A flawed model is driving biofuel policy — and the planet is paying for it

Biofuels—fuels made from plants, agricultural waste, and other biomass—are presented to governments as a climate solution. By burning carbon-rich crops instead of gasoline, the logic goes, governments can reduce greenhouse gas emissions. But this calculation ignores a critical question: What happens after those crops are turned into fuel?

Ultimately, food must come from somewhere else, and that somewhere is often a tropical forest. Biofuels, therefore, not only create food supply problems, but lead to the widespread destruction of forests that would otherwise absorb carbon. Over the last thirty years, this land clearing resulted in three to four times as much carbon released into the atmosphere as we saved from burning biofuels instead of gasoline and diesel.

Unfortunately, governments routinely ignore this evidence and rely instead on a family of complex economic models, led by the popular GTAP model, that predict minimal land use change from biofuels. Variants of this model are used by California's Air Resources Board, by international aviation agreements, and by the U.S. Treasury.

To understand the scope of government error on biofuels, Tobin Center Faculty Director Steve Berry, Senior Policy fellow Tim Searchinger, and Tobin affiliate Anton Yang took a closer look at the GTAP model, running the GTAP code directly to provide new and independent evidence on how it works. They found that it generates physically impossible results, relies on parameters with no econometric basis, and embeds assumptions that guarantee a low estimate of indirect land use change (ILUC) regardless of reality. The authors propose replacing GTAP models with a simpler, more transparent approach to estimating the effect of biofuel production on carbon emissions: the carbon opportunity cost of land.

What We Learned

• The GTAP model creates and destroys fake land. GTAP allocates land by revenue share rather than physical area, meaning any increase in revenue from cropland needs to be matched by a decline in revenue from pasture and forest. Because each acre has a different rent, matching the revenues leads to a mismatch of physical areas—one that ultimately results in the creation or destruction of fake land. To deal with this problem, GTAP modelers have added a pure adjustment factor, which automatically reduces or increases the area of pasture and forest to match the real physical area. A model that requires this kind of correction is not economically valid.
• Thousands of parameters lack any credible empirical basis. GTAP models contain thousands of non-zero parameters. Only a handful are even claimed to have an empirical source, and those few are misapplied—for example, they may be extrapolated from single studies of specific crops in the U.S. to every crop in every country. The sole claimed basis for land conversion elasticities is a study of U.S. land transitions. GTAP distills it into one parameter and applies it globally, in a way that even contradicts that study's own findings. Supply and demand elasticities are also sometimes assumed to be identical, despite this being a fundamental distinction in economics.
• The model's substitution formulas have no economic foundation. A core output of GTAP is predicting which lands replace which when cropland expands—the "diversion ratio" between forest and pasture. These ratios are determined entirely by revenue shares within each agroecological zone. No econometric parameter governs them. This means GTAP's most consequential predictions about forest conversion rest on an accounting identity rather than empirical evidence about land use behavior.
• Multiple specific assumptions stack to guarantee low land use estimates. The paper catalogs a layered system of biases, each reducing projected deforestation: an assumption that 80% of new cropping area comes from double-cropping existing land (contradicted by USDA data showing double-cropping is near historic lows); an assumed yield response to prices for pasture that GTAP's own authors admitted has no empirical basis; the complete exclusion of unmanaged forests—roughly half of all forests globally—from any potential conversion; and trade restrictions that artificially confine the effects of U.S. biofuel demand to the United States, preventing the model from projecting crop price increases in the tropical regions where deforestation actually occurs. These effects compound each other. Fixing any one of them would not resolve the others.
• GTAP's ILUC estimates are a small fraction of the actual carbon cost of land use. Using a straightforward biophysical benchmark—what carbon is released when cropland is produced at average yields—the true ILUC for corn ethanol is approximately 200 gCO₂/MJ, and for soybean biodiesel approximately 330 gCO₂/MJ. GTAP's estimates used by California are 22 and 27 gCO₂/MJ respectively—around 10% of the biophysical benchmark. The version embedded in the GREET model used for federal tax credits is lower still, at 5%. These are not defensible modeling differences. They reflect an accumulation of ungrounded assumptions that collectively ensure a low answer.

Policy Takeaways

• Stop using GTAP to justify biofuel policy. California, the U.S. Treasury, and international aviation bodies are making consequential decisions about biofuel subsidies and tax credits based on a model that cannot credibly estimate the quantities it purports to measure. This paper provides direct, independent evidence—from running the model's own code—that the ILUC estimates GTAP produces are not scientifically valid. Regulatory agencies should withdraw reliance on GTAP-derived ILUC figures until the model's empirical foundations are established.
• Base policy on credible empirical evidence. GTAP is not the only model lacking a clear and credible empirical basis, and replacing it with another model that lacks one will not help. Most economic policy advances are based on a wide range of evidence, not a single model, and it would be better to rely on a suite of evidence rather than a single misleading model. One promising approach is to consider evidence about the opportunity cost of the land used for biofuels.
• Treat biofuels skeptically as a climate strategy. The empirical evidence provides no support for the claim that crop-based biofuels reduce greenhouse gas emissions. The causal chain from biofuel mandates to higher crop prices to tropical deforestation is well-established. The carbon released by that deforestation exceeds the lifecycle savings from substituting biofuels for fossil fuels by a large margin, for decades. Policymakers should not accept modeling stories as a substitute for this evidence—and should be especially cautious about scaling up aviation biofuels, which would require land on a scale that could transform the planet.
• Recognize the stakes of getting this wrong. Between 2004 and 2024, global oilseed cultivation expanded by roughly 250 million acres, with biomass-based diesel accounting for about 40% of the increased vegetable oil demand. Announced projects would more than triple renewable diesel capacity. If vegetable oils were to supply even one-quarter of projected 2050 aviation fuel, they would require 40% of today's global cropland. The models currently used to greenlight these decisions are not fit for the purpose.

Data and Methodology

Berry, Searchinger, and Yang obtained and ran the GTAP-BIO model code directly, enabling independent analysis of the model's internal mechanics rather than relying solely on published documentation or outputs. They examine the 2010 version of GTAP-BIO—the best-documented version, which forms the basis of California's LCFS and subsequent variants—and verify that key structural features persist in later versions. The paper compares GTAP's ILUC estimates against a biophysical benchmark derived from average global carbon losses per hectare of cropland produced, against alternative economic model estimates (Lark et al. 2022; Merfort et al. 2023), and against the empirical literature on yield-price elasticities, commodity price integration, and tropical deforestation. The appendices provide detailed tables showing GTAP's pre- and post-adjustment land use projections in physical hectares and CO₂ equivalents, illustrating the magnitude of the "hand of God" correction and its effect on ILUC estimates.