Weather-induced yield reductions can act like an enforcement mechanism that limits supply to drive up the price, especially if there are land constraints that keep farmers elsewhere from bringing new land into production. A Ricardian analysis of farmland values only measures impacts that are capitalized into farmland values, but does not consider impacts on consumers. This is only appropriate if overall price levels are not impacted, e.g., if gains in one region are outweighed by losses in other regions as found by Rosenzweig and Hillel . Most recent studies combine time series and cross-sectional variation in panel analyses. These studies have linked agricultural outcomes in various locations over time to weather outcomes, including location fixed effects. In a linear model, the variation in a panel again stems from deviations around the mean, comparable to time series studies. A model using location fixed effects is equivalent to a joint demeaning of both the dependent as well as all exogenous variables in each location. A panel combines several time series analyses across different locations and imposes that the effect of a deviation from the mean is the same in all locations. In nonlinear panel models, e.g., one that uses a quadratic specification in temperature,macetas por mayor this is no longer true: both deviations from the mean as well as the mean itself enter the identification.
The reason is that the square of the demeaned variable is different from the demeaned square variable . All three sources of variation: time series, cross-section, and panel analysis have often been used to study the impact in a particular part of the world. As mentioned in the introduction, a key strength of reduced-form empirical studies is that they allow for the proper identification of key parameters, e.g., how weather impacts yields in different locations. On the other hand, they usually omit possible price feed backs that could be crucial in an integrated world market if global production levels were to change. Integrated assessment models might be better suited to address them.There are many more studies linking agricultural outcomes to weather and climate in temperate zones of higher latitude regions. The reason might be threefold: First, agricultural production in higher latitudes accounts for a large share of global production, much larger than its share of the global population. Fig. 1 displays production levels of four key commodities that account for 75% of the calories that humans consumed during the years 1961–2010.1 Production of each commodity is transformed into calories by using the conversion ratios of Williamson and Williamson and then summed across all countries within a continent for the four crops in question. To make the calorie numbers more meaningful, they are displayed as the number of people that could be fed on a 2000 cal/day diet. Production has been steadily increasing everywhere. As a result, the relative shares of production remained relatively constant. Continents with the largest production are Asia followed by the Americas. Table 1 gives not only the fraction of global production at three distinct points in time , but also the share of the global population.
As is immediately apparent, the share of global production in America is significantly larger than its share of the global population. Both the United States as well as Brazil are major exporters of agricultural commodities. At the same time, Asia and Africa, which are predominantly located in tropical areas, produce a smaller share of global production relative to their share of the global population and therefore depend on imports. Further, while there is a general consensus that countries in lower latitudes are likely to suffer from climate change, the sign and magnitude of impacts in higher latitudes is still being debated actively. Impact estimates range from large negative impacts under significant warming to insignificant impacts. Finally, countries in higher latitudes on average have more detailed agricultural data available, which makes empirical estimation easier. Schlenker and Roberts use time-series, cross-sectional as well as panel variation to estimate the effects of temperature and precipitation fluctuations on crop yields. All three sources of variation give similar results if the model allows for nonlinear effects of temperature on yields. They link fine-scale weather data that account for the distribution of temperatures within a day to annual county-level yields for corn, soybeans, and cotton for the years 1950–2005. Yields are increasing in temperature up to a threshold of 29 °C for corn, 30 °C for soybeans, and 32 °C for cotton, when further temperature increases become harmful. The single best predictor of yields is the amount of time that temperatures are above the threshold, summed over the entire growing season. For example, a temperature of 35 °C for a crop with a threshold of 29 °C would give a value of 6 °C. This variable explains almost half of the variation in yields although it completely discards anything that happens below the thresholds.
It also is a much better predictor of yield outcomes than average temperature. Each 24-hour exposure of 1 °Cs above 29 °C decreases annual corn yields by roughly 0.7%. As mentioned above, the same relationship is estimated using time series, cross-section, and panel sources of variation. Further, a similar relationship has been observed outside of agriculture, e.g., in math scores and measures of people productivity and how aggressively they respond to randomized interferences, e.g., a car that stops and blocks an intersection. Hence, one of the key sufficient statistics that integrated assessment models should incorporate is nonlinear effects of temperatures. These non-linearities were only observable when fine-scaled daily weather variables were constructed over the part of a county where crops are grown. Both spatial averaging over a county and temporal averaging over the growing season can hide important nonlinearities. More recently, Fezzi and Bateman obtained individual farm level data and conducted a Ricardian analysis for Great Britain. While farm-level data shows important significant interaction between temperature and precipitation, they disappear if the data is aggregated to the county level, demonstrating the importance of micro-level analysis to identify key parameters.It has been widely noted that agricultural sectors of developing countries are especially vulnerable to climate change. Especially low-lying areas in developing countries are projected to suffer severe damages from climate change over the coming century. Among the more common reasons provided for these statements is the fact that, as Nordhaus shows,blueberry grow poorer countries already have hotter climates. The impact of weather shocks on economic growth has been recently shown to be economically and statistically significant . It has been observed that the link between income and temperature is not only a phenomenon across countries, but can also be observed within countries. At the aggregate level Jones and Olken observe that higher temperatures in developing countries result in lower exports by between 2 and 5.7 percentage points for a year one degree warmer. This effect is not detectable for rich countries. The two sectors which are shown to experience the most significant negative response to a warmer climate are agricultural products and light manufacturing. This is consistent with the findings in Dell et al. , who find a short-term response of decreased growth in agricultural output by 2.66% for each 1 degree Celsius increase in annual average temperature. While these reduced form models do not provide causal evidence of micro-level mechanisms driving these effects, neither do the highly aggregated integrated assessment models . The first thing we learn from the empirical work by Ben Olken and others, is that at the very minimum, the impact of climate on the agricultural sector through temperature is likely to vary by income level of individual countries. Second, the evidence cited above relies on year-to-year fluctuation in weather, which has well understood drawbacks, which we discussed above. Reduced form econometric papers generally acknowledge this fact and attempt to quantify the importance of adaptation, which in some cases results in long run response estimates around 50% smaller than the short run estimates. This suggests that understanding the magnitude of the adaptation response is especially important for the developing world. There is rapid growth in the number of recent papers that study the response of different agricultural crops to changes in climate. Robert Mendelsohn and a number of coauthors have applied the Ricardian method to a large number of countries and regions, including most recently a subset of countries on the African continent .
A second strand of literature use panel data methods . Both sets of papers are very similar in methods to the papers for developed countries discussed above. It is not the purpose of this paper to provide an inventory of the literature, but rather to outline the important issues involved in estimating climate change impacts and capturing adaptation. A recent set of papers on rice production in Asia lend themselves quite nicely to demonstrate the important empirical issues. Peng et al. demonstrated that growing season mean minimum and growing season mean maximum temperature had differential effects on rice yields at their plot using a dataset of 12 observations from an experimental farm. Maximum temperature did not have a detectable impact on yields, while minimum temperature negatively influenced yields. Further, they show evidence of a nonlinear relationship between growing season mean solar radiation and yields. While the sample size is small and plants on experimental farms are grown at close to optimal conditions, which may not be true in the field, this shows that using simple averages of temperature is problematic. Auffhammer et al. picked up on the Peng et al. findings and estimated a two equation system, where famers decide on how much area to plant in a first stage and then harvest at the end of the growing season for rain fed Kharif rice in India. In this first application of the fixed effects approach in the context of climate change, they specify a production function, which models total rice harvested as a function of area and a number of weather variables which are matched to different stages of the rice plant growth cycle. They control for average minimum temperature, rainfall and solar radiation during three growth stages. They show that rainfall and minimum temperature have a statistically significant impact on output but not during all parts of the growing season. Recognizing that area is endogenous, they estimate in a first stage an area demand function, which controls for important input and lagged output prices as well as weather. They show that July–September rainfall has a significant impact on area harvested. An important finding from their aggregate exercise is that it is crucial to properly capture the crop-specific measures of climate when estimating these models. A single temperature measurement, which is calculated over the same time frame for all crops is likely inadequate, especially if the underlying response function is non-linear. In more recent work, Welch et al. use the most extensive farm level dataset covering the main irrigated rice growing regions in Asia to study the climate response of rice at the farm level. The rich dataset from 227 intensively managed irrigated rice farms in six important rice-producing countries contains complete information about all physical and labor inputs applied to the fields, including what strand or rice is planted, how many hours of labor were used in growing season, what pesticides and fertilizer were applied and when. In addition, a weather station delivering daily readings of minimum and maximum temperatures as well as solar radiation was installed at each site. Most farms were observed over a number of growing seasons, which allowed for a fixed effects identification strategy. The econometric estimates show that temperature and radiation had statistically significant impacts during both the vegetative and ripening phases of the rice plant. Higher nighttime temperature reduced yield and higher maximum temperature raised it. The effect of solar radiation varied by growthphase. The authors note that there is evidence that at very high temperatures the impact of maximum temperature flattens out. These findings again stress the importance of properly accounting for temperature changes by crop and growth phase in econometric studies, which is an insight being picked up by some more recent studies .