Competition and prices in USDA commodity procurement.

Plato, Gerald E.

1. Introduction

The United States Department of Agriculture (USDA) is a major food buyer, spending about $3.4 billion in 1999. Over 70% of those expenditures supported international food distribution programs, whereas about $970 million was spent on products for domestic feeding programs, such as the National School Lunch Program.

USDA food procurement relies primarily on auction mechanisms designed to induce "hard" manufacturer price competition, in Sutton's sense (Sutton 1991). The Department limits product differentiation by designing precise product specifications and by requiring USDA labels in place of commercial brands. It runs first-price, sealed-bid auctions each month for products to be delivered in the following month. The Department also reserves the right, albeit rarely exercised, to cancel auctions whose winning bids exceed unannounced reservation prices.

Our report evaluates those auctions. (1) We compare private sector prices to auction bids, and demonstrate that the auctions do obtain products at low prices. But auctions frequently attract only a few bidders, and falling bidder numbers concern policymakers. We therefore assess the effects of changes in the number of bidders on prices in the auctions, and find that bidder numbers matter--prices fall as the number of bidders increases. We also investigate nonlinearity--whether the effect of an additional bidder varies with the number of bidders--and find that bids fall more as bidder numbers increase, when bidder numbers are small (one or two) to begin with.

The results have some general implications. First, they provide evidence on the effects of competitor numbers on prices in markets with easy entry and no product differentiation, that is, in real-world markets that might arguably be called contestable. Second, nonlinearity is important when one wants to evaluate the likely effects of entry, exit, and merger on prices. In evaluating the likely effects of such events on prices, the focus falls to specific questions about the likely change in price attendant upon a move from three competitors to two or from four to three. Because it relies largely on aggregated data, the existing empirical literature on concentration and pricing is rarely able to consider the effects of incremental changes in competitors in already concentrated markets, but that issue is often a key question in policy evaluations.

This study therefore follows on the issues in the collection edited by Weiss (1989)- whether and how concentration changes affect price, particularly in highly concentrated markets. Much of the existing literature on the topic focuses, because of data availability, on industries with histories of regulation, such as commercial banking, railroads, and airlines. But a few empirical studies investigate markets that use auctions.

Auctions are widely used in agriculture and food markets. Examples include import tenders, sales of feeder cattle through video auctions, milk procurement by school districts, and priority rights for rail service for hauling grain. Studies of the role of competition in those markets generally find that bidder numbers matter. Two studies looked at school district milk procurement auctions, in Kentucky (Durham and Babb 1997) and Ohio (Porter and Zona 1999). Each found statistically significant but small effects of competition; Durham and Babb found that prices in Kentucky milk auctions fell by 1.6% with each additional bidder, whereas Porter and Zona found strikingly similar results for Ohio auctions--low bids fell by 1.5% with each additional bidder. Each found only limited evidence of nonlinearity, but nonlinearity may be hard to detect in these markets because bidder numbers are tightly circumscribed; 66% of Kentucky auctions and 97% of Ohio auctions had three or fewer bidders. (2)

Meyer (1988) investigated 1970s Texas rice auctions, in which mills bid to buy rice from growers. He found effects that were statistically significant and fairly large--each additional buyer in an auction drove bids up by 4.6%. He did not test for nonlinearity, and did not summarize the actual variation in bidder numbers across auctions.

Wilson and Diersen (1999) studied bidding by firms seeking to win Egyptian import tenders for the sale of three commodities: sunflower oil, cottonseed oil, and palm oil. Bids fell as the number of rival bidders (sellers) increased. The declines were very small, 0.6%, as a second bidder enters, and showed evidence of nonlinearity in that the effect of a fourth or fifth bidder was distinctly less than that of a second.

Bailey, Brorsen, and Thomsen (1995) reported on the effect of buyer concentration on winning bids for the purchase of feeder cattle in 103 video auctions, covering almost 3 million cattle, in the 1987-1992 period. They used a measure of geographic bidder concentration, the degree to which purchases from a particular county are dominated by buyers from one feeding area, and found that winning bids were lower in those counties facing geographically concentrated buyers. The effects, although statistically significant, appear to be fairly small, with geographic monopsony leading to price reductions of 1.1-1.7%.

Our analysis is based on extensive interviews with auction participants and empirical analyses of a large data set of USDA auction records for five commodities. These five commodities display substantial temporal and cross-sectional variation in bidder numbers, allowing for closer investigation of the effects of competition on bids.

2. The Design of USDA Procurement Auctions

The Department's clients include school systems and social service agencies. The clients receive an annual budget allocation and product information from USDA, and then specify products, package sizes, quantities, desired delivery dates, and destinations in delivery orders sent to the auction operator, USDA's Kansas City Commodity Office (KCCO). KCCO assembles delivery orders into contracts, each to be let in a separate auction (a contract often aggregates orders from different clients for a common item and location). Each contract specifies a precise product type (for flour, the type could be bleached or unblenched, all-purpose or bakery), a package size (such as 5-, 10-, 50-, or 100-lb. bags), a delivery location, and a delivery time window (a two-week period). Contracts specify quantities in truckloads, as well as any distinctive transportation or packaging requirements.

When preparing bids, manufacturers (vendors, in USDA nomenclature) take account of product costs, including labor, energy, packaging, and agricultural inputs, as well as expected competition. Bids also reflect distinctive production requirements, such as USDA labels and required inspections. Because bidders provide transportation to the specified locations, bids take transportation costs into consideration.

Vendors enter sealed bids by a specific deadline, and KCCO selects the lowest bid, subject to certain qualifications. Winning bids that exceed KCCO reservation (maximum) prices are reviewed. (3) The agency may decide to accept the bid, or it may reject the bid and cancel the contract. Orders in cancelled contracts may be reintroduced in the next month (with late delivery to clients), or KCCO may rebid a supplemental contract for the same month.

In summary, a USDA auction typically calls for bids for the delivery of a precise volume (40,000 lb., for example) of a specific product (all-purpose flour, with precise product specifications and tests), with specified packaging (10-lb. bags), to a particular destination (a warehouse in Buffalo), within a rather short time window.

In the lexicon of auction literature, these auctions are more like independent private value (IPV) auctions than common value auctions; any one vendor's valuation of the award is statistically independent of other vendors' valuations, because of vendor-specific differences in material and production costs. Winning bids in IPV procurement auctions should vary inversely with the number of bidders; further, bids should vary nonlinearly with bidder numbers, with bigger effects when bidder numbers are small (McAfee and McMillan 1987).

3. The Data and the Model

Our data set consists of almost 25,000 records of procurement auctions for five commodities (all-purpose flour, bakery flour, pasta products, vegetable oil, and peanut butter) carried out on a monthly basis between January 1992 and December 1996. (4) Table 1 provides summary statistics by commodity and year. The first row in each panel reports the annual number of auctions, which fell sharply between 1992 and 1996. Because some commodity purchases are linked to federal farm programs, which emphasize the purchase of commodities designated as surplus, the overall farm economy and farm policy affect procurement volumes. Domestic food procurement peaked in 1983, a year of large grain and dairy surpluses, at about $2.2 billion, before falling to $800 million in 1996.

The second row shows mean low bids (dollars per hundredweight) in each year. Bids rose sharply between 1992 and 1996 for wheat-based products as wheat prices rose (average wheat prices received by U.S. farmers rose from $3.24 per bushel in 1992 to $4.30 in 1996). Soybean and peanut prices fluctuated sharply during the period, just ahead of observed bid fluctuations for cooking oil and peanut butter.

The third row in each panel reports the median number of bidders across auctions for a given commodity in a given year, whereas the fourth row reports the percentage of auctions that attracted only a single bidder. Each measure varies sharply across commodities. At one extreme, the median number of bidders in peanut butter auctions varied between five and six, with no single bidder auctions. By contrast, vegetable oil, pasta, and bakery flour auctions frequently attracted only a single bidder, and the median number of bidders hovered around two. Bidder numbers also vary over time; the four commodities with single bidder auctions show wide temporal variations in the share of single-bidder auctions from year to year.

Auction bids vary widely over time and across places. Our analysis seeks to identify the effects of competition on low (winning) bids, while controlling for the effects of other factors affecting bids. Other factors include the prices for related agricultural commodities (changes in wheat prices affect flour costs), specified product characteristics (larger package sizes generate lower per-pound costs, and lower bids), delivery points (it costs more to place peanut butter in Maine than in Georgia, where the raw commodity is grown), and purchase volumes.

We assess the factors driving variations in bids in USDA procurement auctions by estimating regressions for each of the five commodities. The models were of the following general form:

BID = f ([PRICE.sub.ag], CHAR, VOL, COMP),

where BID is the lowest bid in the auction, expressed in dollars per hundredweight; [PRICE.sub.ag] is a vector of prices for related agricultural products; CHAR is a vector of product and auction characteristics; VOL is a vector of scale measures; and COMP is a vector of competition measures.

All continuous variables are expressed in natural logarithms. Bids ought to be affected by several important cost considerations, including agricultural prices, delivery locations, and product characteristics. In [PRICE.sub.ag], the series of primary interest is the bid month average cash price for the related agricultural product. (5) We also entered changes in cash prices between the bid month and the following (delivery) month, and changes between the bid month and prior month. (6)

Product and auction characteristics (in CHAR) should matter. Costs should be lower for larger package sizes. (7) Different commodity varieties may also affect costs. (8) We enter separate dummy variables capturing the available variety/package size combinations. Finally, since vendors are responsible for arranging transportation to the client's designated warehouse, we use fixed state effects (for destination states) to capture transportation costs.

Scale effects (in VOL) may matter. First, unit transport costs should decline up to truckload sizes, and may decline with larger quantities, so we include the number of truckloads in a contract (sample quantities vary from one to five truckloads). Second, because it might cost more to ship to lightly used destinations, we entered the total volume of a commodity directed to the specific destination over the five years of the data Finally, total monthly USDA purchase volumes could affect bids. We separately enter monthly purchase volumes for domestic and, where appropriate, international distribution (the Department purchases flour and vegetable oils for international distribution). Because monthly purchase volumes may have nonlinear effects on bids, we also enter squared purchase volumes.

We capture competition (COMP) by entering separate dummy variables for each level of the number of bidders (that is, if there were two bidders, a dummy variable for two bidders would be set to one, and dummies representing other bidder numbers would be set to zero). The specification allows for nonlinearity in the effect of bidder numbers on bids. In contrast, if we simply entered a count of the number of bidders, we would be forcing the marginal effect of additional bidders to be constant.

4. Summary of Regression Results

Appendix Table A1 reports the full regression results. We first summarize the results, then apply the models to the issues at hand. (9) In general, the models show strong fits, with [R.sup.2] coefficients ranging between 0.76 and 0.96 for four commodities (peanut butter, with much weaker measures of monthly agricultural input prices, is noticeably lower, at 0.54). Most coefficients have the expected signs and are statistically significant. Agricultural prices have powerful effects. All current month cash price coefficients are positive, highly significant, and large. Since we don't know the shares of agricultural commodity prices in processed product costs, we cannot precisely test for the degree of pass-through of commodity prices to processed product prices, but the long-term effects are qualitatively consistent with full pass-through. For example, a 10% increase in wheat prices would lead to an estimated 7.8% increase in all-purpose flour bids, consistent with the large factor share that wheat holds in flo ur production. By contrast, pasta bids would only increase by 3.6% in response to a 10% increase in durum wheat prices (pasta embodies another level of processing beyond flour). The coefficients on short-term changes in cash prices, holding current cash price constant, are very small. They suggest, in line with manufacturer interviews, that bid prices react much more strongly to long-term commodity price movements, and that bidders do not pass short-term commodity price fluctuations through to bid prices.

The models show strong locational effects (unreported in the tables). Bid prices rise sharply for delivery to distant states. Transport costs should account for higher shares of delivered prices for less processed products, and those products (flour) have larger state effects. (10) Finally, package sizes matter (larger sizes get lower bids), and variety matters, sometimes by large amounts (reduced-fat peanut butter carries a 38% price premium).

Why Use Procurement Auctions? Low Bids Compared with Private Sector Prices

Because USDA sets tight product specifications and requires specific packaging, bidders compete on homogeneous products. The Department relies on frequent and multiple sealed-bid auctions, and may cancel auctions with unacceptably high winning bids. These techniques are designed to intensify price competition and hence to allow acquisition of products at favorable prices. Evidence presented below indicates that USDA realizes substantial price discounts when compared with private sector prices paid for products of comparable quality.

To compare USDA and private sector prices, we solicited the cooperation of a major food service wholesaler. Such firms take warehouse deliveries from food manufacturers, and provide food products and support services to restaurants and fast food chains, schools, commercial kitchens, hospitals, and other large providers. Food service warehouses are the appropriate point of price comparison since they are located at the same level of the distribution chain as the state and commercial warehouses that receive USDA commodities from manufacturers.

The cooperating firm provided us with data on 1996 manufacturer prices for delivery of truckload quantities to the firm's warehouses. We obtained prices for the highest-quality "private-label" product because that was most comparable. (11) We then compare manufacturers' price quotes to predicted manufacturer low bids from our auctions, delivered to the same states during the same months. Predicted auction bids are based on mean state-specific values for volumes and competition and actual values for agricultural prices. USDA product prices vary by geographic region and by season, so we asked for prices for two time periods (April and September, 1996) and for three states served by the firm (Texas, Illinois, and Massachusetts). (12)

Table 2 compares prices. USDA prices generally fall well below corresponding private sector prices, with typical gaps in the range of 20-40%. Prices narrow only for flour products in the fall, a period when seasonal increases in industry demand lead to noticeable increases in auction bids (this is the only price gap that is not statistically significant). USDA auction mechanisms apparently generate substantially lower prices.

These differences are quite large, and we emphasize that the prices refer to comparable products and services: manufacturers' prices for truckload delivery of high-quality basic food products to a warehouse. The cooperating firm asserts that clients wishing to make significant volume commitments can sometimes obtain lower price quotes from manufacturers, and manufacturers sometimes offer lower prices to certain classes of buyers. But the cooperator price quotes used here represent an appropriate basis for comparison, because we compare cooperator prices with average (predicted) auction prices, not to the lowest observed auction prices, and because the auctions do not entail long-lasting volume commitments. Moreover, the classes of buyers that may be offered lower prices are frequently those with opportunities to purchase through the Agriculture Department, such as local school systems and social service agencies.

The evidence in Table 2 supports a simple proposition: USDA auctions induce low prices. We use it to argue, later in the paper, that competition in the form of bidder numbers matters for prices even in such environments. We do not suggest that private sector procurement practices are suboptimal, or that such firms should follow the government's strategies, but rather that USDA's particular strategies lead to lower prices.

In judging comparative prices, readers should recall some key distinctions between government and private-sector procurement practices. First, the government's strategies suggest that it may be more price sensitive, as a matter of policy, than food service providers. USDA aims to obtain large quantities of a few basic food products, at the lowest possible manufacturers' prices, subject to product specifications. Price is the primary goal, even at the expense of service quality. (13) But private-sector food service firms aim to deliver a wide variety of branded and private-label food products to clients on a timely basis; their clients target several goals, and likely place significant emphasis on the food service provider's variety, timeliness, and responsiveness. USDA and the food service provider each buys the same bundle of goods and services from vendors--delivery of truckload quantities of a private-label product to a warehouse within a specified time window. But because the food service firm then provid es a different package of products and services to its clients, the clients (and the food service provider) are less sensitive to manufacturer prices than is the government. Second, in pursuit of its primary goal, USDA aims to identify and select that vendor with the lowest short-run marginal costs of filling a contract. Marginal processing costs can vary substantially across plants because of differences in plant design, location, and capacity utilization.

Effects of Competition on Low Bids

The procurement auctions are designed to develop aggressive price competition, and Table 2 shows that auction prices fall well below corresponding private-sector prices. Do the number of competitors matter in such an environment? Table 3 reports the results for analyses of low bids in auctions for all-purpose and bakery flour. Three bidders is the base (excluded) variable, so the coefficients report the effects on the low bid as the number of bidders changes from three, and reported t-statistics test for significance compared to the three-bidder base.

Consider first the findings for all-purpose flour. Bidder numbers matter. Low bids increase by statistically significant amounts as the number of bidders falls to two and then to one, and bids decline by small but statistically significant amounts as bidder numbers rise from three to four, five, and six. Moreover, the effects are nonlinear. Low bids change by much more as we go from two bidders to one (8.5%) than they do as we go from three to two (1.6%) or from four to three (1.5%). (14) Similar results obtain for bakery flour; there, single-bidder auctions result in low bids about 8.9% above those in three-bidder auctions, whereas low bids in two-bidder auctions are 2.9% above three-bidder estimates.

The row headed "t-test ordering" summarizes the results of pairwise t-tests for differences in the coefficients on bidder numbers, and requires some explanation. The inequality sign indicates that the difference in coefficients is statistically significant, in the direction indicated. (15) A string of inequality signs before a semicolon ([b.sub.1] > [b.sub.2] > [b.sub.3]) indicates strict dominance: Coefficient [b.sub.1] is significantly greater than [b.sub.2], which in turn exceeds [b.sub.3]. Equality signs indicate that there is no significant difference between coefficients.

For all-purpose flour, strict dominance of t-tests holds across categories from one through four bidders (Table 3). The one-bidder effect ([b.sub.1]) was significantly greater than the two-bidder ([b.sub.2]) effect; each was significantly greater than the three-bidder ([b.sub.3]) effect, and all three were significantly greater than the effects for four ([b.sub.4]) and all greater bidder number effects. Tests among coefficients with many bidders were more mixed. Although [b.sub.4] significantly exceeded five ([b.sub.5]) and seven ([b.sub.7]) bidders, it was not significantly different from the six-bidder ([b.sub.6]) effect, which was itself not significantly different from [b.sub.5] or [b.sub.7]. In bakery flour, strict dominance holds from one through four bidders, although there is no significant difference in low bids between four- and five-bidder auctions.

Our analyses treat the number of bidders as exogenous. That's problematic, in that bidder numbers may be correlated with the error term in our models. That is, unusually profitable auctions, resulting in high bids relative to costs (errors to the econometrician, but perhaps explicable to the bidder), might attract entry, leading to increases in bidder numbers. Ideally, we would deal with that problem by finding an instrument for bidder numbers, highly correlated with bidder numbers, yet uncorrelated with the error term.

Effective instruments are hard to find, but flour provides the best possibility. U.S. Census reports (Current Industrial Reports) provide monthly industry data on capacity and production, from which one can generate estimates of capacity utilization. Because USDA purchases account for a very small share of industry-wide shipments, industry capacity utilization measures should be exogenous to decisions on government auctions. Capacity utilization varies widely over time, and industry participants tell us that excess capacity is an important determinant of the decision to bid on USDA products (because they are low-margin contracts).

We obtained an instrument for bidder numbers in flour auctions by taking predicted values from a regression of bidder numbers on industry capacity utilization and sets of dummy variables for product characteristics, destination states, and year. (16) We then used the instruments in analyses of low bids; the instrumental variable (IV) estimates are also reported in Table 3. The earlier qualitative conclusions hold, in that bidder numbers matter and the effects appear to be nonlinear. The spread between lowest and highest bids in single-bidder and many-bidder auctions widens with JY estimation, to 15% from 10% in bakery flour and to 15% from 14% in all-purpose flour.

The broad patterns--that bidder numbers matter and that monopoly matters most--hold for the other commodities as well (Table 4). Pasta and vegetable oil show statistically significant increases in low bids as the number of bidders falls from three to two and to one, and decreases in bids as bidder numbers increase from three to four and five. The largest individual price effects occur as bidder numbers fall from two to one. Moreover, pasta and vegetable oil show strict order dominance: Each number of bidders coefficient is significantly greater than all effects involving greater numbers of bidders.

Peanut butter, which has no single-bidder auctions, displays the only violation of the expected relation between bidder numbers and price (Table 4). Low bids fall as the number of bidders falls from three to two. Peanut butter bids do increase steadily, by statistically significant amounts, as the number of bidders falls from six to five, from five to four, and from four to three.

There are few two-bidder peanut butter auctions (44, about 0.8% of all peanut butter auctions). Two-bidder auctions overlap with reduced-fat (rf) auctions; most two-bidder auctions are for rf peanut butter, and all but two rf auctions attract only two bidders. Low bids in rf auctions exceed those in other peanut butter auctions by nearly 38%, on average (Appendix Table A1). The small number of two-bidder auctions, and the high correlation with rf contracts, may account for the low coefficient value and high standard error on two-bidder auctions.

The full range of the price effect, from many bidders to one, covers a rather tight range. The maximum price increase, as one goes from many bidders to a single bidder, ranges from 8.4% to 15.0% across the four commodities with monopoly auctions. In peanut butter, where there are no monopoly auctions, the full price effect is 9.8% as one goes from many bidders to three. (17) The results suggest that auctions can induce comparatively low prices even in single-bidder auctions, but that competition continues to have important effects in this environment.

5. Summary and Conclusions

Agriculture Department auctions are designed to obtain large quantities of basic food products at very low prices, by limiting product differentiation and encouraging active price competition. The strategy appears to be successful, albeit at the possible expense of service quality and other goals. The auctions generate low bids that generally fall substantially below private-sector prices for comparable goods.

Yet even in this environment, the number of competitors matters. As the number of bidders declines, the low bids in USDA auctions increase. The largest increases occur as bidder numbers fall from two to one; in the four commodity samples with single bidder auctions, low bids rise by 4.2-8.3%, depending upon commodity and specification. Low bids continue to change, by small but statistically significant amounts, as bidder numbers in our auction samples increase from two to three, four, five, six, and seven bidders. In comparing auctions with the fewest bidders with those with the most bidders in a given commodity category, the aggregate effect of competition ranges from 8.4 to 15.0%, again depending on specification and commodity.

The evidence offered above echoes some earlier summary conclusions in Weiss (1989) in two respects. After a review of a wide variety of studies, Weiss concluded that the effects of concentration, even monopoly, on price were modest and indeed much smaller than he expected. That conclusion holds here; although the effects of changes in bidder numbers are statistically significant, they are not very large.

But Weiss also pointed out that his evidence referred largely to markets that we expect to be reasonably competitive, because of their size and because of the presumed effectiveness of antitrust laws. Similarly, the markets studied here are designed to be contestable; The auctions are designed to generate competing bids on homogeneous products, and entry isn't barred. Yet even here, in a highly transparent bidding environment, competition does matter.

Appendix A

Detailed Regression Results, Low Bids in USDA Procurement Auctions.
Regression Coefficients and t-Statistics are in Parentheses

 All-Purpose
 Flour Bakery Flour Pasta

Agricultural prices
 Cash Price, bid month (1) 0.7755 0.6423 0.3627
 (92.64) (58.32) (62.25)
 Month ahead change 0.2098 0.0971 -0.0018
 (13.16) (2.81) (0.11)
 Month behind change -0.7888 -0.5480 0.3687
 (40.26) (20.01) (20.18)
 Cash price, bid month (2) -- -- --

 Month ahead change -- -- --

 Month behind change -- -- --

Product characteristics
 Base 5-lb. bag bulk 20-lb. spag
 10-lb. bag -0.0207 -- --
 (8.60)
 50-lb. bag -0.0543 0.1165 --
 (19.23) (19.21)
 100-lb. bag -0.0757 0.0852 --
 (2.93) (15.43)
 Bleached 0.0022 -0.0072 --
 (0.30) (2.08)
 Hearth -- 0.0689 --
 (13.29)
 Spaghetti, 2 lb. -- -- 0.0264
 (6.68)
 Macaroni, 20 lb. -- -- 0.0216
 (4.62)
 Macaroni, 1 lb. -- -- 0.0811
 (23.59)
 Rotini, 20 lb. -- -- 0.0760
 (23.36)
 Vegetable oil, 48 oz. -- -- --

 Vegetable oil, bulk

 Shortening, bulk -- -- --

 Shortening, 50 lb. -- -- --

 Shortening, 1 gal.

 Peanut butter, no. 10 can -- -- --

 Reduced fat peanut butter -- -- --

Auction volume
 Truckloads in contract 0.0012 -0.0018 -0.0087
 (1.29) (0.73) (2.09)
 Total volume at location -0.0001 -0.0071 -0.0014
 (0.10) (3.18) (1.23)
 Monthly domestic volume -0.7099 0.1357 -0.0376
 (18.23) (2.17) (12.21)
 Monthly international volume -0.3175 0.1014 --
 (14.92) (10.21)
 Domestic volume squared 0.0122 -0.0025 --
 (9.64) (1.23)
 International volume squared 0.0001 -0.0013 --
 (0.98) (10.82)
 Domestic * international volume 0.0200 -0.0054 --
 (14.64) (7.62)
Competition
 One bidder (three 0.0956 0.0850 0.0586
 is base) (13.55) (13.07) (18.80)
 Two bidders 0.0155 0.0289 0.0170
 (4.38) (7.86) (6.50)
 Four bidders -0.0186 -0.0144 -0.0258
 (7.27) (4.52) (7.73)
 Five bidders -0.0305 -0.0072 -0.0546
 (11.13) (1.97) (12.66)
 Six bidders -0.0263 -- --
 (8.39)
 Seven bidders -0.0358 -- --
 (10.25)
Observations 5726 1711 4660
[R.sup.2] 0.80 0.82 0.77


 Cooking Oil Peanut Butter

Agricultural prices
 Cash Price, bid month (1) 0.4831 0.5005
 (74.36) (26.16)
 Month ahead change 0.1419 --
 (16.33)
 Month behind change -0.1790 --
 (21.39)
 Cash price, bid month (2) 0.3570 --
 (61.97)
 Month ahead change 0.1878 --
 (18.02)
 Month behind change -0.2550 --
 (23.47)
Product characteristics
 Base 1 gal. veg. oi 21 lb. regula
 10-lb. bag -- --

 50-lb. bag -- --

 100-lb. bag -- --

 Bleached -- --

 Hearth -- --

 Spaghetti, 2 lb. -- --

 Macaroni, 20 lb. -- --

 Macaroni, 1 lb. -- --

 Rotini, 20 lb. -- --

 Vegetable oil, 48 oz. 0.2987 --
 (66.01)
 Vegetable oil, bulk -0.2484
 (104.85)
 Shortening, bulk 0.1816 --
 (124.26)
 Shortening, 50 lb. -0.0194 --
 (6.25)
 Shortening, 1 gal. 0.0473 --
 (26.12)
 Peanut butter, no. 10 can -- -0.0125
 (6.97)
 Reduced fat peanut butter -- 0.3210
 (15.25)
Auction volume
 Truckloads in contract -0.0047 0.0021
 (3.13) (0.87)
 Total volume at location -0.0002 -0.0033
 (0.24) (3.64)
 Monthly domestic volume -0.3112 0.0518
 (7.59) (27.92)
 Monthly international volume -0.7709 --
 (16.70)
 Domestic volume squared 0.0068 --
 (5.39)
 International volume squared 0.0196 --
 (14.54)
 Domestic * international volume 0.0068 --
 (10.42)
Competition
 One bidder (three 0.0565 --
 is base) (26.60)
 Two bidders 0.0096 -0.0434
 (7.25) (2.47)
 Four bidders -0.0129 -0.0196
 (5.75) (4.44)
 Five bidders -0.0237 -0.0336
 (6.13) (7.28)
 Six bidders -- -0.0613
 (12.25)
 Seven bidders -- -0.1026
 (17.79)
Observations 7077 5242
[R.sup.2] 0.96 0.55

Dependent variable is the natural logarithm of the lowest bid.
Agricultural prices, total orders at a location, and volumes are in
natural logarithms, and truckloads takes on values from 1 to 5. All
other variables are dichotomous. Models also include fixed effects
state, and estimates in all- purpose and bakery flour models are
corrected for heteroscedasticity. See footnote 5 for related
agricultural price series, and note that cooking oil uses two
agricultural commodities, soybean oil and cottonseed oil (first set of
coefficient is for soybean oil). Because peanut prices only are
available for marketing year months, model includes last quoted monthly
price of marketing year for off-season prices, and then allows for
coefficient on that price to vary with the off-season month (the
adjuster variables).
Table 1

Selected Characteristics of USDA Commodity Auctions

 1992 1993 1994 1995 1996

All purpose flour
 Auctions (no.) 2518 1087 742 809 561
 Mean low bid 13.08 12.27 13.35 15.08 16.60
 Median no. bids 5 3 3 3 4
 Percent with 1 bidder 1.3 15.6 1.4 13.2 3.7
Bakery flour
 Auctions (no.) 722 342 264 231 141
 Mean low bid 12.27 12.50 14.02 14.81 16.81
 Median no. bids 2 2 1 3 3
 Percent with 1 bidder 27.8 13.8 62.9 12.6 19.6
Pasta
 Auctions (no.) 1076 1047 784 893 688
 Mean low bid 27.46 26.27 29.77 30.36 32.24
 Median no. bids 2 2 3 3 2
 Percent with 1 bidder 25.4 33.1 12.7 4.2 19.9
Vegetable oil
 Auctions (no.) 1821 1727 1650 1022 857
 Mean low bid 32.27 36.64 42.59 42.21 39.50
 Median no. bids 2 2 2 2 2
 Percent with 1 bidder 10.3 18.6 12.8 2.2 4.2
Peanut butter
 Auctions (no).) 1711 1348 1046 637 427
 Mean low bid 83.07 78.07 74.62 70.93 78.82
 Median no. bids 5 5 6 6 5
 Percent with 1 bidder 0.0 0.0 0.0 0.0 0.0

Source: USDA Farm Service Agency auction records data, described in text
and footnote 4.
Table 2

Comparing USDA Prices with Comparable Private-Sector Prices

 USDA Price/Private-Sector Price
Product and location April 1996 September 1996

Liquid shortening
 Texas 0.734 0.683
 IL/IN/WI 0.733 0.685
Vegetable oil
 Texas 0.712 0.683
 IL/IN/WI 0.764 0.718
 New England 0.670 0.552
All-purpose flour
 Texas 0.685 0.867
 IL/IN/WI 0.689 0.826
Bread flour-IL/IN/WI 0.793 0.965
Pasta-nationwide:
 Spaghetti n.r. 0.629
 Macaroni n.r. 0.621
 Rotini n.r. 0.622
Peanut butter-nationwide n.r. 0.824

n.r. means that no cooperator price was reported for that month. The
numberator in each cell is the predicted low bid from regressions
reported in Table A1, for single truckload shipments and state- specific
means for volume and competition variables. The denominator is the
manufacturer's price quote to a cooperating food service wholesaler for
truckload delivery of a comparable product to the locations listed.
Table 3

Effects of Bidder Numbers on Low Bids in USDA Flour Auctions

 All-Purpose Flour
Bidders OLS
(Base = 3)

1 0.0956
 (13.55)
2 0.0155
 (4.38)
4 -0.0186
 (7.27)
5 -0.0305
 (11.13)
6 -0.0263
 (8.39)
7 -0.0358
 (10.25)
t-test ordering [b.sub.1] > [b.sub.2] > [b.sub.3] >
 [b.sub.4] >; [b.sub.4] > [b.sub.5];
 [b.sub.4] > [b.sub.7];
 [b.sub.4] = [b.sub.6];
 [b.sub.5] = [b.sub.6] = [b.sub.7];
Obs. 5726

 All-Purpose Flour
Bidders IV
(Base = 3) Coefficients and t-Statistics

1 0.0928
 (8.08)
2 0.0532
 (11.61)
4 -0.0469
 (15.97)
5 -0.0436
 (9.90)
6 -0.0425
 (6.79)
7 -0.0440
 (4.44)
t-test ordering [b.sub.1] > [b.sub.2] > [b.sub.3] >
 [b.sub,4],
 [b.sub.4] = [b.sub.5] = [b.sub.6] =
 [b.sub.7]

Obs. 5726

 Bakery Flour
Bidders OLS
(Base = 3)

1 0.0850
 (13.07)
2 0.0289
 (7.86)
4 -0.0144
 (4.52)
5 -0.0072
 (1.97)
6 --

7 --

t-test ordering [b.sub.1] > [b.sub.2] > [b.sub.3] >
 [b.sub.4],
 [b.sub.4] = [b.sub.5]


Obs. 1711

 Bakery Flour
Bidders IV
(Base = 3)

1 0.0950
 (6.32)
2 0.0301
 (4.96)
4 -0.0130
 (1.93)
5 -0.0467
 (2.45)
6 --

7 --

t-test ordering [b.sub.1] > [b.sub.2] > [b.sub.3] >
 [b.sub.4] > [b.sub.5];



Obs. 1711

Reported coefficients from regressions explaining log of winning bid,
expressed as price per hundredweight. Full results are reported in
Appendix Table A1.
Table 4

Effects of Bidder Numbers on Low Bids in USDA Pasta, Vegetable Oil, and
Peanut Butter Auctions

Bidders
(Base = 3) Pasta

1 0.05867
 (18.80)
2 0.0170
 (6.50)
4 -0.0258
 (7.73)
5 -0.0546
 (12.66)
6 --

7 --

t-test [b.sub.1] > [b.sub.2] > [b.sub.3] >
 ordering [b.sub.4] > [b.sub.5];




Obs. 4660

Bidders
(Base = 3) Vegetable Oil

1 0.0565
 (26.60)
2 0.0096
 (7.25)
4 -0.0129
 (5.75)
5 -0.0238
 (6.13)
6 --

7 --

t-test [b.sub.1] > [b.sub.2] > [b.sub.3] >
 ordering [b.sub.4] > [b.sub.5];




Obs. 7077

Bidders
(Base = 3) Peanut Butter

1 --

2 -0.0434
 (2.47)
4 -0.0196
 (4.44)
5 -0.0336
 (7.28)
6 -0.613
 (12.25)
7 -0.1026
 (17.79)
t-test [b.sub.3] > [b.sub.4] > [b.sub.5] >
 ordering [b.sub.6] > [b.sub.i];
 [b.sub.2] = [b.sub.4], [b.sub.5],
 [b.sub.6];
 [b.sub.2] > [b.sub.1]; [b.sub.2] <
 [b.sub.3];
Obs. 5242

Reported coefficients from regressions explaining log of winning bid,
expressed as price per hundredweight. Full results are reported in
Appendix Table A1.

Received February 2001; accepted November 2001.

(1.) Two USDA agencies purchase food for Agriculture and State Department food support programs. The Farm Service Agency buys dairy, grain, and oilseed-based products, whereas the Agricultural Marketing Service buys fruit, vegetable, meat, and egg products.

(2.) Each study used data collected as part of price-fixing litigation, but the reports analyze data for districts where conspiracy was not alleged. Porter and Zona (1999) estimate that prices in rigged auctions rose by 6.5%, on average.

(3.) In practice, few bids exceed reservation prices (called constructed prices by KCCO). In the sample used in this study, low bids equaled or exceeded reservation prices in 5% of pasta auctions, but in only 0.3% of peanut butter and bakery flour auctions, and 0.1% of all-purpose flour and vegetable oil auctions.

(4.) Auction records were provided to us by the Farm Service Agency, and the data are available from the authors upon request. The five commodities chosen for study were closely tied to related agricultural products and had extensive and accurate auction records.

(5.) USDA (1998a and b) reports monthly average cash prices for agricultural inputs. We use no. 2 soft red winter wheat (St. Louis) for all-purpose flour, no.1 hard red winter wheat (Kansas City) for bakery flour, and no.1 hard amber durum (Minneapolis) for pasta. We use two crude vegetable oil prices: wholesale prices for crude soybean oil (Decatur) and crude cottonseed oil (Mississippi Valley points). USDA only reports peanut prices received by farmers for August through January, when peanuts are harvested and marketed. For peanut butter auction months between January and August, we used the January price (the most recent observed peanut price), hut the model allows the coefficient on the January price to vary with each succeeding month.

(6.) Bidders know current and past commodity prices, and have good forecasts of near future prices. In interviews, bidders explain that they rely on long-term trends in commodity prices and discount short-term spot market fluctuations. By using the current level along with month-before and month-after changes, we can distinguish the effects of short-term fluctuations in agricultural prices from longer-term shifts (results suggest that short-term fluctuations have limited effects on bid prices, whereas longer-term movements have strong effects).

(7.) USDA offers all-purpose flour in 5-, 10-, 50-, and 100-lb. bags, whereas bakery flour can be purchased in 50- or 100-lb. bags, or in bulk. Pasta comes in 1-, 2-, and 20-lb. boxes, whereas vegetable oil is obtained in 48-oz., 1-gallon, or 3-lb. containers, in 50-lb. cubes, or in bulk. Peanut butter comes in 2-lb. cans or #10 cans.

(8.) Clients can purchase vegetable oil or shortening; bleached or unbleached flour; macaroni, spaghetti, or rotini pasta; and reduced-fat or regular peanut butter.

(9.) Estimates were corrected for heteroscedasticity in the bakery and all-purpose flour samples, where Breusch-Pagan tests showed OLS and IV error variances to be inversely related to bidder numbers. Tests found no evidence of heterosecedasticity in the other samples.

(10.) The lowest bids are in Kansas for flour (wheat), Minnesota for pasta (durum wheat), Iowa for vegetable oil (soybeans), and Georgia for peanut butter (peanuts). Highest bids are in Maine (flour), New Hampshire (pasta), Nevada (vegetable oil), and Rhode Island (peanut butter).

(11.) According to our interviews, USDA sets consistently high product specifications, and the avoidance of commercial brands is consistent with a private label program.

(12.) The cooperating firm was able to provide us only with September data for peanut butter and pasta, and only on a nationwide price quote. We therefore compared those prices to nationwide average USDA prices for September 1996.

(13.) KCCO may cancel an auction. Although this strategy may make bidders more aggressive, it also results in late delivery to clients, since cancelled auctions aren't rebid until the following month.

(14.) In semilogarithmic regressions, the coefficient [beta] on a dummy variable only approximates the percentage change in the bid attendant upon a one-unit change in the dummy variable, and the approximation weakens for larger values (Halvorsen and Palmquist 1980). The correct percentage change, given in the text, is ([e.sup.[beta]] - 1).

(15.) The smallest "significant" t-statistic was 2.21 (97.5% confidence); most were much higher.

(16.) The model explained about 50% of the variation in bidder numbers in all-purpose flour auctions, and 40% in bakery flour regressions. Increases in capacity utilization led to reductions in bidder numbers, which also varied sharply across states and product types. Similar attempts outside of flour led to poor fits and weak instruments.

(17.) The range is calculated as the difference in predicted low bids between the maximum number of bidders (six in all-purpose flour, five in the others) and monopoly for those commodities with single-bidder auctions, and the difference in predicted low bids between six and three bidders in peanut butter.

References

Bailey, DeeVon, B. Wade Brorsen, and Michael R. Thomsen. 1995. Identifying buyer market areas and the impact of buyer concentration in feeder cattle markets using mapping and spatial statistics. American Journal of Aricultural Economics 77:309-18.

Durham, Catherine A., and Emerson M. Babb. 1997. The impact of the number of bidders and costs on school milk prices. Unpublished paper, Purdue University.

Halvorsen, Robert, and R. Palmquist. 1980. The interpretation of dummy variables in semilogarithmic equations. American Economi Review 70:474-5.

McAfee, R. Preston, and John McMillan. 1987. Auctions and bidding. Journal of Economic Literature 25:699-738.

Meyer, Donald J. 1988. Competition and bidding behavior: Some evidence from the rice market. Economic Inquiry 26: 123-32.

Porter, Robert H., and J. Douglas Zona. 1999. Ohio school milk markets: An analysis of bidding. RAND Journal of Economics 30:263-88.

Sutton, John. 1991. Sunk costs and market structure: Price competition, advertising, and the evolution of concentration. Cambridge, MA: The MIT Press.

United States Department of Agriculture. 1998a. Oil crops situation and outlook yearbook. Washington, D.C.: Economic Research Service.

United States Department of Agriculture. 1998b. Wheat situation and outlook yearbook. Washington, D.C.: Economic Research Service.

Weiss, Leonard W., ed., 1989. Concentration and price. Cambridge, MA: The MIT Press.

Wilson, William A., and Matthew A. Diersen. 1999. Competitive bidding on import tenders: The case of minor oilseeds. Department of Agricultural Economics, North Dakota State University, Agricultural Economics Report No. 428.

James M. MacDonald, * Charles R. Handy, + and Gerald E. Plato +

* Economic Research Service, U.S. Department of Agriculture, Washington, DC 20036, USA; corresponding author.

+ Economic Research Service, U.S. Department of Agriculture, Washington, DC 20036, USA.

Thanks are due to the editor and referees for useful comments and to Mark Denbaly and Bonnie Tanner for providing data and support for this project. The views expressed herein are our own, and not necessarily those of the U.S. Department of Agriculture.