we would also observe a positive correlation between investment and GDP. And an econometrician who estimated a multiple regression with investment on the LHS, and both GDP and the rate of interest on the RHS, would be unable to tell whether it was GDP or the rate of interest that was driving investment. Because GDP would be perfectly (negatively) correlated with the rate of interest. (How do you spell multicolinearity?)

2. Investment shocks. Assume there are shocks to investment demand, that shift the IS curve up and down. Unless the LM curve is horizontal and fixed (because the central bank sets a rate of interest then falls asleep) we would observe a positive correlation between investment and the rate of interest. (And we would observe a positive correlation between investment and GDP.) And if the central bank were doing its job right, and stabilising GDP in the face of IS shocks, it would make the LM curve vertical at potential GDP (or shift the LM curve in response to IS shocks to keep GDP at potential, if you prefer), and if saving were a positive function of the rate of interest, the econometrician would observe a positive correlation between the rate of interest and investment, but no correlation with GDP.

3. I could do other shocks, but you should get the idea by now:

If you want to observe the effects of monetary policy in the data, what you need is a really stupid monetary policy where the central bank makes big changes to monetary policy for totally random reasons.

That's how controlled experiments work. That's how they test drugs. They flip a coin to decide who gets the drug and who doesn't. Which is a really stupid random drug allocation policy. Except for generating useful data.

If you want to generate some useful data to estimate the effects of monetary policy on things like investment, GDP, inflation, etc., you want a really stupid random monetary policy.

If monetary policy looks irrelevant, or has the "wrong" sign, that probably means it was good monetary policy. The "shocks" to monetary policy weren't shocks. They were responses to other shocks. We want central banks to respond only to other shocks.

This is macro 201 plus econometrics 201.

I have blogged about this many times before. If a central bank is using instrument I to target a variable T at a time-horizon h, then it sets E{T(t+h)/I(t)}=T*, and so T(t+h) should be uncorrelated with I(t) (and uncorrelated with anything else in the central bank's information set at time t). Because the central bank's forecast errors should be uncorrelated with anything in its information set. A non-zero correlation means the central bank is making systematic mistakes. But the correlation could be of either sign, because the central bank might be making mistakes in either direction (over-reacting or under-reacting to shocks).

Comments

[JKH]

In addition:

Just scanning everything quickly, its not clear that this study includes new housing investment.

My guess is that it doesn't. There's too much emphasis on corporate profits and retained earnings as a source of funds for investment.

If so, that's a pretty major omission in terms of the overall interest rate sensitivity of the economy.

[Nick Rowe]

JKH: Well-spotted! And houses being a very long-lived capital good, it ought to be the case where investment would be most interest-sensitive. But whether we would observe that negative relationship in the data depends on the source of the shock. If some non-interest shock causes housing investment to fall, or be expected to fall, we should expect to see the central bank cut interest rates in response.

[Luis Enrique]

I thought it was also odd they compared role of interest rates to things like being very profitable, having your share price rise. Of course these things are going to have far larger effects because they separate out successful growing companies investing in expansion from the rest. Monetary policy is something that should (may) affect all firms changing decisions at the margin, it's never going to be anything like as 'important' in the sense of explaining investment as firm specific characteristics.

[Nick Rowe]

Luis Enrique: Yep. In pooled cross-section/time-series data, it wouldn't be surprising if we saw firm-specific shocks being more important than aggregate shocks. But this study is looking only at aggregate data. It's macro.

[Luis Enrique]

Nick - er, look at first graph in economist article, isn't that comparing effect of interest rates against effects of firm specifics like share price movements?

[Nick Rowe]

Luis: yes, but from the study itself, I interpret that to mean the effect of the share price index (like the S&P500) on aggregate investment.

[Too Much Fed]

"Interest rates do not seem to affect investment as economists assume"

I think you will find firms react to quantities (actual and next year's expected) and not to prices for the most part.

[Miami Vice]

Why wouldn't you expect the same interest rate response from the market to investment "shocks" in the absence of a central bank?

[Brian Romanchuk]

As a person with a background in finance, I feel obligated to make a comment.

I have not had time to think about the article you referenced, but I have a related example. The New York Fed recently published an analysis of their DSGE model. In it, they show the shocks to the economy during the financial crisis. Monetary policy is shown to be stabilising. Fiscal policy (automatic stabilisers) is notable by its absence.

You could explain the tendency for the economy to return to steady state growth as a result of:
(1) monetary policy, which is typically set in a counter-cyclical fashion;
(2) passive fiscal policy, which is counter-cyclical by definition;
(3) a combination of (1) and (2).
(4) None of the above! (More room for research...)

It does not seem obvious that we can easily test which view is correct from observed data, for reasons you describe.

Also I would note that many consider current U.S. monetary policy to be "stupid" if not insanely "stimulative". Therefore, one could argue that a crazy experiment in monetary policy is underway.

It seems that a lot of structural changes are being dreamed up to explain why negative real interest rates are not resulting in accelerating growth in the United States and Canada. Conversely, the view that passive fiscal settings are too tight does not need to keep adding these special factors to explain slow growth. Using Ockham's Razor, one might argue that the hypothesis that monetary policy is dominant could be discarded.

(Note that I am sympathetic to the view in the previous paragraph, but I have not looked at it in enough detail to have much confidence in the argument.)

[Scott Sumner]

Nick, I read the article in The Economist a couple days ago, and thought the same thing. Then I thought the article itself must surely be more sophisticated, everyone knows about the identification problem, don't they? I guess not. Good catch.

Brian, You said:

"Also I would note that many consider current U.S. monetary policy to be "stupid" if not insanely "stimulative"."

I agree. And many others consider those people to be stupid, if not insane.

[Nick Rowe]

TMF: firms probably respond to both prices and quantities. But that makes monetary policy even more powerful, because you get a feedback multiplier effect via quantities.

Miami: Depends on what monetary system we have instead of a central bank. But if we used (say) gold as money, with no central bank, then yes; shocks to investment would cause a positive correlation between investment and interest rates.

Brian: yes, whether or not the economy would be self-stabilising will depend on monetary policy. With a bad enough monetary policy, it won't be stable.

There are degrees of stupidity in monetary policy. The "ideal" monetary policy from the point of view of generating useful data would be extremely stupid from the point of view of having a good economy. Central bankers tossing a coin each year to decide what to do.

Scott: yes, it's a form of the identification problem, but a very extreme form. It's like the very worst possible experimental design anyone could dream up. Like testing a new drug by giving it only to the sickest people. And then observing that the experimental group are no healthier than the control group, and concluding that the drug does not work.

[Miami Vice]

Isn't a finite amount of bank capital enough? Why is a finite supply of money necessary?

[Nick Rowe]

Miami: no, that is not enough to tell me what the LM curve looks like, and what makes it shift. Beta banks (commercial banks) promise to convert their money at par into alpha bank (central bank) money. They follow the lead of the alpha bank. If there is no central bank, what do they promise to convert their money into? Who, or what, do they follow? And what does that alpha person or thing do, when there is a shock to investment? We need to know who is alpha, before we can even think about the question.

[Christiaan Hofman]

When I was first reading the Economist piece the thought that immediately jumped out was "Friedman's thermostat".

[Nick Rowe]

Christiaan: Good! And in this case, even if Friedman's thermostat is working only part of the time, it could still be enough to eliminate the negative correlation between interest rates and investment.

My old post on Friedman's Thermostat

[Dan]

It is curious that three guys at rather distinguished institutions aren't professionally embarrassed to put out a paper like that. And it's curious the economist isn't embarrassed to print a story on it.

I don't know why Thoma linked to it, I couldn't even do more than skim past the first paragraph when its utter silliness became immediately apparent - which seemed certain just from the title.

It looks as if the paper was from 2011? It's curious why now?

[Phil Koop]

You didn't comment on David Glasner's recent post, which in my view made an analogous point: while it's amusing to laugh at the signatories to the open letter warning Bernanke of a looming inflation which never arrived, the real issue is that inflation is exactly what was supposed to have been achieved had the Bank's policy been successful!

Which makes all this pointing and laughing a bit of an own goal. Sure, Krugman et al undoubtedly have a more nuanced position distinguishing between inflation and uncontrolled inflation, but that's not how the message comes through. And arguably, the inability of the Bank to admit its goals is the main reason why they have not been reached. How can the Bank "credibly" promise to be irresponsible, in Krugman's phrasing, if it can't promise to be irresponsible in the first place? The concrete steps are vitiated by the lack of the abstract ones.

In it's own way, central banks of the Argentina and the US are equally compromised; but whereas the former is compromised on inflation, the latter is compromised on disinflation. So far, the evidence suggests that the Argentinian problem is less harmful than the American.

[louis]

Phil - if you read the letter, they said the policy ran the risk of runaway inflation, *without any benefits*!
Far more ridiculous to think that you can create immaculate inflation without closing the output gap than to think that you can get under-trend inflation back up to trend.

[Phil Koop]

@louis, I'm not saying the signatories aren't ludicrous. I'm saying, let's laugh at them for the right reasons.

[Nick Rowe]

Dan: those are good questions. I don't know the answers either, and find it curious too.

It reminds me of an example, from Labour economics I think. I can't remember the details, but there are 3 variables, X, Y, and Z. One group of labour economists regresses X on Y and Z. A second group regresses Y on X and Z. A third group regresses Z on X and Y. They all see just part of the big picture, and want to explain their part.

Phil: Yep, David Glasner's post is related. Central banks usually (Zimbabwe being one exception) don't just print lots of money for reasons unrelated to the demand for money. We can learn more about the effects of monetary policy from watching Zimbabwe than from watching the US. But we need to make a distinction between creating inflation relative to what it was in the past, and relative to what it would have been otherwise.

"And arguably, the inability of the Bank to admit its goals is the main reason why they have not been reached. How can the Bank "credibly" promise to be irresponsible, in Krugman's phrasing, if it can't promise to be irresponsible in the first place? The concrete steps are vitiated by the lack of the abstract ones."

Yes.

[srin]

Nick,

If we were to extend your logic, we cannot run any macro regression unless we know the central bank reaction reasonably well. It is not just the relation between interest rates and real variables, but the relation between employment and inflation. So,all those reams of papers published showing so-and-so calibrated RBC models fitting data well are also comrpomised. Essentially macro empirical research is not possible because you can never recover even the true "shocks" to monetary policy unless the model is correctly specified.

[Nick Rowe]

srin: I worry about that too. How can we ever know anything? I think we have to look at examples where there was (presumably) some exogenous change in monetary policy, and see what happens. Like the 1982 recession. Or gold discoveries.

I had some more thoughts about this towards the end of my Friedman's thermostat post

[Min]

Brian Romanchuk: "passive fiscal policy, which is counter-cyclical by definition;"

What do you mean by "passive fiscal policy"? Do you consider the US sequester, for instance, to be "active"?

[Nick Rowe]

Min: Brian is (almost certainly) talking about "automatic stabilisers": the idea that tax revenues fall in a recession, even with the same tax rates, so the budget goes automatically towards a deficit. (But it's not *strictly* true that it's countercyclical *by definition*.) But yes, the distinction between "active" and "passive" gets very fuzzy, and it's not obvious the distinction really matters.

[Min]

Thanks, Nick. And Brian. :)

[Too Much Fed]

“TMF: firms probably respond to both prices and quantities. But that makes monetary policy even more powerful, because you get a feedback multiplier effect via quantities.”

Let’s go with your definition of monetary policy that I don’t agree with.

I think that will also depend on the hot potato effect working as advertised. I don’t believe the hot potato effect works as advertised.