Minimum Wage and Labor Crises

I dug some more into historical national minimum wage data, and I'm shocked at what I have found.  This will be the first of several posts, tying into the recent conversation about Labor Force Participation and Unemployment.

Here is my post on relationships between historical employment levels and minimum wage hikes.

In that post, I divided each month from 1954 to 2013 into 3 categories:

1. Months that are not within 2 years after a MW hike.
2. Months within 2 years after any MW hike that occurs at least 2 years after any previous hikes.
3. Months within 2 years after any MW hike that occurs within 2 years of a previous hike.

In that post, I found evidence that minimum wage hikes were a significant factor in historical labor shocks, and I outlined an episodic pattern of employment loss after MW hikes.  For this post, I have regressed changes in employment against lagged 6 month changes in MW as a proportion of average wages.  For Group 2 & 3 months (months within 2 years of an initial MW hike and within 2 years of follow-up hikes), the relationship looks like this:

The relationship during Group 1 periods (no MW hikes) is not coherent, since there are no shocks to the MW/(Avg. Wages) measure during those times.  In the Group 2 and 3 periods, the independent variable jumps when MW hikes are implemented, and then recedes in proportion to the growth of average wages.  So, this relationship measures both the effect of the shock (the MW hike) and the mitigating effect of nominal wage increases as the MW remains at the new level.

If the MW is increased by an amount equal to 10% of average wages, it is associated with about a 2% drop in employment in the preceding 6 months and the following 4 six-month periods.  Among the most affected group, 16-19 year olds, the drop in employment is about 5% in each of the periods before and after the hike, with the effect diminishing over the next few periods.

Most MW hikes are implemented as a series, and the Group 3 graph shows the relationship during those follow-up hikes.  The relationship is even stronger here.  This is because follow-up hikes tend to have an increasing effect on MW employment because they tend to push the MW level up to a higher proportion of the average wage, and therefore affect a larger number of workers (the price floor enters a fatter part of the income distribution as the MW/(Avg. Wage) ratio increases).  This stronger relationship also reflects the ability for high inflation or strong real wage growth to push the price floor back to the narrow end of the income distribution, which allows for some catch-up employment growth, especially among the most vulnerable groups, like teenagers.

So, we can construct a forecast of employment trends by using these lagged relationships to MW hikes.  The following graphs compare these forecasts to the actual changes in employment.  The 1976-1982 period is not included in my analysis, because there were a number of MW hikes that roughly matched inflation, so there is no clear pattern of price floor shocks with which to establish a relationship.  For Group 1 (non-MW) periods, the forecast reflects naïve mean values for those periods.  For Group 2 and 3 periods, the forecast reflects the lagged correlations with changes in MW/AW:


Original graph, based on Establishment Survey

Edit:  After I posted this, I made several adjustments to the data.  One was to reduce the effect of the 2007-2009 episode, because in this post I saw evidence of a lesser influence of the MW at the pre-2008 levels.  In addition, after originally using establishment survey data, I had switched to household survey data, in order to have more consistent data between total employment and age-specific employment.  I just realized, while reviewing my spreadsheets, that in the household data, a drop in total employment shows up in 1996, which means that every MW hike is associated with a statistical drop in employment. (Although, one could debate the episodes in the 1960's, where the employment data doesn't seem to convey coherent cyclical trends).

There are 7 MW hike episodes, and 6  all of them are associated with labor shocks or downward changes in trends.  Outside the 1976-1982 period, there are only 3 labor shocks that are not coincident with MW hikes.  The one MW hike not associated with a labor crisis is the 1996 hike, which was a fairly small hike during the hottest labor market of the last half century.  While no negative effect on total employment is visible in that episode, the teen employment level does appear to show a dip followed by a recovery, which is a typical pattern.

We can also see here how MW episodes where nominal wage growth is strong relative to the follow-up hikes can actually be associated with a rebound in employment as employment catches back up to trend.

The minimum wage forecast explains almost all of the teen employment loss of the last crisis, and a surprising amount of the total employment loss.

Here is a graph of detrended changes in total employment in the 2008 crisis, compared to a forecast that is specified by the historical correlations (the 2008 crisis is out of sample):

From July 2007 to June 2011, when the Group 2 & 3 forecasts are active, the forecasted detrended employment decline is -4.24%.  The actual decline in that time period was -8.23%.

Half of the employment loss of the crisis was related to the minimum wage hikes?  Surely not.  This is hard to believe, but there it is.

One caveat is that the measure I use, MW/AW (minimum wage as a proportion of average wages), creates a circular argument, because in the absence of a MW hike, stagnant average wages will cause these forecasts to also forecast lower employment growth.  I believe that this effect is limited, however.  Two points to this effect:

1) During Group 2 & 3 periods, there will be at least one period with a hike in MW.  In the recent episode, these hikes were all in the range of a 3% - 3.5% increase in MW/AW.  During 6 month periods without a MW hike, the ratio has declined by .3% - .6%.  So the effect of the shock period on the forecast is much larger than the effect of the non-shock periods.  This is clear in the forecast graph, where the forecast steps up and down, depending on how many rate hikes were implemented in the preceding 2 years.  Other movements beside these shock movements, related to changes in average wages, are much smaller.

2) A decent portion of the change in nominal average wages is a product of inflation.  In the periods after a MW hike, inflation policy can be correctly considered part of the appropriate policy response to MW hikes.  To the extent that lower wage inflation after a MW hike prevents labor markets from recovering, this is part of the MW policy.  So, to the extent that this effect changes the forecast, it is an effect that is entangled with MW policy in a way that would be difficult to separate.

Addendum:  I messed around with the regressions of Employment growth from MW/AW changes by adding nominal wage growth, real wage growth, and inflation as independent variables.  Nothing made much of a change in the coefficients or significance of the MW/AW variables.  In Group 2 periods, the addition of nominal wage growth does improve the strength of the regression, but, interestingly, inflation seems to have a negative impact on employment growth while real wages have a positive impact.  Neither is statistically significant.  And the addition of these variables causes the coefficients and the significance of the MW/AW variables to increase.  In general, within this context, there is not a clear relationship between inflation or real wage growth and employment growth, but I think I can say that the presence of Average Wages in the denominator of the MW/AW variables is not causing a non-MW factor to falsely inflate the significance of the MW/AW variables.

In the following posts I will look at the effect of MW hikes on labor force participation.  The decline in LFP has been a major concern during this crisis.  Interestingly, given the above finding, my preliminary analysis on the MW shows that most MW employment loss comes from workers leaving the labor force, not from unemployment.  So, in this way also, the signature of the recent crisis has reflected what one would expect from a MW-induced labor shock.