Why Does Unemployment Happen?

Why it rose, why it fell, and what may happen later

Why does unemployment exist? There are workers who are unemployed – they would prefer not to be unemployed. Many would be willing to negotiate down their wages, but are unable to. What gives? What keeps markets from clearing? This type of unemployment is different from that of a recession which is caused by some adverse shock – either inadequate money, as in the case of a demand shock such as the Great Depression or Recession; or a supply shock, like during the oil crisis of 1973. 

What’s more, the heyday of theorizing was in the 1970s and 80s, when unemployment rates were persistently high for many years. The average level of unemployment has clearly decreased since then, even with the slow recovery from the Great Recession. I cannot definitively answer why this decline happened, but I will put together some theories as to why, and speculate wildly as to the prevailing rates of unemployment in years to come. 

i. Labor Search

The starting point for this question should obviously be labor search models. Finding information is costly. If you’re laid-off, it takes time for you to find a new fit. We generally model this as everyone having a reservation wage, below which they will not work, and people search until they find a job with a wage exceeding that. As people search, they update their reservation wage down until they find a new job. Meanwhile, firms do the same thing, changing their reservation wage upwards until they find the workers they need. Unemployment is simply due to friction.

I want to emphasize that this market is two-sided. Employers have “unemployment” too, in the sense that they might have production opportunities which they cannot fill. As the number of unemployed drop, if the costs of filling a job is a negative function of the number of people searching, it will get more and more expensive to fill the positions. Thus, there’s an equilibrium level of unemployment. Pascal Michaillat and Emmanuel Saez suggest we think about the “true rate” of economic misutilization as the geometric mean of unemployment and the vacancy rate, which has stayed remarkably constant over time.

You can also account for externalities. Suppose it takes some fixed cost to enter the labor market, on either side, with the likelihood of you getting a job being dependent upon the investments that other people take. In this case, the optimal investment is dependent upon what you expect other people to do, and there are any number of unemployment equilibria. This was illustrated by Peter Diamond in 1982, in what is called the “Diamond Coconut Model”. There is one good in this world, fancifully taken to be coconuts, which can be harvested by climbing a tree at cost to oneself, and then (on account of the customs of the people) traded with another. If the cost of trading is at all dependent upon the number of people searching, then whether you go up depends upon how many others you expect to do likewise. There are many equilibriums, all of which are perfectly rational.

ii. Efficiency Wages

Another theory of why unemployment exists is called the “efficiency wage hypothesis”. The cleanest exposition is in 1984 from Shapiro and Stiglitz, in their paper “Equilibrium Unemployment as a Worker Discipline Device”, which says that firms pay wages above equilibrium in order to prevent shirking. Workers possess private information about how hard they are working, and monitoring is costly. In a perfectly competitive world, firing a slacking worker is ineffective, because a worker can immediately acquire a job at the same wages as before. In order to make being fired costly, firms pay wages above what clears the market. This is the best state of affairs; all the workers who have jobs prefer that everyone else is kept working to their utmost, and will accept that they, too, must work. Only the unemployed are worse off. Related to this is the idea that paying above market clearing wages is part of a gift relationship, where simply being paid more than the minimum creates a psychological urge to do better and work harder, as advanced by Akerlof. Alternatively, turnover could be especially costly, and if firms cannot bar workers from quitting, they can pay people more to discourage it.

An example of efficiency wages in action would be Henry Ford’s wage-setting decisions, as detailed in Raff and Summers (1987). In 1908, the Ford Company employed 450 workers, of whom two thirds were skilled or very skilled, and produced 10,607 automobiles. In 1913, it employed 14,000 workers and produced 248,307 cars. This was done by increasing the precision of manufacturing so that all cars could be assembled by unskilled workers doing rote tasks, and achieving a frenetic pace. With this came turnover – an annual turnover of 370%, extraordinarily high. Ford hired 50,448 men in 1913 to maintain his force of 14,000. 

Henry Ford chose to increase wages to five dollars, well above what anyone else paid at the time. In exchange, workers took much more intensive monitoring of their personal character and behavior, with the Ford company having the Sociological Department, a quasi secret police, monitor them, and a much faster working pace.This wage was not merely to compensate for a harder job. Far more people were willing to take the job at that price than there were jobs to be had. The New York Times reported two weeks after its introduction that “Twelve thousand men … celebrated with a rush on the plant which resulted in a riot and the turning of a fire hose on the crowd in weather that was little different from zero. … The crowd began forming around 10 o’clock last night in spite of a blizzard. As a last resort at about 8 o’clock this morning the police got out the water hose. … As soon as the job hunters had dried or changed their clothing they came back.” The particular features of the five dollar day also suggest an efficiency wage motivation. It required a six month probation period, and did not apply to women or workers below the age of 22, consistent with them being less likely to change jobs. It was unlikely to be in an effort to improve the quality of labor, as Ford did not hire workers who came from cities outside Detroit, and the workers were largely unskilled and 3/4ths non-English speaking. Nor was it altruism or political grandstanding on the part of Henry Ford. He owned only 54% of the company, with minor partners who could and did sue him on other occasions for not taking their interests into proper account. As best as we can tell, the higher wages were profitable.

Raff and Summers were, on account of the age of the period, unable to bring modern econometric tools to bear on the problem. Later tests of efficiency wages have found mixed results. One method is seeing the results of a fixed wage in different environments. A 1991 paper by Capelli and Chauvin compared auto factories owned by the same company. Wages were set by the union, and were equal across all locations, which meant that the wage relative to outside options was substantially different. Places with a bigger difference from local wages saw less tardiness, fewer citations for shirking, and a higher standard of effort generally. A similar analysis on supermarkets in California found no such effect, however. A 2013 paper from Parsons looks at manufacturing establishments in four states owned by the same firm, although in a bizarre choice, he does not even tell us the product which is being manufactured. In some papers I don’t necessarily believe, Levine (1991) finds that firms which increase their wages tend to have an increased level of productivity, and argues (inconclusively) that this is not purely due to selection. Wadhwani and Wall (1991) use UK microdata on firms, and argue that rises in the unemployment rate elsewhere should increase productivity. (In both, the primary problem is selection on human capital — lower wage workers lose their jobs first, and we know from Rebitzer (1995) that lower wage workers require more supervision). 

The primary trouble for the efficiency wage story, besides the mixed empirical findings, is that a different contractual arrangement avoids the need for unemployment. There’s no reason why constant wages for constant work needs to be the only way that contracts are arranged. If workers paid a fee in order to be employed, then everyone gets the market clearing wages structured such that there are effective penalties to being fired. Workers could post a bond forfeit upon firing for cause. Edward Lazear suggests that seniority based wages are already a solution to this, in which firms pay above market wages to older workers so that newer workers will be willing to work for less early on. There could be problems of moral hazard with this, of course – it is possible workers would not trust their employers to not seize their bonds, or they do not trust the employer to not fire the old, overpriced workers – but this is not a particularly serious objection. People take actions which aren’t maximally beneficial in the short run all the time. In a world of imperfectly specifiable contracts, everything would end. This is not an assumption which can be relaxed in some cases but not others.

iii. Coordination 

I may have been imprecise with what I mean by the natural rate of unemployment. The natural rate of unemployment is simply that at which inflation would not accelerate. It is an ideal, and may not have changed much. At the same time, the rates of unemployment we observe are definitely lower than they were. This could be because we are getting better at returning to the natural rate after negative shocks, or it could be because monetary policy has gotten better.

Martin Weitzman has a thought provoking paper on increasing returns as the foundation of unemployment. The world of constant returns, in which there can be no unemployment, is quite alien from our own. The first thing to realize is that there would not be, in any meaningful sense, “firms”. If returns are constant then it does not matter how we are organized. If someone is fired, then they just go to produce for themselves, and they exchange with others. Why should it necessarily be the case that capital hires labor, and not the other way around? There will always be enough demand to clear all markets, and we can’t have an economy wide shortfall.

An interesting implication of the increasing returns view of unemployment is that beliefs really matter, and the Federal Reserve does not necessarily reduce unemployment through hydraulic means, but through coordinating people’s expectations. This is particularly obvious during recessions, especially the Great Recession. The Federal Reserve repeatedly cut rates, but it wasn’t enough. The market expected even greater changes. It is possible for there to be no growth in money aggregates at any interest rate the central bank sets, and for there to be unbounded money growth at any interest rate.

Once you allow for increasing returns, then no single company can guarantee adequate demand for their products. Think back to Diamond’s coconut model – any individual would find it unprofitable to throw caution to the wind and produce anyway, because they cannot consume only their output. Cooper and John (1988) tie these papers together, and show that any game with strategic complementarity (that is, where your optimal action increases when someone else increases theirs) will have multiple equilibria, and that most of these equilibria will be inefficient. A focus on increasing returns explains slow recoveries from negative demand shocks, why money need not be totally neutral in the long run, and the Keynesian multiplier follows naturally. Monetary policy can continue being important long after we might think that money would become neutral. We had extremely stagnant recovery after the Great Recession, and that was why unemployment was so long in recovering.

But can that really explain the elevated unemployment of the 1970s? After all, there was high inflation at the same time as above normal unemployment. The conventional explanation was the surge in oil prices. This prompts an enjoyable digression into whether changes in factor prices can have an impact greater than their share of the economy. On the face of it, oil prices leading to unemployment to the degree of the 1970s should be impossible. Hulten’s theorem is that the change in output due to a shock is just the size of the shock times the share of output — so very small estimates indeed. Nevertheless, this is conditional upon some assumptions regarding linearity — some things really aren’t as substitutable as others. With moderate confidence, what I think is going on is that oil and gas were not absorbing the full marginal product at the time. Changes in quantity supplied could have effects beyond what could be expected from cost alone.

Juhn, Murphy, and Topel (1991) believe that the increase was due to stagnating wages for the very lowest, pointing out that almost all of the increase in unemployment was among the lowest deciles, and is explained largely by long durations of unemployment for individuals. Time spent jobless increased by an average of 8 weeks, or by 16 percentage points, among members of the lowest decile, with no change for those at the median skill level. It can’t be plausibly explained by an increase in reallocation, as relative mobility rates were constant over time. Nor is it women in the workforce providing more of a family’s income, as women most went to work among the most skilled households, and least among the least skilled. (Which, incidentally, is consistent with the idea that people have fewer kids because they’d have to give up more in income, not that they can’t afford them). I am suspicious of their explanation, though. The idea that wages have been stagnant is generally based on conversions with CPI, which overstates inflation and therefore understates wage growth. Entirely unmentioned is that that break in trend coincided with the expansion of welfare, which would decrease the necessity for the poorest to go to work, but would not affect the highest earners. 

I favor a compromise view. Part of the unemployment of the 70s was due to idiosyncratic oil shocks, which coexisted with a rise in unemployment due to different preferences for work. I think this may have inadvertently caused the Great Inflation, in a thesis which I want to explore later in greater depth. Policymakers could have easily mistaken the rising unemployment for an understimulated economy, when really it was mostly a secular change.

iv.

So why has equilibrium unemployment fallen so? One cannot tell from theory alone, because some of the models have countervailing effects. The most obvious story for why is that it is now easier to find jobs. The internet has surely made it easier to learn about what jobs are available. However, if the need for preventing shirking through wages predominates, we would expect an increased ease of finding jobs to increase the unemployment rate, as it is now less costly to lose your job. Under the efficiency wage model, unemployment might decrease if, for some reason, the cost of monitoring workers fell. It is conceivably possible (though I think unlikely) for the latter to be wholly responsible for the fall. Other possibilities include a decline in union power making wages more flexible. (I am skeptical of this, as unions tend to make wages more flexible, not less. Think of Weitzman — everyone would prefer if wages went down in the face of an adverse demand shock, but it would not profit the individual to unilaterally do so).

I think the simple story — that making information cheaper leads to better matching — is true. There is ample evidence that the roll-out of the internet improved employment, improved wages, and improved job match. Most of it, alas, has to rely on the relatively weak instrumental variable of geographic rollout. A weak instrument is one where it might be related to the characteristics of the place — a company choosing to build internet infrastructure is not doing so at random. Suppose they build in places they anticipate to grow soon, in which case you have an entirely spurious connection between the internet and growth. Thankfully, not all situations are weak. A classic of the genre, Hjort and Poulsen (2019) use the all-or-nothing nature of the arrival of the internet in African countries to find the effect on employment and income. Either the sea cables are connected and work, or they aren’t. They find strongly positive effects on literally everything good that you would care about, with no downside. Bhuller, Ferraro, Kostel and Vigtor (2023), and an earlier paper by the same authors less Ferraro, use the expansion of the internet to Norway, and again find large positive effects. They estimate that it reduced steady-state unemployment by 14%. Denzer, Schank, and Upward (2021) use the rollout of broadband internet in Germany, and find that it increased job-finding by six percentage points. In a study which does not use geographic rollout, Zuo (2021) used a Comcast program which discounted broadband internet for eligible poor families. Again, it caused a decrease in unemployment, and increase in income, fully consistent with the geographic estimates. 

I believe these sorts of studies much more than things like Kuhn and Skuterud (2004), which uses panel data to find the effect on individuals. They’re unable to exclude the possibility that initial users of the technology are worse on unobservables, nor can they even say that it would increase unemployment if their causal story were true. Perhaps searching through the internet improves job match, and people stay with their job longer. In that case, it is perfectly possible for longer unemployment spells to decrease unemployment in the long run. In any case, I think the early studies showing an increase in unemployment duration are dealt with by Stevenson (2006), who chides them for neglecting people searching for other jobs while they are presently employed.

I will confess that I began this blog post as a simple little thing, explaining the rudiments of efficiency wage theories. Of course, one thing led to the other, and now it has swelled to the substantial project which you see now. I am now skeptical that efficiency wages really matter. With that in mind, let’s move to predictions about the future.

v. The Future of the Labor Market

Economics would be a sterile discipline if we cannot make predictions about the future. Will LLMs make finding a job easier? Will it reduce unemployment? Will it improve welfare? It is here that I will betray my poor mathematical ability — this will be entirely verbal — but I hope to show what parameters are to be identified, and having done so, see what we know thus far.

I assume that LLMs allow applicants to apply for more jobs, but that they don’t inform people of more possibilities. I think this is an accurate description of how people use it; people task it with creating a resume and the like, and may set it upon applying to a bunch of jobs, but are unlikely to add much over our present search tools for finding jobs. People apply to the jobs they are best suited for first, but there is, of course, error, and so applying to more jobs is likely going to increase the quality of job matching. However, applying to more jobs has declining marginal returns, because people aren’t applying randomly. 

We have to think about this in a two-sided way, though. Yes, more applications increases the quality of jobs for applicants, but it also increases the cost of sorting through applications for employers. It may be the case that using AI to sort through applications is cheaper than creating them, but I highly doubt that that is the case. One can simply lie, and an employer could never tell. I expect that companies will begin charging for them to review an application, and that if there is strong social pressure to not charge, then AI may reduce social welfare. Networks will matter more than ever before, increasing inequality and decreasing the quality of job matches.

I want to note that this is not a prediction for the entire economy. The gains from AI automating many tasks are surely much larger than any change in labor market sorting. However, in the presence of a distortion, it is possible for increasing efficiency to in one sector to be harmful, and for decreasing efficiency to be an improvement. Just this month, a paper found that the OPEC oil monopoly had the unexpected upside of reducing climate change.