Saturday, January 26, 2013

Three Notes in the Key of NGDP

We all know the line graph which shows the gap between actual nominal gross domestic product and potential, however defined as a constant-growth-rate trend, or the output of an H-P filter, a CBO-OMB assumption, or whatever.

Don't get me wrong, that's an important graph. If I could only have one graph to say "this is the problem" -- this being the recession -- I'd probably choose it over most anything else, such as the unemployment rate.

But I think we need to be giving some other parts of the argument for NGDP targeting a little more attention.

I think it's pretty well-established that expectations matter, that credible commitment matters, that uncertainty matters. In fact, I think it's clear that expectations/commitment/uncertainty -- I'm grouping these all together because they are all related and forward-looking -- all matter more than current-quarter nominal GDP. If you don't agree, I gently point you to work by Michael Woodford, Ben Bernanke, or Lars Svensson.

So if I could add any three other graphs to the canon, it'd be the following three. I'm tipping my hand somewhat for an upcoming Bloomberg post, but I recently discovered that the Survey of Professional Forecasters has been recording NGDP expectations since 1968. Better yet for those inclined -- that is, me -- they have all of the individual anonymized forecast records, mean forecasts, median forecasts, and cross-sectional dispersion statistics on the forecasts. And it's a quarterly forecast for several quarters ahead.

You can do a lot with this. I've actually never seen someone really work with Survey numbers to make the NGDP case, and this is only the tip of the iceberg. (I'm practically pleading with everyone else to write something.)

The first is a graph of mean NGDP expectations, with each line is a time series of expectations for NGDP percent growth one through five quarters out.


You can see the NGDP shock as a shock to expectations. Notice that the effects are noticeable for a full year out. It's not hard to see how a sudden collapse of short-to-medium expectations, with no "bounce-back" recovery seen in the future, could be more important than current-quarter NGDP.

The second is a graph of the dispersion of the quarterly forecasts, the immediate quarter through a year out. The dispersion is expressed as the gap as a percentage of NGDP between the estimate at the 25th percentile and the 75th percentile. (In other words, that volume encompasses the middle half of all the estimates.)


When you re-index all of the initial values to one, what you see is that NGDP uncertainty almost exactly tripled in 2009, at all forecast ranges, relative to its 2005 level.

And the third chart I posted on Twitter earlier today. It combines all of these points. Notice the downturn in expectations combined with the expansion of uncertainty -- I find this chart perhaps as convincing as the canonical graph at the top.

It shows us the forecasted quarterly NGDP growth for one through five quarters out in the first quarter of 2007 versus the first quarter of 2009. The black lines indicate the mean forecast, and the grey bands indicate the 50 percent confidence interval as drawn from the dispersion of the forecasts.

10 comments:

  1. Evan, what do you mean by "highly significant" for one year out? The 4 quarter curve seems to budge the least (as expected) in response to the shock.

    ReplyDelete
    Replies
    1. Yeah, I didn't phrase that right. Editing...

      Delete
  2. Great post and great find, but the final graph needs a little more explanation.

    ReplyDelete
    Replies
    1. Hi,

      I just added this:

      "It shows us the forecasted quarterly NGDP growth for one through five quarters out in the first quarter of 2007 versus the first quarter of 2009. The black lines indicate the mean forecast, and the grey bands indicate the 50 percent confidence interval as drawn from the dispersion of the forecasts."

      Delete
  3. This comment has been removed by the author.

    ReplyDelete
  4. Evan, hopefully I'll have a semi-thorough analysis up tomorrow. I'm doing some relatively intensive excel mining and, so far, my results suggest that the expectations channel is not a particularly robust one. A serious problem is that, for the period of time that data is provided, there are only two monetary "eras". But the inflation under Volcker was primarily due to supply side factors, and the Great Moderation contained little volatility at all. In other words, one time period was broken, while the other one contained little variation that we could use in our tests.

    ReplyDelete
  5. Evan, nice graphs. I too have used the forecasted NGDP series. See here and here (look for impulse response functions)

    Also, see the long-term productivity forecasts in the survey. They show a decline consistent with Great Stagnation.

    ReplyDelete
  6. Evan,

    Nice post (as usual). A couple of years back I wrote a related post on real GDP forecasts. I did not think of looking at the survey's forecast for NGDP. I'll try to update this data (or perhaps you could). One thing I found interesting at the time was my "fact 2"..that these expectations were implicitly suggesting an expected decline in the level of potential GDP. Anyway, in case you're interested, the link is here:

    http://andolfatto.blogspot.com/2010/09/cyclical-variation-in-short-and-long.html

    ReplyDelete
  7. this is likely noobish...

    why in mid-2008, do 1QT and 2QT expectation spike higher right before they plummet? While 3QT seems to more accurately indicate the plummet is coming?

    How does mean that short expectations were headed down, and then suddenly ticked hopeful immediately before he shock?

    ReplyDelete
    Replies
    1. You'd have to ask the forecasters. I imagine it has to do with oil price volatility, but that is a guess I make with low confidence.

      Delete