We design, detail and test an approach which allows optimizing policy instruments to maximize welfare over results emerging from a Computable General Equilibrium (CGE) model. To technically enable policy optimization over a CGE model, we train a deep learning-based surrogate model that approximates the behavior of the CGE model. The final policy optimizing is then subject to the surrogate model. To show case our approach, we optimize the timing of emission abatement and a carbon tax recycling strategy for Tanzania from 2018 to 2050 with a detailed recursive-dynamic single-country CGE model. Our proposed approach and the provided code to train surrogate models for CGE models is quite generic, allowing its application to differently structured CGE models and associated policy instruments. Even though generation of the observation sample is computing time intensive, the surrogate model enables policy optimization not possible by using the CGE model directly. The trained neural network replicates the simulation behavior of the CGE model quite accurately with an average R2 of 99.99% over the outputs. Besides optimization, such a surrogate model representing the key input-output relations of a CGE model could also be easily integrated into other modelling frameworks.
Over the past fiveyears, the Global Trade Assistance and Protection(GTAP)Data Base has been expanded to include 16 additional individual African countries, while datasets for 10 existing African countries have been updated. Most of these efforts collaboration with experts from African National Statistical Offices and the Center for Global Trade Analysis (CGTA), alongside contributions from other partner institutions. As a result, the number of African countries represented individually in the GTAP Data Base has reached 42, enabling researchers and policymakers to conduct more detailed and policy-relevant analysis on priority issues facing African economies. This article documents the progress achieved so far and outlines next steps to address remaining data gaps for African countries in the GTAP Data Base. It also highlights key experiences, lessons learned, and challenges encountered during the process. By linking the work of statisticians and economists, the paper provides valuable insights for researchers, practitioners, and policymakers who rely on the GTAP Data Base for empirical analysis but may be less familiar with the underlying data construction process. Strengthening country-level coverage in the GTAP Data Base is essential to support evidence-based policymaking and unlock Africa's development potential.
This paper introduces MPSGE.jl, a free and open-source package in Julia that facilitates CGE model building from tabular definitions. The package's programmatically generated equations reduce redundant and repetitive code, errors, and development time. The built-in functions simplify troubleshooting, analysis, and reports. Embedding within a general-use scientific language enables streamlined integration with general-use functions and tools. The design combines computational efficiency with intuitive syntax and model formation. The open-source foundation allows for greater access, and user contributions to ongoing development. This work is a contribution to the open-science movement, with the aim of increasing access, robustness, transparency, and collaboration. We give an overview of how to use the package, its construction, and highlight some of its features. We include example models of different forms and scale as illustrations. First, we employ a simple toy model to introduce the basic structure. Then, we illustrate how the flexibility of MPSGE in Julia can facilitate functionality not possible in the GAMS progenitor. With a third example model, we demonstrate the package used at scale by evaluating tariff effects using a model with five household types for each U.S. state. We link to the package and its documentation for further reading and utilization.
Since the 1990s, there have been rapid increases in concentration ratios in many industries in the U.S., Australia and, we suspect, in other countries. Despite this, applications of GTAP continue to be based on pure competition or Melitz-style Large-Group Monopolistic Competition (LGMC). In either case, all firms are small, there is free entry, and industries make zero pure profits. Markusen challenges modellers to move to Small-Group Monopolistic Competition (SGMC) in which industries have high levels of concentration and firms are aware of the likely behaviour of their rivals. By making two generalizations of Melitz-LGMC specifications, we create a version of GTAP in which some industries are modelled as SGMC. First, we treat the demand elasticities perceived by firms for their products as variables. In our SGMC specification, markups over marginal costs, which depend on perceived elasticities, rise when these elasticities are reduced (in absolute terms) by anti-competitive practices. Second, we allow for sticky adjustment of the number of firms in an industry and simulate situations in which entry is blocked or partially blocked, allowing incumbent firms to make positive pure profits. As illustrated in our simulations, the emergence of pure profits has the potential to suppress real wage rates.
This paper provides an overview of the Global Trade Analysis Project (GTAP) Data Base, version 12 (hereafter referred to as GTAP 12). This latest version distinguishes 145 countries and 18 aggregate regions for 7 reference years (2004, 2007, 2011, 2014, 2017, 2019, and 2023) and details the annual value of economic flows, within and between economies, across 65 goods and services sectors at pre-and post-tax valuation. Central to this database are the bilateral trade and international transportation margin flows that link all countries/regions in the world. GTAP 12 also marks the first instance in which the land use and land cover (LULC) data, classified into 18 agroecological zones (AEZs), are incorporated into the standard database construction process, thereby resulting in consistent land information across all GTAP databases. This, along with the greenhouse gas emissions satellite data, will greatly facilitate the use of the GTAP framework in economy-wide studies of trade and environmental issues at the global and regional levels.
This paper describes a method of combining national Global Trade Analysis Project (GTAP) regions with sub-national detail. The approach extends the sub-national TERM methodology to create a family of models named GlobeTERM. In each model, the master database includes 74 sectors, based on GTAP with electricity split into 9 generation sectors plus a distribution sector. The other 64 sectors are those in GTAP Data Base version 11c. In most examples, one country within GTAP is split into sub-national regions, while retaining the other 159 GTAP regions in the master database. Examples include China, Germany, UK and USA. Another version represents Europe's regions at the NUTS-2 level. Using the US version of GlobeTERM, an illustrative simulation examines the impacts of the imposition of large bilateral tariffs between USA and China. The aggregation for this scenario depicts swing states separately. While almost all US regions lose in the short run from the imposition of high bilateral tariffs, there are winning and losing states in the long run amid national losses.
This paper presents gtapshape, an R package that allows the user to flexibly disaggregate the national endowments used in the computational general equilibrium (CGE) models based on the GTAP-AEZ framework. By allowing the user to specify the set of subnational boundaries in the form of a shapefile, gtapshape allows for a richer understanding of how within-country heterogeneity impacts the results of CGE models. gtapshape's modular strategy also allows for fast updating of the database as new sources of data become available. gtapshape is fully written in R and hosted in GitHub as free and open software. This should facilitate its incorporation into specialized workflows.
We introduce endogenous technological change in a multi-sector recursive dynamic Computable General Equilibrium (CGE) model. We consider the optimization problem faced by technology firms in choosing the optimal level of R&D intensity given market conditions. R&D intensity determines the number of innovations and the speed of technological change in each sector. In addition, firms can choose the direction of technological change; for instance, when designing, they may opt for less energy-intensive but more capital-intensive production when the price of energy increases. We also differentiate between local innovations whose outcomes are constrained by the global technology frontier and innovations that transcend this frontier. The incorporation of endogenous technological change has a significant impact on policy simulations. Once it is considered, the model predicts that climate policy (in the form of a carbon tax) has a significantly larger impact on emissions in the long-run than in the short run. Finally, we demonstrate how intertemporal knowledge spillovers lead to path dependency. The introduction of a carbon tax at an early stage induces early low-carbon R&D and early know-how accumulation, which in turn leads to higher productivity in low-carbon sectors and lower long-run costs of decarbonization when compared to the scenario of postponed carbon tax.
In this work, I introduce a formulation of the GTAP version 7 model (Corong et al., 2017) in an open-source algebraic modeling language, JuMP (Lubin et al., 2023), implemented in Julia (Bezanson et al., 2017), that closely follows the specification of the model in GEMPACK (Horridge et al., 2019), including equation and variable names. Unlike the linearized GEMPACK version, my formulation is in levels. However, my formulation of the GTAP model is quite different from the levels formulation in GAMS (Bussieck and Meeraus, 2004) by Mensbrugghe (2018), in following more closely the variable and equation names of the GEMPACK model. I show that my model produces essentially the same results as the GEMPACK model. Because it is expressed in levels, with unabridged functional forms underpinning its behavioral equations (e.g., containing all parameters in the case of CES functions), my model can address a wider range of policy questions, especially those involving parameter changes. Calibrating the model to additional data, e.g., quantities, additionally allows it to be used in scenario analyses involving absolute productivity metrics. As an important benefit, my implementation of the GTAP model using open-source Ipopt solver (Wachter and Biegler, 2006), requires no software license to solve the model.
Shock-intensive simulations can be used to: update computable general equilibrium (CGE) databases; estimate trends in industry technologies and the preferences of households, governments and importers; and generate baselines that incorporate forecasts from organizations specializing in different aspects of economies. We demonstrate the shock-intensive methodology by applying it to the Global Trade Analysis Project (GTAP) model. We update a 2014 GTAP database to 2019 with data-driven shocks to an array of macro and energy variables and describe the simulated shifts in technologies and preferences. Then, starting from the updated database, we conduct baseline simulations for 2019 to 2030, 2030 to 2040 and 2040 to 2050 in which macro and energy variables are driven by forecasts from the International Monetary Fund (IMF), International Institute for Applied Systems Analysis (IIASA) and International Energy Agency (IEA). The simulations connect disjoint years (e.g. 2019 and 2030) and use a smooth-growth assumption for savings in each region to jump over intermediate years. Investors are given forward-looking expectations so that their simulated decisions in 2030, for example, are realistic in light of prospects for 2030 to 2040. Considerable space in the paper is devoted to explaining closure swaps for facilitating shock-intensive simulations.
Mainstream economic wisdom favoring cooperative free trade is challenged by a wave of disruptive trade policies. In this paper, we provide quantitative evidence concerning the economic impacts of tariffs implemented by the United States in 2018 and the subsequent retaliations by partner countries. Our analysis builds on a multi-region multi-sector general-equilibrium simulation model of the global economy that includes an innovative monopolistic-competition structure of bilateral representative firms.
The integration between global and local economic systems has become an increasingly important research topic. This paper presents GTAP-SIMPLE-G, a general-equilibrium framework that extends the existing GTAP model by integrating a gridded partial equilibrium system detailing land use and crop production. This integrated framework links global demand and bilateral trade flows with local level crop supply and land use conversion, accounting for spillover effects across land-using sectors and subnational regions. The paper details the structure of GTAP-SIMPLE-G model, the development of a gridded database for one region in the model - namely Brazil, and the calibration of key parameters that govern the land use conversion and as well as the multi-crop production decisions. For illustrative purposes, GTAP-SIMPLE-G is applied to simulate the impacts of China's retaliatory tariffs on U.S. soybean exports on Brazilian crop production and land use at the local level. Findings show that the tariff shock causes not only an increase in Brazilian soybean production, but also highly heterogeneous responses in the production of other crops as well as land use in the wake of spatially varying multi-crop activities. Finally, this paper discusses the potential extensions of GTAP-SIMPLE-G for future studies and policy assessments on the Global-to-Local-to-Global linkages.
Single country computable general equilibrium (CGE) models often assume price taking behavior in world markets that may miss potentially important terms of trade effects in trade-exposed sectors. In this paper, we assess numerical evidence for modeling large open economies and develop a methodology for parameterizing a reduced form approximation of international trade linkages from a multi-regional global economy model. Simulated export demand and import supply elasticities suggest that assuming price taking behavior (e.g., small open economy assumption) may miss important impacts in export markets and some commodity import markets. We show that a reduced form approach to capturing terms of trade effects can perform well relative to an analogous multi-regional static model with explicit trade linkage. We also illustrate how the calibration procedure can be extended to a dynamic model using U.S. EPA's SAGE model. Our modeling scenarios demonstrate the relative importance of the large open economy assumption in non-trade policy applications, which can be significant.
Instructors of applied general equilibrium (AGE) courses usually face an important challenge: how to effectively and efficiently teach a large number of complex materials that are necessary for AGE analysis, including general equilibrium theory, producer and consumer theory, equational expression and model implementation. To overcome this challenge, in this paper we propose an innovative teaching tool, the SMART handout, which is a Wiki-style interactive handout that connects materials from AGE and several prerequisite courses and forms a knowledge network for easier cross-reference and deeper understanding. We take the Global Trade Analysis Project model as an example to demonstrate how a SMART handout is developed and applied in AGE teaching. This innovative tool can be generalized to other AGE courses, thereby contributing to building the capacity of future AGE researchers.
Growing population and per capita consumption are expected to generate about 3.4 billion tons of waste by 2050. The reuse and recycling of waste reduces the need for landfill, dumping, and incineration, and the extraction of virgin inputs. Such a transition impacts climate change, virgin material providers, producers and consumers. To quantify the direct and indirect impacts of this transition on the economy and environment, we extend a CGE model by developing a method and database including municipal solid waste streams. The waste stream constitutes of five types of municipal solid waste, three types of waste collection services and four types of waste treatment sectors that produce commodities to substitute those made by virgin materials. The model also tracks emissions caused by different waste treatment alternatives. The relationship between consumption, waste generation and waste treatment makes it possible to analyze circular economy policies. A baseline application shows that worldwide waste generation and collection is expected to grow by 45% between 2020-2050. Other waste is expected to grow the most by 53%; food waste is projected to grow the least at 35%. Therefore, without waste management policies, more waste will be incinerated or landfilled, which in turn aggravates climate change.
The first part of the paper written by Ivanic et al. (2023) (hereafter IBN) offers new estimates of the substitution elasticities between inputs for many industries. IBN develop an original econometric framework that solely relies on the GTAP (Global Trade Analysis Project) databases, which lack data on input prices. IBN find, first, that the short run elasticities are often statistically different from zero and of the correct sign; second, that the long-run elasticities are larger than their short-run counterparts. Our comment identifies three concerns with their econometric procedures and results. First, IBN do not acknowledge the Constant Return to Scale (CRS) assumption of the Constant Elasticity of Substitution (CES) function. Second, IBN 's statistical inference fails to correct the p-value problems associated with large samples. Third, the dynamic specification developed by IBN, where decisions are a function of price changes and not price levels, lacks theoretical justifications. We propose simple remedies to these three issues and, in some cases, find that many elasticity estimates are no longer statistically different from zero or of the correct sign. Moreover, we do not find different levels of significance between short- and long-run elasticities.
We propose a method for calibrating an industry-level technology to engineering (bottom-up) estimates with a particular focus on abatement opportunities. As a demonstration, substitution elasticities across inputs are adjusted in the nested cost function for the electricity sector to best fit a target marginal abatement cost (MAC) curve derived from engineering assessments of available technologies. Elasticities are optimized over an entire relevant range of the MAC, whereas current techniques use local point estimates under little or no abatement. In the context of fitting to a given MAC we evaluate alternative nesting structures and find that, while complexity in nesting improves the fit, even relatively simple nesting structures can reasonably approximate the target MAC. In our example, focused on the electricity sector, we find standard elasticities adopted in top-down models moderately overstate abatement costs relative to the engineering targets. In our preferred specification the most important adjustment is to escalate the substitution elasticity between energy and value-added inputs. This is consistent with an argument that the current set of point estimates fail to properly account for new capital-based technologies. These conclusions, however, are sensitive to our assumption about output-intensity abatement and consumer price responsiveness, both of which are not delineated in engineering estimates.
We propose a method for calibrating an industry-level technology to engineering (bottom-up) estimates with a particular focus on abatement opportunities. As a demonstration, substitution elasticities across inputs are adjusted in the nested cost function for the electricity sector to best fit a target marginal abatement cost (MAC) curve derived from engineering assessments of available technologies. Elasticities are optimized over an entire relevant range of the MAC, whereas current techniques use local point estimates under little or no abatement. In the context of fitting to a given MAC we evaluate alternative nesting structures and find that, while complexity in nesting improves the fit, even relatively simple nesting structures can reasonably approximate the target MAC. In our example, focused on the electricity sector, we find standard elasticities adopted in top-down models moderately overstate abatement costs relative to the engineering targets. In our preferred specification the most important adjustment is to escalate the substitution elasticity between energy and value-added inputs. This is consistent with an argument that the current set of point estimates fail to properly account for new capital-based technologies. These conclusions, however, are sensitive to our assumption about output-intensity abatement and consumer price responsiveness, both of which are not delineated in engineering estimates.
The incorporation of increasing returns and imperfect competition into applied general-equilibrium (AGE) models, beginning with Harris (1984), led to much larger welfare effects from changes such as trade liberalization. But the imperfect competition side of these IO developments has often failed to incorporate meaningful strategic behavior, largely ruling out firm-level productivity and scale effects. I show here that the incorporation of theory-based endogenous markups into AGE models is not difficult in spite of the added simultaneity of the system. I first derive the optimal markup equations for Nash Cournot and Nash Bertrand competition in a CES environment with free entry and exit. Then I code a simple numerical model using non-linear complementarity. Three alternatives are considered: large-group monopolistic competition (LGMC), small-group Cournot (SGC) and small-group Bertrand (SGB). Growth in the economy is the experiment used to compare these specifications. While the overall effects of growth on welfare are qualitatively similar, the gains to initially small economies are much larger under either small-group assumption relative to LGMC, but diminish relative to LGMC as economies grow large. Secondly I show how the contributions of variety (entry), firm scale (productivity), and markups (distortions) to welfare changes differ substantially among the three alternatives.
