The best thing Milton Friedman ever wrote

From the Epilogue of Money Mischief:

... Pierre S. du Pont, a deputy from Nemours to the French National Assembly. Speaking on a proposal to issue additional assignats -- the fiat money of the French Revolution -- he said: "Gentlemen, it is a disagreeable custom to which one is too easily led by the harshness of the discussions, to assume evil intentions. It is necessary to be gracious as to intentions; one should believe them good ...

Excessive finance

At FiveThirtyEight: The GOP’s Corporate Tax Cut May Not Be As Big As It Looks.

Whatever. But they show this graph:

As you go from left to right along the horizontal axis, the number of firms increases. Also, the dots get bigger: more pixels, more firms, I guess.

Anyway, what's the biggest, rightmost dot? Banks. We have more banks than anything.

It's not subtle at all, really, and anyway I caught you

The Fed's Inconsistent Numbers at the Peterson Institute.

From the footnote:

CPI inflation has generally been slightly higher than PCE inflation.

No, that's not right.

CPI was the official measure until the year 2000. Since the year 2000, PCE has been the official measure. The way one describes the difference between them must not contradict the change that occurred in the year 2000. If the CPI is generally higher than the PCE, and the official measure changed from the CPI to the PCE, then the correct way to describe the relation is to say

PCE inflation has generally been slightly lower than CPI inflation.

Otherwise, you are embedding a subtle lie into your story.

Got that, Peterson?

The mind works in mysterious ways

I caught this scribble out of the corner of my eye ...

... and it made me think of this:

Such brilliance -- yet so down to earth

Best of all that we should know the future. But if not, then, if we are to control the activity of the economic system by changing the quantity of money, it is important that opinions should differ.

- J.M. Keynes

On Protectionism

During the fiscal controversy of the first quarter of the present century I do not remember that any concession was ever allowed by economists to the claim that Protection might increase domestic employment.

- J.M. Keynes, 1936

Got it? Now try a bigger bite:

During the fiscal controversy of the first quarter of the present century I do not remember that any concession was ever allowed by economists to the claim that Protection might increase domestic employment. It will be fairest, perhaps, to quote, as an example, what I wrote myself. So lately as 1923, as a faithful pupil of the classical school who did not at that time doubt what he had been taught and entertained on this matter no reserves at all, I wrote: “If there is one thing that Protection can not do, it is to cure Unemployment. ... There are some arguments for Protection, based upon its securing possible but improbable advantages, to which there is no simple answer. But the claim to cure Unemployment involves the Protectionist fallacy in its grossest and crudest form.” As for earlier mercantilist theory, no intelligible account was available; and we were brought up to believe that it was little better than nonsense. So absolutely overwhelming and complete has been the domination of the classical school.

- J.M. Keynes, 1936

And just one more bite:

But if nations can learn to provide themselves with full employment by their domestic policy (and, we must add, if they can also attain equilibrium in the trend of their population), there need be no important economic forces calculated to set the interest of one country against that of its neighbours.

Q:

If corporations are people, and slavery is illegal, then
why are corporations allowed to own other corporations?

vigor

FRED says:

Business Sector Price Indexes

This graph, from 18 December, shows real and nominal Labor Share. Does it remind you of anything?

Graph #1: Labor Share, nominal (red) and real (blue), Indexed to 1947 Q1

Hint: It shows a gap that opens in 1961 and increases relentlessly in size thereafter.

You got it -- We see the same gap between productivity and compensation:

Graph #2: Real Output per Hour (blue) and Real Compensation per Hour

Interesting. I wonder why that is.


On Graph #1, the red line (labor share in the business sector) shows compensation relative to output. The blue line shows the same. The only difference is that the red line is a ratio of nominals and the blue is a ratio of reals.

If you have a knee-jerk reaction to this, it is probably that the ratio of nominals and the ratio of reals should be identical. I've seen it said about Debt-to-GDP, for example, that the ratio of nominals and the ratio of reals are identical.

Sometimes yes, sometimes no. For debt relative to GDP, the ratio of reals and the ratio of nominals are identical if you use the same calculation for debt as for GDP -- the calculation that converts nominal values to reals. But if you use one calculation to convert GDP to real values, and some other calculation to convert debt, you are almost guaranteed to get ratios that are not identical.

Something even simpler explains the difference visible on Graph #1: Real output and real compensation are figured using different price indexes. Here are the price indexes for Business Sector Output and Business Sector Compensation:

Graph #3: Business Sector Price Indexes for Compensation ( blue) and Output (red)
The Compensation Index is Higher than the Output Index, and Comes Down More when Deflated

I didn't know. I never saw them before. But the two price indexes had to be different: The arithmetic screams it. Why different indexes are used, I wouldn't know. You'd have to ask an economist.

On Graph #3 you can see a gap opening between the lines in the 1970s. And you might be able to see the upper edge of the blue line creeping out from under the red line some time around 1970. To get a better look at what is going on, I took Graph #3 and cut off everything after 1970, letting the rest expand:

Graph #4: The Business Sector Price Indexes, 1947-1970

It is easy to see now that the gap between the red and blue lines goes all the way back to 1962. And again you can see the blue line creeping out from behind the red -- this time during the 1960-61 recession. Call it 1961.

That year sound familiar? 1961 is also the year that marks the start of the gap between real and nominal labor share on Graph #1 above. And 1961 marks the start of the gap between productivity and compensation on Graph #2.

All these graphs show a gap opening in 1961 and expanding relentlessly thereafter. I will come back to this thought shortly.


For all these graphs, I indexed both lines on their start values so we can see how things change from the beginning. Economists don't seem to get that. If you don't index 'em on their start values, the two lines appear to run roughly parallel since 1947, and then the gap closes in 2009 -- because the source data is indexed on that date. Here's what Graph #3 looked like before I indexed the two lines on their start dates:

Graph #5: Before I Indexed Graph #3 at 1947, It Looked Like This

The only difference between Graph #3 and Graph #5 is that in Graph #3, the two lines are pinned together in 1947 by indexing. In Graph #5, the two lines are pinned together in 2009 by default.

If we get rid of all the data before 2009 where the lines are pinned together, and then wait 70 years or so while more data is gathered and added to Graph #5, the graph will give you an idea of how the two price indexes have changed since 2009.

I get a similar effect on Graph #3 by pinning the lines together at 1947. We don't have to wait 70 years because the time has already gone by. We can just look at Graph #3 and get an idea how the two price indexes have changed since 1947.

The purpose of a price index is to show how much prices have gone up over a period of time. On Graph #5, both lines show that prices have gone up. But because the two lines are pinned together near the end, rather than at the beginning, the graph creates the false impression that there is little difference between the two lines. Pinning the lines together where they start -- as on Graph #3 -- is a way to correct this false impression.


The gap on Graph #3 opens in 1961 and expands relentlessly thereafter, exactly like the gap on the first two graphs. Is this a coincidence?

It is not coincidence.

Graph #1 shows a gap between nominal labor share and real labor share. The only difference in the calculations is that real labor share uses the price indexes, and nominal does not. So it has to be the price indexes that create the gap on Graph #1.

Since there are no other differences in the calculation, it has to be the price indexes that create the gap. The fact that the labor share gap and the price index gap both start in 1961 is evidence that the price indexes create the gap in the labor share graph. The fact that both graphs show similar gap growth is additional evidence. That's evidence, not coincidence.

If I take the higher line on the labor share graph as a percent of the lower line, and do the same for the price index graph, I can put the two gaps together and compare them:

Graph #6: Gap Size Comparison for Labor Share (blue) and the Price Indexes (red)
Click the graph for a larger view

Identical. The gaps are identical.


The story of the gap in the Productivity vs Compensation graph is not so neat and tidy. Even with nominal values there is a gap. Compensation per Hour and Output per Hour run more or less parallel, but are not identical:

Graph #7: The Nominal Version of Productivity (blue) and Compensation (red)
Note: Productivity (real output per hour) is a measure of economic growth,
and economic growth is correctly measured in "real" rather than "nominal"
values. This graph shows nominal "output per hour" values that will become
"productivity" when the inflation is removed.

Compensation falls behind Output per Hour, even in the nominal data. When we bring price indexes into the calculation, the price index gap is combined with the gap in the nominals. The gap we so often see on graphs of Productivity vs Compensation is partly due to differences in the price indexes, and partly due to the differences between the nominals.

How much of the Productivity vs Compensation gap is due to price index differences? How much is due to differences between Output per Hour and Compensation per Hour, price indexes aside?

We can answer that question by using one price index instead of two. On the graph below, blue is productivity and red is compensation, figured as always, with the two different price indexes. The green line is compensation again, figured using the same price index that is used for the productivity numbers.

The gap between green and red is due to the price index. The gap between green and blue is due to compensation falling behind productivity.

Graph #8: Productivity (blue), Compensation (red), and Compensation Deflated by
the Same Price Index as Productivity (green)

If I take the red line as a percent of the blue line, we get a look at compensation relative to productivity, based on the way the two are always figured -- with different price indexes. I will show that in red on the next graph.

On the same graph I will show, in green, the green line as a percent of the blue, to see compensation relative to productivity when the two are figured with both using the same price index. This may give us a better measure of the gap between them.

I don't know yet if it's a better measure. I do know it's an issue. So I look.

Graph #9: Two Measures of Compensation Falling Behind Productivity

The gap between the green and red lines on Graph #9 should remind you of the gaps on the first three graphs of this post. Graph #1 in particular is strikingly similar to #9.

The blue line on Graph #9 is the "goal" -- the level of Productivity, which is the level we expect Compensation to reach. Productivity is 100% of Productivity. Both measures of Compensation are generally less than 100% of Productivity, and show decline.

The red line is the compensation measure you would generally see on a "Productivity and Compensation" graph. The gap between blue and red shows how much Compensation has fallen behind. In the late 1950s, Compensation had not fallen behind at all. Since 2010, Compensation has been down to about 60% of Productivity; so, about 40% behind.

But the red Compensation uses a different measure of inflation than Productivity. The different measure may exaggerate the extent to which Compensation has fallen -- I'm not sure. (Whether it is "exaggeration" is a separate question.) The green Compensation line uses the same measure of inflation that is used for Productivity. Using the same price index for Compensation and Productivity eliminates this potential exaggeration.

The gap between blue and green shows how much Compensation has fallen behind without exaggeration. In the late 1950s, Compensation had not fallen behind at all. Since 2010, Compensation has been down to about 86% of Productivity. So, about 14% behind. About one-third of the gap.

But that's only if we figure it using the Output price index. If we figure it using the Compensation price index, Compensation is about 29% behind Productivity. Twice as much. Twice as far behind. Two-thirds of the gap.

But either way, Compensation per Hour is an average that does not count factors like rising inequality.

I'm gonna go take some aspirin now. Please feel free to check my work.


I should add that at FRED, if I change the units for a data series to "Index (Scale value to 100 for chosen date)", the chosen date defaults to the data-start date 01/01/1947, which is exactly what I want.


This series of posts begins on 14 December.

Productivity and Compensation -- The Early Years

I put output-per-hour and compensation-per-hour on a graph, for both the Business Sector and the NonFarm Business Sector. Indexed all the lines to the start date so they all start at the 100 level. Then I subtracted 50 from the Nonfarm Business numbers to shift them down on the graph. Now we can compare productivity-and-compensation for the business sector (blue and red) to productivity-and-compensation for nonfarm business (green and purple). Through 1970 only:

Graph #1: Productivity and Compensation 1947-1970 for Business & Nonfarm Business

The separation between productivity and compensation is greater for the business sector than for nonfarm business in these early years. Since I want to see differences, I'm going to use the business sector data and set the nonfarm data aside.

As I said before, the numbers are indexed. Both business-sector lines start at 100. But we know compensation was actually less than the value of output, because somebody made a profit.

We know compensation was less than the value of output, but we don't know how much less. From the indexed data, we cannot tell.


I messed with that all day, how much less, and made no progress. So I'm moving on.

I'm looking at the early years. Real compensation and real output run together for a while, then separate for a while, then run together, then separate. There's no inflation in these numbers, but they do show compensation falling slightly behind output, or catching up, or sometimes maybe gaining on it.

And I remember James Forder saying

Samuelson and Solow published a widely read paper in the May issue of the American Economic Review of 1960. It discussed the causes of inflation, the Phillips curve, and related matters. Discussion of their paper frequently says that it presented the Phillips curve as a stable, exploitable relation, and hence played an important role in the development of inflationary policy. This is hardly so...

The question [Samuelson and Solow] were addressing was that of the explanation of the inflation of the 1950s – particularly the period 1955-57 – and the implications it had for macroeconomics. Mild though that was later to seem, this 'creeping inflation' as it was called was, at the time, a source of much anxiety.

I think Samuelson and Solow had caught a case of cost-push inflation. Ah, the great ignored cost-push inflation. It fascinates me. That's where science advances, in the ignored areas, once they stop being ignored. It's where the interesting questions are.

I'm wondering about the cost-push inflation of 1955-57 and what I might find in the early years of the business sector output and compensation data. My table of FRED data lists both compensation-per-hour and real compensation-per-hour; I can divide the nominal by the real to get a price index; I wonder how that index compares to the CPI and the Deflator. And the new one, PCE.

Graph #2: The Business Sector Compensation Price Index (blue) and other Price Measures

Well that's interesting: No match. The "business sector compensation" measure falls about midway between the CPI (red) and the GDP deflator (purple). PCE, the new one, runs lowest of all. Yeah that makes sense, because we solve our economic problems by changing the way we measure them, so you'd expect the new measure of inflation to be the lowest one.

Focus, Art. The measure of inflation in business sector pay is different from the others. I wonder what the rate of inflation looks like...

Graph #3: The Rate of Inflation in Business Sector Compensation

There it is! The inflation Samuelson and Solow were looking at, 1955-57. Inflation jumps to one percent per quarter -- four percent per year, about. Out of nowhere, after inflation had died down nicely following the Korean war.

The "Great Inflation" shows up right on schedule: Compensation inflation increases beginning in the mid-60s. And it peaks with the 1970 recession, the 1974 recession, and the 1980 recession. Plus you can see the near-recession of 1967 in a brief, sudden fall from the trend of increase. It's all there, all of it. But, like Samuelson and Solow, I am fascinated by the inflation of 1955-57.


I notice FRED has data for both current dollar output and real output for the business sector. So I can get another price index, a "business sector output" price index. I'll put that inflation rate on the graph along with Business Sector Compensation inflation. And again I'll cut off everything after 1970, to get a close-up view of the early years.

Graph #3: The Rate of Inflation in Business Sector Compensation

The first thing I notice is that, after the 1954 recession, Output Inflation went up earlier and faster than Compensation Inflation. Oh, that's interesting!

If it's true, I would expect to see on Graph #1 that the "output" line rises relative to the "compensation" line coming out of the 1954 recession. It does. The output line was already above the compensation line during the '54 recession, but the gap opened more in '55.

The gap closed in '56 as compensation inflation scooted up to the 1% level. So I can imagine that prices (i.e., output prices) were rising after the '54 recession, and it took until '56 for wages to react.

I want to subtract Compensation Inflation from Output Inflation of the business sector. This will show me the portion of rising business sector prices that is not due to labor.

I told ya: The ignored areas, that's where the most interesting questions are.

For both the Business Sector Compensation Price Index and the Business Sector Output Price Index, I take the nominal data at FRED and divide it by the inflation-adjusted data. I use the default units (Index 2009=100) and the default frequency (Quarterly).

After the division, in the "Finally, you can change the units of your new series" part of the FRED window, I set the units to "Percent Change". In order to do more arithmetic after that, I have to export the data to Excel.

In Excel I receive the quarterly inflation rates for business sector compensation and business sector output. I subtract the compensation number from the output number. This gives me the inflation in the price of business output which is not due to the cost of labor. I put that on a graph.

Then I go back to FRED for the quarterly inflation rate as measured by the GDP Deflator. I add this to my graph in Excel so I have a context: something to compare against business non-labor inflation. Here is the result:

Graph #5: Non-Labor Inflation in the Business Sector

The red line is the Deflator. It is a measure of inflation relevant to GDP as a whole. (By contrast, the Consumer Price Index (CPI) is more relevant to consumer purchases. I don't want that narrow focus here.)

The blue line is the inflation in business output, with the inflation in employee compensation subtracted out of it. What remains in the blue line is the inflation in business output which is not due to labor.

The blue line runs mostly below the red. This stands to reason, if only because labor-cost inflation does not show up in the blue line. In the early '60s, for example, the trends of red and blue run parallel, with red a little higher. Then in the latter 1960s as the Great Inflation took hold, business non-labor inflation falls behind overall GDP inflation.

The first years on the graph show inflation immediately following the Second World War. The red runs much higher than the blue. This could be attributed to government, rather than business or labor. Similarly, in 1951-52 the red runs extremely high, related this time to the Korean War.

Now look at 1953-1956. In '53 the blue is flat at zero. The red is flat and a little higher, comparable to what we saw for the early 1960s. Then in 1954 and 1955 the lines move upward from the zero level, reaching the 1.0 level. The 1.0 level is 1% inflation quarterly, or about 4% per year. That's what Samuelson and Solow were looking at.

Notice that during this transition the red and blue run together, sometimes touching, and sometimes even with the blue above the red. There is essentially no gap between the lines during this time, from before 1954 to after 1955 (1953 Q4 to 1956 Q1).

In other words, during all of 1954 and all of 1955, business sector non-labor inflation was responsible for inflation as measured by the GDP Deflator. Labor costs are excluded from the calculation. Labor cannot be held responsible for the inflation of those years.

I think that's pretty interesting.

Productivity and Compensation -- The Early Years -- the first take

I put output-per-hour and compensation-per-hour (two lines) on a graph, for both the Business Sector and the NonFarm Business Sector (four lines total). Indexed all the lines to the start date (1947 Q1) so they all start at the 100 level. Then I subtracted 50 from the Nonfarm Business numbers to shift them down a bit on the graph. So we can compare productivity-and-compensation for the business sector (blue and red) to productivity-and-compensation for the nonfarm business sector (green and purple):

Graph #1: Productivity and Compensation 1947-1970 for Business & Nonfarm Business

The separation between productivity and compensation is greater for the business sector than for nonfarm business in these early years.

...and then I got distracted...

I suppose, if you subtract nonfarm business from business, you're left with farm business. I have to look.

I should say first that I'm not sure of this calculation

Business - NonFarm Business = Farm Business

It certainly seems right. But it might be right conceptually, and wrong mathematically. I'm only looking at this graph because I want to see what it looks like:

Graph #2: Productivity and Compensation 1947-1970 for Business & Farm Business

Almost flat. Farm productivity -- and farm wages -- are almost flat. Well, on Graph #1 the lower lines are almost parallel to the upper lines. If we subtract one from the other, the result should be almost flat.

But does it make sense? I don't think so. I added population to the graph, indexed it, and moved it down near the farm productivity line. Population grows faster than my measure of farm productivity. So that would mean farm employment would have to rise to keep up with demand. But farm employment has gone down, right?

I googled farm employment. Google thought I wanted a job.

I had a thought

Yesterday I showed "labor share" and "real" labor share (figured from inflation-adjusted values) together on a graph, set equal at the start by indexing, so we could compare the changes. Graph #4 from yesterday.

I'm adding two circles to that graph:

Graph #1: Labor Share, nominal (red) and real (blue), Indexed to 1947 Q1

On the red line, the circle shows the increase of labor share during the "Goldilocks" of the latter 1990s, and the return to trend after 2000.

Point of interest: On the blue line, the ratio of reals, labor share does not increase. It only runs flat. So, whatever it was that was happening in the latter 1990s, that's what was needed to prevent the decline of labor share.

What was happening in the latter '90s? Debt-per-Dollar (DPD) was rising rapidly, returning to trend after a decline early in that decade:

Graph #2: Debt per Circulating Dollar, Below Trend in the 1990s
The ratio is an Indicator of Financial Cost in the Economy as a Whole

As soon as DPD got back to trend in 2000, the rate of increase slowed. As soon as the rate of increase slowed, labor share fell.

What these graphs tell me is that, by the 1990s, our economy required an abnormally rapid increase in debt to prevent labor share from falling. In the 1950s, by contrast, a very slow increase in debt was sufficient to keep labor share running flat, and at a high level besides.

I'm gonna have to look into that.

I took the DPD ratio (blue on Graph #2), made it green, and put it on the graph with real and nominal labor share so you can get a better look:

Graph #3, Combining Graphs #1 and #2
The link gets you the series data, but FRED can't remember user-defined line settings

Oh and by the way, I slept on it. I don't see anything wrong with figuring labor share as a ratio of reals.

Exploring labor share

To figure labor share they take compensation, divide it by current dollar output, and multiply by 100.

This is what labor share looks like:

Graph #1: Labor Share (red) and the Calculation that produces it (blue)

Pretty, huh. Except it's going downhill. But hey, you ain't seen nothin yet.

I got onto this topic a few days ago after looking at productivity and compensation. Productivity is output per hour worked. It is a way to evaluate economic growth. Because it is a way to look at growth, the "output" number has to have the inflation component removed. Otherwise, if prices are changing, you get a false reading of how much the output component actually changed.

Hours are hours. Inflation doesn't make the "hours worked" number different. But compensation is money. If we are using real output to figure productivity, and we are comparing it to compensation, then the compensation number also has to have the inflation component removed. Otherwise, inflation distorts the result.

But you don't always have to remove the inflation component. Labor share, for example, is not a measure of economic growth. So it is okay to leave inflation in the numbers used to calculate labor share, as long as you leave it in both compensation and output.

FRED uses the words "current dollar output" in Graph #1, to make it explicit that the data is a "nominal" series and the inflation component has not been removed. Compensation, the other series used in the calculation, is also nominal. The version of that series that has the inflation component removed would be called "real compensation" if FRED had that series. (They do have "real compensation per hour" data sets.)

Anyway, I started out looking at productivity and compensation, where inflation-adjusted data is used. Then I changed to nominal (inflating) data and discovered that compensation relative to output gives me labor share.

So now I'm looking at labor share, and I want to see how it looks if I use inflation-adjusted data for the calculation. I still can't find a "Business Sector: Real Compensation" series at FRED, but I can make one. I can take their "Real Compensation per Hour" series and divide it by "Compensation per Hour" to get a price index for the compensation data, then multiply by Compensation to get Real Compensation.

Hours cancels hours, comp cancels comp, and what remains is real compensation.

Then divide that by Real Output to get the inflation-adjusted version of Labor Share:

Graph #2: "Real" (Inflation-Adjusted) Labor Share, calculated as
((Business Sector: Real Compensation Per Hour / Business Sector: Compensation Per Hour) * Business Sector: Compensation) / Business Sector: Real Output

You were looking for downhill? There's downhill for ya.

You can eyeball it and see that I got a value of 1.0 for the base year 2009. That's because I didn't multiply by 100 because, you know, you multiply by 100 to get percent, and labor share is definitely not percent. As we discussed the other day.

But I can multiply it by 100 to make my number compatible with FRED's labor share. Then I'll add FRED's series to the graph in red so we can compare the two:

Graph #3: Labor Share (red) and "Real" Labor Share (blue)

Nice, huh? Now you can see what downhill really looks like.

The ratio of reals is not the same as the ratio of nominals because, in the business sector numbers, the inflation measures for compensation and for output are not the same.

Note that the two lines cross in the base year (2009) at the level 100. Or in that neighborhood; it's hard to see exactly.

But you know, this is why I hate those updates they do every few years, where they change the base year to a more recent date. Everything after the base year on Graph #3 is all bunched together. The blue is not any lower than the red, really, at the end, the years after 2009.

But it has to be lower, because it falls more. The graph is deceiving.

Yeah, now I sound like those guys that whine about how much value the dollar has lost since 1913. But you just don't get a good picture, when the base year is repeatedly changed to a more recent date. Maybe that's why those updates are done: to blur the picture and keep us confused. Oh Art, that's not a nice thing to say.

Maybe they do the updates because they are confused.

Sorry, that's as nice as it gets.

I'm going to index these two data sets using the data-start date as the base year. So they will start out, the two lines, together at the 100 level. And what happens after that, well, we'll see.

The start date for these data sets is January 1947. Back then, my dad was a young guy getting his start in the world, and I was still a couple years away from being born. The graph will show things from my dad's perspective, how things changed since that early point in his life.

The graph shows how "the greatest generation" got screwed. It shows a 50-year decline of labor share that could and should have been stopped somewhere along the way:

Graph #4: Labor Share, nominal (red) and real (blue), Indexed to 1947 Q1

It's interesting, I think, that the "real" line (the blue line) suggests that the decline in labor share definitely started in 1961. Some people say the red line shows no decline until 1970.

Some say labor share was flat until around 2001. Those guys are 40 years off.


I spent a lot of time on this "real labor share" thing. I don't mean to suggest that labor share should be figured using inflation-adjusted quantities. I haven't yet had so much as a thought about that. I'm just looking, to see how things look.

The blue line shows a decline with a remarkably straight trend. That's gotta mean something. But again, I haven't yet had so much as a thought.

Labor share: Where do they get those numbers?

So I can answer a question that's been in the back of my mind for a long time now.

Why is labor share more than 100%?

It's not percent, Art. It's an index.

Yeah I know. But how did it get so high? It's useless. It tells me labor share is going down, but it doesn't tell me what the share is, the share that labor gets. It should be a percentage. It's not, I know, but it should be.

So where do they get those numbers?

Labor share, as I learned a couple days ago, is calculated by taking compensation as a percent of current-dollar output.

It is a percentage -- see? I was right. It has to be a percentage.

But the thing is, it takes one set of indexed values as a percent of another set of indexed values. Indexed values are not the actual values. Indexed values are like a price index (duh, Art). They pick one year to be the base year, and they figure all the values as a percent of the base year value. So right away when they do that, the actual values are gone. You get a line on a graph that is the exact same shape you get from the original values, but the whole shape has been moved up or down until the base year value is equal to 100.

Seems harmless, right? Except of course the original numbers are gone. So you cannot look at labor share and find compensation as a percent of current dollar output, because you don't have those numbers anymore.


The name "labor share" sounds like it would give you compensation as a percent of current dollar output. But it doesn't work that way. When they pick a year to be the base year, something they do again every few years, they pick a date from the recent past. An arbitrary choice, let's say. Then they re-figure the data for all the years so that the new base year gets the value 100. All the numbers get changed.

One thing that's for sure is that the base year value is 100. If you have two data sets like compensation and current-dollar output, you can be sure that when you plot them on a graph, the two lines will cross in the base year. Because both data sets have the value 100 for the base year.

I don't know what compensation is, as a percent of current-dollar output. Maybe it's 80%. Maybe it's six. But I know for sure that compensation is going to be 100% of current-dollar output in the base year of a graph, because the numbers are indexed and the base year values are equal.

It's ridiculous to think that compensation and output are equal. That would mean labor share is 100% and capital share is zero, and I'm sure that's not the case. But that's all we can get from the indexed series called "labor share". That's all we can get even if we go back to the data that is used to calculate labor share, because that data is indexed, too.

So anyway, we know that labor share is compensation as a percent of current-dollar output, except the two lines cross at the 100 level in the base year. We also know that labor share has been going downhill for a long time.

Labor share goes downhill for a long time, and then the lines cross at the 100 level. So that means labor share has to be higher than the 100 level in the years before the base year. And sure enough, it is.

Reversal of fortune

Yesterday, looking at compensation and current-dollar output for the business sector, I said "we are looking at income and cost."

I was thinking of compensation as income, and the purchase of output as a cost.

I'm pretty sure economists have it the other way around.

Intuitive, but only after I saw it

Starting with the "productivity versus compensation" thing, I decided to look at aggregate totals rather than "per hour" numbers, and switched from "real" to "nominal" data. We're not looking at growth now, but we are looking at income and cost.

I figured compensation as a percent of current dollar output. Guess what it looks like: Labor share.

Graph #1: Compensation as a Percent of Output, nominal Business Sector index values

Exactly like labor share:

Graph #2: The Graph #1 Data (blue) and Labor Share (red)

I had no idea. I love it when that happens.

Since when? The Productivity–Pay Gap

Noah Smith at Bloomberg: Workers Get Nothing When They Produce More? Wrong. Noah shows this graph:

Graph #1: Employee Compensation is Falling Behind Already by 1962

Noah's article is about the way people interpret the graph. As he puts it, some people see compensation falling behind output and ask:

If productivity improvements don’t actually get translated into wages, what’s the point of making the economy more efficient?

They say the graph shows there is no sense in boosting productivity. Noah disagrees, as you can tell from the title of his article. He cites a study that looks at short-term changes and sees "a correlation between productivity and wages -- when productivity rises, wages also tend to rise." But over the longer term, compensation falls behind productivity because of "forces pushing in the other direction" he says.

That's interesting. It leads to questions: What forces? And: Why? And: What can be done about this? It opens doors.

Before we come up with a solution to the problem of lagging compensation, we have to know the cause of the problem. We don't yet know the cause. We don't even know when the problem started, so how can we know the cause?

If you don't know when the problem started, how can you possibly know the cause? And if you don't know the cause, how can you possibly know the solution?

How can you have any pudding if you don't eat yer meat?


When did the problem start? Noah doesn't say:

Since the end of World War II, productivity, in terms of economic output per hour, has grown by a factor of five, while compensation has only tripled. Since 1980 the divergence has been especially stark ...

More stark divergence may indicate a new phase, but does not mark the start of the problem. In any case, on Noah's graph I don't see a more stark divergence beginning around 1980. I see a kink in both lines shortly after 1972, and a kink in both lines around 1997. Nothing around 1980, other than a brief dip in productivity. Not even a wiggle in compensation. Look at the graph.

It looks to me that, for Noah's data at least, productivity and compensation ran neck-and-neck from 1947 to just before 1962. Call it 1960. A gap opens after 1960. This is where the problem begins. Not 1980.

There is a kink around 1973, and then compensation slows more than productivity. Maybe the second stage of the problem begins in 1973.

There is another kink around 1997, after which productivity accelerates upward and compensation does not. This would be the third stage.

If you hover a mouse over Graph #1 you will see the trends I'm looking at.

Around 1980 or 1982 there is nothing but a wiggle in productivity which is surely the result of the double-dip recession that occurred at that time. No wiggle is visible in the compensation data for those years.

Perhaps Noah sees stark divergence since 1980 because he has heard people speak of it a million times. Or perhaps he is thinking of this graph from Robert Reich:

Graph #2. Source: the Preservation Institute Blog.
Ritholtz at The Big Picture shows this graph as part of a larger image
and links to an article by Robert Reich at the New York Times.

Reich puts a white stripe down the graph at 1980. The white stripe is a conclusion imposed on the data, a conclusion that intrudes upon unbiased evaluation. The data on Robert Reich's graph is not the same as that shown on Noah Smith's graph, yet both graphs show productivity and compensation running together from 1947 to about 1960, and productivity gaining on compensation since that time.

You could say that Robert Reich put the white stripe at the point where compensation peaks and starts to fall. This would explain why the white stripe appears at 1980. But by 1980, compensation had been falling behind productivity for 20 years, as you can plainly see on Reich's graph.

Noah's data is not the same as Robert Reich's. Noah's graph shows a change around 1973, and a change around 1997. It shows nothing but a recession-related wiggle in productivity around 1980. And yet Noah's "stark divergence" begins in 1980, at the same moment that Robert Reich's white stripe appears. Why?

Our views are influenced by the views of others. Perhaps Noah sees stark divergence since 1980 because he has heard people say it a million times, or has seen it on Reich's graph.

At the Preservation Institute where I found Reich's graph, the evaluation of the graph is based more on the white stripe than on the data:

From 1947 to 1979, the compensation of non-supervisory workers increased by almost as much as productivity.

From 1980 to the present, compensation stagnated as productivity continued to grow - partly because of deliberate economic policies that were adopted by the Reagan administration ...

Indeed, pointing the finger at Reagan may be the real reason for locating that white stripe at 1980. But even if that is not the case, Reich's white stripe evaluates the data for us, before we can evaluate it for ourselves. No thank you, Mr. Reich.

I don't see a "more stark divergence" after 1980. But don't think that I am defending Ronald Reagan. Reagan's policies did nothing to close the gap between compensation and productivity. Reagan did his share to keep the gap growing. But Reagan did not do more than his share. Not according to Noah's graph.


Now comes the part where I duplicate the original graph so I can examine the data.

I wasn't sure what data series Noah used in his graph, but he does say Source: Federal Reserve Bank of St. Louis so I'm thinkin FRED. FRED doesn't have a lot of data series that match "Real output per hour" or "Real compensation per hour" and go all the way back to 1947. I only find data for the "business sector" and the "nonfarm business sector". So I could throw darts at this, blindfolded, and not be very far off.

I duplicated Noah's graph once for "business" and once for "nonfarm business". Both were close, but neither exactly matched what Noah showed. Maybe there was a recent revision? No matter. I'll go with "business sector" and index the values on the first quarter of 1947. That'll be close enough.

Here's what I got:

Graph #3: My Attempt to Duplicate Noah's Graph -- A Pretty Good Match

My lines both start at 100, like Noah's. One line ends at 300 and the other at 500, as Noah shows. And I got colors comparable to his. So far, so good. The gap on my graph starts just after 1960 and shows trend changes around 1973 and 1997, just like Noah's graph. But the wedge opens up more slowly on my graph. Between 1960 and 1970 I show a narrow gap between the two lines; Noah's gap looks wider. Or maybe it's just my eyes. Whatever, I'm going with this data.

Data in hand, I want to look at the gap between productivity and compensation. A simple way to do that is to subtract compensation from productivity. If productivity is higher than compensation -- which it is -- then the graph will show how much higher the productivity number is. It will show the difference, the gap.

Here's what I get:

Graph #4: The Size of the Productivity-Compensation Gap

"Since 1980," Noah says, "the divergence has been especially stark". I don't see it. The gap is obviously bigger in the later years. But there is no "kink" in the blue line around 1980, and no acceleration thereafter. The line just goes up, the same after 1980 as before.

Just after the year 2000, yeah, there does seem to be a kink there. And the line goes up faster after that, for a while. But it's 20 years too late. You can't blame Reagan. There is no kink and no acceleration around 1980 on this graph. There is no sudden change in the growth of the gap. What was Noah thinkin?

Nor is there a dramatic increase since 1973, contrary to what EPI claims. The line is obviously higher after 1973 than before, but there is no sudden change in the upward trend. The pace is constant since 1961.

Hey, when you're looking at the gap between two jiggy lines, it is difficult to see exactly the size of the gap. But when you subtract the one line from the other, as on Graph #4, there is nothing left to look at but the gap. And then it is easy to see the size and shape of the gap.

The gap opens around 1961, or possibly earlier. The gap shows consistent increase. There is no sudden acceleration after 1973. There is no sudden acceleration after 1980. There is just consistent increase. I moved the data into Excel and put a trend line on it:

Graph #5: A Trend Line Added to the Data from Graph #4

If anything, the line looks a little low in the 1980s and '90s, relative to trend.

No sudden surge around 1980. No sudden surge around 1973. Only a continuous and stable increase in the gap since 1961.

What are the forces pushing compensation down while productivity rises? The Reagan revolution may be part of the problem. The economic slowdown after 1973 may be part of it. But the problem didn't begin in the 1970s or '80s. It began in the 1960s, or before. To discover the troublesome forces, we have to look to that earlier time.

Labor Productivity since 1988

I like this graph of labor productivity. The red and blue data lines look like trend lines for the jiggy green:

Graph #1: Labor Productivity

See how productivity has been really low most of the time since 2010? You know about that.

It was very low like that for two or three years before 1995. Then all of a sudden it went high and the economy was good for a while. Yeah, I expect that to happen again soon and I think we're getting there now.

What else do you see on the graph? Productivity always goes high after a recession. After the 1991 recession on this graph, and after the 2001 recession, and even after the Great Recession. A good big fat spike after the Great Recession. And then nothing.

Any minute now it will go up, like in the mid-90s. If you run that red line out to the end of the graph, it's going up.

Any minute now ...

Recommended reading: Steve Denning

Prior to 1982, large stock buybacks were illegal because they constituted obvious stock price manipulation. But the SEC in the Reagan administration introduced a new rule—Rule 10b-18—which creates “a safe harbor” for firms to buy back as many shares as they like. This effectively opened the floodgates.

- Steve Denning

Well that was quick!

Goldilocks is back -- The Economist, 17 October 2017

Time is running out on 'Goldilocks' -- CNBC, 5 December 2017

Twin Peaks

It’s no mere coincidence that over the last century the top earners’ share of the nation’s total income peaked in 1928 and 2007 — the two years just preceding the biggest downturns.

- Robert B. Reich

Quantum Unemployment

Call me the duplicator: See a graph, make a graph.

Look up Okun's Law, and this graph comes up:

Graph #1: Okun's Law. Source: Wikipedia

I always want to see how it works: What is the data? What are the units? Can I duplicate the graph? It's an automatic process. I don't decide that I want to do it. I just do it. I don't know why. But anyway, the data series are Real GDP, and UNRATE at FRED. The values are quarterly. The unit for GDP growth is "percent change", and the unit for unemployment is just "change" because it starts as percent. So yeah, I can do that.

Here's what I got:

Graph #2: Okun's Law at FRED

The horizontal black line on the second graph is the X-axis, not the trend. Ignore it. Look at the shape of the cluster of blue dots: high on the left, low on the right. A "best fit" trend line would run from high on the left to low on the right, just like the trend line on the first graph. So I'm satisfied: I duplicated the graph.

I even got blue dots! My dots are all the same color, all faint blue. Some of them look dark because they overlap. There are more dots in the middle of the cluster than elsewhere, and they overlap and it makes them look dark blue. If you look at the first graph, there are more dots in the middle of the cluster there, too.

So that's what I saw when I looked at my graph. And then I noticed something odd: My dots are all arranged in columns, with blank space between the columns. You probably noticed it before I did. It's like unemployment always makes a quantum leap from one level to the next, as you go from left to right.

The other graph doesn't show that.

No, I know what it is: FRED rounded the unemployment numbers to one decimal place. You can see it, from the way the columns are spaced. And sure enough, if you hover over the graph at FRED it shows the unemployment values rounded to one decimal place.

Damn, I thought I was on to something with "quantum" unemployment!