Brad DeLong finds a post on the White House website that makes him believe.
Looks good, doesn't it? Kind of makes me want to believe too. But something's fishy here. As I tell my students, always be conscious of the scale of the chart. And the corollary, always be conscious of the time period chosen.
Thanks to the St. Louis Fed (FRED database), I had a chart with a bit more historical context on my screen in about 30 seconds. I then merged the Edmunds data from the White House chart and produced this:
Two things emerge which are worth pointing out. The first is that the difference between the forecast and the counterfactual is extremely small. Repeat after me, forecasts without confidence intervals are not very useful. The press release from Edmunds.com back in October doesn't report a confidence interval. My take is as follows. Any reasonable confidence interval would have mostly likely encompassed the counterfactual. But since only one side of the confidence interval really mattered to them, they were essentially gambling that the actual number would come in at the lower end of the interval and they would look smart without having to reveal how much guesswork was really involved. Oops!
The other feature of the chart is something that we've known for some time. Namely, that the spike due to Cash for Clunkers did not even make a dent in reversing the cumulative deficit in auto sales that had built up for the last year-and-a-half.
But only when you put those two observations together do you realize just how silly it is for either "side" to claim "victory" in this argument. Look at the chart (either one). Project out the counterfactual and the forecast to continue out on the same trajectory for a few more months. Guess what, the actual data bounce above and below both. That's not too surprising, is it? Auto sales tend to do that. They are quite volatile. This exercise really doesn't tell us much at all about whether the program "succeeded" or "failed." That judgment rests on your definition of "succeeded" and "failed."
Edmunds and others (myself included) argued that the program would simply shift sales from the future. People who might have purchased in the next few months would purchase during the program and this would decrease sales in the months that followed. I still think this is true to a certain extent--and not completely undesirable either. However, because the time horizon over which this will take place is not something that we can know with any certainty, any forecast is going to be imprecise.
Furthermore, the auto sales are very heavily dependent on the state of the economy as we recover. Construction of the counterfactual needs to acknowledge this, and would need to be updated periodically to reflect how the recovery is progressing. Maybe the counterfactual was overly pessimistic. What if the economy turns up in the second half of this year and autos continue to lag? There's a lot more going on here than these charts show.
Let's revisit some of my own favorite statements about Cash for Clunkers and see if they still ring true.
On August 9, I said:
The next day, I said:
And two days after that:
Indeed. And so in closing, on a scale of 1 to 10 with 1 being "causing great damage and remembered in the annals of policy blunders forever" and 10 being "give this man a Nobel prize," I would give CARS a solid 5. Incidentally, this is the same rating that I would give to spitting into the wind or pouring a glass of water into the ocean. All of which reminds me of another great policy debate in which I said:
That will continue to be my rallying cry.
Wow! Cash for Clunkers Worked!! Graph of the Day for April 7, 2010: That surprises me. But Christy Romer and Chris Carroll have a graph:And here is that graph...
Looks good, doesn't it? Kind of makes me want to believe too. But something's fishy here. As I tell my students, always be conscious of the scale of the chart. And the corollary, always be conscious of the time period chosen.
Thanks to the St. Louis Fed (FRED database), I had a chart with a bit more historical context on my screen in about 30 seconds. I then merged the Edmunds data from the White House chart and produced this:
Two things emerge which are worth pointing out. The first is that the difference between the forecast and the counterfactual is extremely small. Repeat after me, forecasts without confidence intervals are not very useful. The press release from Edmunds.com back in October doesn't report a confidence interval. My take is as follows. Any reasonable confidence interval would have mostly likely encompassed the counterfactual. But since only one side of the confidence interval really mattered to them, they were essentially gambling that the actual number would come in at the lower end of the interval and they would look smart without having to reveal how much guesswork was really involved. Oops!
The other feature of the chart is something that we've known for some time. Namely, that the spike due to Cash for Clunkers did not even make a dent in reversing the cumulative deficit in auto sales that had built up for the last year-and-a-half.
But only when you put those two observations together do you realize just how silly it is for either "side" to claim "victory" in this argument. Look at the chart (either one). Project out the counterfactual and the forecast to continue out on the same trajectory for a few more months. Guess what, the actual data bounce above and below both. That's not too surprising, is it? Auto sales tend to do that. They are quite volatile. This exercise really doesn't tell us much at all about whether the program "succeeded" or "failed." That judgment rests on your definition of "succeeded" and "failed."
Edmunds and others (myself included) argued that the program would simply shift sales from the future. People who might have purchased in the next few months would purchase during the program and this would decrease sales in the months that followed. I still think this is true to a certain extent--and not completely undesirable either. However, because the time horizon over which this will take place is not something that we can know with any certainty, any forecast is going to be imprecise.
Furthermore, the auto sales are very heavily dependent on the state of the economy as we recover. Construction of the counterfactual needs to acknowledge this, and would need to be updated periodically to reflect how the recovery is progressing. Maybe the counterfactual was overly pessimistic. What if the economy turns up in the second half of this year and autos continue to lag? There's a lot more going on here than these charts show.
Let's revisit some of my own favorite statements about Cash for Clunkers and see if they still ring true.
On August 9, I said:
As policies go, it's probably better than some ways to spend money and worse than others. But to those who think it is undeniably a net benefit to the economy, then I would ask, why stop at cars? Why not distort the prices of some other things that could be replaced with more energy efficient versions and let the government pick up part of the tab? Why not tear down rodent and asbestos infested old inner-city school buildings and replace them with safer high-tech environmentally sound buildings? Sure it would cost more upfront, but the energy savings and the environmental impact would be enormous. And think of the jobs!I really like that last part. I've got to use that more often.
The next day, I said:
Basically I get really irritated when any politician from any party tries to score points with a policy that is bound to be popular for obvious reasons even though it produces little or no real positive effect. It irritates me that so many people can't see through the smoke and mirrors. It irritates me that politically popular but nearly pointless policies crowd out other policies that are less politically popular but could produce far better results.
And two days after that:
And that means that any meaningful increase in production going forward would have happened anyway. The increased sales from CARS could not possibly explain even a return of production to the levels of a year or two ago. The sales declines have been too large, and the CARS program too small.
But I have to give the politicians credit for setting up the illusion that they made the recovery happen. Sales (and production) will turn up eventually. They have nowhere to go but up. And when they do, CARS will get the credit, just you watch.
Indeed. And so in closing, on a scale of 1 to 10 with 1 being "causing great damage and remembered in the annals of policy blunders forever" and 10 being "give this man a Nobel prize," I would give CARS a solid 5. Incidentally, this is the same rating that I would give to spitting into the wind or pouring a glass of water into the ocean. All of which reminds me of another great policy debate in which I said:
My criteria for good public policy is that it be well out of the neighborhood of "pointless."
That will continue to be my rallying cry.
What you are ignoring is that the objective of cash for clunkers was to eliminate the inventory overhang at the dealers so that firms would resume auto production.
On that basis cash for clunkers achieved exactly what it set out to do.
You are evaluating the program on a completely different basis than it was designed to achieve.
It is like saying the doctor set my broken arm so it would heal properly but he failed because he did not make me look like a movie star.
CARS accounted for about 700,000 unit sales. A typical summer month before the recession would see about 1.4 million unit sales. CARS accounted for 15 days of pre-recession sales. The month after CARS clearly saw a significant drop relative to the counterfactual and the forecast. In the months that followed, sales bounced above and below both the counterfactual and the forecast.
Skeptics claimed that it would shift sales from the future to the present. It did. But to the surprise of many, the largest effect was at only the one month horizon. Of course, since the size of the program was about 15 days with of pre-recession sales, this should not be too surprising.
Edmunds made a mistake, which I call them on in the post. They made a forecast that was statistically almost the same as the counterfactual and hoped that the data would come in on the right side of the prediction. Given the variability of auto sales, that seems like a risky statement to make. It's a great cautionary tale for an applied econometrics course. They made a prediction that, upon closer inspection, was bound to fail.
And so when Edmunds' prediction failed, the White House gets to crow about it. They can say that sales came in above the predicted amount (for a couple of months anyway). The naysayers were wrong! It didn't cannibalize future sales!
Except that's not the whole truth either. When you look at the normal monthly variance in sales, you can see that the data could have been drawn from a stochastic process described by EITHER the counterfactual OR the prediction. And it did cannibalize future sales, but since we're only talking about 15 days worth of sales, the fact that the largest effect (which is undeniable in the data) played out over just a few weeks.
So what?
I have never said that CARS is some sort of horrendous blunder. Just that it's only slightly better than pointless. And I will give it credit for being slightly better. Surely there were some auto workers who were saved from a temporary layoff because of CARS.
But I simply do not think that the program was large enough to be seen as a successful sort of Keynesian "pump-priming" that jump started the industry.
It moved two weeks of sales forward a few weeks. We're not even out of the woods yet, and the effect of CARS has been played out for some time.
And yet, if the economy picks up steam tomorrow, auto sales should jump by multiples of what CARS accomplished. And what will get the credit? You guessed it. People will credit CARS for the recovery, and I don't believe it deserves the credit.
This may surprise you, but I would have supported spending more money on programs that had a larger impact. The real problem I have with this and so many other policies is that (nearly) pointless policies like these politically crowd out better policies. But then, if you read what I wrote, that's pretty much what I said back in August.
You're right about one thing (which I also point out in the post). It all hinges on your definition of success or failure. I just have this feeling that certain politicians are going to attribute more success to this program than it deserves. The error in your analogy is that when the economy improves it is going to "look like a movie star" again. And as much as the doctor who set the broken arm would like to take credit for that, he can't. Unless you've got a really gullible public who doesn't know the difference.