A model for low inflation due to low unit labor costs and labor share
Inflation is low and has been declining for years. I want to present a model to explain a dynamic between inflation and unit labor costs.
First let’s look at a chart of the two of them over the years. (semi-annual data, % change from year ago)
They move together with more noise in the unit labor costs (blue line). I hope you notice that unit labor costs tend to peak before a recession. That is an important clue to seeing the connection between the two, because the natural level of output is reached before a recession.
Setting up the model
The model to explain this effect is an alternative version of the AS-AD model (aggregate supply – aggregate demand). The model uses effective demand instead of aggregate demand. So it is the AS-ED model. I am the only one who has ever used it.
The AS-ED model still puts price level on the y-axis and output on the x-axis. Yet the AS curve and the ED curve have different equations than what is normally seen.
Inflation rate for AS curve = ulce(3400 + real GDP – Potential real GDP)/(3400 * cu) – 1
Inflation rate for ED curve = real GDP * ulce/(ED*cu*(1-u)) – 1
Both equations use unit labor costs (ulc). In this model, the effective unit labor cost is used, which is the index of ulc * 0.76. I won’t go into this conversion here. cu is capacity utilization. u is the unemployment rate. ED is the effective demand limit. 3400 is a business cycle amplitude constant in 2009 billion dollars. Potential real GDP refers to the center of the business cycle and is determined by effective labor share.
Note: Inflation rate for AS curve uses real GDP on x-axis. As real GDP increases, inflation would rise holding all else equal. Inflation rate for ED curve uses effective demand (ED) on x-axis. As ED increases, inflation rate would fall holding all else equal.
The model looks like a normal AS-AD model. Yet the lines always cross at the LRAS curve, which is the natural level of output. In the first graph above I pointed out that the natural level of output is reached before a recession. These lines determine that natural level.
GDP moves with the AS curve below the ED curve. GDP does not like to rise above the ED curve. Profits decrease. The implication is that inflation is thus limited below the ED curve. So when we drop unit labor costs, holding all other variables constant, the top limit upon inflation drops.
As unit labor costs fall, the natural level of output does not change, but the inflation limit of the economy will fall. So as we see in the first graph, the trend of lower unit labor costs has brought down the inflation upper limit.
I will point out that unit labor costs have a connection to labor share too. I add yearly % change of labor share in this graph.
In some years, the movement of unit labor costs (blue line) moves with inflation (red line) and in other years moves with labor share (green line). We could generally say that when inflation expectations are low or falling, unit labor costs tend to move with labor share.
The result is stubborn low inflation
Currently we have a situation where labor share has been falling since 2001. Inflation expectations are still low, even though they are higher than actual inflation. Unit labor costs are trending low with labor share. Unit labor costs went negative after the last 2 recessions. The effective demand limit is weighing down upon inflation as shown by the AS-ED model. The result is a dynamic for stubborn low inflation.
Lowering wages is not the solution
The problem is that business has the attitude that it must lower unit labor costs. Here is common advice to small businesses from an online source… (Bryan Keythman)
“An increase in your small business’ unit labor cost means your productivity is decreasing. Productivity is the efficiency with which you generate output using certain inputs. The increase in unit labor cost may result in higher expenses compared to revenue. Unless you decrease other expenses or increase your selling prices to increase your revenue, your profit will decrease. Investigate the cause to determine how you can decrease your unit labor cost and regain your previous level of productivity.”
The statement in bold is wrong… decreasing unit labor costs does not mean raising productivity. It may just mean lower wages.
unit labor costs = wage rate / productivity
It is true that raising productivity will lower unit labor costs, but many businesses just hold down wage increases thinking that will increase productivity. But that is not correct.
The key is to raise productivity, which many times involves raising compensation to workers… as this video refreshingly shows. (from Labor Performance) The video says that it is a “common misconception that if you pay more money, your labor costs go up.”
Compensation of Employees reached its peak share (67.7%) of National Income in 1980:
http://research.stlouisfed.org/fred2/graph/?graph_id=132281&category_id=0
And that happens to be the peak year for inflation as well. Coincidence?
The labor share of income during the age of disinflation has declined pretty much everywhere in the advanced world.
Peak Core CPI Rate*, Peak and Recent Labor Share of Income (Total Economy) (*Except Portugal)
Nation——CPI-Year– Peak-Year-Recent-Year-Change
US———-12.4-1980-69.6—1980-63.7—2010–5.9
Japan——-20.2-1974-72.5—1977-56.6—2009-14.9
Germany——6.8-1974-76.3—1974-68.5—2011–7.8
UK———-22.1-1975-75.6—1975-71.3—2010–4.3
France——12.7-1980-79.2—1981-68.4—2010-10.8
Italy——-22.3-1980-83.4—1971-68.1—2010-15.3
Spain——-26.4-1977-76.4—1977-59.9—2011-16.5
Canada——11.1-1980-68.1—1971-59.8—2008–8.3
Australia—12.8-1977-75.5—1975-61.3—2006-14.2
Neth.——-10.5-1975-77.1—1975-68.5—2010–8.6
Sweden——12.5-1980-77.9—1978-63.3—2011-14.6
Switzerland–8.9-1974-67.5—2002-65.9—2010–1.6
Austria—–11.1-1981-98.5—1978-66.3—2011-32.2
Norway——12.2-1981-73.7—1977-55.4—2011-18.3
Portugal*—33.1-1977-83.9—1975-66.4—2010-17.5
Denmark—–10.6-1978-73.7—1980-69.5—2011–4.2
Finland—–17.5-1975-76.9—1991-66.1—2011-10.8
Ireland—–21.2-1981-79.3—1980-60.9—2010-18.4
New Zeal.—–17.2-1982-60.7—1975-49.0—2006-11.7
The international labor share data comes from the OECD and is not consistent with BEA data:
http://stats.oecd.org/Index.aspx?queryname=345&querytype=view
Select *Total Economy*.
Although there are many reasons for the increase in inequality that we have seen in the US and in other parts of the world (regressive taxation, weaker unions, lower minimum wages, globalization, Skills-Based-Technological-Change (SBTC) etc.) the leading hypothesis for why there has been such large scale declines in the labor share of factor income is disinflation (i.e. tight monetary policy).
Disinflation during the eighties and the nineties was accompanied by a significant rise in the profit share of national income in most OECD countries or, equivalently, by a reduction in the labor share. This suggests that changes in the rate of inflation are non-neutral with respect to the distribution of factor income. The consequences of inflation upon inequality thus may largely be the indirect result of the effects of inflation upon factor shares. The mechanism by which this comes about is fairly simple. Accelerating inflation is correlated to falling unemployment rates, falling unemployment rates lead to greater labor bargaining power, and greater labor bargaining power is correlated with lower markups. Furthermore, higher inflation rates create greater price dispersion leading to greater competition among producers to limit markups. This hypothesis was tested with a panel of 15 OECD countries over the period from 1960 to 2000 and a robust positive relationship between inflation and the labor share was obtained:
Inflation and Factor Shares
Francisco Alcalá and F. Israel Sanchoy
August 2000
Abstract: “We use results from the literature on the determinants of price-cost margins to derive an equation relating labor’s share of national income to the inflation rate (as well as to the output gap, the unemployment rate and the capital stock per worker). The equation is tested with a panel of 15 OECD countries. We obtain a robust positive relationship between inflation and the labor share. Our results suggest that disinflation is not distributively neutral, provide empirical support for the distinct concern about price stability shown by trade unions and employers’ organizations, and help explaining the negative impact of inflation on growth.”
http://pareto.uab.es/wp/2000/46000.pdf
Edward,
I’ve run a Granger causality test using the Toda and Yamamato Vector Autoregression (VAR) technique on PCEPI and Compensation of Employees as a percent of National Income over 1947Q1 through 2013Q2. The results show inflation Granger causes labor share at the 5% significance level, but that labor share does not Granger cause inflation.
As I’ve said to you many times before on this issue, the direction of causality matters.
Mark,
Your point is a good one. However, Granger causality has many limitations and assumptions which I’m sure you’re aware of. Additionally, if you used first differences then you’re removing a significant amount of information.
Anyway, it’s good to see others discussing this work.
Take care.
Mark,
Again it is great to see your ideas. Granger does have its limitations, but it still needs to be considered.
You might say that inflation causes labor share changes, but there are other factors, like the effect of imports from China. And the effects of inflation around the natural level of output. Inflation takes its own course at that moment, and labor share acts to set a top limit. Labor share does not so much cause inflation, as it more sets a limit upon it.
So when you say that inflation Granger causes labor share, inflation rises a bit ahead of labor share. If you look at the 3rd graph above that is pretty much true before a recession. Yet, labor share has more noise in it.
What I am saying is that there is a dynamic of limits around the movements of these factors. Limits don’t really Granger cause anything in a correlation time period. So a Granger causality test may not be the most appropriate test for a limit. But I listen to a Granger test because I like them, and feel they add good information.
SJ,
“Additionally, if you used first differences then you’re removing a significant amount of information.”
The Toda and Yamamato technique involves estimating a VAR in levels, declaring extra lags of each variable as exogenous variables in order to correct for non-stationarity.
Mark,
Good to know. Thank you.