The Little-Known Programming Problem Inside Dynamic ScoringBy
Dynamic scoring, as adopted on Jan. 6 by the House of Representatives, seems like the ultimate no-brainer. Reasonably enough, it requires the House to take long-run macroeconomic effects into consideration when deciding how to vote on tax and spending bills. Dynamic scoring makes tax cuts look better from a budgetary standpoint because it assumes they will stimulate economic activity. You don’t have to be an acolyte of Arthur Laffer to believe that lower income tax rates will cause people to work at least a little harder and pay a little more to the Internal Revenue Service, helping to offset the revenue lost from the tax cut.
The problem with dynamic scoring isn’t political. It’s actually a problem of computer programming that’s well known to experts in the field, Republicans and Democrats alike, but hard to convey to the general public. It has to do with something that economists call the “transversality condition”—easier to remember if you think of it as the kerflooey problem.
Here’s the idea: Certain computer models that are used to predict the impact of tax and spending bills won’t produce an answer—any answer—unless they’re allowed to assume long-run stability in the budget outlook. To run, the programs must assume that federal debt as a share of gross domestic product flattens out at some point. The transversality condition is the constraint in the program that nothing can be projected to go to infinity—i.e., go kerflooey. If debt’s share of GDP is projected to keep growing, the program won’t “solve.” It will spit out some modern-day equivalent of “It does not compute,” or maybe, “Danger, Will Robinson.”
The problem is that under current projections, the federal debt as a share of GDP is indeed headed toward infinity. Check out this chart from the Congressional Budget Office, which assumes current laws stay on the books. Notice that it’s rising all the way out to 2089, with no end in sight.
It’s obvious that in reality the line will level off at some point, probably well before 2089. (I will be dead in 2089, so don’t come knocking if I’m wrong.) After all, even the man-eating plant in The Little Shop of Horrors didn’t keep growing forever. The only real question is how stability will be reached: Either spending’s share of GDP will shrink, or revenue’s share will increase, or some combination of the two.
That’s where dynamic scoring comes in. The scorers are forced to exercise judgment and insert some assumptions about what will happen in the future to stabilize federal debt as a share of GDP. Let’s say they’re modeling the impact of a tax cut. They could choose to assume that taxes’ share of GDP will rise in the future. Or they could assume spending’s share will fall. Whichever choice they make will have a huge impact on the score they give to the tax-cut bill. Imagine that they choose to assume future tax hikes. The headline could be: “Tax-Cut Bill Will Shrink Budget Deficit Assuming Tax Rates Rise.”
These complexities don’t arise today because the Congressional Budget Office and the Joint Committee on Taxation use only “static scoring,” which sticks to microeconomic effects such as the impacts on what crops farmers will grow, how much medical care will be provided, and the uptake of various government programs. To quote the CBO: “CBO’s cost estimates generally do not reflect changes in behavior that would affect total output in the economy, such as any changes in the labor supply or private investment resulting from changes in fiscal policy.”
The difficulty of carrying out dynamic scoring isn’t just a talking point of liberal Democrats who are looking for an excuse to oppose tax cuts. It was explained by Douglas Holtz-Eakin, a former director of the Congressional Budget Office who advised Arizona’s Republican Senator John McCain on his 2008 presidential campaign. Here’s Holtz-Eakin testifying in 2011 to the House Ways and Means Committee:
“At the moment, we have two difficulties in this modeling. I want to emphasize this. The first is actually quite remarkable, and that is for some of these models, particularly for ones where there is tremendous foresight, those models simply cannot be calculated, meaning the computer algorithms will not run if the Federal Government’s budget is on an unsustainable trajectory. Our Federal budget is on an unsustainable trajectory. So in order to actually do the analysis you have to make some assumption about how to get the debt stabilized relative to GDP, and that is even before you can do the analysis of the tax reform. The second piece is that when you do the tax reform analysis you have to have a regular and predictable offset for any budgetary gains or losses. Will it be spending cuts? Will it be tax increases out further in the future? The economy will react very differently depending upon how you do it. You have to decide upon a set of procedures which may seem arbitrary but which allow you to do the business on a regular fashion.”
Case in point: Last year the Joint Committee on Taxation, which scores tax bills for both houses of Congress, gave an unofficial dynamic score to tax legislation pushed by Representative Dave Camp (R-Mich.), who was chairman of the House Ways and Means Committee. (Camp has since retired from the House.) Eight projections using different assumptions and models of the economy produced wildly different results.
The one that was most favorable to Camp projected that the U.S. economy would be 1.6 percent larger in 2023 because of the tax plan than otherwise. The least favorable projected a gain of just 0.1 percent. (The scorers at the Joint Committee on Taxation chose to assume that Congress would bring debt under control over the long term by reducing the growth rate of transfer payments, not by raising taxes.)
Interestingly for the debate over dynamic scoring, the model projecting better results from the Camp plan was what economists call “forward-looking”—namely, it assumed that Americans have perfect foresight about the future of the economy and policy. It’s this type of model that’s vulnerable to the transversality condition/kerflooey problem. Another kind of model, called the macroeconomic equilibrium growth model, isn’t vulnerable to going kerflooey. Then again, that’s the kind that showed less upside from the Republican tax plan.