Under the 1996 farm bill, producers have increased planting flexibility, which. Producer Ability to Forecast Harvest Corn and Soybean Prices

Transcription

1 Review of Agricultural Economics Volume 23, Number 1 Pages Producer Ability to Forecast Harvest Corn and Soybean Prices David E. Kenyon Harvest-price expectations for corn and soybeans were obtained in January and February each year from Producer expectations on average missed actualcorn and soybean prices by $0.41 and $0.67 per bushel, respectively. Producer price expectations each year had a range of over $1.00 per bushelfor both crops. Producer price distributions were skewed toward higher prices, and they consistently underestimated the probability of large price changes from January until harvest. Under the 1996 farm bill, producers have increased planting flexibility, which may increase price risk. Since 1996, Virginia harvest prices for corn have ranged from $2.05 to $3.36, and soybean prices have varied from $5.61 to $7.13. The elimination of acreage restrictions, target prices, and storage programs will likely make such price swings common in the future. Price risk management strategies and supply response models developed under previous government programs will need to be reexamined. How producers form price expectations is critical to price risk management, supply analysis, and policy formulations. However, few studies are available of actual producer price expectations, the distribution of these expectations, and producer estimates of the distribution of likely price outcomes. Prior studies of actual producer expectations have reached the following conclusions. Heady and Kaldor analyzed the price expectations of 200 farmers in Iowa every six months from December 1947 to June Producer expectations missed actualannualcorn prices by $0.32 in 1948 and $0.01 in 1949, and missed soybean prices by $0.39 in 1948 and $0.05 in Only 52% of individual expectations fell within ±10% of actualprices. In both 1948 and 1949, producer price expectations were skewed toward higher prices. Fisher and Tanner conducted an David E. Kenyon is professor of Agricultural and Applied Economics at Virginia Polytechnic Institute and State University in Blacksburg, Virginia.

2 152 Review of Agricultural Economics economic game with 55 wheat growers in Australia to determine the methods they were using to develop price expectations. The adaptive expectations model fit the data best, but only explained 20% of the variation in actual price expectations. The rationalexpectations modelperformed badly, but the poor performance may be related to producers having been given supply and demand information and prices for commodities they did not grow. In the Fisher and Tanner study, farmers suggested the best strategy was to take an arithmetic average of past prices. Eales et al., using producers expectations from 1987 and 1988, found producers price expectations were not significantly different from futures prices, but that producer estimates of market price variance were always lower than the markets estimate based on option market implied volatilities. In 1996, Schroeder et al. asked 55 Kansas crop producers to Rank the top five sources you use to formulate price expectations where 1 = most important and 5 = least important. The top four information sources and their ratings were: market advisory service, 3.1; futures market, 3.3; electronic information provider, 4.3; and university outlook meetings, newsletters, 4.4. Unlike the previous three studies, the Schroeder et al. study does not analyze actual expectations, but reports on how producers say they are forming price expectations. In the absence of recent studies that analyze actual producer expectations, many different hypotheses of how producers formulate price expectations have been posited. Some of these hypotheses include market prices adjusted by government program provisions (Houck and Ryan; Kenyon and Evans), futures prices (Gardner), adaptive expectations (Nerlove), and various combinations of these hypotheses. Shideed and White compared six acreage response models for corn and soybeans using various price expectation hypotheses. They conclude that no unique specification emerged as the best hypothesis for both commodities. Given the limited number of actual studies of producer price expectations and the wide range of theoretical models of price expectation formulation, the current study of actualproducer price expectations over an eight-year time period may produce some new insight into how producers are actually forming price expectations for corn and soybeans. Analysis of their actual expectations and the distribution of their expectations may provide usefulinformation to economists delivering price outlook and price risk management strategies information, and to researchers estimating acre response models in the more market-oriented agricultural economy under the 1996 farm bill. Data Producer price expectations for corn and soybeans were obtained by survey at the annualvirginia Corn Soybean Conference held in January or February each year starting in The survey was conducted during the first afternoon of the program before any price outlook information was presented. A short explanation (less than five minutes) was made explaining what information was being requested. Producers were asked for three cash prices for each commodity the most likely harvest price, the price with a 1 in 10 chance of prices falling below at harvest, and the price with a 1 in 10 chance of prices rising above at harvest. Producers were not allowed to give price ranges. Each producer was asked to specify the market where the majority of each crop was sold. Individual producers were

3 Producer Ability to Forecast Prices 153 not identified, but most of the responses were from the same producers over the eight-year period. Many of these producers are industry leaders and may not be representative of all Virginia corn and soybean producers. On average, they produce 456 acres of corn and 633 acres of soybeans. Each year, questionnaires were completed, representing between 5% and 10% of Virginia corn and soybean acreage. The corn and soybean industry in Virginia is very diverse with very different regionalsupply and demand characteristics. The grain-producing regions have surplus grain, while the livestock and poultry production regions are grain deficient. In this study, only those producers located in the grain surplus areas selling in two major markets were selected for analysis. In the grain deficit regions, the reported cash prices had a wide range, making it difficult to determine the harvest price received by producers. Therefore, the deficit market areas were not included in the study. The elicitation date and the number of producers each year are shown in table 1. The elicitation date is before any USDA supply and demand estimates or planting intention reports are released for the upcoming production season. Some private market forecasts of acreage and price for the next season may be available. The harvest date was not specified on the survey form. The actual harvest price used in table 1 is for September 15 or the closest day thereafter for corn and November 15 or the closest day thereafter for soybeans. These two dates coincide with 50% completion of harvest in the two markets analyzed according to the Virginia Department of Agriculture and Consumer Services. The forecast length is approximately 7 to 7 1/2 months for corn and 9 to 9 1/2 months for soybeans. Average Price Expectations The average price forecast and actualharvest price for both commodities are in table 1. The years 1991 to 1998 were a difficult period over which to forecast price, with wide fluctuations in weather conditions and the initiation of a very different farm program. Virginia harvest corn prices ranged from $2.05 to $3.36 per bushel, with price changing at least $0.30 a bushel each year compared to the previous year. Soybean prices at harvest ranged from $5.27 to $7.13 a bushel, although harvest price was between $5.27 and $5.29 per bushel in three of these years. Producer price expectations generally missed harvest prices by a substantialamount. Their average expectations for corn ranged from $0.65 under to $0.69 over harvest price in seven of the eight years. Their price expectations for soybeans ranged from $0.93 under to $1.04 over harvest price in seven of the eight years. These differences between expectations and actualprices were comparable in magnitude to those found by Heady and Kaldor in 1948 and In general, producers tended to overestimate low harvest prices and underestimate high harvest prices. These large errors were generally related to years with exceptionally good or poor weather. Their average expectations during January and February were very close to average harvest prices across the eight years $2.57 versus $2.54 for corn and $6.16 versus $6.06 for soybeans, respectively. Over these

4 154 Review of Agricultural Economics Table 1. Elicitation dates, number producers, average producer price forecast, actual harvest price, and forecast error Average Actual Average Elicitation Number Price Harvest Expectation Date Producers Expectation Price a Error Corn ($/bu) 2/5/ /4/ /19/ /18/ /6/ /5/ /3/ /9/ Average b Variance RMSE c 0 46 Soybeans ($/bu) 2/5/ /4/ /19/ /18/ /6/ /5/ /3/ /9/ Average b Variance RMSE 0 74 a September 15 for corn and November 15 for soybeans. b Average absolute error. c Root-mean-squared error. few years, there does not appear to be a consistent bias in expectations relative to harvest price. Research by Kenyon, Jones, and McGuirk indicates that in the spring, December corn futures and November soybean futures were not good price forecasters of harvest prices over the years 1974 to Table 2 records the ability of the futures market to forecast harvest prices over the same time period producers were asked to estimate cash prices (1991 to 1998). Based upon a comparison of root-mean-squared errors (RMSE), there appears to be very little difference in the ability of the futures market and producers to forecast harvest price (tables 1 and 2). Overall, neither producers nor the futures market seems to be very good at forecasting harvest prices during January and February. Uncertain weather conditions make harvest prices difficult to predict so early in the season.

5 Producer Ability to Forecast Prices 155 Table 2. Futures market ability to forecast harvest price Elicitation Elicitation Harvest Forecast Date Day Price Price a Error December corn futures ($/bu) 2/5/ /4/ /19/ /18/ /6/ /5/ /3/ /9/ Average b Variance RMSE c 0 42 November soybean futures ($/bu) 2/5/ /4/ /19/ /18/ /6/ /5/ /3/ /9/ Average b Variance RMSE 0 73 a September 15 for corn and November 15 for soybeans. b Average absolute error. c Root-mean-squared error. Distribution ofprice Expectations Individualproducers had a wide range of price expectations each year. Individualcorn producer expectations had a range of approximately $1.00 per bushelor more each year (table 3). In a typicalyear, two-thirds of the expectations were within $0.25 per bushelof the average expectation. In 1992, 1994, 1995, 1996, and 1998, less than 5% of the producers had price expectations within the $0.20 price range of the eventualharvest price. In these years, the producers price expectations were substantially skewed above or below the harvest price. Only in 1991 were the producer expectations centered near the eventualharvest price. Individualsoybean producer expectations had a range of $1.50 per bushelor more each year except 1995 and 1998 (table 4). In some years, the distributions were very skewed. In 1991, 11 out of 30 producer expected prices were between $5.76 to $6.00 per bushel. Only 4 producers expected higher prices, but 15 producers expected lower prices, with 8 producers expecting prices as much as $0.50

6 156 Review of Agricultural Economics Table 3. Distribution ofproducer most likely corn price expectations: Range ($/bu) Number of producers < Expectations ($/bu) Average Minimum Maximum Range Standard deviation Actualharvest to $1.00 lower than the most frequently expected price range. The same general pattern holds in 1992 and 1994, where most of the price expectations are below the most frequently expected price range. In 1995, 21 of the 28 expectations are between $5.01 and $5.75. Yet, six producers expected prices above $6.01. In 1993, 1996, and 1997, the price expectations are more evenly spread around the most frequently expected price range(s). Like corn, only one or two soybean producers price expectations in January or February were within the $0.25 price range of the eventualharvest price. Only in 1996 were the price expectations centered around the eventualharvest price. In all the other years, the eventualharvest price was in the tails of the producer price expectation distributions. Heady and Kaldor found that 48% of the producer price expectations were more than ±10% from the actualprice. In this study, 58% of the corn and 45% of the soybean expectations were ±10% from the actualprice. Based on this small sample, producer ability to estimate actual prices does not appear to have improved much over time. Untilthe ability to forecast weather severalmonths into the future improves, producer ability to forecast prices is not likely to improve. Comparison of the two distributions of expected prices seems to indicate producers have more difficulty reaching a consensus of expected price for soybeans than for corn. This apparent difficulty may be related to the two-month longer forecast period for soybeans and the weather uncertainty associated with the South American soybean crop through April. For soybeans, producers must

7 Producer Ability to Forecast Prices 157 Table 4. Distribution ofproducer most likely soybean price expectations: Range ($/bu) Number of producers < Expectations ($/bu) Average Minimum Maximum Range Standard deviation Actualharvest anticipate the weather for two crops: the South American crop from January through May and the United States crop from May through September. Producer Distribution Estimates Producers were asked to estimate harvest prices that reflected only a 10% probability of going below or above these prices at harvest. The average low and high price estimates for each year are reported in table 5. For corn, the average spread between the low and high price is $1.05 per bushel. The distributions are skewed to the right with more probability of higher than lower prices. Compared to the most likely price, the average high price is $0.59 per bushel higher, while the low price is $0.46 per bushel lower. The lower bound price could be affected by the USDA loan rate, but the lower corn price estimates of some producers were only near the loan rate in The skewness toward higher prices versus lower prices occurred each year. For soybeans, the average difference between the low and high price expectation is $1.85 per bushel. The average low price is $0.78 below the most likely price, while the average high price is $1.07 above the most likely price. Like corn, the price distribution is skewed to the right each year. The greater probability of higher compared to lower prices reflects producers general optimism

8 158 Review of Agricultural Economics Table 5. Producer expectations ofmost likely, low, and high price at harvest compared to actual harvest price Producer expectations Actual price Average Average Average Actual < low (L) or most likely low high harvest > high (H) price Year price price a price b price expectation c % Corn ($/bu) H L H L H H H L Average N.A. d Soybeans ($/bu) L L H L H L H L Average N.A. a Ten percent probability going below this price. b Ten percent probability going above this price. c Percent of producers whose actual harvest price was either less than their low price expectation (L) or more than their high price expectation (H). d N.A.: not applicable. toward higher prices. This optimism may be appropriate given that the elasticity of demand for these grains becomes more inelastic as use declines and prices increase. The average most likely, average low, and average high price obscures the wide variation in the level of the upper and lower 10% price bounds elicited from individualproducers. To demonstrate this wide variation, table 5 reports the percentage of time the harvest price exceeds the 10% lower or upper bound price each year. For corn, the actualprice exceeded either the expected upper or lower bound price of 50% of the producers in 1992, 1994, 1995, and For soybeans, the actualprice exceeded the expected upper or lower boundary price for 35% or more of the producers in five of the eight years. These percentages indicate that producers have a strong tendency to underestimate the potentialchange in price from January or February untilharvest. In other words, producers had a

9 Producer Ability to Forecast Prices 159 strong tendency to underestimate the variance in harvest prices. These results are similar to those of Eales et al. based on data for one year. Expected price volatility is one of the major variables determining option premiums. When prices are expected to be highly variable, premiums increase and vice versa. Producers subjective probability distributions have smaller variances than the market. Hence, producers are likely to view option premiums as too expensive, because producers underestimate the probability of large price changes. Comparison Expectation Models How are these producers forming their price expectations? Are they based on cash prices, futures prices, government support prices, errors in previous estimates, or some combination of the above? Many theories have been suggested, but rarely are these theories tested against actual producer expectations. Although the sample is extremely small (n = 8), actualaverage producer expectations were regressed against expectations from various producer price expectation models. The first comparison is to December corn and November soybean futures on the elicitation day adjusted for expected basis in the two markets. Expected basis is the average harvest basis from the previous five years. The second comparison is cash harvest prices from the previous year. The third comparison is the current cash price on the elicitation day. The fourth comparison is the average harvest price the previous five years. And the fifth comparison is to Nerlove s adaptive expectations model: (1) P t = P t 1 + α 1 + β 1 (A t 1 P t 1 ) where P t = expected price in year t, and A t 1 = actualprice in year t 1. Equation 1 can be written as (2) P t P t 1 = α 1 + β 1 (A t 1 P t 1 ) and was estimated using data from 1991 to The dependent variable in each modelis the actualaverage producer price expectation. The independent variable is the price expectation of the various models. The estimated equations are in table 6. The producer price expectations are most closely correlated with current cash prices for both corn and soybeans. Producer expectations are also highly correlated with current futures prices for December corn and November soybean futures. Based on an F-test for equalmean-squared error, the current cash and futures price models are not statistically different at the 5% level. Producer expectations are also correlated with harvest prices from the previous year, but statistically they are less accurate in estimating expectations compared to current prices. The historicalfive-year average and the Nerlove adaptive expectations prices are far less correlated with actual producer expectations. These producers appear to be relying most heavily on current cash prices and current harvest month futures prices. These results are consistent with those of Eales et al. and Gardner, who found price expectations were highly correlated with futures prices.

10 160 Review of Agricultural Economics Table 6. Regressions between producer price expectations and alternative models ofprice expectation formation: a Coefficients Statistics Model b α β R 2 F Corn Dec. futures (0 47) c (0 174) Harvest cash (0 41) (0 158) Current cash (0 21) (0 072) 5-year average (1 32) (0 543) Adaptive (0 86) (0 306) Soybeans Nov. futures (0 67) (0 108) Harvest cash (0 59) (0 097) Current cash (0 24) (0 037) 5-year average (3 35) (0 568) Adaptive (1 67) (0 268) a Dependent variable is producer price expectation. b See text for modeldescriptions. c Standard error. They are not consistent with the experiments of Fisher and Tanner that suggest producers use weighted averages of historicalprices. Shideed and White found that corn price expectations based on futures prices were best at estimating corn supply response. However, they found that soybean price expectations based on the previous year s prices were superior to futures prices in estimating soybean supply response. Given the small difference in the correlation between current cash prices, futures prices, and to a lesser degree previous harvest prices, it is reasonable to assume that alternative hypotheses of producer price expectation formation based on these three models would produce similar statistical results in acreage response functions for corn and soybeans. In terms of economists working with producers, these results indicate producer price expectations can best be approximated by current cash prices or futures prices.

11 Producer Ability to Forecast Prices 161 Summary Producer price expectations in January or February of each year have large errors compared to actualharvest prices. From 1991 to 1998, the average expectation of producers missed the harvest price by $0.41 per bushelfor corn and $0.67 a bushelfor soybeans. Price errors of this magnitude generate revenue errors that are equalto or greater than typicalprofits per acre in Virginia. Producers had a strong tendency to overestimate low prices and underestimate high prices. The futures market during this time period did no better than producers at forecasting harvest corn and soybean prices. Producers have a wide range in price expectations at a given point in time. For corn, producer expectations in January and February consistently varied by as much as $1.00 per bushel, while soybean expectations varied by $1.70 per bushel. Less than 5% of the producers correctly anticipated within a $0.20 to $0.25 price range the actualharvest price in five of the eight years. Soybean producer price expectations were frequently bimodal or highly skewed toward higher prices. These skewed distributions of soybean expectations may be related to weather uncertainties associated with both the South American and United States crops. Producer price expectations were most highly correlated with current cash prices, futures prices, and previous harvest prices. Price expectations were equally correlated with current cash prices and futures prices, and less correlated to harvest prices the previous year. These results should be viewed with some caution because they are only based on eight years of data, but this is a larger data set of actualexpected producer prices than that used in prior research. Each producer s distribution of prices was skewed to the right each year. Producer expectations of price increases were consistently $0.20 to $0.30 a bushel greater than their anticipated price decreases compared to the most likely price. Over 50% of the actualharvest corn prices were in the right or left tailof the individualproducer s expectations in 1992, 1994, 1995, and For soybeans, the harvest price was in the tailof 35% to 68% of the producer distributions in five of the eight years. These results indicate that producers are substantially underestimating the probability of large price changes. Implications Producer price expectations for corn and soybeans were not substantially better or worse that the futures market estimates of harvest price. Since severalresearch studies (Kastens, Schroeder, and Plain; Brorsen and Irwin; O Brien, Hayenga, and Babcock; Zulauf and Irwin) report that econometric model forecasts cannot outperform the futures market, it appears that using current futures market estimates of harvest price of corn and soybeans is a good alternative for economists providing outlook information. Futures market quotes are readily available to producers, they are updated each day (unlike most advisory services, USDA estimates, and econometric models), and they are relatively inexpensive. Of course, producers need to understand they are not very accurate at forecasting harvest price 6 to 12 months in advance. But many producers have price expectations that vary substantially from the futures market prices estimates. Extension outlook programs should encourage producers with expectations substantially different than the futures market to carefully evaluate their expectations. Outlook presentations

12 162 Review of Agricultural Economics should concentrate on the potential price impact of changing acres, yields, and use. Over time such presentations should improve producers analytical skills and understanding of the markets. The producers in this study substantially underestimated the potential change in price from January/February untilharvest. When producers underestimate price risk, they may not obtain adequate downside price protection through forward pricing or crop insurance programs. Underestimation of downside price risk is a serious potential problem for lending agencies, since producers may get into financial difficulty much more rapidly than anticipated. These risk considerations indicate that a substantialportion of extension risk management programs should concentrate on the magnitude of price risk and their economic implications for the farm business. Economists may not be able to consistently forecast commodity prices better than the futures market, USDA, or private consulting firms, but as educators we should be able to point out the full impact of price risks and the advantages and disadvantages of various methods of managing these risks. Most educators assume farmers understand the substantialamount of risk involved in farming, but one of the most important findings of this study is that producers still substantially underestimate the price risks associated with producing corn and soybeans. References Brorsen, B.W., and S.H. Irwin. Improving the Relevance on Price Forecasting and Marketing Strategies. Agricultural and Resource Economics Review 25 (1996): Eales, J.S., B.K. Engel, R.J. Hauser, and S.R. Thompson. Grain Price Expectations of Illinois Farmers and Grain Merchandisers. Amer. J. Agr. Econ. 72 (August 1990): Fisher, B.S., and C. Tanner. The Formulation of Price Expectations: An Empirical Test of Theoretical Models. Amer. J. Agr. Econ. 60 (May 1978): Gardner, B.L. Futures Prices in Supply Analysis. Amer. J. Agr. Econ. 58 (February 1976):81 4. Heady, E.O., and D.R. Kaldor. Expectations and Errors in Forecasting Agricultural Prices. J. Pol. Econ. 62 (February 1954):34 7. Houck, J.P., and M.E. Ryan. Supply Analysis for Corn in the United States: The Impact of Changing Government Programs. Amer. J. Agr. Econ. 54 (May 1972): Kastens, T.L., T.C. Schroeder, and R. Plain. Evaluation of Extension and USDA Price and Production Forecasts. J. Agr. Res. Econ. 23 (July 1998): Kenyon, D.E., and R.S. Evans. Short-Term Soybean Acreage Projection ModelIncluding Price and Policy Impacts. Virginia Agr. Res. Bul. 106 (1975). Kenyon, D.E., E. Jones, and A. McGuirk. Forecasting Performance of Corn and Soybean Harvest Futures Contracts. Amer. J. Agr. Econ. 75 (May 1993): Nerlove, M., The Dynamics of Supply: Estimation of Farmers Response to Price. Baltimore: The Johns Hopkins Press, O Brien, D., M. Hayenga, and B. Babcock. Deriving Forecast Probability Distributions of Harvest-Time Corn Futures Prices. Rev. Agr. Econ. 18 (May 1996): Schroeder, T.C., J.L. Parcell, T.L. Kastens, and K.C. Dhuyvetter. Perceptions of Marketing Strategies: Producers versus Extension Economists. J. Agr. Res. Econ. 23 (1, July 1998): Shideed, K.H., and F. White. Alternative Forms of Price Expectations in Supply Analysis For U.S. Corn and Soybean Acreages. West. J. Agr. Econ. 14 (December 1989): Virginia Department of Agriculture and Consumer Affairs. Virginia Agricultural Statistics, Zulauf, C., and S.H. Irwin. Market Efficiency and Marketing to Enhance Income of Crop Producers. Rev. Agr. Econ. 20 (Fall/Winter 1998):