Sunday, February 21, 2016

Implausible plausibility

Likely in response to Paul Krugman's post titled "Plausibility", JW Mason put out his own post titled "Plausibility". However, the data shown is from 1925 to 2005 -- and includes the Great Depression and the mobilization for WWII:

Without those periods, the Friedman/Sanders point is well outside the mean (it is a 4.0-sigma deviation along the x-axis and a 2.8-sigma deviation along the y-axis) -- which was exactly Krugman's point:



JW Mason has updated his post with regard to this objection -- and thinks a WWII level of mobilization is possible. He now adds a linear fit ... but that fit likely leaves out some of the data (marked in red, including it gives us the red line):

But the other thing Mason suggests is that we can select a point along this line, which is basically an assumption about the current output gap and deviation from the trend.


Update, the second

Unfortunately I had to attend to some errands, and so didn't complete my thought above. Additionally, Bill in comments below hits on something relevant that I want to fold in:
However, since Sanders's proposals are radically different from US government policies in the post-WWII period, why should we expect their results to be the same?

The key point to understand is that the US economy is assumed to have a log-linear RGDP per capita that holds from 1925 until today. John Handley discusses some of the issues with that assumption at his blog here (he sent a link in comments below) -- for example demographics.

Let's suspend our disbelief for a second and take it on its face: there is a trend of 2.3% log-linear RGDP per capita growth. This trend exists from the roaring 20s, though the depression, the New Deal, WWII's top-down industrial policy, the Great Society, deregulation, Reaganomics, the 90s tech boom, George W. Bush, the housing crisis, the financial crisis, the ARRA and Obamacare.

So is the trend the result of policy?

I noted before that the recovery in unemployment has a remarkable regularity:

As I said there:
This regularity over several decades would imply that any mechanism that explains the rate likely has nothing to do with the internet, inequality, jobless recoveries, war, government spending, unemployment benefits, Keynesianism, monetarism, technology, ... etc. It is doubtful these different forces conspire in differing degrees to achieve approximately the same result every time.

It would be doubtful that government policy has much impact on a trend growth that purportedly holds for a hundred years (1925 - 2026). An anonymous commenter below asks:
If Sanders is planning a WWII-like mobilization, why wouldn't the data from the last one and the years preceding it be relevant?
Mason seems to agree with the anonymous commenter; they believe this kind of mobilization is plausible (such that it should be the null hypothesis -- you have to explain why it is implausible).

However, they can't have it both ways. Either 1) there is reversion to a 2.3% trend that holds over a hundred years, making the Friedman/Sanders claims plausible, or 2) policy is relevant. If it is just reversion to trend, then Trump's policies would get us there too. The plausibility claim is either banal or a massive WWII-sized mobilization.

If policy is relevant and we could implement WWII-sized mobilizations whenever we'd like, then what is keeping that 2.3% log-linear growth trend in force over a hundred years? What steps in to say: Your mismanagement of the Great Depression has gone too far from trend. Please start a world war now. What steps in to say: Your WWII expansion gone too far above trend. Please end the war now. Did WWII start because the global depression was mismanaged? I have to say: that is an interesting theory. Maybe there is something to it. But the flipside that WWII ended because we had re-established real output per capita is kind of silly: we got tired of fighting and decided to make money instead.

I mentioned the other day that the Friedman/Sanders growth claims smack of neoliberalism -- neoclassical economics in service of progressive goals. A constant growth rate of 2.3% is basically a neoclassical growth model. In that view, we let the market do its dirty business, and growth happens. At least in that case it is explicit: it is a balancing of current consumption versus future consumption that creates an equilibrium growth rate.

These are two possible implicit theories behind the prima facie "model free" acceptance of the plausibility of a 2.3% growth rate.

I have a hard time accepting simultaneously a hundred-year-long 2.3% growth trend and relevant policy. With the IT model, I actually go for the former! There are significant long-run trends in the IT model in which policy plays little-to-no role. It's fairly empirically successful, but also says post-financial crisis growth is on-trend:

This isn't constant 2.3% RGDP per capita growth, though.


Update, the third

If we extrapolate from our current (post-financial crisis) trend, we'd be here:

This is another way of illustrating that the assumed output gap is huge -- Friedman/Sanders takes us from the red dot to the orange dot.

You may ask why the red dot is so far from the data -- that's really an indication of how implausible the 2.3% growth trend is. If we assume a trend of 1.3% (red dot appears close to the data) or 3.3%, we get completely different figures:


Update, the fourth (22 Feb 2016)

JW Mason stops by in comments below; I thought I'd illustrate the idea that the positive demand shocks have to balance the negative demand shocks to maintain a constant growth rate. I used an ARIMA process to simulate output. In one case I added large negative demand shocks (e.g. Great Depressions ... which are persistent in this model since trend growth will only slowly undo a large demand shock), and in another I added large positive demand shocks (e.g. WWII-mobilizations, which are also persistent).

Here is the first case:

And here is the second:

In the first case, the growth rate is on average lower with the large negative shocks (yellow) than without the shock (blue). In the second case,  the trend is on average about the same with the large balanced shocks (yellow) and without (blue).

The negative shocks correspond to the points labelled "Great Depression" and the positive shocks correspond to the points labelled "WWII" in dark blue above.


  1. I think that economic projections are mostly GIGO, so I am not defending Friedman's or anyone else's. However, since Sanders's proposals are radically different from US government policies in the post-WWII period, why should we expect their results to be the same?

    1. I added a second update that addresses your question.

      I wouldn't put these arguments in the same class of GIGO economic forecasts, though. This is literally looking at a data set and extracting a trend growth number. Sure that means you have an implicit model (as I mention in the update) ... but what is that implicit model?

    2. Well, someone -- I can look up the link, if you like -- compared recent projections of the type that Friedman did, including projections by his detractors, and the errors were pretty horrendous. So much for "evidence based policy". ;)

    3. I have seen that link (if you're talking about the same link):

      There are a few kinds of errors -- model, empirical and fundamental. None of the model errors shown at the link are "horrendous" unless you think the RGDP measurements represent zero empirical and zero fundamental error.

      The IT model says the uncertainty (either measurement or fundamental error) in RGDP growth is larger than any of the model errors shown.

      ... but the Friedman projection is well outside the error bands. It really says something different.

      Those are not necessarily horrendous -- and we can't expect more precision than exists. It is like saying the error on the measurement of the temperature of a 1000-atom system is horrendous -- it's not. It is fundamental error due to the fact that 1/sqrt(1000) is not small.

      This is what happens because economists don't use error bands.

    4. "This is what happens because economists don't use error bands."

      Amen to that. :)

      P. S. I say horrendous because they go out 10 years or more. That's what multiplies the errors beyond reason.

  2. Jason,

    Irrespective of the bad regression and use of irrelevant data, Mason (among every other person I've seen coming to the defense of the Friedman analysis of Sanders' plan) makes this mistake of assuming real GDP per capita follows a linear trend. More on that here.

    PS I feel weird suggesting you read my own post, but I think you'll enjoy it.

    1. That's what comment sections are for!

      I agree -- however I have an additional issue with the projections even after ceding that a trend RGDP per capita growth rate exists. If it does, then is policy relevant at all?

    2. It could be argued that policy effects the speed of retrenchment to trend, but your unemployment graph above basically disproves that hypothesis. Otherwise, it could be that policy permanently shifts the level of potential output up and down, leaving a semi-constant trend rate of growth that is exogenous to policy. I'm more sympathetic to this view, but I don't think that it really stands up for the current scenario -- I don't see any of the policies post 2008 causing a reduction in the level of potential output; hysteresis is unconvincing and nothing supply-side has happened that would warrant such a large effect. Given this, I think that the primary reason for slow post-2008 growth is demographic, which is why I focus on it so much in my post.

    3. I agree -- and it explains more than just the US growth slowdown. It explains the EU and Japan as well!

    4. Hi, John! :)

      Bill Mitchell has a blog post dealing with Friedman's projections and the dispute. ( ) He divides the decline in the participation rate between ageing and cyclical factors, and estimates that ageing accounts for 53% of the decline and cyclical factors account for 47% of the decline. For reference, he talks about his methodology at this post:


  3. If Sanders is planning a WWII-like mobilization, why wouldn't the data from the last one and the years preceding it be relevant? It seems that you should explain why such a mobilization isn't conceivable, not whether someone should believe the data selection decisions of one or the other of the debaters.

    1. That's really the nub of it: does the 2.3% RGDP per capita growth trend exist because of massive policy endeavors or in spite of them? It seems implausible for 2.3% growth to hold because of policy across the New Deal, WWII, Reaganomics and the 1990s tech boom. It's much more plausible for either:

      2.3% growth holds regardless of policy (Sanders or anyone)
      2.3% growth isn't actually a stable trend (meaning assuming the trend doesn't represent plausibility).

      We all like to make a play for the null hypothesis (H0 = "WWII mobilization is possible" versus H0 = "WWII mobilization isn't possible"), but given the Congressional situation that possibility isn't so much an economic question, but political.

    2. " It seems implausible for 2.3% growth to hold because of policy across the New Deal, WWII, Reaganomics and the 1990s tech boom...."

      It seems eminently plausible to me. If you assume that the long-run growth path is only achievable and sustainable within certain inequality limits , and that when those limits are exceeded a period of mitigation is possible via increasing economy-wide leverage , then you can explain why FDR's policies re-established the trend ( a world war is unnecessary , unless you're in a hurry ) , and you can explain why Reaganomics demolished the trend , but with a lag due to the masking effects of debt , with ultimate failure as debt saturation takes hold :

      In other words , both Volcker and Greenspan had a copy of this FRED graph on their desks , and when they felt like they were in danger of falling away from the red-dashed trend line , they opened the debt spigots. The dotcom boom provided a brief respite from increasing leverage , as the one-percenters funded it with equity ( out of generosity rather than greed , of course ).


    3. I'd agree that is a better (and more plausible) model, but still would have to apply across the gold standard, the Treasury controlled Fed of the 1940s and 50s, the Keynesian Fed of the 60s and 70s, the Volcker Fed of the 80s and the great moderation period.

      During these periods, the Fed targeted different things -- the price of gold, whatever the Treasury told them, unemployment, inflation and interest rates.

      It would be astounding if RGDP per capita was stabilized by these different efforts.

      Additionally, Sanders is not advocating monetary policy interventions -- which would probably make Matthew Yglesias drool ... :)

  4. Thanks for the response. But I don't think it's so complicated. Population growth, technological change etc. create a steady growth in potential output. Big demand failures can pull actual output below that trend. Big demand-boosting policies than are needed to (or at least can) return to trend. Why did the mobilizaiton happen to stop at just the point the earlier trend was restored? Actually it didn't -- it overshot quite a bit but that additional spending wasn't sustainable, was increasingly inflationary even with price controls, etc. This seems both straightforward and consistent with standard theory.

    1. Hi JW,

      "Big demand failures can pull actual output below that trend ..."

      I agree that persistent demand shortfalls exist -- but that means when one happens, the trend growth level will be reduced. And it will be reduced proportionally to the length of time the persistent demand shortfall exists.

      "Why did the mobilizaiton happen to stop at just the point the earlier trend was restored? Actually it didn't -- it overshot quite a bit ..."

      That overshoot was required to maintain the trend. If RGDP per capita just returned to trend after a persistent demand shortfall, the average growth value would be lower.

      I added some graphs above to illustrate my point. But the basic idea is that the large negative shocks need to be exquisitely balanced by large positive shocks in order to maintain a constant trend growth rate. Any imbalance leads to too persistently high or too low growth. That means assuming the existence of a trend growth rate implies all persistent demand shortfalls are perfectly balanced by overshooting (with the size of the overshoot proportional to the length of the demand shortfall).

      The larger issue is that persistent demand shortfalls are inconsistent with the idea that trend growth rates aren't impacted by persistent demand shortfalls. Saying we can return to the level implied by 2.3% RGDP per capita trend growth is inconsistent with the existence of persistent demand shortfalls.

      Personally, I would like to hold onto the idea that persistent demand shortfalls exist ... I don't want to cut off our rosy forecast noses to spite our Keynesian demand management faces.

    2. I guess a lot depends on how you think about trends. The vision I've had is more along the lines of Friedman's "plucking" model , where a persistent shortfall doesn't require an overshoot to reacquire the trend line. It's a lot like Sumner's argument about ngdp rate vs. level targeting , I suppose.

      It is a distinction that needs to be made in order to make sense of these discussions. Your added graphs helped in that regard.


    3. That is true, but in the case of a plucking model, the average growth rate is lower than the trend growth rate -- and the difference depends on the number (and size) of "plucks".

      You could define average growth rate by simply the slope between the end points -- in that case you would get exactly the trend growth rate as long as your end points aren't at a "pluck". I discuss this a bit here:

      I am sympathetic to a plucking model (the IT model should have the characteristics of one)! It just means average growth is impacted by recessions, though. Average growth with recessions excised will be exactly trend.

    4. With a plucking model the trend should be based upon the upper envelope, right?

    5. Given measurement and intrinsic error, it should be based on the mean growth rate minus the "plucks".

    6. I'm not sure that subtracting the "plucks" yields Gaussian error (which the use of the mean implies).

    7. Using a mean makes no assumption about the distribution. You probably want a symmetric distribution for it to make sense above ... And it is symmetric:

    8. Sorry. What I had in mind was a distribution for which estimation via minimizing the mean square error is appropriate.

      Moi, for most human data I think that minimizing the maximum error is appropriate, following Chebyshev. That's because I am interested in positive feedback, which I think is important in human systems. OC, positive feedback leads away from equilibrium and towards chaos or the edge of chaos.

  5. Update and question:

    The arguments have continued. James Galbraith, who has defended Friedman, offers a summary here: . One of his points is that Friedman believes that "public activity" can move the economy permanently away from an "underemployment equilibrium", as happened with the New Deal and WWII. As I would put it, the government could move the economy from a suboptimal equilibrium to a better equilibrium.

    Which leads to my question. I have been surprised that, since the recent financial crisis and Great Recession (Not So Great Depression, New Normal), economists have not seemed to talk much about suboptimal equilibria. They do acknowledge that multiple equilibria may exist, but don't say much about moving between them. By contrast, many family therapists view their main job as moving the ailing family system from a suboptimal equilibrium to a better one, although they use different language.

    IIUC, there are multiple economy-wide IT equilibria, which tend to change over time with κ. I get the impression that you do address the question of moving between equilibria, without using that language. Is that correct? Thanks. :)

    BTW, in your interesting graph of post WWII unemployment recoveries, the definition of unemployment changed during the Reagan administration such that it is lower than it would be by the previous definition (and that was probably a big reason for the change). That change, however, did not alter the slope of the recovery gradient. (!)

    And something I just noticed. Even with that change, the "recovered" level of unemployment has increased. The bottom envelope has an upward slope. And under the old definition of unemployment, that slope would be even greater. Doesn't that suggest a shift towards less optimal unemployment equilibria? And the possibility, whether via Sanders's proposals or not, of shifting towards better equilibria?


Comments are welcome. Please see the Moderation and comment policy.

Also, try to avoid the use of dollar signs as they interfere with my setup of mathjax. I left it set up that way because I think this is funny for an economics blog. You can use € or £ instead.