Ricardian Estimates for Developing Countries and Canada

For the Ricardian estimates used in this study, the Mendelsohn and Schlesinger (1999) cross-section model estimated for the United States provides a default model for application to climate projections for countries and regions in which no directly estimated function is available, as discussed later. More recent models estimated explicitly for several important developing countries and regions provide a preferable basis for the estimates developed in the present study for those countries and regions.

4. In 2003, GDP originating in agriculture was $98.5 billion (appendix table E.1). "Net cash income of operations" in 2002 was $40.5 billion (USDA 2004). On this basis, net revenue was 41 percent of output as measured by agricultural value added.

Mendelsohn, Dinar, and Sanghi (2001) provide estimates for India; Ku-rukulasuriya et al. (2006) provide estimates for Africa; and a series of studies sponsored by the World Bank provide new estimates for major Latin American countries (see appendix G).5 For all three sets of estimates, the model structure is as follows:

where z is the measure of agricultural productivity (net revenue per hectare for Africa, natural logarithm of net revenue per hectare for India, and land value per hectare for the Latin American studies), T is average temperature, P is average monthly precipitation, i refers to the season, and K is a composite variable that reflects the regression constant as well as the influence of other control variables in the particular model esti-mated.6 The impact of business as usual global warming through the 2080s is then obtained using this equation to estimate the difference between agricultural productivity using the base period (1961-90) and future period (2070-99) climate estimates of this study. Application of these models requires applying the relevant seasonal monthly averages for future temperature and precipitation from the climate models, rather than the annual averages. The effect of carbon fertilization is not incorporated in these regional Ricardian estimates and must be added subsequently to obtain the overall impact of future climate change.

It is necessary to translate the change in net revenue from climate change to the corresponding percent change in output from the base level of output. In principle the change in net revenue will be the same in absolute terms as the change in output.7 In order to estimate this change as

5. Note that the Mendelsohn, Dinar, and Sanghi (2001) study also provides a model for Brazil. However, application of the model results in estimates of complete shutdown of agriculture from global warming, which as in the Brazil finding with the MS functions strains credibility. In part because of ambiguities in the data (including nonavailability of average land price for the study), the more recent World Bank study for Brazil is used instead as the preferred estimate for that country, as discussed later.

6. Note, however, that in the India model, the underlying variables are expressed as differences from their means (for example, T-T, for temperature, or (T-T)2, for temperature squared, where T is base average temperature). This approach has the property that the coefficient on the linear term shows the marginal impact of the climate variable (e.g., temperature), because the square term causes symmetric damage for either a rise or a decline in temperature and has a marginal impact of zero at the original base temperature, where the influence of an increase in temperature is shifting from positive to negative. In contrast, in the Africa and Latin America models, the levels rather than differences from means are the underlying variables.

7. That is, NR = Q - X, where NR is net revenue per hectare and X{ is the amount of purchased input i per hectare (mainly hired labor and fertilizer). With such inputs held constant, a yield shock from climate change translates directly into the same change in net revenue: ANR = AQ.

a percent of output, it is necessary to know the base level of output that corresponds to the base level of net revenue.

India

Table 5.2 reports the results of applying the Mendelsohn, Dinar, and Sanghi (2001) model for India to the base and 2080s climate variables identified in the present study. The model applies seasonal monthly climate data (see appendices G and H). The table first reports the levels and change in the dependent variable, which is the logarithm of net revenue per hectare. It then identifies the corresponding percent change from the base level of net revenue.8 The final column restates the change as a percent of base output.9

The results for India are sobering, with reductions in output potential ranging from about 30 to 35 percent in the southern regions to about 60 percent in the northern regions. As discussed later, this model does not include the favorable effect of carbon fertilization. Even after inclusion of carbon fertilization effects, however, the losses would be severe.10

Africa

The World Bank has recently carried out a massive farm survey in Africa to examine the relationship between agricultural productivity and climate

8. Given an initial actual net revenue or land value of q0 and base level model-estimated logarithm zo, the implied value of the missing constant is K = ln(q0) - zg. With the change in logarithm resulting from change in temperature and precipitation estimated as zt - zg, the absolute level changes from q0 to = exp(K + z) The proportionate change is then (q - q0)/q0. For moderate changes, this proportionate change will be approximately equal to zt - zg.

9. A rough estimate for India is that average net revenue per hectare in the estimation period amounts to two-thirds of output per hectare. This estimate is obtained as follows: According to Dinar et al. (1998, 98), average net revenue in the India sample was 1,424.7 rupees of 1980 per hectare. The data referred to the period 1966-86. Total farm area in India amounted to about 170 million hectares. In 1976, the midpoint of the period, agriculture accounted for 47 percent of GDP, or $43.7 (World Bank 1978, 80). By 1982, agriculture's share of GDP was down to 33 percent, amounting to $49.8 billion (World Bank 1984, 222). Taking the average of these two estimates, and using the 1980 exchange rate (7.86 rupees per dollar), agricultural value added in the base period was 367 billion rupees of 1980, or 2,160 rupees per hectare. Average net revenue was thus 1,425 / 2,160 = 66 percent of agricultural output. Correspondingly, for a given estimate of the percent change in net revenue, the appropriate estimate for percent change in output will be only two-thirds as large.

10. The counterintuitive greater losses in the higher latitude regions appear to stem from the following influences. First, the increase in temperatures in the northern regions is greater than that in the southern regions, even though the base temperatures are higher in the south. Second, the impact of changes in precipitation turns out to be positive in the south but negative in the north.

Table 5.2 Impact of global warming by the 2080s on Indian agricultural productivity3 using the Mendelsohn-Dinar-Sanghi modelb

Region

Present climate

Implied constant K for other variables

Future climate

Change in log

Percent change Net revenue Output

Northeast

0 0

Post a comment