Using the GTAP Model to Simulate Second Quarter Real GDP
According to the advance GDP release (4/29/20), the U.S. economy contracted by 4.8 percent during the first quarter of 2020. Though not as bad as the 8.4 percent quarterly decline that occurred during the Great Recession, performance in the second quarter could be even worse. Forecasts available on the subscription service Haver Analytics were predicting second quarter declines ranging from 5.9 percent to 10.7 percent.
This blog presents a rough estimate of U.S. real GDP for 2020 Q2 using the Global Trade Analysis Project (“GTAP”) model and database.[1] The GTAP model is a computable general equilibrium model widely used in the analysis of trade policies such as free trade agreements and tariff increases. The GTAP model is operationalized using the GTAP database, which contains data on trade, production, protection, and other economic variables for 65 sectors in 141 countries and regional groupings. You should visit the website for a more thorough explanation of the model, database, and additional applications than the gross simplifications I provide here.
Although the model’s purpose is trade policy analysis, it is possible to simulate the effects of other types of economic changes, such as an increase in national labor productivity caused by improving human capital, an expansion of the labor force, or a reduction in the demand for oil. The flexibility of the model got me thinking it might be interesting to use the model to estimate the economic effects of COVID-19. There are many ways to do so in the GTAP framework. One could work from the bottom up, reducing employment in various industries and/or taking into account reduced productivity caused by remote working or lost economies of scale. Here, I take a simple top down approach by reducing the quantity of labor supplied. I aggregate the database into 11 regions, treating the United States as a separate region, and 11 commodity sectors. I then reduce labor supply in each region by 12 percent.
This approach, to be charitable, is simplistic. The job losses resulting from COVID-19 will vary significantly by country and no one knows precisely what they will be. However, the GTAP suite of programs allows for systematic sensitivity analysis (“SSA”), which varies the size of the employment shock in each country over multiple simulations. Based on these simulations, the SSA module calculates mean results and standard deviations for each variable. I used SSA to impose labor reductions ranging from 6 percent to 18 percent independently across all regions and to generate ranged outcomes for real GDP, my main variable of interest.
The mean outcome for the U.S. economy is an 8.6 percent reduction in GDP. My confidence interval is + 4 standard deviations from the mean to be conservative, which yields a range of minus 14.4 percent to minus 2.8 percent, consistent with the recent estimates I found on Haver.
One could pore through results for dozens of variables for each region and glean insights into how the reductions in employment may affect the U.S. and global economies. As shown in the table below, the contraction of U.S. GDP is predicted to be larger than those of other regions. This makes sense because U.S labor productivity is higher than labor productivity of the other regions/country groupings.
Another interesting variable is the price of oil. Several different price variables show significant price effects for energy products. For example, the supply price for natural resources, which includes oil and coal, declines by 26.8 percent to 33.1 percent in the United States. Thus, even without considering the price war between Russia and Saudi Arabia, the model predicts that the large decline in global economic activity would have an out-sized impact on energy prices.
Table 1. Simulated changes in 2020:Q2 Real GDP, by Region
Source: Author’s calculations, GTAP 10 Data Base and RunGTAP 3.70.
These simulations are not meant to be precise, but rather to compare results from the GTAP model with existing forecasts, and to show how the SSA module can be an efficient tool for dealing with uncertainties. In this case, the uncertainties relate to the level of job loss that will occur across the various regions. In other cases involving tariff reductions, the main uncertainties frequently relate to key parameters, such as the elasticity of substitution. Either way both the modeler and the client are better served when ranges are presented using SSA.
[1] AGUIAR, Angel et al. The GTAP Data Base: Version 10. Journal of Global Economic Analysis, [S.l.], v. 4, n. 1, p. 1-27, june 2019. ISSN 2377-2999. Available at: https://www.jgea.org/resources/jgea/ojs/index.php/jgea/article/view/77