When these separate effects are added up and the total summed over the 2,860 US counties in our sample, the net effect of the 5 degree increase in temperature and 8% increase in precipitation is to decrease agricultural profits by $1.9 billion, which is 5.3% of the $36.0 billion in annual profits. The predicted change in precipitation play a small role in this overall effect, underscoring that the temperature change is the potentially more harmful part of climate change for agriculture. These predicted changes are a function of 16 estimated parameters, so it is not especially surprising that the estimated decline is not statistically different from zero, regardless of whether the heteroskedastic-consistent standard errors or larger spatial standard errors are used to judge statistical significance. Further, separate tests cannot reject that either the change in temperature or the change in precipitation have no impact on agricultural profits. There are a few other noteworthy results. First, the marginal effects in and differ from those in and which indicates that there are state-level, time-varying factors that covary with the county-specific deviations in weather. The changes in the marginal effects are occasionally large, but they tend to cancel each other out.For example, the overall predicted change in profits is -$3.5 billion in column , but the difference with the -$1.9 billion effect in column is modest in the context of the standard errors. This finding that the estimated decline in profits is smaller with state by year fixed effects suggests that price changes do not appear to be a major concern in this context. Second, the marginal effects are virtually unaffected by the inclusion of the controls for soil characteristics. This is reflected in the F-statistics,growing blueberries in pots which fail to reject the null that the soil characteristics are jointly equal to zero at the 1% confidence level.
This suggests that the fixed effects approach is successful in eliminating any confounding due to time invariant determinants of agricultural profits.In this respect, it is preferable to the cross-sectional hedonic equations where the estimated effect of the climate variables on land values was sensitive to the inclusion of these controls. Table 6 explores the robustness of the results to alternative specifications.The entries report the estimated impact of a uniform 5 degree Fahrenheit and 8% increase in precipitation on agricultural profits calculated with the marginal effects of the weather variables. The last column normalizes this predicted impact by total profits for the relevant sample. The true functional form of the weather variables is unknown and thus far we have assumed that these variables are adequately modeled with a quadratic. We have also experimented with modeling them with a cubic. Panel [A] reveals that the estimated impact of climate change on agricultural profits is -$4.1 billion with the cubic approach. The point estimate is larger than in the comparable specification in Table 5, but this difference is small relative to the standard error . The Census profit variable is based on revenues and expenditures from the Census years . To this point, we have used weather measures from those same years but it is possible that lagged weather affects current profits.For example, poor weather in October 1996 might affect yields in 1997. Consequently, panel [B] includes contemporaneous and lagged weather and uses both sets of these variables to estimate the impact of climate change. The resulting predicted impact is essentially unchanged. Panel [C] explores the possibility that the results in Table 5 are driven by outliers. Specifically, it presents the results from a robust regression routine. This routine begins by excluding outliers, defined as observations with values of Cook’s D>1, and then weights observations based on absolute residuals so that large residuals are down weighted.The entry indicates that the estimated impact is modestly smaller in magnitude, but the qualitative findings are unchanged. The bottom panel separately reports the predicted change in profits when equation is fit separately on the 1,064 irrigated counties and 1,796 non-irrigated ones.These results may be of particular interest because Schlenker, Hanemann, and Fisher and others argue that climate change will have especially harmful consequences in irrigated areas of the country.
The predicted change in irrigated counties is –$4.4 billion, and it is $2.2 billion in the non-irrigated counties. Although statistical significance remains elusive, these findings are consistent with the view that parts of the country that rely on irrigation will be hardest hit by climate change. Panel [E] summarizes the results from estimating separate versions of equation for each of the 50 states. Thus, all the parameters are allowed to vary at the state-level. The sum of the state-specific estimates of the impact of the benchmark climate change is roughly -$3.4 billion. However, the meaningfulness of this estimate is undermined by its poor precision, which reflects the heavy demands placed on the data by this approach. Figure 2 graphically depicts the predicted impacts of climate change across states of the country. Here, the shadings reflect different categories of the predicted impact. It is evident that there is tremendous heterogeneity in the impacts. When the state-specific point estimates are taken literally, California and Texas are the two hardest hit states with losses of $1.7 billion and $1.8 billion in agricultural profits, respectively. Generally, the estimates suggest that global warming is most detrimental in states with a substantial number of counties in the USDA’s “Fruitful Rim” region33 and those that are heavily dependent on irrigation. The two states that benefit the most are North Carolina and Wisconsin with predicted increases of profits of $1.9 billion and $1.2 billion, respectively. It is not surprising that Wisconsin would benefit from milder winters, but the North Carolina result is less intuitive. All in all, the state-specific estimates should be viewed cautiously, because in general they are estimated imprecisely due to the over-parameterization of these specifications. Overall, the estimates in this subsection suggest that uniform increases of 5 degrees Fahrenheit and 8% precipitation would modestly reduce annual agricultural profits. The point estimates ranges from -$2 to -$4 billion, which is approximately 5% to 10% of annual agricultural profits. It is important to recall that these figures are likely downward biased relative to estimates that allowed for the fuller range of adjustments available to farmers over longer time horizons. This subsection tests the extent of adaptations available to farmers in response to county-specific annual deviations in weather. This is an interesting exercise in its own right but it also provides an opportunity to understand the source of the results in the previous section. Table 7 reports the results from this exercise for total sales, total expenditures, and a wide variety of subcategories. Column presents means of the total values of each of the variables in the U.S. agricultural sector across the three Census years.
Columns and report the predicted impacts and associated standard errors from the version of equation that controls for soil quality and includes state by year fixed effects. Column presents the estimated effect as a percent of the annual total in that category. The entries are derived from a balanced sample of 2,384 counties with non-missing values for each of the variables listed in the table. These counties account for 90% of total agricultural profits. The first panel reports the results from a constructed measure of net cash return,drainage gutter which is measured as the difference between farm revenues and expenditures. It is roughly 6% larger than the net cash return variable available in the Censuses. With this constructed measure, the predicted change in profits is -$1.4 billion with a heteroskedastic consistent standard error of $3.8 billion. These results are very similar to the ones in Table 5. The remainder of the table allows for an examination of the subcategories of profits. The revenue results suggest that the benchmark doubling of greenhouse gases is predicted to lead to a $7.5 billion increase in revenues, and that approximately 80% of this is due to higher livestock sales. On the cost side, total expenditures are predicted to increase by $8.8 billion. The subcategories reveal that the predicted change in costs is positive for virtually every input. The largest increases in dollar terms are for “feed and seed,” “purchases of livestock/poultry,” and “fertilizers and chemicals.” The “feed and seed” and “purchases of livestock/poultry” reinforce the credibility of the higher livestock sales and are large relative to the mean of these variables. However, none of the estimates in this table would be judged statistically significant by conventional criteria.The final row provides the predicted impact of the benchmark doubling of greenhouse gases on agricultural subsidies to farmers under the system of agricultural programs that prevailed in the 1987-97 period. When viewing these results, it is important to bear in mind that government payments are not included in net cash returns so the findings do not shed light on this paper’s earlier results.The estimate indicates that total government payments would increase by a dramatic and statistically significant $4.4 billion, which is 65% of mean government payments in this period. This would more than offset the predicted decline in profits. This finding is an important reminder that the agricultural sector is heavily subsidized and that the net cash return results are conditional on the planting decisions that are affected by the existing set of agricultural programs. They would likely differ under an alternative set of subsidy programs. Overall, the expenditure results provide modest evidence that farmers are able to undertake a limited set of adaptations in response to weather shocks. Profit maximizing farmers will only choose to incur these extra expenditures if the benefits exceed the costs, so it is reasonable to assume that the predicted losses would be even larger if these adaptations were unavailable. Although our estimates are surely downward biased from those that a correctly specified hedonic model would produce, they appear to account for a limited range of behavioral responses by farmers. In this respect, they are preferable to the production function approach. Optimal decisions about climate change policies require estimates of individuals’ willingness to pay to avoid climate change over the long run. The above analysis has developed measures of the impact of climate change on the profits from agricultural operations that accrue to the owners of land. Since land values ultimately reflect the present discounted value of land rents, or profits from land, we use the estimates from the previous section to develop measures of the welfare consequences of climate change. Table 8 reports estimates of willingness to pay to avoid climate change. The entries are derived by taking the parameter estimates from column in Table 5 and assuming that the predicted change in annual agricultural profits holds for all years in the future. We then apply a discount rate of 5% to determine the present value of this change in the stream of land rents. Some readers will prefer a higher discount rate, while others will prefer a lower one, and the entries can easily be adjusted to reflect alternative assumptions . The table is structured so that the predicted impacts of alternative global warming scenarios are readily evident. For example, the table indicates that the benchmark increases of 5 degrees Fahrenheit and 8% in precipitation imply that agricultural land values will decline by roughly $39 billion. Recall, the total value of agricultural land and its structures is approximately $1.4 trillion, so this is a roughly 3% decline in land values. It is also possible to estimate the effect of scenarios not reported in this table. The impact of such alternative scenarios can be determined by multiplying each 1-degree increase in temperature by -$7.0 billion and each 1% increase in precipitation by -$0.46 billion.36 There are a number of important caveats to these calculations and, more generally, to the analysis. First, some models of climate change predict increases in extreme events or the variance of climate realizations, in addition to any effects on mean temperature and precipitation. Our analysis is uninformative about the economic impact of these events. If the predictions about these events are correct, a full accounting of the welfare effects of climate change would have to add the impacts of these changes to the impacts presented here. Similarly, it is thought that permanent changes in climate will disrupt local ecosystems and this will affect agricultural productivity.