Historical Revisionism

Okay, I got a few "vintages" of Potential GDP from ALFRED, going back as far as 1991. I got 1991, I got every five years from 1995 to 2010, and I got 2013.

These are "real" values, inflation adjusted values. And as it turns out, there are all different measures of correction-for-inflation used for the various vintages. The README page in the ALFRED file says the 1991 series uses 1982 dollars; the 1995 series uses 1987 dollars; the 2000 series uses 1996 dollars; the 2005 series uses 2000 dollars; and the 2010 and 2013 series use 2005 dollars.

Why make it simple, right?

The first thing I want to do is convert everything to the same dollars, so that when I put them on a graph it's apples-to-apples. I did the work in a Google Drive spreadsheet this time.

I got quarterly data, but the vintage series are expressed in dollars of some year, not some quarter. So I'm taking an average of quarterly values for each base year value that I will need.

I'm using the current GDP Deflator values to convert the vintages. For example, for the data given in 1982 dollars I'm dividing by the 1982 value of the current Deflator series and multiplying by the 2005 value of the current Deflator series. For data given in 1987 dollars I'm dividing by the 1987 value of the current Deflator series, and like that. So I end up with everything in 2005 dollars, based on the most recent data on prices.

(I tell you these things because my method makes sense to me, but I don't know if it would make sense to economists or to others.)

Here's the graph:

Graph #1:
From the "In Uniform Dollars" Sheet of this Google Drive Spreadsheet

If you have trouble telling one series from another on this graph, it helps to know that the data series end in date order. See how that red line splits off low, and ends around the year 2000? That red line is the 1995 series, and it is shadowed by the 1991 series which ends a few years earlier.

Those two series, by the way, were based on economic performance and economic thinking before the macroeconomic miracle of the latter 1990s. By the year 2000, evidently, economists had factored the "miracle" years into their thinking. All of the Potential GDP series shown for the year 2000 and after group together at a much higher performance level than the 1991 and 1995 series.

Were the numbers influenced by the improved economic performance? I would say so, yes. If you put your mouse cursor on the red line and follow it back to where it joins the others, you can see that the numbers changed as far back as 1982. The 1995 prediction still shows no performance improvement, but the 2000 prediction shows a performance improvement that goes all the way back to 1982!


The four later series that all showed improved economic performance through the 1980s and 1990s suddenly break apart in the mid-2000s. You can see the little golden stub of the 2000 series in the crotch there where the others veer apart. That golden stub stops too soon, but the next estimate by date, the 2005 estimate -- green -- runs even higher than the 2000 estimate, and shows no indication of weakening.

That estimate, the green one, was made before the global financial crisis.

The next estimate, the purple one, made after the onset of the global financial crisis, is as much lower than the golden stub as the green line is above it. And what could have caused this change?

The global financial crisis, perhaps?

The most recent estimate of potential GDP, the 2013 estimate, is lower yet. And I think you can see it lower than the others as far back as 2005 -- two or three years before the crisis.

Isn't that odd?

Cooperating with Alfred

I finally figured out where to get the "vintage" data that ALFRED offers. After you find the series you want (in my case, Potential GDP) and after you get a graph on the screen, that's when you click download.

ALFRED then gives you a selection window with several old versions of the data -- all the versions they have, I guess. You just highlight the vintages you want (in my case, all of them) and the thing is ready to send you a file (in my case, a zipped Excel file).

I wanted to look at Potential GDP because, as you know, it has been revised down.

Turns out, the "real" PGDP data from ALFRED comes in not only several vintages, but also several different "base year" series. To whit:

Data Values	Start Year
Billions of 1982 Dollars	1991-01-30
Billions of 1987 Dollars	1992-01-22
Billions of Chained 1992 Dollars	1997-01-28
Billions of Chained 1996 Dollars	2000-01-27
Billions of Chained 2000 Dollars	2004-09-13
Billions of Chained 2005 Dollars	2010-01-26

If you select one year's data from all the different vintages and make a graph of it, the graph shows an upward-stepping pattern. Upward-stepping, because every time they pick a different base year and adjust for inflation, the number gets bigger. So, if I want to compare vintages across base years, I'll have to adjust for base years. I didn't get into any of that today.

Just for the base-year 2000, ALFRED offers eleven different vintages. The oldest is dated September 13, 2004, and the newest August 27, 2009. That period includes the time of the financial crisis, so I figured I'd look at that.

Graph #1

Move your mouse up and down over the legend to highlight different lines on the graph. You will see that as the mouse moves down toward newer vintages, the highlighted Potential GDP line on the graph is most often lower also.

My Google Drive Spreadsheet is available.

Revising the past, in the past

As an afterthought to this morning's post, the graph below is reproduced from the Economic Report of the President, 1977 (PDF) from FRASER:

Potential GNP 1964-1977, Old and New ("new" = 1977)

The thing is so old it uses GNP, not GDP. Note the "old" trend line and the "new" one. The new one is lower, just like with Altig and Gavin's graphs. For the 1977 Report of the President, they revised the past back to 1964.

We've always been at war with Eastasia

The protagonist of the novel, Winston Smith, is a member of the Outer Party who works for the Ministry of Truth (Minitrue), which is responsible for propaganda and historical revisionism. His job is to re-write past newspaper articles so that the historical record always supports the current party line.

- Wikipedia

Let me start with the justification. In What Is Potential GDP and Why Does It Matter? (PDF, 2 pages), the Fed's Willian T. Gavin writes:

Looking back in time, potential output is relatively easy to measure because we have reliable methods to extract smooth trends from historical data. However, measuring potential output in real time is more difficult because only past data are available to estimate the trend.

PGDP is a trend line. As a trend, it must summarize the present and the future as accurately as it summarizes the past. But it is difficult to know the present trend, because it depends on the future. And it is even more difficult to know the future.

Fair enough. But what comes of these difficulties is a policy of historical revisionism:

Graph #1, Source: What Is Potential GDP and Why Does It Matter? (PDF)

The estimate created in 2011 changes the past back to 2005 or before.


Recently, the Fed's David Altig showed a similar graph:

Graph #2, Source: Nature Abhors an Output Gap (Macroblog)

Altig's graph begins in the first quarter of 2009.

For the estimate created in 2011, the value for 2009 is lower than the 2009 value from the estimate created in 2010. For the estimate created in 2012, the 2009 value is lower still. And the estimate created in 2013 shows an even lower value for 2009. Every year, we tell a different story about the past. We've always been at war with Eastasia.

Mr. Gavin's explanation notwithstanding, I'm not sure this practice is reasonable.

Looking Failure in the Face

The St. Louis Fed's Featured Economist on 25 April morning was William T. Gavin, whose latest work is
What Is Potential GDP and Why Does It Matter?, Federal Reserve Bank of St. Louis Economic Synopses, Apr 20

Short PDF, page and a half, my kind of stuff.

Gavin says PGDP is a measure of the best we can do in an imperfect world. Fair enough. He says policymakers estimate PGDP by smoothing out actual GDP. And he admits that "the accuracy of our estimate depends on the accuracy of our long-term forecast."

He also points out the importance of PGDP: Policymakers use it to set policy.

Gavin provides a graph showing two versions of PGDP: one from just before the crisis, and one from a few years after:

The higher level of potential GDP was estimated in 2007 and the lower level in 2011. The reduced 2011 estimate reflects the impact of sluggish GDP growth over the past three years.

Since they don't know how to bring conditions up to meet expectations, policymakers are lowering their expectations to match conditions.

Source: What Is Potential GDP and Why Does It Matter? (PDF)

See that box there on Gavin's chart? The box shows that the 2007 estimate was wrong by almost 10% in 2009, and by more than 11% in 2011. But the 2011 estimate was off by only about 7% in 2009, and by about 5½% in 2011. The implication is that the newer estimate is better: If the forecast is less wrong, it must be better, right?

Lest I am too subtle: The reason we study the economy is to make things better. Not to put the best face on failure.