Skip to main content
McMaster University Menu Search
News
The three flags of NAFTA, USA, Mexico and Canada

We do not have good models to predict the impact of protectionism

A recent article by economists Timothy J. Kehoe (University of Minnesota and Federal Reserve Bank of Minneapolis), Pau S. Pujolas (McMaster University), Jack Rossbach (Georgetown University-Qatar) concludes that there is no good reason to believe that existing trade models can help shed light on the impact that a renegotiation (or even dismantling) of NAFTA can have on Canadian industries.

Oct 29, 2018

The following is a revised and translated version of an article that first appeared in Foco Económico.

Donald Trump’s term as President of the United States has prompted many countries to worry about the possible consequences of new trade barriers.  Because it could not be otherwise, the North American Free Trade Agreement (NAFTA) makes Canada (and Mexico) one of those countries.  Being worried about new trade barriers is normal: within the economics profession there is a strong consensus, supported by a vast array of empirical evidence, that a trade liberalization, while possibly entailing some costs, generates more gains than losses.  Hence, it is reasonable to conclude that doing the opposite will generate more losses than gains.  

Unfortunately, the bad news continues.  As we show in a recent article (Kehoe, Pujolas, and Rossbach, 2017), we currently do not have good models to accurately predict the impact of a trade liberalization.  In particular, predictions fail miserably when we try to assess how liberalization affects different industries.  Thus, we have no good reason to believe that existing trade models can help shed light on the impact that a renegotiation (or even dismantling) of NAFTA can have on Canadian industries.

Starting in the 1970s up until the mid-1990s, many economists, pioneering the use of computers to develop economic theory, started building applied general equilibrium (AGE) models and using them to predict the impact that certain policies would have on the economy.  When AGE models are used to analyze the impact of fiscal policy, they work quite well.  Unfortunately, when they are used to analyze the impact of trade policy, they perform quite poorly.

The key question, then, is why do AGE models perform so poorly when used to analyze international trade? A key observation is that international trade models, with one notable exception (Arkolakis, 2010), do not accurately capture the idea that  least traded products (LTP) grow much more than more heavily traded products after a trade liberalization (Kehoe and Ruhl, 2013). AGE models, when used to analyze issues regarding international trade, are based on the economic theory of international trade. Unfortunately, the LTP margin does not appear even in the most up-to-date models of international trade (e.g., Eaton and Kortum, 2002; Melitz, 2003).  In these models, a trade liberalization makes some firms move from a position of not exporting to a position of beginning to export.  In other words, in these models it is not the LTP that grow but the nontraded products.  Older trade models, based on love-for-variety preferences, assume that all firms are the same and, in fact, the number of varieties falls — not increases — after a trade liberalization.  Even older models emphasize comparative advantage as the engine of trade, which means that they predict that trade liberalization increases the trade of already traded products.  Hence, AGE models cannot accurately predict the impact of a trade liberalization because the models they are based on are unable to capture the role that LTP play.

In this article, we conduct a very simple exercise to illustrate why AGE models perform poorly when used to predict which industries grow and which ones fall after a trade liberalization.  First, we compute the predictive power of LTP at explaining changes at the industry level (following a methodology similar to the one in Kehoe,  Rossbach, and Ruhl, 2015): we order all products from least to most traded and add up all those that are least traded until we get 10 percent of the total value of trade. For each industry, we look at the share of trade accounted for by LTP.  Then we show that the correlation between this share and the observed change in the data is quite high.  This result suggests that a model that uses LTP information to predict the impact of trade policy will be successful.

After that exercise, we evaluate how well AGE models perform when predicting the impact of trade policy.  To this end, we use the Global Trade Analysis Project model (GTAP; see Hertel, 2013) and the model developed by Caliendo and Parro (2015, henceforth CP).  GTAP covers 140 countries and 57 industries for 2005, 2007, and 2011, and is the framework used by many countries (including Canada) and international organizations (such as the World Trade Organization and the World Bank) to assess the impact of a trade liberalization.  We evaluate the predictive ability of GTAP for the trade liberalizations of Australia and the United States (2005), Chile and the United States (2004), Chile and China (2006), and China and New Zealand (2008).  CP is the absolute best version of an AGE model using state-of-the-art models of trade (this one is based on Eaton and Kortum, 2002) and includes a detailed input-output structure that connects all the sectors of all countries involved.  It is a model designed for NAFTA, so we use it to evaluate how well it predicts the changes in Canadian, American, and Mexican industries when NAFTA was enacted. We then compute the correlation between predictions of GTAP (table 1) and CP (table 2) and the actual changes, and compare them to the correlation between the percentage of LTP and the actual changes.

Table 1: GTAP model

versus Least Traded Products methodology comparison

Exporter a

Importer

Correlation

GTAP and data

Correlation LTP and data

United States

Australia

0.27

0.55

Australia

United States

−0.14

0.53

United States

Chile

0.08

0.55

Chile

United States

0.03

0.48

China

Chile

0.14

0.61

Chile

China

0.04

0.07

China

New Zealand

−0.36

0.61

New Zealand

China

−0.09

0.48

Simple average

0.00

0.49

a Excluding outliers in the data: petroleum exports from United States to Chile and beef exports from Australia to United States.

Table 2: Caliendo and Parro model (2015)

versus Least Traded Products methodology comparison

 

Exporter

Importer

Correlation CP and data

Correlation LTP and data

Canada

Mexico

−0.46

0.27

Canada

United States

0.36

0.19

Mexico

Canada

−0.68

0.83

Mexico

United States

−0.17

0.33

United States

Canada

0.35

0.28

United States

Mexico

0.54

0.16

Simple average

−0.01

0.33

 

It is clear that LTP information correlates better than GTAP predictions for all trade liberalizations.  In fact, GTAP predictions are quite dire: on average, the correlation between data and predictions is zero; if predictions were done at random, the correlation would be similar. By comparison, the correlation using LTP is substantially larger at 0.49.  Similarly, LTP information correlates better than CP predictions, with two important characteristics: the correlation is always positive for LTP, and CP’s predictions negatively correlate with the data for half of the six exporter-importer pairs. Not surprisingly, then, the average of CP correlations is close to zero (and, in fact, is negative). By contrast, for LTP, the correlation is higher, at 0.33.

From this exercise, we learn two things: first, standard models are not good predictors of the impact that a trade liberalization has on the industries of a country, even when using state-of-the-art models.  Second, making good predictions in the near future will be possible as long as LTP information is used.  Our current work is precisely to develop a model that we can use to make these predictions.

To conclude, we would like to share our personal views on this topic.  Since we have virtually no data on what happens after trade barriers increase, as researchers, we think that what President Trump is willing to do has some interesting value from a scientific viewpoint.  As citizens of the world, however, we think it is quite depressing.