Determining Returns to Storage: Does Data Aggregation Matter?

Journal of Agricultural and Applied Economics, Dec 2007 by Klumpp, Joni M, Brorsen, B Wade, Anderson, Kim B

Aggregate data are commonly used to determine returns to storage. However, recent studies have shown that aggregating data may lead to underestimated returns. This article compares aggregate and elevator data from Oklahoma to determine if aggregate data underestimate returns. We find no difference between the mean returns estimated with aggregate data and the mean returns estimated with transaction level data from grain elevators in Oklahoma.

Key Words: aggregate data, data collection, information loss, returns to storage

JEL Classifications: Q13

There has been a long debate in both finance and agricultural economics literature as to when it pays to store grain (i.e. Chang; Musser, Patrick, and Eckman; Zulauf and Irwin; Schroeder et al.). While researchers typically agree that there is little gain in trying to predict prices because markets are efficient, they disagree on how well producers are actually performing. For example, Anderson and Brorsen (2005) found that producers tend to perform above the market average. However, Hagedorn et al. found that farmers underperformed the market. An important difference between these two studies is the data used by the researchers. While Hagedorn et al. used aggregate data from the U.S. Department of Agriculture (USDA), Anderson and Brorsen used transaction-level data from grain elevators in Oklahoma. So could data aggregation explain the difference in the findings in these two studies?1

Brennan, Williams, and Wright argue that "no stocks are held at a monetary loss" and that "any apparent loss is an illusion from spatial aggregation" (p. 1009). Therefore, based on this view, what may appear to be storage at a loss in Hagedorn et al. could be the artifact of data aggregation (Benirschka and Binkley; Brennan, Williams, and Wright; Wright and Williams). An alternative explanation of why storage at a loss might occur is convenience yield (Kaldor; Working 1948, 1949).2 But the concept of convenience yield is being challenged since an inverse carrying charge can be explained by risk aversion (Chavas) and by transactions costs (Benirschka and Binkley; Brennan, Williams, and Wright; Chavas, Despins, and Fortenbery; Wright and Williams).

The objective of this article is to determine differences in returns to storage estimated with transaction-level data and returns to storage estimated with market-level data. Market-level data are from the Oklahoma Department of Agriculture data, and microlevel data are from three grain elevators in Oklahoma. We also look at frequency of sales over time to determine if producers continue to store after prices have peaked. Results indicate that storage at a loss does occur in Oklahoma and that it cannot be explained by data aggregation.

Theory

Equation (1) suggests that as long as c^sub A^ > 0 and c^sub B^ > 0, then the "cheapest" suppliers will sell first. Since transportation costs and returns to storage increase with distance, the "cheapest" suppliers are producers closer to the market. Thus, producers closer to the market sell early in the marketing season, and those farthest from the market sell later in the marketing year.

To illustrate how data aggregation underestimates returns to storage, we consider a two-period grain market supplied by two regions A and B. Region A is assumed to be closer to the market than location B, so the price at the closer location A is higher than that at the farther location B in both time periods (see Table 1). Assuming an interest cost of 10% and storage cost of $0.20 at both locations, net returns to storage at location A (-$0.01) are less than net returns at location B ($0.01). This is consistent with the notion that returns to storage increase with distance from the market. However, if all of location A sold in period 1 and all of location B sold in period 2 and the data are aggregated, then results are much different. The aggregate price is $3.20 in period 1 and $3.51 in period 2, and the net return to storage calculated with aggregate data is -$0.21. The example demonstrates how data aggregation may lead to underestimating returns to storage. But the interest costs assumed are larger than historical and the difference in prices across locations is larger than actually occurs within a state, yet the difference in storage costs is only two cents. Thus, the example also suggests that the effect of aggregation on data within a state may be small.

Data and Procedures

The microlevel data for this study come from three elevators located in the southern, central, and northern regions of western Oklahoma. The data span nine crop years, from the spring of 1992 through the spring of 2001 and record transactions of individual producer wheat sales at each elevator.3 Each transaction includes the number of bushels sold, the nominal price received per bushel, and the date of the sale.

Harvest is a three-week period with beginning and ending dates that vary by elevator as well as by year. The harvest start date was determined by reviewing the daily transactions that occurred around the end of May or beginning of June. The beginning harvest date was identified as the date when the number of bushels sold increased noticeably and stayed relatively high for an extended period of time. The southern elevator has an earlier harvest that typically begins around the end of May.4 Harvest at the central and northern elevators is slightly later, beginning around the first of June and the middle of June, respectively. Commercial storage is widely available near producing areas, and over 90% of the grain produced in Oklahoma is stored commercially (National Agricultural Statistic Service). Small amounts of grain are held on farm mainly for seed.