« Tim Duy's Fed Watch: Soft Jobs Report Likely Welcomed by Fed | Main | Tim Duy's Fed Watch: Neutral or Not? »

Monday, May 07, 2007

Rodrik: The Globalization Numbers Game

Dani Rodrik responds to my challenge. This is how it's done:

The globalization numbers game, by Dani Rodrik: Many many years ago ..., I heard Gary Hufbauer tell an anecdote at a conference on international trade. A government economist is called in by his superior, who tells him "Look, I have to make a case for this policy in front of Congress, and I need a number real bad." The economist responds, "well, I haven't done a proper analysis, so I can give you a real bad number." Perhaps it was a true story based on Gary's own government experience.

I am reminded of this story by Mark Thoma's post, which focuses on the magnitude of the gains from globalization. He says "there's something important that's generally missing from the attacks on globalization's supporters, actual evidence." He refers to a Bernanke speech and at length to a paper which Bernanke cites by Bradford, Grieco, and Hufbauer (yes, the same Hufbauer). The Bradford et al. study argues that removing all remaining barriers to trade would raise U.S. incomes anywhere from $4,000 to $12,000 per household (or 3.4-10.1% of GDP). That is a whole chunk of change! Thoma writes:

Whatever the theory says, the evidence in this paper and the evidence more generally is pretty clear, globalization has large net benefits....

If you disagree that trade benefits the US overall, what are the problems with the econometric methodology used to produce these results or the results in other papers coming to similar conclusions? Saying the results must be wrong because they don't support your point is not an argument. What specifically in the data, estimation procedures, etc., do you think is problematic and leads to the wrong result? Are there other notable academic papers that come to different conclusions? If so, what is the source of the difference in the estimates? Is it the data, the estimation technique, the theoretical assumptions, or what? Help us to understand why we should be doubtful about the results Bernanke cited, or about the results of other papers reaching similar conclusions.

As Thoma says, we cannot dismiss "evidence" just because we disagree with it. ...[W]hile I would not quarrel with the assertion that globalization increases the size of the pie for the U.S., I do have a big quarrel with the kind of numbers presented by Hufbauer and company. They seem to me to be grossly inflated. Let me take up Thoma's challenge and explain why.

First a reality check. The standard partial equilibrium formula for calculating the gains from moving to free trade is 0.5 x [t/(1+t)]^2 x m x e, where t is the tariff rate, m is the share of imports in GDP, and e is the (absolute value of the price elasticity of import demand). In the U.S. average tariffs stand in low single digits and imports are less than 20% of GDP. There is no way of tweaking this formula under reasonable elasticities that would get us a number anywhere near the Bradford et al. estimates. For example, using the generous numbers t = 0.10, m = 0.2 and e = 3, the gains from moving to complete free trade are a meager 0.25% of GDP (compared to Bradford et al.'s lowest estimate of 3.4% of GDP).

Now of course this is a back-of-the envelope calculation, and in particular it ignores general-equilibrium interactions--both across sectors within the U.S. and among countries. What if we take those into account? There is a cottage industry of computable general-equilibrium models which attempt to do just that. ... The most accomplished work along these lines takes place at the World Bank and at Purdue. A representative recent estimate comes in a paper by Anderson, Martin, and van der Mensbrugghe--hardly a bunch of rabid anti-globalizers. Their bottom line for the U.S.: full liberalization of global merchandise trade would eventually increase U.S. income by 0.1% by 2015 (see their Table 2). No, you did not read that wrong. The gain amounts to one-tenth of one percent of GDP! (And the bulk of it comes not from the liberalization in the U.S., but from the terms-of-trade effects of other countries' liberalization.) Is this number right? I have no clue. But at least it passes the test that it is consistent with first principles.

So how come Bradford et al. get wildly different numbers? Some of the difference is due to Bradford et al.'s attempt to take into account liberalization in services as well as in merchandise trade. But that is only a small part of the story. The real action is in the non-standard methodological choices made by Bradford et al.--choices which are designed to generate large numbers.

Bradford et al. report three different methods of arriving at their estimates. One is based on a University of Michigan computable general-equilibrium model with increasing returns. As I have already discussed (see point 5 in this post), increasing returns greatly enlarge the range of possibilities from trade opening. You can easily magnify the gains from trade, as well as produce losses, depending on how you build your model. The model Bradford et al. rely on is designed to do the former (surprise, surprise!). The results are entirely model-driven, and it is hard to see how they could count as "evidence."

The second approach is based on estimating the gains from price convergence. The basic idea is that removing remaining restrictions on trade in goods and services should equalize prices at home with those abroad (making due allowance for transport costs). The difficulty with this is that the bulk of these price discrepancies today arise not from trade restrictions per se, but from jurisdictional discontinuities that arise from differences in legal regimes, regulatory systems, and currency arrangements across countries. If you want to believe that globalization will yield full price convergence, you must also believe that it is practical and desirable for the U.S. to harmonize its laws and regulations with those of its trading partners and to join a currency union with them. (To imagine what this might look like, you may want to think of the contortions that the EU is going through--with its common currency, European Court of Justice, 80,000-plus pages of regulations, and huge inter-regional transfers--in order to achieve a "single market" on a much smaller scale and among an already like-minded set of countries.)

The final calculation is based on an econometric estimate by Andy Rose of how much a regional trade agreement raises the volume of trade. At least that's what Bradford et al. say. When I checked out Rose's paper, I could not find the number that Bradford et al. use. (Rose says "belonging to a regional trade agreement raises bilateral trade by (exp(1.17)-1»)222%," whereas Bradford et al. use 118%, without explanation). In any case, Rose explicitly discounts his own estimate (pp. 9-10), saying that it is too large to be credible and too large by the standards of other findings in the literature.

I don't really mean to pick on Bradford et al. (although as a particularly egregious example, their paper fully deserves it). But as Thoma rightly says, we can move the discussion forward only when we present specific critiques of the work out there that disagrees with our own conclusions. So consider this an exercise in that vein.

What puzzles me is not that papers of this kind exist, but that there are so many professional economists who are willing to buy into them without the critical scrutiny we readily deploy when we confront globalization's critics. It should have taken Ben Bernanke no longer than a few minutes to see through Bradford et al. and to understand that it is a crude piece of advocacy rather than serious analysis. I bet he would not have assigned it to his students at Princeton. Why are we so ready to lower our standards when we think it is in the service of a good cause?

Come to think of it, did I not write about this earlier?

Paul Krugman wrote about this last point too, Repeating from an earlier post:

As for the whole Noble Lie theory of selling free trade with happy stories, this has a tendency to backfire. Case in point: there was and is a good argument for NAFTA, but the way supporters made the case back in 1992-3 - highly dubious arguments that the US trade surplus with Mexico would rise, creating jobs - was disreputable (I said so at the time) and quickly became a sick joke after the 1995 peso crisis. I actually covered that debate in my trade policy class last week, and felt ashamed all over again.

I can't say whether Bradford et al. intentionally made "non-standard methodological choices ... designed to generate large numbers" in order to sell their case rather than the choices they believe give us the best answer to the question. They will have to speak to that themselves. But let me emphasize that we are not debating whether benefits from trade exist for the U.S., but rather about the size of the gains.

Understanding why different empirical approaches to estimating the gains from trade arrive at different conclusions is important in moving this debate forward, and thanks to Dani for taking us a long way in that direction. For now, I am going to leave it at that, but I hope to have more to say about this topic soon.

    Posted by on Monday, May 7, 2007 at 11:10 AM in Academic Papers, Economics, International Trade | Permalink  TrackBack (1)  Comments (35)


    TrackBack URL for this entry:

    Listed below are links to weblogs that reference Rodrik: The Globalization Numbers Game:

    » Cui Bono? from Political Animal

    CUI BONO?....According to Dani Rodrik, a study by Bradford, Grieco, and Hufbauer suggests that removing all remaining barriers to trade would raise U.S. incomes anywhere from $4,000 to $12,000 per household. Wow! Let's do it! I could use the extra... [Read More]

    Tracked on Tuesday, May 08, 2007 at 12:28 PM


    Feed You can follow this conversation by subscribing to the comment feed for this post.