New CPA Economic Model Answers Sen. Warren’s Request; Moves Modeling Closer to Reality

Economic models of international trade have come in for a lot of criticism in recent years. These models incorporate data on imports, exports, consumption, gross domestic product (GDP) and other variables for over 100 different countries and claim to forecast the results of a trade policy “shock” such as a free trade agreement or a change in tariff rates.

These models were widely used and quoted in the years when the U.S. moved towards ever-freer trade. Free trade agreements have fallen out of favor in the U.S. but elsewhere they are still popular. The European Union’s recent trade deal with four Latin American nations was modeled by European economists, who unsurprisingly found it was a “win-win” agreement for both sides. (French President Macron and French farmers have loudly disagreed, citing damage to French agriculture from cheap Latin American beef, and halted European implementation of the deal for the time being.)

CPA has been a prominent critic of these models. Years ago, we pointed out that these models have been wrong time and again. We have argued that the models have an inbuilt bias towards free trade. This stems largely from the insistence on “equilibrium” in a so-called computer-generated equilibrium model. Equilibrium means that all workers who are made unemployed by rising imports are immediately re-employed in a different industry, at the same or a higher wage. That’s not realistic. Another problem with these models is that they fail to capture the impact of investment flows. Trade agreements often act as starting pistols for multinational companies to offshore production from high-cost regions to low-cost regions and those billion-dollar investment flows are usually excluded from such models.

We showed that models can be done differently in 2019 when we published an exercise based on a modified model which showed that 25% tariffs on all U.S. imports from China could raise U.S. GDP and create over 900,000 new jobs. This paper shocked the economics profession and was honored with an award from the National Association of Business Economics. Since then, we have published other modified models which try to integrate real-world effects into the standard trade model, which is known as GTAP (Global Trade Analysis Project). GTAP was developed at Purdue University and is open source, i.e. any economist can download it and use it. Other economists have also developed variations and modifications which increase the capabilities of the model and take it in new directions. But when governments and politicians use it, they typically use the standard version which will generate “win-win” results to support whatever free trade action they advocate.

In 2021, five Democratic senators led by Sen. Elizabeth Warren published a scathing letter to the U.S. International Trade Commission attacking trade models for “deep flaws” and “an analysis that uses wholly unrealistic models of the U.S. labor market and that systematically ignores the negative effects of trade agreements on workers, particularly workers of color, and on the overall health of the American economy.”

The letter further stated:

“The USITC report also ignores how trade impacts workers differently based on their region, race, and other characteristics…. The effects of trade are not distributed evenly across the country, but instead are geographically concentrated in a small number of places; similarly, the effects of trade are also distributed unevenly by race: among U.S. workers hurt by trade policies, “Black and Latino workers have suffered disproportionate injury.” As a group of Howard University economists recently explained, trade shocks have a `cascading effect . . . on the economic security of Black workers as a result of broader structural disadvantages in the U.S. labor market that many workers of color face.’ ”

The senators were absolutely correct in pointing out the wide variations in how trade (and many other economic policies) impact different segments of the population. We responded to this critique with a new modification of our version of the GTAP model which incorporate a new, granular analysis of the U.S. labor force.

Using Bureau of Labor Statistics data, we broke the U.S. labor force down by occupation, race, gender, and industry. For our analysis we used two genders, four races (Black, White, Hispanic, Asian), 65 industry groups, and five occupations. The end result is a labor force broken down into 2600 different categories.

We used this granular labor force with our previous modified model. The previous model, named GTAP-FP, allows for the flexible response of domestic production to limitations in import such as those from a tariff or quota. We call this new model GTAP-USL, to indicate that the U.S. labor force is analyzed on a granular basis.

We looked at the impact of a hypothetical 20% across-the-board tariff on all U.S. goods imports. The result of such a trade shock would be substantial economic growth. The showed GDP rose by 7.5%, total employment rose by 4.4% or 7.54 million jobs, and household income rose by 14% or $11,700 in 2025 dollars. Drivers of the economic growth were firstly the increase in domestic production as imports fell due to the tariff, and secondly, the multiplier effects as the growth in the domestic manufacturing industries impacted other industries that supplied or benefited from the increased incomes in the manufacturing industries.

Figure 1. Substantial economic growth from a 20% tariff in new GTAP-USL model

The labor force breakdown by industrial sector shows us that manufacturing sectors gained more from the tariff in both employment and wage levels than service sectors. As Table 1 shows, the two sectors that gained most were apparel manufacturing and computer/electronics manufacturing. That’s due to the fact that imports held large shares of each of those markets before the tariff. After the imposition of the tariff, imports were still significant, but the domestic production in those sectors saw double-digit increases in output, employment and wage levels.

Table 1. Substantial job gains and wage increases in manufacturing sectors from the tariff.

Finally the model allows us to look at the differential impact of the tariff shock on different segments of the workforce by race, gender, and occupational sector. Table 2 shows that professional workers (“Office managers and professionals”) tend to best out of the economic expansion. Retail (“Shop and other service workers”) tend to do worse, with wage increases over two percentage points less than the 8.08% achieved by the professional workers.

Table 2. All groups gain from economic expansion but professionals gain more from and retail workers gain less.

Occupational Group	Race	Gender	Wage increase (%)
Office managers & professionals	White	Male	7.17%
Technical workers	White	Male	6.42%
Clerks (office workers)	White	Male	6.32%
Shop (retail) and other service workers	White	Male	5.33%
Agricultural & other low-skilled workers	White	Male	6.45%
Office managers & professionals	Black	Female	6.19%
Technical workers	Black	Female	5.51%
Clerks (office workers)	Black	Female	5.28%
Shop (retail) and other service workers	Black	Female	4.52%
Agricultural & other low-skilled workers	Black	Female	5.36%

The model does not distinguish significantly between Black and White workers at the current level of granularity. The levels of income of the different races and genders come direct from the BLS and are accurate. In other words, the starting values in the database reflect the fact that Whites’ average incomes are higher than Blacks’, and men’s average incomes are higher than women’s. However, to analyze and estimate how an economic shock affects each of those different groups requires finer analysis of changes in employment and wages than the GTAP model currently offers. This will require further development of the model.

Another important addition to the model would be greater regional granularity. Import shocks such as massive job loss following a surge in imports tend to be regionally concentrated. BLS data divides the U.S. into 430 metropolitan statistical areas. The breakdown of the GTAP database into all those areas with accurate information on workforce race, gender, occupation and industry in each one would be a very worthwhile goal.

Conclusion

The development of GTAP-USL marks another step forward in our efforts to make the GTAP trade model more realistic and a better predictor of the real-world effects of trade policies or trade shocks. It’s critical to build models that provide a better understanding of how policies impact people, families, racial groups, gender, cities and regions. There is still more work to be done.