Do Sticky Wages Matter?

Do they even exist?

I recently read Basil Halperin’s blog post on price-stickiness vs wage stickiness from a few years ago, and was struck by one of its points on wage stickiness. Paraphrasing Pissarides 2009, most of the stories people give for wage stickiness causing unemployment don’t make sense. It isn’t the salary of incumbent workers that matter, but those of new hires. But first, let’s back up and give you all the context you need for that point to be meaningful.

People have expectations over what the future course of prices are. Generally they expect for them to keep rising at around 2% per year, with around 2% real growth as well, and make contracts on that basis. If those expectations are foiled, then someone loses out. If inflation is unexpectedly high, then lenders lose out because people are able to pay back loans with dollars that can now buy less stuff. If prices are unexpectedly low, people lose their jobs, because now their jobs are paying them more in real terms. This pushes some people over the edge of profitability, and so they laid-off. 

Recessions can be due to either real, or nominal, negative shocks. A real shock is something like OPEC deciding to cut back oil production for essentially political reasons. Oil is an input into many products, often without good substitutes, and we are simply able to produce less than we could before. The 1970s had several oil crises, which resulted in disappointing levels of growth. Real shocks are unfortunate, but largely inevitable, so their existence doesn’t really bother me. 

Nominal shocks, however, annoy me because they really shouldn’t be allowed to exist. The central bank – the Federal Reserve in the US, the European Central Bank in the EU – determines the rate at which people can borrow money from them, and can also simply “print” money. They ultimately have total control over the course of nominal GDP. Let’s suppose they screw up though. People expect prices to continue by 2%, but instead they don’t rise at all. This’d be like everyone getting a two percent raise, but with no commensurate increase in productivity. In the worst cases, like the Great Depression, the price level actually plummets – prices fell by at least a third in the US. 

The critical assumption for nominal shocks leading to unemployment is that wages are sticky. We really aren’t sure that they are, though. The plain theory, where the vagaries of price changes leads one to one with unemployment, is clearly too strong, but very few would argue that it does not explain the basics of why unemployment happens during very large negative nominal shocks, such as the Great Recession and the Great Depression. In other words, during large shocks wages are countercyclical (they rise when the economy falls, and vice versa), and not procyclical (they fall when the economy falls), but during real shocks, they are mildly procyclical. This has changed over time, toward becoming more procyclical. (Though note that this article was written before the Great Recession! Which was a proto-typical nominal shock leading to unemployment, after decades of real shocks dominating.) Tyler Cowen surveys the evidence here: I recommend it for its readability, and links to other papers. (It is certainly no substitute for a thorough survey if you are interested in knowing more). Sumner and Silver 1989 argue that the procyclicality of wages depends on the sample period chosen, and that while supply shocks lead to procyclical wages, aggregate demand shocks – that is to say, nominal shocks – lead to countercyclical wages. This is consistent with recessions occurring due to monetary policy. In short, how wages respond to recessions is an enormous muddle with no clear answer, and this is one of the better arguments for an RBCesque view of the business cycle, at least most of the time. I will note that during recessions, it will be biased toward procyclicality if you aren’t careful, simply because the lowest wage jobs tend to be jettisoned first. 

Most of the stories of sticky wages rest upon psychological reasoning. Truman Bewley, for example, simply went out and asked people for his book, “Why Wages Don’t Fall During a Recession”. Managers and business owners believed that giving out broad cuts in income will negatively impact morale, more than a few layoffs will. It might also be individually optimal to not allow your wages to fall. The hope is that everyone else takes a pay cut, while you are able to keep your wages high. This can work if hiring and retraining a new worker is costly, or if it is difficult to tell who is responsible for productivity. In other words, there may be no incentive compatible way of getting people to agree to wage cuts. (To illustrate, with an example from Myerson 1979: suppose two people want to build a bridge with costs 100. Player 1 always values it at 90, while Player 2 values it at 90 90% of the time, and 30 10% of the time. Obviously, the bridge should always get built in a perfect world, with Player 1 paying 70 every time, but Player 1 could increase their gains by not having the bridge be built some percentage of the time. Thus, optimal negotiating with asymmetric information leads to social loss). Getting laid off is similar to the bridge being built – everyone would prefer if wages got lowered, but maximizing your revenue is done by being the one

Once someone is fired, though, their position is no different than any other person. We should expect them to negotiate just the same as any other unemployed person, and for there to be no more unemployment than during normal times. This is the observation of Pissarides 2009 – sticky wages are dependent upon the wages of new hires, not that of those within the firm. After all, after someone is laid off, what’s to keep them from being hired back at lower wages? Any explanation of sticky wages must explain this, and cannot rely too heavily on risk-aversion or endowment effects. This doesn’t mean that sticky wages don’t exist. They (probably, with caveats) do. We just don’t have good explanations as to why. 

And why does this matter? After all, as per Friedman, the assumptions of a model don’t matter, only its predictive power. I think Friedman is mostly wrong, though. If we are repeatedly predicting the same things, then it doesn’t matter how we arrived at the predictive model. If the number of times we are predicting is small, though, then we have no way of avoiding spurious overfitting except through having reasonable priors. Moreover, we can not be sure that we are making predictions about the same thing every time. 

Having better microfoundations of sticky wages would help us predict how it will change in the future. Will better AI in job search make wages stickier or more elastic? Let’s assume that the effect will be to make changing jobs easier. In our prior stories of nominal wage rigidity, we would expect wages to become more elastic. If wage stickiness is not dependent on people not changing jobs, then we would predict no change at all. The whole point of economics is to allow us to make better decisions in the present day, and without sound microfoundations, we are likely to be in error.