Thursday, September 27, 2012

Oh NGDP, is there anything you can't do?

In a long post over at Free Exchange, Ryan Avent passionately defends the honor of NGDPism against any and all heretics. This time it's GMU wonder boy Eli Dourado in the cross-hairs.

But Mr. Avent's analysis, to speak plainly, makes no sense at all. Let's break it down.

Mr. Avent starts with this chart, which he argues shows how NGDP growth is strongly related to job growth and thus crucial to economic performance:

But people here's the thing. NGDP is PY, the price level times real income. And one of those components, real income, is indeed highly correlated with jobs. The other component, prices, is not.

Let's look at how inflation (growth in the price level) and real GDP growth each chart up with jobs growth over Mr. Avent's sample period, starting with inflation:

The linear correlation between inflation and jobs growth shown here is 0.22, which is INSIGNIFICANTLY DIFFERENT FROM ZERO (t-stat = 1.01). In other words, one component of NGDP, prices, is not significantly correlated with job growth over this period (not even at the 0.20 level). Another way to see the lack of correlation is to note that if we ran a regression of jobs growth on inflation here the r-square would be around 0.05.

Now let's look at the other component, Real GDP growth:

The linear correlation between real GDP growth and jobs growth shown here is 0.64 which is SIGNIFICANTLY DIFFERENT FROM ZERO (t-stat = 8.2) at the 0.01 level. The r-square of this regression would be around 0.41.

In other words, Mr. Avent is using fluctuations in the real economy to explain fluctuations in the real economy. When real output tanks, jobs tank. It does appear from the graph that GDP growth might lead job growth, but we are really just graphing the same thing twice here.

People there's plenty more to come after the jump:

Mr. Avent then doubles down and says the close relationship between NGDP growth and jobs growth doesn't always hold, and shows another chart from a period where NGDP growth was "too high":

Mr. Avent considers this 1960-1980 period to be a different world than the more recent period discussed above. 

But, you know what people? It's actually exactly the same world, and that is a world where Mr. Avent is wrong.

Let's go to the breakdown:

Here's the inflation - job growth graph for the period Avent chose:

Here the linear correlation between inflation and jobs growth is negative -0.15, but it's still NOT SIGNIFICANTLY DIFFERENT FROM ZERO, just like in the more recent period Mr. Avent initially used. The r-squared of a regression would be just about 0.02.

Now the Real GDP growth - job growth graph:

The linear correlation between real GDP growth and job growth here is 0.66, virtually indistinguishable from the 0.64 correlation in the first time period, and significantly different from zero at the 0.01 level.

Both periods Mr. Avent uses are periods where job growth is uncorrelated with inflation and very highly correlated with real GDP growth. It's only the misguided insistence of the NGDPist in treating these two very different things equally that fools Mr. Avent into seeing two different worlds.

In sum, Mr. Avent's charts and post are simply nonsense.

There are two major problems here.

The first is that using real growth to explain job growth, and make no mistake, Mr. Avent is doing exactly that, is like explaining Y with Y. They are two measurements of the same concept; the overall level of economic activity. One is not explaining the other, they are moving together expressing the same overall economic concept.

The second, and overall more troubling problem, is this strange tenet of NGDPism, that one shouldn't separate nominal income growth into real growth and inflation, that NGDP fluctuations somehow have the same effect on the economy whether they come from inflation or real growth. That is just an elementary economic error that the NGDPers are somehow trying to turn into a theorem!

Think about it like this. Your boss tells you that NGDP is surging and wants to know if it will be profitable to hire more workers and expand the business. To give a good answer, as I've just illustrated, you absolutely want to know WHICH COMPONENT, P or Y is surging.  If you just go by the NGDP aggregate, you are really hurting your chances of keeping your job.

Another way to say this is that there are many ways to achieve any rate of NGDP growth. But not all of them are equally desirable.  Say we want 7% NGDP growth. Are we really indifferent between 2% real GDP growth and 5% inflation vs. 5% real GDP growth and 2% inflation (yes I know I'm taking a minor math liberty in these examples)?


John Thacker said...

However, isn't there a problem that if the central bank is actively targeting inflation, as it is in the latter time period, one would expect no correlation between inflation and jobs (or nearly anything else)? Indeed, the only time when there appears to be a correlation is when the Fed let inflation drop significantly below the target of 2%.

You can't easily throw around correlations and call them meaningful when an actor is specifically adjusting one. It would be more interesting to see the correlation between inflation and jobs if the Fed didn't act to maintain inflation in place.

An analogy-- suppose one is traveling in a car on a very hilly road. The driver targets a constant speed, applying brakes on downhills and the accelerator on uphills. The result, if graphed, will be little to no correlation (depending on driver skill) between gradient and speed, or between brake and acceleration and speed.

Mind you, I don't like Ryan Avent's argument either.

John Thacker said...

In fact, I believe that by their own argument, you would get exactly the results you depict. They would say that "the Fed is targeting inflation (and only inflation, contrary to its dual mandate) so we expect no correlation between inflation and jobs or growth in the economy, since Fed action smooths out inflation and removes any overt relationship. There may have been times when inflation would have happened if not for Fed action, which should be labeled in any chart. However, targeting based on both price and income would be superior, even though that would further reduce correlation."

I don't know if I trust the last bit, and I don't find this chart of his particularly illuminating, but the lack of correlation anything with inflation isn't meaningful when the Fed is consistently and successfully targeting inflation.

Anonymous said...

@ John Thacker-

The argument made that is being rebutted is that (quick and dirty version) raising NGDP will raise the real economy. If the Fed is "smoothing" inflation and inflation "smooths" employment then there should still be a correlation between the two even with intervention. NGDPers are in fact making the claim that inflation as a tool does impact real employment (yes their argument is more subtle than this) and this is their basis for arguing not only for a adherence to the trend but to hold that trend positive. Consider the fact that NGDPers never argue for a predictable, stable changing NEGATIVE NGDP. If all that is important is the stability or smoothness then they should have no preference for a positive or a negative sign.

James Oswald said...

I was planning to comment that NGDP correlates with employment closer than GDP (R2 of .49), but John Thacker is correct, neither fact is meaningful. You'd have to compare unemployment instability in a inflation targeting regime to a NGDP targeting regime, not simply run a regression between the two variables. Until someone actually tries NGDP stabilization, you can't empirically say which is better.

Also, the Fed can't target RGDP anyway. If they could target real variables, they might as well target employment directly.

kebko said...

Isn't one of the arguments that NGDP targeting is basically the implementation of the Taylor Rule, but with a stable target and less discretion? Isn't that basically what you're seeing here?

John Thacker said...

Imagine that inflation targeting were the perfect policy for the Fed, and could smooth out all economic shocks by itself. Then negative economic shocks would be associated with the Fed missing its inflation target, and there would be some sort of strong correlation (not necessarily linear-- job loss or lower real GDP growth could be associated with misses in both the high and low direction) between inflation and other economic variables.

Now suppose that inflation targeting is insufficient but the Fed engages in it. Then we should see some economic shocks even with very stable inflation, and, as a result, a very low correlation between the (flat) inflation and various economic variables. This is arguably what we see.

The NGDP targeting advocates in fact argue that inflation targeting is not the correct target. Therefore, they would argue that the result as Angus has presented it is consistent with their theory-- the Fed is engaging in inflation targeting, that doesn't prevent all shocks and negative consequences, and thus the correlation is low.

They then go on, however, to claim that NGDP targeting would be superior. Sumner himself does not claim that NGDP targeting would prevent all shocks; he argues that supply-side changes, including, say, the minimum wage increase and expansion in UI duration, had some effects. Other NGDP fans make stronger claims, I believe, principally those of a more left-Keynesian bent.

However, importantly, the data and charts do not prove that their claim that NGDP targeting would be superior; it is easily possible to argue that while inflation targeting can't prevent declines in real GDP and the inevitable business cycle, etc., NGDP would do no better and perhaps worse of a job. But neither are they inconsistent. They simply demonstrate that: the Fed targets inflation, and inflation targeting doesn't prevent the real business cycle or various shocks from affecting the economy.

John Thacker said...

"Consider the fact that NGDPers never argue for a predictable, stable changing NEGATIVE NGDP. If all that is important is the stability or smoothness then they should have no preference for a positive or a negative sign."

Correct. However the NGDPers all believe in money illusion and the stickiness of wages at the zero bound. They argue, persuasively, that people hate pay cuts of 1% in a time of 2% deflation more than 3% raises in a time of 4% inflation. They also argue, again persuasively, that people will not accept a nominal negative return on their bank accounts so long as cash exists.

Andy Harless said...

"Your boss tells you that NGDP is surging and wants to know if it will be profitable to hire more workers and expand the business. To give a good answer, as I've just illustrated, you absolutely want to know WHICH COMPONENT, P or Y is surging."

AFAICT you haven't demonstrated that at all. All you've talked about is which component is correlated with employment, not which one justifies employment. The only reason I can think of that it should matter to your employer which one is surging is that, since the Fed is likely targeting inflation, it will tend to tighten in response to surging inflation but not so much in response to surging real growth. But of course that doesn't support your argument: it just says that, if the Fed is targeting inflation, then private sector actors should behave as if the Fed is targeting inflation.

More generally, if the Fed is just behaving randomly, for example, if it has a target for NGDP that it randomly changes, or if it raises the monetary base at a constant rate regardless of the demand for base money (and holds IOR constant), then, to the extent that you can identify it as a demand-side phenomenon, an increase in the price level is just as good (from the point of view of your hypothetical boss) as an increase in real output. If people are willing to spend more money for the same goods, that means that the demand curve has shifted. You should advise your boss to expand his business to the same extent that you would if the increase in NGDP was due to an increase in real output. Of course, in reality you might have information about the shape of the supply and demand curves that would enable you to exploit the distinction between real output growth and price changes, but without having that specific information, there's no obvious reason that one or the other should be more encouraging to your boss.

Take this example: I know the surge in NGDP is all about real output and not prices. OK, I advise my boss to cut production, because this means his competitors have gotten more productive and will undercut his prices if he tries to sell more. I can come up with 3 other examples, to get into every box of the 2x2 matrix, but the only general point will be that: if you know the surge in NGDP is a demand-side phenomenon, then your boss should expand, whether it is prices or output that is rising.

Andy Harless said...

There are a couple of reasons it makes sense to regard NGDP as a causal factor:

1. The Fed influences NGDP more directly than it influences RGDP or prices. In Keynesian terms, we say that macro policy determines the position of the aggregate demand curve. But NGDP is a reasonable approximate statistic for the position of the aggregate demand curve. In order to get from there to real output and prices, you need to know the position of the aggregate supply curve, which, I would suggest, is a taller order than we need to give the Fed.

2. You can, in fact, aggregate nominal value added, and you can do so without any inherent conceptual problems (although there are measurement problems in some cases). You can't really aggregate prices or real value added because that requires you to add apples and oranges. You can come up with various ways to say what 3 apples plus 2 oranges equal, but they are ugly, really ugly. And you get people like Niall Ferguson telling you you've done it all wrong, that in fact 3 apple + 2 oranges = 2 peaches, not the 5 peaches you thought.

Mark A. Sadowski said...

It would seem to me your argument consists of showing that two real variables are highly correlated (absolutely amazing), assuming that money is totally non-neutral in the short run (in which case why does the Fed have an employment mandate?) all while totally ignoring the core of the Avent-Dourado argument which implicitly concerns the extent to which unemployment is structural.

Congratulations on raising obtuseness to levels hitherto unseen in the econblogdom.

P.S. Keep in mind that it is real GDP that is the derived quantity. The fact that it correlates so well with employment is a testament to our ability to estimate something so contrived as an aggregate price level.

Adam said...

Is it really that hard to understand that NDGP is not the price level times real GDP?

The "real" number in that jumble is NGDP. Real GDP and the price level are made up numbers (I'm sorry "estimates").

Once you start with that "reality" in mind, it's much easier to avoid the sort of confusion on display here, and the temptation to start thinking about policies that influence only one of the made up numbers.