If only we could apply dynamic scoring to the rest of life

February 19th, 2017 at 9:51 am

“Dynamic scoring” is one of those phrases that sounds way more innocent than it is. It’s the process of guesstimating what impact your budget proposals will have on economic growth, and in turn, revenues flowing into the Treasury.

For example, if your budget includes big tax cuts, as Trump’s will, that’s obvious a revenue loser, which is exactly what the “static” scores show. But with dynamic scoring, you can claim to make back some share of that loss due to the growth effects spun off by your awesome, pro-growth tax-cut plan.

You see the problem. Economic models are dumb, or at least compliant, beasts who will give you whatever answer you want. Put such models in the hands of the purveyors of alternative facts, and the outcome is predictable, as the WSJ reported on Friday and budget nerd extraordinaire Stan Collender takes apart here. Depending on your willingness to torture the model, that “some share of the loss” you can allegedly get back approaches 100%.

This is a serious problem, and I’m not sure what the rest of us can do about it. In normal times, the scoring of the Trump budget by the nonpartisan Congressional Budget Office, which would surely show it to cause deep pools of red ink, would pose at least somewhat of a constraint. But expect team Trump to be closer to the heavenly figure below (h/t: R Kogan, who has this cartoon on his office wall).

Source: New Yorker

In the meanwhile, consider how great it would be if we could use dynamic scoring in the rest of our lives:

Diets: This salted caramel milkshake with extra whip cream has a static calorie score of 800. But when I factor in the efforts expended in 1) taking the paper off the straw 2) drawing the thick shake through the straw (which really is exhausting) 3) stirring in the extra whip cream, the net caloric intake is -60.

Dating apps: “Statically scored, I probably don’t seem that appealing. But once you dynamically account for certain attributes, you’ll want to swipe right. I mean, who else up here is going to regale you with in-the-weeds facts on budget processes? If you’re looking for a pro-growth guy, that’s me!”

Sports outcomes: The static box score had us losing the basketball game 100-40, but once you dynamically model the counterfactual that their 7-foot center played for our team instead of theirs, that score flips and we win.

Elections: Yes, Trump won the electoral college, but he lost the popular vote, and if we dynamically score the possible damage to our fiscal accounts by putting him in charge, especially given the extent to which he will abuse dynamic scoring, he loses. Yes, that logic uses dynamic scoring against dynamic scoring, but what are you gonna do about it?!

A look at a few recent articles that caught our attention: immigration, SNAP, ACA repeal.

February 16th, 2017 at 10:11 am

First, Eduardo Porter of the NYT wrote a controversial piece about the negative impacts of immigration (not Porter’s view–he’s reporting, not endorsing). I’ll have a lot more to say about the research in the piece, but to put it mildly, I’m unconvinced.

The piece reports on research suggesting the increase in low-skill immigration has put downward pressure on productivity growth, by lowering the skill level of the workforce. This immediately triggered my BS meter, as no one really knows what makes productivity growth go up and down. Given the sharp slowdown in this key variable in recent years–which really is a problem–our ignorance enables people to plug in the thing they don’t like as the cause.

Neither does the pattern of immigrant flows make much sense in this regard, at least from 40,000 feet up. As immigrant flows from the south (of persons with relatively low education levels) accelerated in the 1990s, so did productivity growth. As southern flows sharply decelerated, productivity growth slowed as well.

Where immigration increasingly matters in macro terms is around issues of labor supply. Our aging demographics is one reason for slower labor force growth, which in turn is a main factor in slower output growth. Diminished immigration plays a role in that, as this research note I got just this AM from the Goldman Sachs econ team (no link) shows.

Reduced immigration would result in slower labor force growth and therefore slower growth in potential GDP—the economy’s “speed limit”. In addition, academic studies suggest there could be negative knock-on effects on productivity growth. As a result, we see immigration restrictions as an important source of downside risk to our 1.75% estimate of potential growth…

Given that my BS meter is symmetrical, I’m skeptical of this upside productivity claim as well, but the labor-force-growth part of this sounds right.

The other really troublesome bit of Porter’s reporting comes from this bit of what seemed much more like racism than economics to me and to Ben Spielberg, who tweeted:

Like I said, more analysis to come on this. I fully admit a BS meter doesn’t take the place of doing the work. But I also submit that my BS meter rarely fails me.


Next, Chuck Lane objects to a recent piece by Ben S and me on SNAP, or food stamps. We made two arguments. Our main point was that the NYT mis-reported on a study that the paper suggested showed SNAP recipients:

…buy very different, and nutritionally much worse, food than households that don’t use food stamps.

In reality, here’s the study’s headline finding: “There were no major differences in the expenditure patterns of SNAP and non-SNAP households, no matter how the data were categorized.” A related finding — one that reflects an important truth that comes out of the Times piece — was: “Less healthy food items were common purchases for both SNAP and non-SNAP households.

We then argued:

American diets could surely use some improvement. But the improvement mechanism the Times’ reporting discussed — paternalistic bans on the types of food low-income people are allowed to buy with food stamps — is the wrong way to promote healthier eating.

Lane glossed over our first point, though he shouldn’t. Mis-reporting on that study is a big deal and even the NYT public editor got into the mix, supporting the case made by me and Ben. Especially with facts on the run these days, it’s really important to get this sort of thing right.

But Lane disagreed with our paternalism point, and he’s not alone. I’ve long heard the argument that as long as taxpayers are footing part of SNAP recipients’ grocery bill, we should have some say on what’s in their basket.

End of the day, this just comes down to how you feel about paternalistic policies. Lane makes a fair case with which some readers will agree. We strongly object to imposing such rules on one group of people–the poor–because we can, while the same behaviors by more affluent people are out of the reach of policy makers.

But Lane likely goes too far when he suggests that our anti-paternalism will “undermine” food stamps, or that a more paternalistic program would be less vulnerable to cuts. I’d love to see some evidence to support that claim and fear that those who would cut or “block grant” the program would not be moved by restrictions on what recipients could buy.

Finally, I’d urge Lane and others to consider the evidence we show:

Anderson and Butcher analyzed the relationship between SNAP benefits and both food spending and food-related activities. As the figure below shows, they estimate that a $30 boost to SNAP benefits would increase vegetable consumption by about 1.5 percent, increase the time spent on food shopping and preparation by 2.5 and 3.5 percent, respectively, and decrease fast food consumption by about 2.5 percent.

More evidence, less chin-stroking!


Finally, and this deserves more attention than I can give it right now, but some of the gears of “repeal and replace” Obamacare are moving and must be scrutinized.

Today’s papers have articles about President Trump’s executive order on the ACA that is making some changes to IRS rules and to requirements of insurers in the exchanges. My CBPP colleagues Aviva Aron-Dine and Edwin Park take you through some of these changes and their impact on consumers.

The Trump Administration’s new proposed rule on health care would raise premiums, out-of-pocket costs, or both for millions of moderate-income families. If finalized as proposed, the rule would reduce the amount of health care that marketplace plans have to cover. That would allow individual-market insurers to offer plans with higher deductibles and other out-of-pocket costs than they can now sell through the marketplaces. It would also have the hidden impact of reducing the Affordable Care Act’s (ACA) premium tax credits, which help moderate-income marketplace consumers afford health care. As a result, the rule would force millions of families to choose between higher premiums and worse coverage.

I’ve got one point about this, and it’s one that seems fundamental in whatever’s coming in this space of ACA repeal, replace, delay, or repair. It’s my belief, based on focus groups and polling, that there is a large gap between what health consumers, especially those with low or moderate incomes, want from Trump, Ryan, et al, and what they’re likely to get. What’s on offer–high deductibles, less helping paying for coverage, more paperwork, more power to insurers, less comprehensive coverage–isn’t at all what people were looking for when they complained about Obamacare.

More to come…

Slack, inflation, and the Fed

February 15th, 2017 at 12:02 pm

What with Chair Yellen testifying in Congress over the last couple of days, I’ve been trying to dig into the case for forthcoming Fed rate hikes. Clearly, there is more price pressure in the system than in recent months, but there are also these factors to consider:

–The data-driven case for rate hikes is far from a slam dunk, according to evidence I present in the WaPo today.*

–Today’s inflation report is interesting in that it gives ammo to both hawks and doves. Here’s the figure from BLS:

Source: BLS

What’s happening here is that energy costs are normalizing, after being freakishly low for awhile (see figure below). From the perspective of Fed and monetary policy, the key insight here is that energy costs are set on global markets and thus not a reflection of US capacity constraints. That’s why they’ve largely focused on the core, which reveals little acceleration in the above figure. (Dean Baker also pointed out a spike in prices of new cars last month–that’s a monthly anomaly.)

I should also note that the CPI core runs about 0.5 ppts above the PCE core, so think of the Fed’s target for CPI core as ~2.5 percent (as the PCE is their preferred inflation gauge).

Yr/yr energy costs

Source: BLS

So while I understand the case for tightening, and have tried not to get too wound up about small rate hikes, neither do I see a compelling, data-driven case for rate hikes, and that judgement includes this morning’s CPI report.

*For econometric types, there are different ways of running “rolling regressions” of the type I used in the figure in the WaPo piece. Considering that we have T observations (calendar-year quarters in this case), the approach I used estimates the equation using the sample T-n, where n is about half the sample, and then adds back one quarter at time until n is used up. The figure, which plots the coefficient on the slack variable, thus shows how that variable evolves as I add quarterly observations.

But you can also do a “fixed window” approach where you take, say, 20 year samples, for example, and then move the full window up one observation at a time (so, in this case, each regression will have 80 observations). That gives you the picture below (using the same data on core PCE as in the WaPo piece) which tells a similar story re the decreasing sensitivity of inflation to slack.

Source: my analysis (see data note in WaPo piece)

CBO: There’s more slack in the labor market than you thought

February 14th, 2017 at 4:58 pm

[co-authored with Ben Spielberg]

OTE readers know that in the quest for full employment, we pay a lot of attention to the extent of slack in the labor market (and btw, our next podcast episode will focus on this concept of labor market slack—what is it, how much is there, etc. We know…we can hardly wait, too!). One way economists try to gauge this concept of slack is to compare the actual value of labor market indicators to the theoretical value of where they’d be at full employment.

So when a recent CBO revision pretty significantly raised the budget office’s estimate of the potential labor force, we took notice: their revision suggests there’s more slack in the job market than they previously thought.

What is “potential labor force?” It’s CBO’s estimate of what the labor force would be if the economy were at full capacity and the unemployment rate was at its “natural rate” (the lowest rate consistent with stable inflation). In order to project budget outlays and receipts, CBO needs to guesstimate various quantities in the economy, like employment, the labor force, unemployment, and average compensation. They’re often revising such estimates, as they should when new info comes upon the scene, and in their latest revision, they tweaked up the size of the 2016 and 2017 potential labor force by almost one million for this year, or 0.6% of the total. CBO’s incorporation of race (and hence the rising share of the population that is Hispanic) for the first time and increases in educational attainment were two of the revision’s main drivers.

The following figure shows actual and potential labor force participation rates, which we’ll call LFPR and LFPR* (the rates are just the actual and potential labor force as a share of the working-age population). The gap between the two has diminished as the job market has tightened, but it has yet to close.

Source: CBO, our analysis (see data note)

So what does that mean for labor market slack? While there’s no single, reliable indicator that can tell us that, we have found Andy Levin’s total employment gap to be a useful composite of three measures: unemployment, involuntary part-time employment (IPT), and labor force participation. Following Andy’s method, we compare each measure to its estimated potential at full employment.

The unemployment rate—which for many economists is “all she wrote” when it comes to evaluating slack—is just about equal to CBOs “natural rate,” so if that’s all you were looking at, you might think we were at full employment. But IPT is still elevated, and, given CBOs LFPR* upward revision, there’s more room for improvement there, too. The figure below plots Levin’s slack measure before and after the CBO revision, which raised last month’s result from 1 percent to 1.6 percent (an increase of over 800,000 full-time-equivalent-jobs worth of slack).

Source: Andy Levin, our calculations

Caveats abound. Every one of the potential measures has a large confidence interval around it (see here for what we mean re the “natural rate” of unemployment). For years now, economists have been trying to figure out how much of the decline in the LFPR is a function of the aging workforce and how much is a function of insufficient labor demand, and such estimates are always speculative.

But many economists use these data to calibrate their models of the labor force, and we wanted to make sure that everyone was aware of this revision. It suggests that, while we continue making steady progress towards full employment, we’re not there yet.

Data note: CBO does not publish their estimate of LFPR*. They do publish the numerator–LF*–and they use a smoothed version of the BLS 16+ non-institutional population to calculate the rate (smoothing is necessary because the BLS often updates the population weights). Using BLS’s adjustment factors, we created our own smoothed series of the 16+ population, and thus our version of CBO’s LFPR*. Also, we interpolated CBO’s annual labor force estimates to make them monthly.