Monday, June 15, 2009

Not all forecasters are alike 

I don't record news programs (except for Special Report, to feed my Krauthammer addiction) but between Twitter and friends I know some things about what happens on the Sunday shows that church keeps me from. �Apparently my favorite VPOTUS was on the air yesterday, and Mark Halperin provides some key quotes.
"No one realized how bad the economy was. The projections, in fact, turned out to be worse. But we took the mainstream model as to what we thought -- and everyone else thought -- the unemployment rate would be."
"Everyone guessed wrong at the time the estimate was made about what the state of the economy was at the moment this was passed."
Two things about that: First, we were forecasting in a period where there was little good past experience to work form. Almost all forecasting involves taking a set of data that you think represents the state of the economy, looking at the most recent experience and finding "comparable" periods of history where the same conditions essentially occurred. Now of course the data is never identical -- we don't live in test tubes. I refer to macroeconomic data as "in the wild", impure, contaminated by hundreds of other influences that we never measure. You assume the stuff you're not measuring isn't important. Good emphasis on this comes from Ed Leamer, whose book "Macroeconomic Patterns and Stories" includes this paragraph:
You may want to substitute the more familiar scientific words �theory and evidence� for �patterns and stories.� Do not do that. With the phrase �theory and evidence� come hidden stow-away after-the-fact myths about how we learn and how much we can learn. The words �theory and evidence� suggest an incessant march toward a level of scientific certitude that cannot be attained in the study of the com plex self-organizing human system that we call the economy. The words �patterns and stories� much more accurately convey our level of knowledge, now, and in the future as well. It is literature, not science.
This was the essence of what I was trying to say Thursday. We try to express that lack of certitude by providing a confidence interval, which when done properly makes most of the so-called "created or saved" jobs into something that could well be just noise. �But even confidence intervals give the forecast a patina of scientism that has no business in economic forecasting. �I'm not confident of confidence intervals (in short, here's a one sentence explanation: I have to have some degree of knowledge of the shape of the distribution of the errors I make in forecasting ... and I question whether or not I can possibly know that.) �When Biden also said�""The bottom line is that jobs are being created that would not have been there before" or that it "clearly has had an impact", we have to respond that we don't know that.

Second, it's foolish to think nobody knew the recession would get as bad as this. �Think about the WSJ Board of Economists forecast, for instance. �The June unemployment rate median forecast from there in Feb 2009 (pre-stimulus) was 8.3%, then 8.7% in March (taken about two weeks after signing of ARRA) and 9% in April. �The forecast has been catching up to reality -- as is true in just about every recession. �In February, at least six forecasters had unemployment forecasts at 8.9% or 9%. �In March, David Rosenberg of BankAmerica (since moved to Canada) was at 9.3%. �Many had forecasts of unemployment in February at or about 10% for the December 2009 report.

All of this from the Obama Administration has been to provide some justification for the size of the stimulus package, which has scared market participants into fears of inflation via monetization. �I don't think any of the tacks taken by the team so far have gained traction.

Labels: , , ,

Thursday, June 11, 2009

Admitting you don't know 

I may be a member of a small group, but I find Joe Biden endearing. During a conference call with media Monday he expressed his own confusion over the "saved or created" meme and said "I'm sorry, I'm not an economist."

During the call, Biden said that the stimulus package, which was signed into law in February, saved or created 150,000 in its first 100 days, and he outlined White House plans to accelerate the pace so that 600,000 more jobs will be saved or created in the second 100 days this summer.

Asked by Jonathan Riskind of the Columbus Dispatch to explain how the White House determined these numbers, Biden said that the Council of Economic Advisers makes its estimates based on measuring what the U.S. employment level would have been without the stimulus, and then comparing it to the nation�s actual employment level.

�I�m sorry I�m not an economist,� Biden said as he was describing the methodology. �My background is foreign policy and the constitution. �

The White House estimates also consider the number of jobs needed to complete certain projects funded by the stimulus, and take into account the �spinoff effects� of spending programs. For instance, tax credits and subsidies for weatherization of homes and wind farms boost contracts for businesses that may not be receiving stimulus money directly, the vice president, who is spearheading the implementation of the program, said.

�I�m a little above my pay grade here as I try to explain in more detail how they count spinoff effects of actual jobs created, � Biden said.

�It�s complicated,� he acknowledged. �But the fact is that there has been no challenge to the methodology the Council of Economic Advisers has come up with, known to national economists as being reasonable to the estimates we have as to the actual jobs saved or created.�

I don't blame the Vice President for being too busy to read this humble blog, but I think I already did that. The calculation is from CEA chair Prof. Christina Romer's testimony:
I have been told by the Office of Management and Budget that approximately $75 billion in spending under the ARRA has been obligated and almost $14 billion in outlays have already occurred. During the first 100 days in office, which the Administration marked yesterday, we estimate that the ARRA has already saved or created 150,000 jobs.
That comes out to $93,333.33 in outlays per job created. Since the Administration has since spent another $32 billion (and obligated another $71 billion), why are they stuck on 150,000 jobs saved or created? If they believed the methodology Prof. Romer testified in April, the number should be significantly higher. This is the nature of Ed Morrissey's first paragraph here this morning. (I think, unlike the Vice President, Ed reads here occassionally.)

Let's then tie this to the unemployment rate: How can they claim this "created or saved" figure as the unemployment rate continues to rise as the new "hockey stick" graph says? People making that tie in either direction -- to support or refute the claims of the Obama Administration -- should treat the data with care. Unemployment rates come from a household survey which first has to establish who is in the labor force and who is not. You could be creating jobs and have more people unemployed (and a higher rate) because that nascent job growth got a lot of people off the couch and into job search. For example, last month the payroll survey says 345,000 fewer people were on U.S. payrolls, but the household survey says 787,000 more people were looking for work. Some of that is differences in surveys, but part of that is the household survey's finding that 350,000 more people were looking for work in May than April. Make that last number zero, and the unemployment rate only goes up to 9.1% rather than 9.4%. You could claim that last 0.3% is the 'hope' that stimulus created.

Joe's right: It's complicated.

It's too simple therefore to go from employment levels or employment rates to unemployment rates. The mistake with the new hockey stick graph is one made by Romer and Biden economic advisor Jared Bernstein back during the transition. The "saved or created" claim is a different claim. (Their report is the one with the "spinoff jobs" that Biden talks about.)

On the other side, though, 150,000 in a total employment pool of 140,000,000 is barely more than 0.1%, which is the relative sampling error of the employment survey. (See technical notes.) I don�t see how the Administration could prove in a statistical sense that job levels are higher than where they�d be in the absence of the stimulus. If you know a little statistics, what I�m saying is that the confidence interval � margin of error, when done with polling data you may know better � is too wide to say we know with 90% or 95% confidence that there are more jobs. Any job level they claimed would have happened without the stimulus so far could be random variation or sampling error to be corrected when they benchmark the employment rate next year.

I say again what I've been telling anyone who'll listen for the last two months: You can't say the stimulus has worked, and you can't say it hasn't. You don't know yet. It won't take as long as the French Revolution to figure out if it worked, but I think it's better for supporters and critics of the plan to admit they don't know than to start firing off their rhetorical guns wildly.

Take the summer off and get back to me in October on this. We may be able to put a reasonable guess on ARRA's efficacy by then, but not before. You're better off worrying about the bills to come than the ones that are behind us.

Labels: , ,