Scott Sumner  

Keep it simple, and pay attention to the key stylized facts

Selgin on Money... The Diction of Social Desirabi...

Rajat recently left the following comment, over at TheMoneyIllusion:

Scott, I know you don't like Twitter, but there has been a few exchanges going on about your argument: One between Kocherlakota and Vaidas Urba. Kocherlakota seems to think that although negative supply shocks would reduce current AS, it would raise future expected growth in AS for some reason (?) and this would help boost current consumption. In a separate exchange, Tony Yates seems to agree with Kocherlakota, saying that "using Eggertsson like arguments", lowering AS now would be stimulative of actual output growth and would not hurt in the long run if you later reversed the harmful policy once the output gap closed. It's all a bit esoteric for me.
Rajat is right that I don't like Twitter; so without responding to any specific individual, let me state my general views on negative supply shocks.

1. The AS/AD model is still our best model of the business cycle. It's never been surpassed. If the AS/AD model suggests that negative supply shocks are contractionary, then that is in fact the most likely outcome.

2. Any contrarian theoretical model should be built to address key stylized facts. The stylized facts suggest that negative supply shocks are contractionary, even at the zero bound. Thus economists should not try to build models where negative supply shocks are expansionary, because we have no persuasive evidence that this is the case.

In my first week of blogging I said:

There were actually five wage shocks, four of which are easily dated. As part of the National Industrial Recovery Act, FDR ordered an across the board 20% hourly wage increase in late July 1933, and then further increases in the spring of 1934. At the same time the workweek was reduced about 20%. The NIRA was declared unconstitutional in 1935, but a minimum wage was instituted in November 1938, and raised a year later. To say the IP data is bad for the Krugman interpretation would be an understatement. These numbers are horrendous:

Table 12.2: Four month (nonannualized!) growth rates for industrial production

Timing of Wage shock - Before -- After
July 1933 wage shock +57.4% -18.8%
May 1934 wage shock +11.9% -15.0%
Nov. 1938 wage shock +15.8% +2.5%
Nov. 1939 wage shock +16.0% -6.5%

You'll notice that I left out the fifth wage shock, but its no better for Krugman's view, just messier. Historians argue that the huge union drives of late 1936 and 1937 were due to both the Wagner Act and FDR's massive election victory. Whatever the cause of the union gains, they led to rapid wage increases in late 1936 and much of 1937. This time, monthly industrial production did not fall immediately, as prices were also rising fast in late 1936 and early 1937. But when prices stopped rising, industrial production began falling sharply under the burden of high wages.

My new book on the Great Depression has a lot of evidence supporting the claim that supply shocks were contractionary during the 1930s. If you prefer more technical studies, Cole and Ohanian have provided them, as have many other researchers. Even Keynes was critical of the NIRA. The evidence was pretty obvious for contemporaneous observers; attempts to artificially raise wages during the 1930s led to lower output. Any model that suggests otherwise should be abandoned.

And it's not just the 1930s. Early in 2014, Paul Krugman suggested that the sudden end of the extended unemployment benefits could provide a test of whether they created jobs or cost jobs. In fact, payroll employment rose by over 3 million in 2014, more than 700,000 higher than in 2013, an already pretty good year. And of course job growth increased in 2013, as compared to 2012, after Krugman's claim that 2013 would be a "test" of the market monetarist claim that monetary stimulus would offset fiscal austerity. So many tests, so many examples of where the more exotic forms of the Keynesian model fall short of market monetarism.

Comments and Sharing

COMMENTS (4 to date)
Gordon writes:

Scott, I can understand your reluctance to using Twitter if you're viewing it as a means of personal interaction. I appreciate the time you take to respond to comments both here and over at The Money Illusion. And I imagine you don't want to have to add responding to people on Twitter to that.

But social media like Twitter now serves as people's information channel. This is why RSS has fallen by the wayside. Content creators and publishers have Twitter and Facebook accounts simply for the purpose of increasing awareness of what they have published. These accounts do not exist for personal interaction. So you could have a Twitter account for the sole purpose of disseminating links to your latest Econlog or Money Illusion post. And both your blogs have tools for sharing links to Twitter. You can make it very clear that you will not respond to any comments on Twitter and that people must use the comments sections on your blogs.

James writes:


What is your basis for saying this: "The AS/AD model is still our best model of the business cycle. It's never been surpassed."

Are you are saying the AS/AD is the best because it gets the signs of some correlations right? That is a low bar and one that many models can satisfy by matching moments on a computer.

Are you saying there is some AS/AD model that results in lower out of sample forecast errors than any alternative? I'm sure that it's possible to generate smaller one period ahead forecasts of GDP, inflation and unemployment with a computer and historical NIPA data.

When I was a student I was told that vector autoregressions were "only good for forecasting" but other models are better because they allow for real understanding of the business cycle. Understanding, in this context, meant attributing unexplained stuff like high unemployment to some other unexplained stuff called "shocks." Is that what AS/AD offers?

Scott Sumner writes:

James, I mean that is the best framework for understanding the stylized facts of the business cycle. It allows us to understand the roles of nominal shocks (AD) and real shocks (AS). Other models like RBC only focus on real shocks. IS-LM does a very poor job of handling interest rates, and also real shocks.

Prediction is not the criterion, as the business cycle is probably unforecastable. It's about understanding what is occurring, or perhaps making conditional predictions based on various shocks.

Scott Sumner writes:

Gordon, Thanks for that info. At some point I might have to do that.

Comments for this entry have been closed
Return to top