Valuing High Precision, Long-Term Weather Forecasts

Department of Agricultural, Environmental, and Development Economics The Ohio State University Working Paper: AEDE-WP-0029-03 Valuing High Precision, Long-Term Weather Forecasts Brent Sohngen (The Ohio State University) Ted Napier (The Ohio State University) Mark Tucker (The Ohio State University) February 2003 Abstract: This paper explores how farmers use weather information in current production decisions, whether farmers would adopt improved weather information, and their willingness to pay for the information. A sample selection model is developed to estimate the adoption and willingness to pay decisions. Farmers are shown already to rely on long-term forecasts of weather information for a variety of farming decisions, and 47% of the sample suggested that they would adopt improved forecasts for additional decisions if available. Mean (median) willingness to pay among individuals who would use the improved forecasts is $104 ($75) per year. When aggregated across Ohio farmers, aggregate willingness to pay is $2.1 million per year. Keywords: Value of Information, Contingent Valuation, Willingness to Pay, El Nino Introduction Weather risk is clearly one of the most important factors farmers face as they make decisions from year-to-year. In addition to the concerns farmers face with weather variation during “average” years, large scale events like the El Nino-Southern Oscillation (ENSO) phenomena can have longer-term, wider ranging impacts. Adams et al. (1999) suggest that the impacts of one ENSO cycle in the United States could range from $1.5 to $6.5 billion. Brunner (2002) suggests that ENSO could have broader implications by affecting real world commodity prices. Given the large implications of ENSO for the agricultural sector, and potentially other sectors, significant effort has gone into developing long-term weather forecasts to reduce uncertainty (Mjelde et al., 1998). Recent research suggests that society could gain substantially from improved long-term weather forecasts. Solow et al. (1998) find that the annual value of a perfect forecast of ENSO could be as large as $323 million per year in the U.S. agricultural sector, while Mjelde et al (2000) find that over a 10-year period, net social welfare could increase $1.2 to $2.9 billion with improved long-term weather forecasts. While the potential pay-off for society with improved weather forecasts is large, many studies assume that uncertainty revolves entirely around the forecasts themselves. That is, they measure the yield effect or the cost savings associated with improvements in forecast accuracy, presumably estimated by a technical measure of forecast variables (i.e. average temperature predictions or millimeters of precipitation during some time period in the future) against realizations of these variables after the fact. Little consideration is given to the uncertainty associated with farmer adoption and use of the forecasts in decision- 1 making. Potential adoption or non-adoption, however, could have a large effect on the benefits that farmers actually receive. Adoption depends mainly on farmer perceptions about forecast accuracy, so that even if the benefits could be large, these benefits may never materialize, or they may take years to occur if farmers are slow to adopt improvements in weather forecasts (Cohen and Zilberman, 1997). Furthermore, many studies simply assume how farmers will respond to the improved weather forecasts, and the results often imply reductions in producer surplus when the price effects of improved yields are evaluated in market models (i.e., Mjelde et al. 2000). Farmers, however, may benefit from the improved forecasts in many ways other than through yield changes if the forecasts alter how they handle risks inherent in farm production. Farmers already have a wide array of risk controlling devices available to them, including forward contracting, crop insurance, alternative seed varieties, genetically modified seeds, etc. Weather information could substitute for these other risk controlling devices, or it could complement them by helping farmers make better decisions about these other factors in the production function. The existing set of studies have not explored how farmers may actually use improved weather forecasts to adjust their production decisions, and ultimately their net revenues (i.e., Solow et al, 1998 and Mjelde et al., 2000). It is thus important to consider not only whether farmers will use the improved information, but it is also important to control for the use of other risk reducing technologies when considering potential adoption of improved forecasts. This study explores farmer decisions over adopting improved long-range weather forecasts. Another study (Kenkel and Norris, 1995) has examined the value of short-term (i.e., less than a day to several days in advance) forecasts, but no studies have used survey 2 data to explore the value of longer-term forecasts such as those offered by using information on El-Nino events. Using results from a recently conducted survey of Midwestern farmers (Tucker et al., 2002), we estimate the potential adoption of longterm weather forecasts, namely forecasts of temperature and precipitation variables 6 – 9 months in advance of important production decisions in which farmers can place 75% 100% confidence. Using contingent valuation, we also estimate farmer willingness to pay for the improved weather information. Our model assumes that farmers make two decisions, the adoption decision and then the willingness to pay decision for the weather information. A sample selection model is developed to estimate the two decisions jointly. Estimating the Value of Weather Information in Production Decisions Farmers make numerous decisions many months in advance of actually planting or harvesting crops, including forward contracting of crops, purchasing crop insurance, determining the types of crops or seeds to plant (i.e., genetically modified seeds), among other decisions. To the extent that farmers trust long-term weather forecasts for decisionmaking, they are likely to adopt the forecasts if they reduce risks or enhance yields. As argued by Solow et al. (1998), farmer profits could rise with improved weather forecasts as well. For this study, we are interested in assessing whether farmers would actually use improved long-term weather forecasts in which they can be more confident than currently, and if so, how much they would be willing to pay to obtain forecasts that provide them with this improvement in confidence. For the purposes of this study, long- 3 term forecasts are assumed to provide information on monthly precipitation and temperature for the farm in question 6 to 9 months in advance of important production decisions. Many factors are likely to affect the decision to adopt and pay for improved weather forecasts. Some farmers already use long-term weather forecasts for some decisions. These farmers may be more likely to adopt improved forecasts for other decisions, and further, they may be willing to pay for information in which they can have more confidence. Other farmers, however, may have had bad experiences with existing forecasting information, and they may be skeptical of introducing any improved weather forecasts into their decision-making. The adoption and willingness to pay decisions are likely to be linked, as they involve farmers assessing not only the validity of the weather information, but also how they would use the improved weather information if it was available to them. We assume that the decision to adopt improved weather forecasts relies on the underlying profit functions. Expected profits in a baseline case under existing confidence levels in long-term weather forecasts is: (1) Eπ0 = E[P0X0 – C(Z0; X0,P0) – I(L0; X0, P0)] The subscript “0” indicates the baseline scenario. P0 is a vector of prices for the outputs; X0 is the vector of crop outputs produced; C(Z0; X0,P0) is the cost of purchasing inputs, Z0, for the production process; I(L0; X0, P0) is the cost of risk reduction through traditional financial instruments, such as crop insurance or forward contracting. At the 4 beginning of the season, farmers will choose their input levels, Z0 and L0, contingent on the expected weather and prices for the entire season. While crop insurance and forward contracting are important financial instruments, farmers may also shift other inputs, such as seed varieties or nitrogen applications, in order to reduce risk. Under the hypothetical scenario, farmers are given the option to use improved long-term weather forecasts in which they can have high confidence. The information can help them adjust crop input decisions, and their expectations on future prices and future profits. While individual farmers may believe that the new information can benefit them, they may also recognize that the same information can benefit their neighbors and thus influence prices. This may be particularly important in the weather context because weather information has traditionally been provided freely as a public good. Farmer’s expected profits with the weather forecasts in which they can have high confidence are given as: (2) Eπ1= E[P1X1 - C(Z1; X1,P1) – I(L1; X1, P1)] where the subscript “1” indicates the scenario with weather forecasts in which they have high confidence. Given a sample of farmers, the probability of adopting the use of long-term forecasts with high confidence is: (3) Pr(Adoption) = Pr(Eπ1 + e1 > Eπ0 + e0) = Pr(E(π1 - π0)+ e > 0) 5 For the decision over whether to adopt improved long-term weather forecasts, individuals will adopt improved long-term weather forecasts if the expected profits under the forecasts in which farmers can have high confidence outweigh the expected profits under the original forecasts. Note that it is also possible that individuals get no improvement in expected profit, but they instead get smaller tails for the distribution of the error term. In other words, improved weather information could help farmers reduce the likelihood of large negative tail events. Risk averse farmers may place significant value on reducing the probability of large negative events. In this research, we do not actually observe the change in profit for individuals, but we instead observe an indicator, T. T is “1” if (4) T*= E(π1 - π0) + e > 0 and 0 otherwise. For this research, T* is parameterized as (5) T* = γZ + e, where e~N(0,1), where Z is a vector of farmer attributes, and γ is a vector of parameters to be estimated. The adoption decision can be estimated as a standard probit model if farmers are given the option to adopt or not adopt improved long-term weather forecasts. The second decision of interest is willingness to pay for weather information in which farmers can have high confidence in the accuracy of the forecast. Farmers choose 6 their willingness to pay for the improved weather forecast, WTP, such that their utility without the improved weather forecasts equals their utility with the weather forecast: (6) U(Eπ0) + u0 = U(Eπ1- WTP) + u1. Solving (6) for willingness to pay, we derive: (7) WTP = f(E(π1 - π0), u; T*>0) In our sample, willingness to pay for improved long-term weather data is only observed for individuals with T*>0, i.e., individuals who are willing to adopt and use the improved forecasts. Willingness to pay can be parameterized as follows: (8) WTP = B’X + u, where u~N(0,σ2) The resulting distribution of willingness to pay from the individuals who would use the improved weather forecasts could be biased upwards if those individuals respond strategically. For example, individuals who state that they would use the improved weather forecasts may think that by responding with high values, they can motivate policy makers to provide funding to make the information public. We do not suspect high degrees of strategic bias in our survey for two reasons. First, we focused broadly on farm decision-making rather than narrowly on willingness to pay for improved weather information. Thus, farmers were not asked to consider whether or not they would need to 7 pay for the new information. Second, we did not introduce the idea that the results would be used by policy makers to decide whether or not to develop improved weather information, and whether or not to provide the data publicly or privately. Farmers had little reason with the survey to try to bias the results by bidding too high or too low. While we cannot definitively state whether or not the responses are biased by strategic voting, by estimating the adoption and willingness to pay models jointly, we can test for correlation in adoption and willingness to pay responses. Depending on the strength and direction of correlation, we can assess the potential for strategic bias to exist. To estimate the adoption and willingness to pay decisions for improved weather forecasts, we rely on a recent survey of farmers in Ohio (Tucker et al., 2002). This survey explored how farmers use existing weather information to make farm production decisions, as well as how they might use improved weather forecasts. The survey and responses are discussed in more detail below, however, of interest here is the set of questions related to the adoption and willingness to pay decisions discussed above. Given the wide variety of weather information currently available, including both short-term and long-term forecasts from various channels of information, the adoption and willingness to pay components of the survey focused farmers on forecasts of specific duration and quality. The survey was designed to first assess how farmers use weather information and how much confidence they place in forecasts for different periods in the future. Forecast lengths range from long-term (6 – 9 months in advance) to short-term (one day to one month in advance) and we use the following confidence bands: “no” = 0 – 25% confidence; “little” = 26 – 50%; “considerable” = 51 – 75%; and “high” 76 – 100%. As suggested by Tucker et al. (2002), farmer adoption of weather information 8 appears to be driven heavily by their confidence in those predictions, and the risks they sense are imposed by weather, income and education. In this survey, farmers appear to have fairly high confidence in short-term forecasts, with confidence declining substantially for forecasts more than a week in advance (Table 1). Farmers rank weather conditions as the third most important risk factor they face each year, behind output and input prices, but well ahead of other factors such as world market fluctuations, regulations, pollution, urban sprawl, and public concern with genetically modified foods. We then asked farmers the following question: If you could access weather information in which you could place 75 – 100 percent confidence of predicting monthly temperature and precipitation for your farm 6 – 9 months in advance, would you use such weather information to make crop production decisions? Unlike many existing studies that focus on forecast accuracy or precision (i.e. Solow et al., 1999; and Mjelde and Hill, 1999), this study explores farmer confidence in forecasts. Even though forecasts may become highly accurate or precise as measured by comparing forecast values and realizations after the fact, it may take time for farmers to gain confidence in them. Given that most farmers appear to place little or no confidence in long-term weather forecasts currently (see Table 1), the hypothetical improvement in the question above would provide a substantial increase in confidence for most farmers. Unfortunately, the estimates cannot be compared directly to production oriented studies like Solow et al. (1999) and Mjelde and Hill (1999) that explore relatively narrow bands of forecast accuracy, but the results will allow us to assess whether having additional confidence in weather projections would even be valuable to farmers. 9 Respondents were allowed to answer yes, no, or uncertain to the adoption question. For individuals who responded yes, we asked a follow-up question on their willingness to pay each year for the service. The follow-up used a payment card with $50 increments in annual payments, starting at $0, rising to $400+ per year. Only individuals who would use the information were asked about their willingness to pay to obtain the information. As a consequence, if true willingness to pay is WTP*, we observe only an indicator variable such that Y = 1 if WTP* = $0 or Y = 2 if WTP* is between $1 and $49, etc. Willingness to pay is assumed to have the following functional form: (9) WTP* = B’X + u, where u~N(0,σ2). Y then takes on categories from 1 to 10 for 10 possible categories ($0, $1-$49, $50 – $100,$100 - $149,$150-$199,$200-$249,$250-$299;$300-$349,$350-$399,$400+). Following Bhat (1994), the likelihood function for the combined model can be written as: (10) Log-L = T =0 ∑ log[1 − Φ(γ ' Z )] + ∑∑ log Φ 2 (Y j − B' X ,γ ' Z ,− ρ ) − Φ 2 (Y j −1 − B' X , γ ' Z ,− ρ ) T =1 j =1 J [ ] Where Φ is a normal distribution, ~ N(0,1), Φ2is a bivariate normal distribution, ~ N(0,0,σ2,1,ρ), ρ is the correlation between the the errors in the two models, and Yj is the 10 upper or lower limit of the payment card category. For the limiting cases where j = 1 or j = J, the second part of the likelihood function takes on the following form: j = 1: ∑ ln[Φ T =1 2 (Y1 − B' X , γ ' Z ,− ρ )] (YJ − B ' X , γ ' Z ,− ρ )] . j = J: ∑ ln[1 − Φ T =1 2 A recent paper by Cameron et al. (2002) explores an alternative method for handling payment card data that assumes a logistic distribution for the responses. That study, however, does not consider a problem like this one where landowners can select themselves out of the market for the improved weather data by claiming that they would not use it even if it was available. The statistical approach used for estimating the model above allows for correlation between the adoption and willingness to pay decisions. Many willingness to pay studies have a large mass of respondents at $0 willingness to pay, and it is sometimes useful to understand why individuals bid $0 versus some positive amount. In the context of our study, there are at least two reasons why individuals may be observed to have $0 willingness to pay. First, they may not be interested in adopting the data at all for production decisions, or second, they may be willing to adopt the data for decisions, but only if they do not have to pay. By estimating the adoption and willingness to pay decisions simultaneously, we are able to assess which factors are most important for the two decisions, and we are able to assess the correlation between the two decisions. 11 Data and Results The data for this study is based on a survey of 2,160 farmers in 2001. Approximately 936 farmers responded. However, many of the farmers who did respond omitted critical information necessary for estimating willingness to pay. For reporting in this study, we focus attention on 417 of the surveys where individuals answered all of the questions necessary for our estimation. As shown in table 2, the group of farmers in our sample is slightly different than the average farmer in Ohio. For example, farmers in our survey tend to have a higher average annual gross farm income, work fewer days off the farm per year, have more years of education, more experience farming, and more acres owned than the typical Ohio farmer. It appears that our survey is weighted more heavily by individuals who rely on farming for a larger proportion of household income. It will be important to account for these distinctions when aggregating the willingness to pay results below. Before estimating the value of weather information to farmers, it is useful to explore how farmers actually use existing weather products. Table 3 shows the proportion of individuals in the survey indicating that they use long-term or short-term weather forecasts for important production decisions on the farm. Not surprisingly, a large proportion (>70%) of farmers use short-term weather forecasts for harvest, planting, pesticide, nutrient, and tillage decisions. Alternatively, crop selection, crop insurance, forward contracting, and seed variety and rate decisions are not heavily influenced by short-term weather forecasts. 12 There is considerably less variation across responses for long-term weather forecasts, and a much smaller proportion of farmers currently uses long-term forecasts for any production decisions. Farmers appear to rely most heavily on long-term forecasts for forward contracting of crops, seed variety and rate decisions, as well as to schedule some production decisions (i.e. planting, harvesting, and nutrient applications). Long-term weather forecasts also appear to influence some farmer decisions related to crop insurance. Crop type and tillage decisions do not appear to be heavily influenced by longterm forecasts. We estimate three models, a probit model to explain adoption decisions (equation 5), a grouped regression model to explain willingness to pay (equation 9), and a model that combines the two decisions and accounts for correlation across the error terms (equation 10). We begin with the probit model that explores the decision to use improved long-term weather information. The raw data indicates that 47% of farmers would use long-term weather forecasts in which they can place high confidence if that information was available. We hypothesize that the main factors affecting farmer decisions to adopt this information in decision-making are income (INCOME), age (AGE), years of education (EDUC), how important long-term weather information is to farmers currently (COUNTLT), percent of row crops lost as a result of drought in the last 5 years (PCTLOST), percent of losses insured over that same time period (INSURED), and use of genetically modified seeds for drought resistence (GENSEED). COUNTLT is the number of farm production decisions in table 2 where farmers indicate they currently use long-term weather data. In addition to these variables measured at the observation level, we also use two county-level average variables for the county where the farmers in the 13 survey are located: Average yield of crops in the county (AVYIELD), and variance of crop yields between 1997 and 2000 in the county (VAR9700). The probit model (columns 2 and 3 in table 4) performs reasonably well, with 65% of the responses predicted correctly. Income does not have a significant effect on the probability of adopting improved weather forecasts. More educated farmers are more likely to adopt the information, while older farmers are less likely to adopt it. Farmers who have adopted long-term weather information for a larger number of farm decisions today are more likely to adopt the improved weather data we offer them. Greater losses from drought likewise increase the probability of adopting the improved weather information, whereas the use of insurance and genetically modified seeds to reduce susceptibility to droughts does not seem to influence the adoption decision. Average yield and variance in the yields for the county where the farmer is located do not have significant effects on the adoption decision. We next turn to estimates of willingness to pay (columns 3 and 4 in table 4). For willingness to pay, we use a grouped tobit model, where the groups were defined as above, and we focus on the 219 farmers who indicated that they would use the improved weather data. In this model, income, education, age, and use of genetically altered seeds for drought all have a significant influence on willingness to pay. Income has the strongest effect. Average yield is close to being significant at the 5% level, suggesting that farmers in regions with higher yields in general are willing to pay more for weather information. Older farmers are willing to pay less for improved weather information, while farmers with more years of education are willing to pay more. The percentage of 14 crops lost to drought over the past 5 years and the use of insurance to mitigate these effects do not have a significant influence on willingness to pay. The final model we test combines the adoption and willingness to pay decisions in a sample selection model (Table 5). Maximum likelihood techniques are used to estimate the combined decisions. The results are generally stronger for the selection model compared to the two models independently run. The correlation parameter, “Rho”, is negative and significant, suggesting that there is correlation across the decisions to adopt and willingness to pay for the information. Income, the proportion of drought losses insured, and average corn yield for the county are included in the willingness to pay equation while the rest of the variables are included in the adoption decision. From the models above, income has its most important influence on willingness to pay and not on adoption. As shown above, income positively influences willingness to pay. While the size of drought losses would be expected to influence the decision to adopt, the use of insurance to mitigate those losses is expected to influence willingness to pay for weather information. Although the sign on insurance is significant only at the 10% level, the positive sign suggests that insurance and weather information are complements. Farmers who use insurance appear to be willing to pay more for weather information, so that weather information and insurance appear to be complements. Individuals who currently use insurance likely would gain from improved weather information because they could alter the insurance product they buy based on improved knowledge about potential outcomes. The adoption model has similar results to those in table 4, although the use of genetically altered seeds is significant and positive in the combined model. Farmers who 15 use genetically altered seeds are more likely to adopt improved weather information. This makes sense if the seeds altered to be drought resistant are more expensive than seeds without the altered genes. Variance in crop yields for each county have surprisingly little effect on the adoption decision. Average and median predictions of willingness to pay are presented in table 6. As expected, mean willingness to pay is greater than median willingness to pay, suggesting that a number of individuals are willing to pay a fairly large amount for the weather information. The selection model provides similar estimates for mean and median willingness to pay. Note that the predictions for the sample selection model are based only on the individuals who stated that they would actually use the information for farm production decisions. Aggregating these estimates is complicated in part because there are differences between the sample in our survey and the population of farmers in Ohio. It would not be appropriate to simply extrapolate our estimates to all 68,611 farmers in Ohio. Agricultural census data suggests that 35% of farmers in Ohio have gross annual sales less than $5000, while our sample does not include any farmers in this category. This category of small farmers, however, represents only 11% of farmland in Ohio. In our sample, more than 50% is composed of farmers making $50,000 a year or more in annual sales. According to census data, this group of farmers account for 73% of farmland in Ohio. Thus, while our sample contains a larger proportion of large farmers than represented in Ohio, our sample appears to be broadly representative of the individuals who control most of the farmland in Ohio. 16 To aggregate the estimates, we employ the following procedure. First, we break our predicted willingness to pay estimates for individuals into nine income and acreage groups, and determine the median willingness to pay for each group. Second, we determine the proportion of our sample that stated that they would use the improved weather information in each of these nine categories. Third, we determine the median willingness to pay for the average farmer in each of the groups by multiplying median willingness to pay for each group by the proportion stating that they would use the information. Fourth, we determine the population proportions for each of these nine groups with census data, and use this to weight willingness to pay for individuals in each group. The second column of table 7 presents estimates of the population proportion for each income and acreage group derived from census data. The proportions in this column do not include the small farmers who are not in the sample, i.e. farmers who have less than $5000 gross annual sales. The third column presents the proportion of individuals in our sample who stated that they would use the new weather information for decision-making if it was available. The final column is the median willingness to pay for each of the income and acreage groups. The willingness to pay for improved weather information weighted by proportion yes’s and population is $40.71 per year. This number is multiplied by 52,644 farmers in Ohio in these categories (more than $5000 in annual sales) to derive aggregate annual willingness to pay of $2.1 million. An alternative calculation would be to take the median willingness to pay from entire sample, $75.14, and use that directly to calculate the aggregate WTP. Since this median applies only to individuals who state that they would use the data, 47% of the 17 sample, one would multiply the median by 24,755 (= 52,644*0.47), to obtain aggregate willingness to pay of $1.9 million. This is less than the aggregate amount obtained above because the methods above assign more weight to high income farmers with larger acreages who have higher willingness to pay. Conclusion This study explores farmer willingness to adopt and to pay for improved longterm weather forecasts in which they can have high confidence. A two stage model is developed in which farmers first decide whether to adopt improved long-term weather forecasts. In the second stage, farmers who are willing to adopt the improved long-term weather forecasts state their willingness to pay for that information. Farmers are assumed to be profit maximizing, so that paying for improved weather forecasts increases annual costs, but also potentially increases annual production of crops or livestock, and consequently profits. Improved weather forecasts may not increase crop production directly, but instead they may increase production or profits by enhancing the farmer’s ability to use other risk management tools, such as crop insurance or genetically modified seeds. To estimate farmer adoption and willingness to pay, a survey of farmers in Ohio was conducted to assess how they used current weather information, and how they might use improved long-term forecasts. The survey asked a range of questions, including farmer willingness to use long-term weather forecasts in which they can place high confidence, i.e. forecasts 6-9 months in advance. Farmers who indicated they were 18 willing to use the forecasts were further asked to state their willingness to pay for this information. The results suggest that 47% of farmers would likely adopt weather forecasts in which they can have high confidence for farm decision-making. While we do not explore specifically the improvement in forecast accuracy it would take to give farmers high confidence, our results do indicate the type of improvement it would take to obtain widespread adoption. Currently 48% of farmers have high confidence in short-term forecasts and 72% of farmers use these forecasts for five or more production decisions on their farms. Our results indicate that if farmers could place high confidence in long-term forecasts of temperature and precipitation variables, there would be wide-spread adoption for farm production decisions, but it would not be as wide-spread as the current adoption and use of short-term forecasts. This is encouraging for agencies such as the National Oceanic and Atmospheric Administration because it suggests that achieving the forecast accuracy of short-term forecasts with long-term forecasts would be valued by many farmers. The primary decisions for which the information is used are choosing seed varieties, scheduling planting, determining forward contracts, scheduling harvests, and scheduling nutrient applications. A sample selection model was applied to estimate the decision to use and to pay for the improved weather information jointly. The results suggest that farmers who are more educated, who have experienced larger drought impacts in the last five years, who currently use long-term forecasts in production decisions, and who use genetically altered seeds to avoid drought are more likely to use improved long-term weather forecasts in production decisions. A surprisingly large number of farmers are already using long- 19 term weather forecasts for farm decisions, and these farmers in particular would begin adopting improved weather forecasts if available. The use of genetically altered seeds seems to be complementary to weather forecasts in that farmers who have used the seeds in the past are more prone to adopt improved weather information. As expected, older farmers, however, are less likely to use the improved weather forecasts. Willingness to pay for the weather information is most closely associated with income. Interestingly, the percent of drought losses covered by insurance increases willingness to pay, suggesting that crop insurance is also complementary with weather forecasts. Improved weather information could enhance the farmer’s ability to use other risk reduction technologies. Mean willingness to pay among the individuals who would use the improved weather forecasts is $104 per year and median willingness to pay is $75 per year. When weighted across the income and acreage sizes observed in our sample, aggregate willingness to pay is $2.1 million per year for improved weather information in the state of Ohio. This amounts to a value of approximately $0.15 per acre of cropland in Ohio. Although other authors have not predicted the impacts of improved forecasts specifically for Ohio, Mjelde et al. (1998) suggest no benefits for cropland in a nearby state, Illinois. Unfortunately, our results cannot be extrapolated to the entire United States. However, they do suggest that not all farmers would adopt the forecasts immediately. This suggests that other studies, such as Solow et al. (1998) and Mjelde et al. (2000), likely overestimate the potential societal benefits. Further, the results of Mjelde et al. (2000) suggest that when price changes are considered, farmer welfare could decline while society (consumers) gains. Our results suggest that farmers would gain from the 20 improved forecasts, potentially by being better equipped to use existing risk reducing technology like crop insurance or genetically altered seeds. 21 REFERENCES Adams, R.M., C.C. Chen, B.A. McCarl, and R.F. Weiher. 1999. The economic consequences of ENSO events for agriculture. Climate Research. 13:165-72. Bhat, C. 1994. Imputing a continuous income variable from grouped and missing income observations. Economics Letters 46: 311-20. Brunner, A.D. 2002. El Nino and world primary commodity prices: Warm water or hot air? Review of Economics and Statistics. 84:176-83. Cameron, T.A., G.L. Poe, R.G. Either, and W.D. Schulze. 2002. Alternative Nonmarket Value-Elicitation Methods. Journal of Environmental Economics and Management (Forthcoming). Cohen, D.R. and D. Zilberman. 1997. Actual versus Stated Willingness to Pay: A Comment. Journal of Agricultural and Resource Economics. 22(2): 376-381. Kenkel, P.L. and P.E. Norris. 1995. Agricultural Producers Willingness to Pay for RealTime Mesoscale Weather Information. Journal of Agricultural and Resource Economics. 20(2): 356-372. 22 Mjelde, J.W., J.B. Penson, C.J. Nixon. 2000. Dynamic aspects of the impact of the use of perfect climate forecasts in the corn belt region. Journal of Applied Meteorology. 39: 67-79. Mjelde, J.W., S.J. Hill Harvey, J.F. Griffiths. 1998. A review of current evidence on climate forecasts and their economic effects in agriculture. American Journal of Agricultural Economics. 80: 1089-95. Solow, A.R., R.F. Adams, K.J. Bryant, D.M. Legler, J.J. O’Brien, B.A. McCarl, W. Nayda, R. Weiher. 1998. The value of improved ENSO predictions to U.S. agriculture. Climatic Change 39: 47 – 60. Tucker, M., T. Napier, B. Sohngen, C. Henry. 2002. Adoption of Long-Range Climate Forecasts as a Farm Management Decision-Making Tool. Mimeo. Department of Human and Community Resource Development. The Ohio State University. 23 Table 1: Confidence in weather forecasts of alternative lengths. No Little Considerable High Confidence Confidence Confidence Confidence Time Length of Forecast (0 -25%) (26 – 50%) (51 – 75%) (76 – 100%) 12 months 47.4% 47.4% 4.7% 0.4% 6 months 31.8% 54.3% 12.9% 1.1% Monthly 15.8% 48.4% 33.0% 2.8% Weekly 5.3% 19.9% 58.3% 16.5% Daily 3.6% 9.6% 38.0% 48.8% 24 Table 2: Demographic characteristics of the sample of farmers used in this survey. Variable Annual Gross Farm Income Days worked off-farm each year Age Education (years in school) Years Farmed Acres Owned Total Acres Cultivated Percent Income from: Corn and Soybeans Cover Crops and oilseeds All Row Crops All Animals Fruits and Vegetables Average (this survey) $113,013 95 50.8 13.6 24.5 206 433 57.3% 14.9% 72.1% 23.2% 1.5% Census (1997) $68,293 114 53 12 21 103 206 36% 25% 61% 24% 3% 25 Table 3: Proportion of farmers responding that they use long- or short-term weather information to influence farm production decisions (highest 5 proportions are italicized) Long Term Short Term (6 – 9 Months) (1 Day to 1 Month) Select Crops 0.19 0.06 Schedule Pesticide Applications 0.19 0.71 Schedule Nutrient Applications 0.26 0.63 Schedule Machinery Maintenance 0.19 0.34 Conduct Crop Inspections 0.15 0.39 Purchase Crop Insurance 0.20 0.06 Schedule Harvest 0.26 0.80 Schedule Planting 0.32 0.83 Schedule Tillage 0.22 0.64 Choose Seed Varieties 0.17 0.33 Choose Seeding Rates 0.24 0.18 Determine Forward Contracting 0.13 0.28 26 Table 4: Regression Results for Adoption and Willingness to Pay Models Estimated Separately. Use Weather Information (Probit) (N = 417) Variable Coef. SE Constant -0.574 1.159 INCOME 0.000 0.001 AGE -0.016 0.005 EDUC 0.111 0.031 COUNTLT 0.074 0.030 PCTLOST 1.784 0.658 INSURED 0.040 0.376 GENSEED 0.301 0.228 AVYIELD -0.006 0.008 VAR9700 0.000 0.000 Sigma --Log-L -262.61 -Bold = Signficant at the 0.01 level Bold/Italics: Significant at the 0.05 level Willingness to Pay (Tobit) (N = 219) Coef. SE -370.067 195.631 0.550 0.101 -1.992 0.951 12.544 4.753 13.838 97.453 75.717 2.665 0.041 152.0925 -428.36 106.587 56.880 34.844 1.464 0.037 9.782686 -- 27 Table 5: Results of sample selection model to estimate adoption and willingness to pay for improved weather forecasts. Variable Willingness to Pay Constant INCOME INSURED AVYIELD Sigma Coef. -180.015 0.554 79.692 2.608 182.790 SE 164.606 0.103 43.964 1.401 17.381 0.545 0.005 0.032 0.522 0.028 0.194 0.000 0.12369007 -- Selection Model Constant -1.388 AGE -0.016 EDUC 0.117 PCTLOST 1.559 COUNTLT 0.082 GENSEED 0.400 VAR9700 0.000 Rho -0.680 Log-L -693.06 Bold = Signficant at the 0.01 level Bold/Italics: Significant at the 0.05 level 28 Table 6: Predicted Willingness to Pay for Improved Weather Information Mean WTP Median WTP Tobit Alone $104.88 $75.05 Tobit with Sample Selection $103.99 $75.14 29 Table 7: Aggregation of willingness to pay. Population Proportion Income 5 – 25 Acres <100 Acres < 220 Acres > 220 Income 25 – 100 Acres <100 Acres < 220 Acres > 220 Income >100 Acres <100 Acres < 220 Acres > 220 0.28 0.13 0.04 0.05 0.13 0.13 0.03 0.02 0.20 Proportion Yes’s 0.38 0.35 0.49 0.56 0.57 0.59 0.60 0.49 0.52 0.47 Median WTP $25.87 $25.11 $74.79 $75.18 $26.00 $149.64 $75.92 $125.44 $175.13 $40.71 30