John Taylor likes this paper by Leduc & Liu, which argues that the unemployment rate would be 1.3 percentage points lower right now were it not for policy uncertainty.
He writes:
The following chart from the paper shows that their measure of recruiting intensity has been low while the measure of policy uncertainty has been high.
…and here is the chart:
In other words, employers tried less hard to recruit new workers (the green line) at the same time that policy uncertainty (the red line) went up. So policy uncertainty and a wishy-washy attitude towards hiring went together.
Where did they get that measure of recruiting intensity? From the blue line, basically, which is the Beveridge curve shifter.
The Beveridge curve shifter is a measure of how much the Beveridge curve has shifted — when it moves up, the Beveridge curve has shifted rightward; when it moves down, the Beveridge curve has shifted left.
The Beveridge curve itself relates job openings (on the y-axis) to unemployment (on the x-axis):
During good times (pre-2007), there were lots of job openings and low unemployment; and during bad times (post-2007), there have been very few job openings, and very high unemployment.
The fact that the curve has shifted rightward, since 2007, means that there have actually been quite a lot of job openings since then, and yet we have had high unemployment anyway. This is an Econ Riddle if there ever was one.
That rightward shift in the Beveridge curve — the combination of many job openings and high unemployment — is currently a mystery to the economics profession, as far as I know. There are a few explanations, but there’s no consensus on any of them being correct (if anything it’s the opposite).
One theory is that job applicants just don’t have the skills that employers are looking for. The employers are posting lots and lots of job descriptions, and doing lots of interviews, in search for just the right kind of skills, but the applicants don’t have the skills, can’t do the jobs, and so they don’t get hired.
A second theory is that generous unemployment benefits have coddled the unemployed. Jobless folks are so comfortable on their UI checks that they’ve been lazy in their job searches — maybe they’re only applying for the really good jobs, and passing over the bad ones, and only sending out one application per week.
But as Leduc and Liu point out, the data don’t seem to support either theory. Even highly skilled workers are inexplicably unemployed (including many a recent college graduate), and indeed all industries are suffering high unemployment, which suggests a skills mismatch isn’t the problem; and even though unemployment benefits have been cut back in the last two years, unemployment has not fallen commensurately. In fact unemployment seemed to be falling faster before unemployment benefits were cut.
So Leduc and Liu propose a third explanation, which is that policy uncertainty made firms reluctant to hire new workers.
In the model they’re proposing, policy uncertainty will prevent you from hiring workers if you are a potential employer, but will not prevent you from posting job descriptions and doing interviews:
When uncertainty rises, businesses become more hesitant to hire. They reduce recruiting efforts by raising hiring standards, increasing the number of interviews, or simply not filling vacancies. For instance, some businesses may interview candidates multiple times and end up deciding to postpone hiring altogether (see Rampell 2013).
I don’t really buy that model, even on theoretical grounds. Why, if you are an employer, would you open lots and lots of job openings, and interview a whole bunch of people, and then hire nobody? Surely if you were uncertain about tax policy (or whatever), then you would simply not post any job openings in the first place. Then the curve wouldn’t have shifted rightwards, we’d just be way over on the right side it — i.e., we’d have few job openings and high unemployment, NOT lots and lots of job openings and high unemployment, as we have now.
If you were an employer and you’d already been interviewing multiple candidates, but then you unexpectedly became uncertain about policy, then you might end up hiring nobody. That would cause a rightward shift in the Beveridge curve, right? Lots of interviews, but no hires…
Wrong. Yes, there’d be a shift in the Beveridge curve, but it’d only be temporary. When policy uncertainty spikes upwards, the Beveridge curve would shift rightwards as employers fail to hire the people they’d been interviewing, but then as soon as the policy uncertainty leveled off (or continued growing), employers still might still not be hiring people, but they wouldn’t be interviewing people either, so there’d be no shift in the Beveridge curve beyond a short period in which policy uncertainty suddenly struck during the interview process. But that’s not what we see in the data, nor is it what Leduc and Liu are proposing with their model — instead, the Beveridge curve shifts rightwards, and stays rightwards, which suggests employers are continuing to post job openings and interview people even long after the initial rise in policy uncertainty.
Furthermore, the above story would only happen if policy uncertainty was unexpected – because, if you’d seen it coming, then you wouldn’t have been doing interviews in the first place. The policy uncertainty index is composed of three factors, one of them being the existence of soon-to-be-expired tax cuts. But you’d have known about those since the early 2000s in the case of the Bush tax cuts, and since 2009 in the case of the stimulus. Those would not really have been surprising. The other thing that causes sudden uncertainty is Presidential elections — but surely people saw that one coming…
But never mind the theory; let’s turn to the evidence.
What Leduc and Liu’s graph above shows, is that recruiting intensity has fallen at the same time that policy uncertainty has risen, and their measure of recruiting intensity depends in turn on the Beveridge curve shifter (i.e. the phenomenon that employers are posting job descriptions but not hiring anyone).
Here’s the problem with a causal interpretation that runs from <uncertainty> to <unemployment> of that correlation between rising policy uncertainty and reduced hiring intensity — it’s just that: a correlation. It’s just two or three things happening at the same time — uncertainty, job openings, and unemployment. I’m not saying that uncertainty doesn’t matter, just that this graph on its own tells you very little about whether it does matter. To show that a causal interpretation that runs from <uncertainty> to <unemployment> is correct, you’d need more information than this.
And what information do Leduc and Liu provide to show that this is a causal relationship that runs from uncertainty to unemployment? Basically just some model they made up and wrote down on paper:
So far, we have demonstrated a correlation between heightened policy uncertainty and the outward shift in the Beveridge curve. The next step is to assess how much heightened policy uncertainty may have contributed to this shift. To answer this question, we use a statistical model to explore the relationships between changes in policy uncertainty and the other variables in our model, including the unemployment rate, the job vacancy rate, and our measure of the Beveridge curve shifter. We use our model to estimate the extent to which surprise changes in policy uncertainty produced movements in the Beveridge curve shifter.
In other words, they simply assume into existence a model in which unemployment is a function of policy uncertainty (among other things). There is no actual evidence here (the title of Taylor’s post notwithstanding) — only a model that happens to be not inconsistent with the data.
But there are lots of other, competing models that also happen not to be inconsistent with the data. For example, the uncertainty-unemployment-vacancies correlation is 100% consistent with a story in which 1) we had a huge bubble, a huge bust, and a huge spike in un(der)employment, 2) we had a big fall in nominal spending (no more customers), 3) we had lots of household debt left over from the bubble years, which further depressed nominal spending, 4) we had contractionary fiscal and/or contractionary monetary policy, which depressed nominal spending even more (OK, seriously no customers), and 5) all of the above in turn created lots and lots and lots of policy uncertainty, as the government and businesses and civil society flounder about trying to deal with this mess.
They could have plugged the numbers above into that model, but they didn’t. And there are all sorts of other models you could use.
To be clear, I am not suggesting that that particular model is the correct one to use (nor even that the Leduc & Liu model is incorrect). I’m only saying that all that UncertaintistasTM have going for their argument — on the basis of this one paper anyway — is a simple correlation, and a theory that I don’t think is coherent.
If you want to see the evidence (and theory) that policy uncertainty is positively not to blame — that in fact the causal, uncertainty-to-unemployment interpretation is simply wrong — then you can read Ezra Klein, Mike Konczal, and Mike Konczal again. (That third link is pretty devastating…)