Simple models and simple-minded reasoning
A blog called Canucks Anonymous has a new post up that exemplifies everything wrong with modern macroeconomics. The basic argument is that we market monetarists have fled in terror from Lawrence Ball’s demonstration that NGDP targeting would lead to greater economic instability:
Greg Ip refers us to this excellent paper from Laurence Ball which shows that targeting NGDP is a pretty lousy idea. One can see immediately that Ball must be spot on from that fact that none of the usual suspects, Scott Sumner or Nick Rowe for example, has even acknowledged the existence of the paper even though Nick actually did a post responding to Ip’s article!
I don’t respond to everything I read, as I don’t have time. And I’m especially unlikely to respond if the macroeconomist I most respect (Bennett McCallum) thinks the argument has no validity. But let’s take a stab at it anyway:
Ball writes down the simplest model you can imagine:
y = -b*r(-1) + l*y(-1) + e
pi = pi(-1) + a*y(-1) + nHere, y is real income, pi is inflation (both as deviations from trend) and e and n are shocks. The notation y(-1) denotes one period lagged income, etc.
I’m already highly irritated. What makes this canuck think that “the simplest model you can imagine” has any bearing on whether NGDP targeting is a wise policy for our highly complex economy? Theoretical macroeconomists love to play with their little toy models, but unfortunately these models don’t actually describe the world we live in. In their models the IS curve slopes downward. Output depends on real interest rates. Inflation depends on output gaps. I don’t know about you, but that’s not the world I inhabit.
Haven’t we seen ultra-low real interest rates for several years? Where is the robust growth? I know, the “error term” takes care of all these pesky little problems. And why is inflation not falling right now? That’s right, the error term explains why. And why did prices rise sharply after March 1933, despite the biggest output gap in US history? Yes, the error term explains why. I prefer models where all the really big phenomena that I’d like to understand don’t have to be explained away by references to error terms.
Balls’ model has monetary policy impact Y with a one period lag and P with a two period lag. In fact, monetary policy affects P with a lead. Or if you consider expectations of future monetary policy to be “policy,” then the effect is contemporaneous–as when rumors of QE2 drove up commodity prices, and hence the headline CPI.
I don’t think the real interest rate measures monetary policy. To the extent that interest rates matter at all, I believe it is the nominal rate minus expected NGDP growth. But I’d rather just leave interest rates out of my “simplest imaginable model,” and stick with NGDP shocks and sticky wages as my explanation of changes in real output. And I’d rather model inflation (or better yet NGDP growth rates) via expected future NGDP, and hence expected future monetary policy. The hot potato effect. I’m not saying the output gap plays no role, but it’s much more complicated than the second equation implies.
Come back to me later when Ball has a model in which output fluctuations are caused by NGDP shocks and sticky wages, not real interest rates.
I also found this amusing:
You can’t cheat this, inflation targeting is simply better.
So inflation is best. Or is Canuck saying that inflation is best in Ball’s model? I can’t quite tell. Canuck doesn’t seem to realize that inflation is something that exists only in the minds of economists, not out there in “reality.” Or perhaps I should say there are as many inflation rates as there are economists. There is the US CPI inflation, which is built on the assumption that using housing prices have risen 7.5% in the past 5 years. Is that the inflation that Canuck wants us to target? Then there is the inflation rate that uses Case-Shiller data, which shows housing prices down 32% in the past 5 years. Given that housing is 1/3 of the CPI, it might be nice to know which of those two inflation rates we should target model. But in my search of Ball’s model I didn’t find any answer to that question. Indeed there are a million inflation rates; are we to believe that all of them are “optimal” policy targets? Is that what Ball’s equations prove?
I don’t work with toy models; I try to stay grounded in the real world. I notice that periods of above 5% NGDP growth (like the 1970s) are viewed as periods where monetary policy was too expansionary. And when NGDP plummets, like in the 1930s, money was too tight. And when NGDP grew at a steady rate of 5%, we achieved the best macro performance in history, the so-called “Great Moderation.” And when we let NGDP collapse in late 2008 and 2009, we had a very severe recession.
I also notice that those who slavishly follow inflation targeting seem to often give the “wrong” advice. And I don’t just mean wrong in the sense that I don’t agree, but also wrong in the sense that almost all sensible economists, even those who don’t believe in NGDP targeting, would beg to differ. Like right now in the UK, when an inflation target would call for much tighter monetary policy. Or indeed even in the US, where the implications of current inflation are ambiguous, but most economists (including I’d guess Ball) think that more demand is obviously desirable.
And that’s not even touching on the political problems associated with inflation targeting. In theory inflation and NGDP are similar targets; whenever either aggregate is too low, the Fed calls for an increase. But as I pointed out in this paper, calling for higher inflation led to a political firestorm in late 2010. In contrast, most Americans would be accepting of a policy aimed at higher NGDP. (An argument I have made many times, and which was recently picked up by Paul Krugman.) But the messy complications of the real world don’t matter to those living in a world of pure mathematics.
I make no apologies for ignoring these little toy models, and having my policy analysis incorporate a complex mixture of politics, macroeconomic history, well-established basic economic principles, and logic.
HT: Bill Woolsey
Update: Josh Hendrickson sent me some comments that might be a more effective rebuttal:
Four points:
1. In my paper on nominal income and the great moderation, I show that a strong responsiveness to nominal income reduces the variability and the output gap in both a New Keynesian model and McCallum’s P-bar model.
2. It’s hard to assess nominal income targeting when nominal income isn’t defined in the model. Monetary transmission has nothing to do with nominal income in these models so we shouldn’t expect nominal income targeting to be optimal.
3. A long standing reason to support nominal income targeting is because there is uncertainty about how fluctuations in nominal income are divided between output and prices. Ball’s model assumes a specific process for inflation.
4. I continue to go back to the point I made regarding Svensson’s paper. The transmission of monetary policy in the model is through output in the first period and only effects inflation with a one period lag. This is the source of instability and I don’t think its a realistic assumption.
Tags:
5. November 2011 at 13:00
Scott, I’m a strong advocate of anything 4% or under level targeting beginning now (or Q1 2011 when the Tea Party GOP took over)…
But one thing your ideas as stated do not have is a complex mixture involving politics. You are to politics, what I am to Monetary Theory, a mere dabbler.
5. November 2011 at 13:03
And you are good at the logic, but you have OBVIOUS extrapolations that you have to be forced to state out loud.
5. November 2011 at 14:00
Morgan, OK, Mr. Perry supporter, I defer to your expertise on politics.
5. November 2011 at 14:05
Scott, I’m on your side but isn’t this: “Come back to me later when Ball has a model in which output fluctuations are caused by NGDP shocks and sticky wages, not real interest rates.”
assuming your main result?
There are a ton of papers that do just that. Bean (1983), Taylor (1985), Mankiw and Hall (1994), Asako and Wagner (1992), Frankel and Chinn (1995) and that’s just off the top of my head. They use a pretty conventional AD-AS model with sticky wages, similar to what you yourself implicitly use, and give qualified support for NGDP targeting against monetary targets and exchange-rate targets. But they don’t address inflation targeting.
Virtually all of the inflation targeting/NGDP targeting debate now occurs in the context of IS-PC models. If NGDP targeting is desirable, shouldn’t that result emerge from a wide class of models? Why is it that the IS-PC-MR models typical come out in favor of inflation targeting over NGDP targeting, or at least give ambiguous results, even when you have a full-blown model with consumer optimization and sticky prices?
I think Ball’s model is basically useless in assessing the relative merits of NGDP targeting versus inflation targeting. It has too many holes, in particular a complete lack of optimization and forward-looking behavior. But even more sophisticated IS-PC models only give qualified support for NGDP targeting.
I’m working on some of these questions myself but it’d be nice to hear from the man himself.
5. November 2011 at 14:16
Integral, You said;
“I’m working on some of these questions myself but it’d be nice to hear from the man himself.”
Thanks for working on this problem, but just to be clear, the “man himself” is Bennett McCallum, not me.
I’m not too worried about the fact that NGDP targeting doesn’t look good in models that I find totally implausible. You seem to suggest I should be worried, and perhaps you are right. Maybe it’s a sort of confirmation bias that leads me to overlook these results. But I don’t think so; I think you go with the models you think are best, and also look at the empirical evidence. I don’t see what more I can do if the models conflict.
I’ve published a paper in the Berkeley Electronic Contributions to Macroeconomics (2006) on targeting CPI futures contracts. So I’d be on board for that policy if someone could convince me it was best, and was politically acceptable. I have no ideological hostility to inflation targeting, as I am basically right wing, and it’s a right wing idea. Instead my arguments for NGDP are pragmatic.
5. November 2011 at 14:31
Integral:
Who are you?
Please keep us informed on your work.
I favored a stable price level for many years. I was persuaded that stable nominal GDP growth is better, because of supply shocks.
Maybe inflation targeting is better with unanticipated supply shocks, but the last four years suggest it is not effective with large aggregate demand shocks. It looks to me that it has allowed us to settle into an equilibrium with output growing below trend, inflation close to target, and output way below potential.
For many years, I was willing to accept that inflation targeting worked well enough. The Great Moderation looked pretty good. While I thought nominal GDP targeting made more sense, the status quo worked ok, (and delivered a stable growth path of nominal GDP.)
Now, we had a shock where it makes a diffence. Inflation has stayed pretty stable really. Sure, the inflation rate was low for a time. The GDP deflator is a bit below trend. And everything is great, right?
5. November 2011 at 14:54
deep in the bowels of the paper, notice that a different Phillips curve produces different results: Indeed, recent work by McCallum (1997) and Dennis (1998) shows that the result is not robust. Reasonable modifications of the Phillips curve, equation (2), produce finite variances of output and inflation under income targeting.
So there no reason to believe the results of this paper are robust. This model is backward looking, and, yes, I’ve found that driving while looking in the rearview mirror can lead to instability. My #1 pet peeve these days is that we have models, we have data, but we seem to have lost the ability to do something scientific, like test models against the data.
The sad truth is that a lot of bank models that are used to price things important, like credit risk, are not much better and just as backward looking. ahem.
5. November 2011 at 14:57
incidentally, I am very good at modeling. tell me what coinclusion you want I’ll put together a model that gets you there.
5. November 2011 at 15:00
Scott: I’m “worried” primarily because inflation targeting is rather entrenched as a macroeconomic paradigm and academics are only going to be convinced by NGDP targeting if it can be shown to outperform inflation targeting in models they’re comfortable with. That means engaging IS-PC seriously. The case is already pretty clear in an AS-AD framework.
Of course if you can get the Fed to embrace NGDP level targeting, I’m sure a flurry of academic work will emerge to justify the policy ex-post.
Bill: I’m a graduate student of economics, fields of interest being time-series econometrics and monetary economics, particularly monetary policy evaluation.
5. November 2011 at 15:08
Apologies for the double-post.
dwb: Ball (1997) has two main results: (1) NGDP targeting is unstable/explosive, and (2) it is inferior to inflation targeting.
Conclusion (1) is fragile, as McCallum and Dennis point out. But Ball claims that (2) is robust to a variety of specifications and it is (2) that Canucks Anonymous is highlighting (as he has stressed to me in a comment on his blog). I have yet to be convinced of this robustness, but that simply requires more digging and actually working through the paper by hand.
5. November 2011 at 16:02
I get the sense the anti-Market Monetarists are just gold nuts, hiding behind the facade of a monetary policy that targets inflation.
The peevish fixation with inflation, really an unhealthy obsession, undermines sensible monetary targets.
To paraphrase a great economist, “You know I think about sex all the time but I try to keep it out of my blog comments. Other people think about inflation all the time, and it becomes their idea of monetary policy.”
5. November 2011 at 17:04
Cole:
Not Adam P.
He favors the dominant approach of setting interest rates to target inflation.
5. November 2011 at 17:48
[…] you’ve been reading this blog for at least two months, you know who said it. No Responses to “The Most Ironic Statement I’ve Heard All […]
5. November 2011 at 18:07
Wow! Now we’ve all seen Angry Scott.
5. November 2011 at 18:12
Integral wrote:
Scott, I’m on your side but isn’t this: “Come back to me later when Ball has a model in which output fluctuations are caused by NGDP shocks and sticky wages, not real interest rates.”
assuming your main result?
Bingo. By the same token, I could insist that the Fed target the price of peanut butter (creamy of course). Then when a macroeconomist challenges me, I say, “Come back to me later when you have a model in which output fluctuations are caused by peanut butter shocks and sticky fingers.”
And speaking of modeling realism, who here remembers all those panicked housewives and day traders running around in September 2008 yelling, “Oh my gosh! The Fed has caused me to lower my NGDP forecast for 1q 2009!!” ?
5. November 2011 at 18:34
People may not have NGDP forecasts; but they don’t have inflation forecasts either, in the sense that price indexes measure inflation, since everyone has their own particular set of prices that they monitor. However, prices and nominal incomes ARE part of people’s planning, and they are hardly disconnected from NGDP.
So some tighter thinking is perhaps needed on everyone’s part here, even those possessed of a caustic wit and ample self-assurance.
5. November 2011 at 18:40
(NGDP also tracks monetary policy expectations.)
5. November 2011 at 19:22
W. Peden wrote:
People may not have NGDP forecasts; but they don’t have inflation forecasts either, in the sense that price indexes measure inflation, since everyone has their own particular set of prices that they monitor.
Two things:
(1) People *were* flipping out about “inflation” in late 2008, when they saw those FRED graphs. Some even foolishly put a prefix before the term. In contrast, nobody was worried about “NGDP expectations” except Scott and three other people who visit this blog. (The number has grown extraordinarily in the meantime.)
(2) If the Canuck or Larry Balls went around ripping Scott for the unrealism of his model, then yes it would be fine to point out that their own models were unrealistic too. But that’s not what happened here. Scott is making it a strength of his approach that it is more realistic than people who rely on this mysterious thing called “inflation.” (As if it’s straightforward to calculate NGDP.) So my point isn’t that NGDP targeting is more or less realistic than inflation targeting, my point is that this isn’t a particularly strong leg for Scott to stand on.
So some tighter thinking is perhaps needed on everyone’s part here, even those possessed of a caustic wit and ample self-assurance.
Indeed, and it should be shared with those speaking in the passive voice too.
5. November 2011 at 19:46
Bob Murphy,
(1) I didn’t say that people weren’t concerned about “inflation” in any sense of that term. It’s important to look at the subordinate clauses in sentences. If the prices of things that you buy were all going up and the price of some other things in CPI that you don’t want were falling dramatically such that the CPI was falling, in what sense would you have low inflation expectations?
(2) That point only stands if economists’ sense of inflation connects in some important way with things that people actually do monitor.
“Indeed, and it should be [i]shared with[/i] those speaking in the passive voice too.”
One should be so lucky.
5. November 2011 at 20:06
“Empirically” NGDPTLT gives the “best picture” being the only period during which both real growth and inflation were stable!
http://thefaintofheart.wordpress.com/2011/11/06/another-nail-in-the-%E2%80%9Ccoffin-of-inflation-targeting%E2%80%9D/
5. November 2011 at 20:14
Perhaps yet more reminding folk of Australia is needed.
The explicit target of of the Reserve Bank is:
The Governor and the Treasurer have agreed that the appropriate target for monetary policy in Australia is to achieve an inflation rate of 2-3 per cent, on average, over the cycle. This is a rate of inflation sufficiently low that it does not materially distort economic decisions in the community. Seeking to achieve this rate, on average, provides discipline for monetary policy decision-making, and serves as an anchor for private-sector inflation expectations.
The RBA’s target of stable growth in prices (P) over the business cycle means that if output (y) falls, then there is more scope to let P rise. Conversely, if y surges, then that means that there is scope for less rise in P, evening out movements in P over the business cycle as per the RBA’s target. In effect, the RBA has been stabilising growth in Py (i.e. NGDP).
And Australia’s macroeconomic stability speaks for itself.
5. November 2011 at 20:29
If you want a simple model of why what the RBA does works, then there is a simple nominal wage story. Rising P when y falls pushes down “real wages” compensating firms for falling sales and so reducing negative effects on employment. Conversely, lower rises in P when y surges passes benefit from output growth on to wage-earners and so reduces upward pressure on wages, beyond actual labour shortages, leading to more growth in employment.
Folk know what is going on so there are much less expectation failures in the Australian economy compared to what the Fed imposed on the US economy when it decided to disinflate surreptitiously.
6. November 2011 at 00:19
[…] Scott Sumner in a recent comment also discuss the relationship between NGDP, prices and quantities in Keynesian and (Market) […]
6. November 2011 at 04:18
[…] 1. Sumners e o modelo de Ball. […]
6. November 2011 at 05:43
Here’s my take and please correct me if I’m wrong: is not the labor function (measured in hours) the most reliable constant? Please remember I only wish my brain was still good enough to absorb the math I should have had. Anyway, it seems that the labor function is the most reliable constant that can possibly be measured, in that – while it steadily decreases as a total portion of production efficiency, it nonetheless does so at a slow rate. Whereas, other parts of the equation such as capital (in the present) are gyrating all over the map until they actually settle into a new equilibrium of resource utilization.
6. November 2011 at 05:48
Integral and dwb:
When you say the IS-PC model, do you mean the model with (1) a forward looking IS curve, (2) a Phillips curve, and (3) a monetary policy reaction function? If so, isn’t true that though (1) can be derived from microfoundations, the Phillips curve is still rather arbitrary, especially it formulating it so that there is persistence?
Finally, how would one do a Phillips curve so that it would fit Sumner’s view of the world?
6. November 2011 at 08:25
Anon1:
When I think IS-PC, I think of
(1) an IS schedule that falls out of the consumer’s intertemporal MRS condition, perhaps with a lag component stemming from habit formation. It looks like
c(t) = Et c(t+1) + b*c(t-1) – (1/d)*r(t) + u(t)
Variables: c is consumption, r the one-period real interest rate, u is a disturbance, Et is the expectations operator. Coefficients: “b” is some function of the coefficient of habit persistence and “d” is the constant of relative risk aversion. This is easy to derive, particularly the case of b=0. Just use a two-period model of consumption with CRRA utility, set up the MRS condition and rearrange.
(2) A Phillips Curve that stems from supply behavior. What we want is a function like
p(t) = a*Et p(t+1) + (1-a)*p(t-1) + (1-a)*k*y(t) + v(t)
where p is inflation, y is output and v is a disturbance. Constant “a” determines the degree of forward-looking behavior and “k” is some function of the optimal markup. If y is contemporaneous this is just a modified AS schedule. In some cases y is lagged by a period.
It is really, really hard to get a decent Phillips curve from the microfoundations. You need some sort of staggered price setting (ad-hoc Calvo pricing) and getting persistence is even more ad-hoc.
Notice that expectations are key in both equations.
How do we fit Sumner into this framework? That’s a big topic and not something I can cover in a comment.
🙂
6. November 2011 at 08:33
Thanks Integral. You answered (I think) my biggest question on whether the Phillips Curve is somewhat arbitrary. And that to me seems the key problem with Adam P. and Ball’s critique.
6. November 2011 at 09:27
This paper:
The IS-LM Model and the Liquidity Trap Concept: From Hicks to Krugman
http://www.econ.ucdavis.edu/faculty/kdhoover/pdf/hope/boianovsky.pdf
It contains this 1969 quote from Hicks:
“Expectations of the future (entirely rational expectations) are based upon the data that are available in the present. An act of policy (if it is what I have called a DECISIVE action) is a significant addition to the data that are available; it should result, and should almost immediately result, in a shift in expectations. This is what I mean by an announcement effect (p. 315; italics in the original [replaced by caps for this comment]).”
There’s a lot more besides on modern [up to 2003] developments in the IS-LM model.
There are basic problems with all such models, such as the Sonnenschein-Mantel-Debreu theorem.
http://en.wikipedia.org/wiki/Sonnenschein%E2%80%93Mantel%E2%80%93Debreu_theorem
Still there’s an advantage to refuting arguments in their own terms, even if you have problems with the terms. Hick’s had a few things to say about the IS-LM model in post-1937 work. The original paper doesn’t represent his final views.
6. November 2011 at 11:05
Anon1,
Yeah, there’s no real consensus on the Phillips Curve, even now. The use of lags, contemporaneous, or expectations for the output-gap term is arbitrary and varies from author to author; similarly with inflation persistence elements. Theorists leave out persistence entirely, empiricists leave it in to better fit the historical data.
While the Phillips Curve ought to evolve from the firm’s FOCs, and Gali (2008) does a good job motivating one specification, the PC as a whole is not on solid microfounded grounds, handwaving about Calvo pricing aside.
I could teach any undergrad who’s had intermediate macro to derive a modern, dynamic IS curve and it’d make sense to them, I couldn’t do the same for the PC. That’s not good in a “standard” model.
6. November 2011 at 16:48
“Come back to me later when Ball has a model in which output fluctuations are caused by NGDP shocks and sticky wages, not real interest rates.”
I love many things on this blog, but I have to admit, I am sad to see such an intellectually dishonest sentence here. You cannot just destroy other researchers’ arguments by telling them to come back once they have a model that fits nicely your own personal view of the world. I’ll show it to my students as a warning…
7. November 2011 at 05:25
Just curious what would be a good guiding principal for calculating politics into an economic model? Thoughts anyone?
7. November 2011 at 05:55
[…] Sumner is rightfully put off by this assertion Greg Ip refers us to this excellent paper from Laurence Ball which shows […]
7. November 2011 at 06:45
[…] Sumner gives a good example. Advertisement Eco World Content From Across The Internet. Featured on EcoPressed IBM's […]
7. November 2011 at 09:26
Adam P has a response up:
http://canucksanonymous.blogspot.com/2011/11/simple-minded-non-response-from-scott.html
7. November 2011 at 12:58
and another:
http://canucksanonymous.blogspot.com/2011/11/ngdp-targeting-and-supply-shocks.html
7. November 2011 at 13:43
The problem is, the formulation of NGDP targeting in Ball’s paper seems to be wrong. Although he defines y and pie as deviation from trend, he seems to forget his own definition in the NGDP targeting formulation. If you count that in, NGDP targeting becomes one of his optimal solutions.
I showed that in the comment section of
http://canucksanonymous.blogspot.com/2011/11/monetarists-misunderstanding-other.html
9. November 2011 at 07:09
It’s really easy to propose a theory that you refuse to submit to a quantitative test. It just isn’t science.
9. November 2011 at 14:33
dwb, Good points. Remember the old acronym: GIGO
Integral, You said;
“Scott: I’m “worried” primarily because inflation targeting is rather entrenched as a macroeconomic paradigm and academics are only going to be convinced by NGDP targeting if it can be shown to outperform inflation targeting in models they’re comfortable with. That means engaging IS-PC seriously. The case is already pretty clear in an AS-AD framework.”
I’m not expert enough to do this, but obviously a skilled theorist could do so, as the IS-LM model allows for any result (once you let monetary policy shift the IS curve.)
Ben, That’s too tough, the NKs don’t favor gold.
Bob, Just wait until January!
I didn’t just say I don’t like Canuck’s model, I explained why. If he doesn’t like my model, he can explain why–then I’ll have something to respond to.
You keep saying this:
“(1) People *were* flipping out about “inflation” in late 2008, when they saw those FRED graphs. Some even foolishly put a prefix before the term. In contrast, nobody was worried about “NGDP expectations” except Scott and three other people who visit this blog. (The number has grown extraordinarily in the meantime.)”
And you keep failing to address my answers. That won’t work when we debate.
Marcus, Thanks, I find your empirical evidence to be much more persuasive than those toy models.
Lorenzo, Yes, Australia’s a good example.
Becky, I see hours worked as the best indicator of the real economy over the busienss cycle. Is that what you meant?
Peter, I don’t have any problems with those Hicks comments–and i like his 1937 paper.
Turpentine, You said;
“I love many things on this blog, but I have to admit, I am sad to see such an intellectually dishonest sentence here. You cannot just destroy other researchers’ arguments by telling them to come back once they have a model that fits nicely your own personal view of the world.”
If you read my entire post you’ll notice I made several specific criticisms of the Canuck model. It’s up to others to criticize my model–then I can respond. That’s what I meant.
RHD, That would be hard.
Jason and Adam, Thanks, I’ll take a look.
Himaginary, Interesting. BTW, I don’t know if you read all my posts, but I mentioned one of your comments a few weeks back.
ScottSumneryoufool, You said;
“It’s really easy to propose a theory that you refuse to submit to a quantitative test. It just isn’t science.”
I agree. And love the name.
11. November 2011 at 06:33
“Himaginary, Interesting. BTW, I don’t know if you read all my posts, but I mentioned one of your comments a few weeks back.”
Yes, I read it. Thank you for mentioning.
14. January 2012 at 10:48
[…] the first claim to exhaustion. For a review see Adam P’s first post on the paper, replies by Scott Sumner and Bill Woolsey, Adam’s rejoinders (1,2), Adam again, and a contribution from Nick […]