The issue: When the congressional budget office "scores" legislation, figuring out how much it will raise or lower tax revenue and spending, it has been using "static" scoring. For example, it assumes that a tax cut has no effect on GDP, even if the whole point of the tax cut is to raise GDP.
This is obviously inaccurate. But, as Greg points out, there is a lot of uncertainty in dynamic scoring.
How much will a tax cut raise GDP, and thus potentially not cost as much in tax revenue? (Tax revenue = tax rate x income, so if income rises a given reduction in tax rate costs less in tax revenue.)
By what mechanism? Keynesians will analyze the issue through a multiplier. The tax rate cut puts money in people's pockets, they spend the money, that raises income, and so forth. Other economists focus on the incentives of a tax cut rather than the income transfer. A tax rate cut can induce people to work, save, invest, go to school, etc. They will come to different answers, especially for policies that emphasize transfers (often with bad incentives) or that emphasize incentives.
How much will policy change growth rates? Long run growth really swamps everything. And the connection between policy and growth is especially hard to nail down.
Greg doesn't really come down on how to solve this issue. I have two suggestions:
1) Embrace uncertainty. It's a fact, we don't know the elasticities, multipliers, and mechanisms that well. So stop pretending. Don't produce only a single number, accurate to three decimals. Instead, present a range of scenarios spanning the range of reasonable uncertainty about responses. The CBO presents a range of fiscal scenarios already.
2) Transparency. Calculations should be utterly transparent and reproducible. If you don't like the labor supply elasticity assumption, you should be able to change the number and produce a new forecast. Scoring should capture "if you think x, then the answer will by y."
Good policy will not result from the illusion of certainty.
Greg also opined on the second round effects, how policy might change economic outcomes which might change future policy. Here I'll side with the old fashioned approach -- let's not go there! The science of forecasting future congressional reactions to events is, let us say, a bit less certain (even) than that of assessing private-sector behavioral responses.
Dynamic scoring requires the solution of a general equilibrium model. To solve a dynamic GE model, you need to specify how the government is going to satisfy its present-value budget constraint. You might be tempted to ask the model what happens if the government cuts taxes and never does anything else. But you won't get very far. The model will tell you that the government has to do something else eventually, and it won't tell you what will happen if the government tries to do something impossible.Greg is right. Though this hasn't bothered CBO scoring yet. Year after year the CBO releases budget forecasts in which debt to GDP ratios climb inexorably; the CBO proclaims this "unsustainable," and life goes on.
Let's try to compromise. A rule that "dynamic scoring models must satisfy a long run restriction in which debt/GDP is no greater than 100%" might work. But one does not have to do huge changes to many models to accomplish that fact. It would be good to have a common benchmark assumption about long run policy so different short run policies can be compared. For example, score all policies in the first 20 years with a common assumption about how debt / GDP at the end of 20 years is resolved.
Where I would rather not go is more detailed political modeling of future congressional actions, especially ones with large distortions.
And for many policies this will not be a huge issue. For example, if we get rid of energy tax boondoggles, one can calculate many interesting behavioral responses, but it is a drop in the bucket of the big social security/medicare/pensions/slow growth debt nexus.