There is a growing consensus that emissions of greenhouse gases due to human activity will lead to higher temperatures and increased precipitation. It is thought that these changes in climate will impact economic well being. Since temperature and precipitation are direct inputs in agricultural production, many believe that the largest effects will be in this sector. Previous research on the benchmark doubling of atmospheric concentrations of greenhouse gases is inconclusive about the sign and magnitude of its effect on the value of US agricultural land . Most previous research employs either the production function or hedonic approach to estimate the effect of climate change.Due to its experimental design, the production function approach provides estimates of the effect of weather on the yields of specific crops that are purged of bias due to determinants of agricultural output that are beyond farmers’ control . Its disadvantage is that these experimental estimates do not account for the full range of compensatory responses to changes in weather made by profit maximizing farmers. For example in response to a change in climate, farmers may alter their use of fertilizers, change their mix of crops, or even decide to use their farmland for another activity . Since farmer adaptations are completely constrained in the production function approach, it is likely to produce estimates of climate change that are biased downwards. Its clear advantage is that if land markets are operating properly, prices will reflect the present discounted value of land rents into the infinite future. In principle, this approach accounts for the full range of farmer adaptations. The limitation is that the validity of this approach requires consistent estimation of the effect of climate on land values.
Since at least the classic Hoch and Mundlak papers,growing blueberries it has been recognized that unmeasured characteristics are an important determinant of output and land values in agricultural settings.2 Consequently, the hedonic approach may confound climate with other factors and the sign and magnitude of the resulting omitted variables bias is unknown. In light of the importance of the question, this paper proposes a new strategy to estimate the effects of climate change on the agricultural sector. We use a county-level panel data file constructed from the Censuses of Agriculture to estimate the effect of weather on agricultural profits, conditional on county and state by year fixed effects. Thus, the weather parameters are identified from the county specific deviations in weather about the county averages after adjustment for shocks common to all counties in a state. This variation is presumed to be orthogonal to unobserved determinants of agricultural profits, so it offers a possible solution to the omitted variables bias problems that appear to plague the hedonic approach. Its limitation is that farmers cannot implement the full range of adaptations in response to a single year’s weather realization, so its estimates of the impact of climate change are biased downwards. Our analysis begins with a reexamination of the evidence from the hedonic method. There are two important findings. First, the observable determinants of land prices are poorly balanced across quartiles of the long run temperature and precipitation averages. This means that functional form assumptions are important in this approach. Further, it may suggest that unobserved variables are likely to covary with climate. Second, we replicate the previous literature’s implementation of the hedonic approach and demonstrate that it produces estimates of the effect of climate change that are very sensitive to decisions about the appropriate control variables, sample and weighting.
We find that estimates of the effect of the benchmark doubling of greenhouse gasses on the value of agricultural land range from -$420 billion to $265 billion , which is an even wider range than has been noted in the previous literature. Despite its theoretical appeal, the wide variability of these estimates suggests that the hedonic method may be unreliable in this setting.The results from our preferred approach suggest that the benchmark change in climate would reduce annual agricultural profits by $2 to $4 billion, but the null effect of zero cannot be rejected. When this reduction in profits is assumed permanent and a discount rate of 5% is applied, the estimates suggest that the value of agricultural land is reduced by $40 to $80 billion, or –3% to –6%. Notably, we find modest evidence that farmers are able to undertake a limited set of adaptations in response to weather shocks. In the longer run, they can engage in a wider variety of adaptations, so our estimates are downwards biased relative to the preferred long run effect. Together the point estimates and sign of the likely bias contradict the popular view that climate change will have substantial negative effects on the US agricultural sector. In contrast to the hedonic approach, these estimates of the economic impact of global warming are robust. For example, the overall effect is virtually unchanged by adjustment for the rich set of available controls, which supports the assumption that weather fluctuations are orthogonal to other determinants of output. Further, the qualitative findings are similar whether we adjust for year fixed effects or state by year fixed effects . This finding suggests that the estimates are due to output differences, not price changes. Finally, we find substantial heterogeneity in the effect of climate change across the United States. The largest negative impacts tend to be concentrated in areas of the country where farming requires access to irrigation and fruits and vegetables are the predominant crops .
The analysis is conducted with the most detailed and comprehensive data available on agricultural production, soil quality, climate, and weather. The agricultural production data is derived from the 1978, 1982, 1987, 1992, and 1997 Censuses of Agriculture and the soil quality data comes from the National Resource Inventory data files from the same years. The climate and weather data are derived from the Parameter-elevation Regressions on Independent Slopes Model . This model generates estimates of precipitation and temperature at small geographic scales, based on observations from the more than 20,000 weather stations in the National Climatic Data Center’s Summary of the Month Cooperative Files during the 1970-1997 period. The PRISM data are used by NASA, the Weather Channel, and almost all other professional weather services. The paper proceeds as follows. Section I motivates our approach and discusses why it may be an appealing alternative to the hedonic and production function approaches. Section II describes the data sources and provides some summary statistics. Section III presents the econometric approach and Section IV describes the results. Section V assesses the magnitude of our estimates of the effect of climate change and discusses a number of important caveats to the analysis. Section VI concludes the paper. The production function approach relies on experimental evidence of the effect of temperature and precipitation on agricultural yields. The appealing feature of the experimental design is that it provides estimates of the effect of weather on the yields of specific crops that are purged of bias due to determinants of agricultural output that are beyond farmers’ control . Consequently, it is straightforward to use the results of these experiments to estimate the impacts of a given change in temperature or precipitation. Its disadvantage is that the experimental estimates are obtained in a laboratory setting and do not account for profit maximizing farmers’ compensatory responses to changes in climate. As an illustration, consider a permanent and unexpected decline in precipitation. In the short run,square plant pots farmers may respond by increasing the flow of irrigated water or altering fertilizer usage to mitigate the expected reduction in profits due to the decreased precipitation. In the medium run, farmers can choose to plant different crops that require less precipitation. And in the long run, farmers can convert their land into housing developments, golf courses, or some other purpose. Since even short run farmer adaptations are not allowed in the production function approach, it produces estimates of climate change that are downward biased. For this reason, it is sometimes referred to as the “dumb-farmer scenario.”
In an influential paper, Mendelsohn, Nordhaus, and Shaw proposed the hedonic approach as a solution to the production function’s shortcomings . The hedonic method aims to measure the effect of climate change by directly estimating the effect of temperature and precipitation on the value of agricultural land. Its appeal is that if land markets are operating properly, prices will reflect the present discounted value of land rents into the infinite future. To successfully implement the hedonic approach, it is necessary to obtain consistent estimates of the independent influence of climate on land values and this requires that all unobserved determinants of land values are orthogonal to climate.4 We demonstrate below that temperature and precipitation normals covary with soil characteristics, population density, per capita income, latitude, and elevation. This means that functional form assumptions are important in the hedonic approach and may imply that unobserved variables are likely to covary with climate. Further, recent research has found that cross sectional hedonic equations appear to be plagued by omitted variables bias in a variety of settings .5 Overall, it may be reasonable to assume that the cross-sectional hedonic approach confounds the effect of climate with other factors . This discussion highlights that for different reasons the production function and hedonic approaches are likely to produce biased estimates of the economic impact of climate change. It is impossible to know the magnitude of the biases associated with either approach and in the hedonic case even the sign is unknown. In this paper we propose an alternative strategy to estimate the effects of climate change. We use a county-level panel data file constructed from the Censuses of Agriculture to estimate the effect of weather on agricultural profits, conditional on county and state by year fixed effects. Thus, the weather parameters are identified from the county-specific deviations in weather about the county averages after adjustment for shocks common to all counties in a state. This variation is presumed to be orthogonal to unobserved determinants of agricultural profits, so it offers a possible solution to the omitted variables bias problems that appear to plague the hedonic approach. This approach differs from the hedonic one in a few key ways. First, under an additive separability assumption, its estimated parameters are purged of the influence of all unobserved time invariant factors. Second, it is not feasible to use land values as the dependent variable once the county fixed effects are included. This is because land values reflect long run averages of weather, not annual deviations from these averages, and there is no time variation in such variables. Third, although the dependent variable is not land values, our approach can be used to approximate the effect of climate change on agricultural land values. Specifically, we use the estimates of the effect of weather on profits and the benchmark estimates of a uniform 5 degree Fahrenheit increase in temperature and 8% increase in precipitation to calculate the expected change in annual profits . Since the value of land is equal to the present discounted stream of rental rates, it is straightforward to calculate the change in land values when we assume the predicted change in profits is permanent and make an assumption about the discount rate. Since climate change is a permanent phenomenon, we would like to isolate the long run change in profits. Consider the difference between the first term in equation in the short and long run in the context of a change in weather that reduces output. In the short run, supply is likely to be inelastic , which means that Short Run > 0. This increase in prices will help to mitigate farmers’ losses due to the lower production. However, the supply of agricultural goods is more elastic in the long run, so it is sensible to assume that Long Run is smaller in magnitude and perhaps even equal to zero. Consequently, the first term may be positive in the short run but small, or zero in the long run. Although our empirical approach relies on short run variation in weather, it may be feasible to abstract from the change in profits due to price changes . Recall, the price level is a function of the total quantity produced in the relevant market in a given year.