Economics Letters 107 (2010) 4245

Contents lists available at ScienceDirect

Economics Letters
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e c o l e t

Last-minute bidding in eBay charity auctions

Olin Business School, Washington University in St Louis, United States University of North Carolina, Chapel Hill, United States

a r t i c l e

i n f o

a b s t r a c t
Relative to typical eBay auctions, bidding in eBay charity auctions occurs earlier and also less frequently in the auctions' nal seconds. This suggests that some charity bidders benet from driving-up others' payments, as when auction revenue is a public good. 2009 Elsevier B.V. All rights reserved.

Article history: Received 16 April 2008 Received in revised form 12 December 2009 Accepted 15 December 2009 Available online 21 December 2009 Keywords: Charity auctions Bid timing Bid sniping Preferences for giving Online auctions JEL classication: D44 H41 L81

1. Introduction Consumers are frequently presented opportunities to purchase products that are linked to charitable causes, green production, or fair trade practices. In Elfenbein and McManus (forthcoming) we demonstrate that consumers are willing to pay a premium for these items by analyzing a hand-matched data set of identical items sold in charity and non-charity auctions on eBay. This premium may occur because consumers receive a private benet from winning a charity auction, such as egotistical warm glow (Andreoni, 1989), because bidders value charity auction revenue as a public good, or both.1 Models of bid timing in online consumer auctions suggest that bid times may reveal some aspects of bidder preferences, specically, whether bidders view charity auction revenue as a public good. In particular, sophisticated bidders who value overall auction revenue will be less likely to bid at the last minute (snipe) and may be inclined to bid earlier to elicit more bidding from nave bidders. While the data in Elfenbein and McManus (forthcoming) allow some coarse measurement of bid timing, they are insufcient to examine truly last-minute bidding, as sniping is generally described.
Corresponding author. Campus Box 1133, One Brookings Drive, St. Louis, MO 63130-4899, United States. Tel.: +1 314 935 8028; fax: +1 314 935 6359. E-mail address: elfenbein@wustl.edu (D.W. Elfenbein).
1 See Engers and McManus (2007) for a model of equilibrium bidding in charity auctions.

For the present paper we construct a new data set specically designed for this purpose. 2. Background Last-minute bidding is widespread in online consumer auctions. Leading explanations for late bidding include: delaying the release of private information in common value auctions (Bajari and Hortacsu, 2003), efforts to avoid price wars among strategic bidders (Roth and Ockenfels, 2002), and attempts to avoid having nave bidders drive up the price (Ockenfels and Roth, 2006; Ariely et al., 2005; Ely and Hossain, 2009). Ely and Hossain (2009) compare sniping (submitting one bid very late) to squatting (submitting one bid very early) in eBay auctions. They consider the optimal action of a strategic (sophisticated) bidder facing nave bidders who, rather than bidding their willingness-to-pay rst, follow incremental bidding strategies, starting at the current price and submitting a series of bids, each slightly higher than the last, until they either lead the bidding or learn that the current price exceeds their willingness-to-pay.2 By sniping, sophisticated bidders avoid giving nave bidders a high hurdle to clear when they choose to end their bidding, thereby reducing the price at which the sniping bidder can win the auction. Squatting, on the other hand, can benet a bidder if it deters some
2 See Elfenbein and McManus (forthcoming) for institutional details about the eBay bidding system which support this practice.

0165-1765/$ see front matter 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.econlet.2009.12.021

D.W. Elfenbein, B. McManus / Economics Letters 107 (2010) 4245 Table 1 Summary statistics on charity auctions. Charity auctions Obs. Auction level data Price ($) Donation (%) Number of bids Seller rating (/1000) Percent positive rating (%) Power seller (dummy) Length (days) High bid came in last 30 s (dummy) 1 min (dummy) Bidder-level data Time of highest bid before end of auction (days) Incremental bidder (dummy) Bidder Rating (/1000) 7654 7654 7654 7654 7654 7654 7654 7654 7654 Avg. 76.45 92.1 5.67 5.54 99.5 .330 7.04 .136 .171 Median 15.50 100 3 .85 99.7 0 7 0 0 Std. dev. 354.64 24.8 6.56 19.85 .8 .470 1.26 .343 .376 Non-charity auctions Obs. 18,347 18,347 18,347 18,347 18,347 18,347 18,347 18,347 18,347 Avg. 78.25 0 5.19 4.77 99.5 .347 6.26 .142 .179 Median 15.82 0 3 .737 99.9 0 7 0 0

43

Std. dev. 893.68 0 5.88 18.443 1.6 .476 1.69 .349 .383

21,298 21,298 20,831

2.34 .188 .290

.978 0 .077

2.67 .390 .765

50,629 50,629 49,584

1.79 .168 .239

.639 0 .067

2.27 .373 .619

other bidders from participating in the auction. Ely and Hossain show that an individual's private net benet from sniping and squatting is nearly equal; however squatting is associated with greater overall auction revenue and expected payments from other bidders. For bidders who care about auction revenue as a public good, these differences may then affect bid timing choices.

Finally, we note that more than 90% of observed charity auctions donate 100% of revenue to charity. About 7% of charity auctions donate 10%, and fewer than 3% dedicate a share of revenue between 15% and 95%, inclusive, to charity.

4. Analyses 3. Data We examine eBay charity listings ending in September and October 2006. We identied the 48 eBay-dened product categories in which charity auctions were most common, and then collected auction details and bidding histories for all charity listings in these categories. We then downloaded the same information for up to 2000 non-charity auctions in each category during the same months. We restrict our analysis to true auctions. We exclude all listings that did not result in a sale, buy-it-now listings, and all auctions that ended with prices below $2 or which lasted more than 10 days. Our data set contains about 26,000 completed auctions.3 Selected summary statistics on these data are presented in Table 1. Average and median prices of the charity and non-charity samples are similar. Completed auctions in the charity sub-sample received slightly more bids on average than those in the non-charity subsample. Seller attributes in the two groups are also similar, although charity auctions' sellers are slightly more experienced and slightly less likely to be a Power seller, an eBay designation for sellers with both high feedback scores and percentages of positive feedback. The median auction length for the two sub-samples is identical, although the average charity auction length is about three-quarters of a day longer than the mean non-charity auction. Between sub-samples there are clear differences in the timing of bidders' nal (i.e. highest) bids. The fractions of winning bids submitted in the last 30 s and 1 min are lower in the charity subsample (13.6% and 17.1%, respectively vs. 14.2% and 17.9% in the noncharity sub-sample), and bidders' nal bids occur earlier on average (2.34 days vs. 1.79 days, representing 33.2% and 28.5% of time remaining in the auction). Additionally, bidders are more likely to bid multiple times in charity auctions, despite being more experienced, on average, than bidders in non-charity auctions.
3 By contrast, the data in Elfenbein and McManus (forthcoming) contain 2433 auctions, organized around 723 distinct groups of auctions for identical products. The frequency of within-group variation in last-minute bidding is too low to permit analysis of true sniping within those data.

4.1. Days between bid and auction close We begin by estimating the lapsed time (in days) between a bidder's nal bid and an auction's closing. The econometric model is:
daysij = c + CHARITYj + AUCTIONj + SELLERj + BIDDERi + ij :

The dependent variable daysij is the number of days between the nal bid of bidder i in auction j and the auction closing time. CHARITYj is a vector of indicator variables for whether an auction donates 10%, 100%, or some share strictly between 10% and 100%. These dummy variables are, respectively, 10%-SHARE, 100%-SHARE, and MID-SHARE. AUCTIONj is a vector of auction characteristics including dummy variables for the scheduled duration, ending time of day (in four-hour blocks), and ending day of the week. SELLERj describes the seller's percentage positive rating, logged feedback score, and Power seller status. The vector BIDDERi includes quadratic terms for the feedback score of bidder i to control for prior experience in eBay auctions. The model includes a set of category-level dummy variables in c to account for differences in bidding practices across product classes. Finally, we cluster the error term, ij, on auction identier (j) because multiple bidders (i) are observed for each auction. We estimate Eq. (1) with OLS and report the results in Table 2. We suppress the coefcient estimates for AUCTION, SELLER, and BIDDER in this and all subsequent tables, focusing exclusively on the three variables in CHARITY. When we include the nal bids of all bidders in our sample (column 1), we nd that bidders in each class of charity auction submit bids signicantly earlier than in non-charity auctions. We repeat the analysis in column 2 after dropping bidders who bid more than once. The retained bidders may be more likely to be sophisticated and aware of the strategic value of shifting bid timing. While the magnitudes of our coefcient estimates change between columns 1 and 2, we again nd that bidding is substantially earlier in charity auctions.

44 Table 2 Timing of a bidder's nal bid. Dependent variable Included bidders Column Charity variables 10%-SHARE 100%-SHARE MID-SHARE R2 N

D.W. Elfenbein, B. McManus / Economics Letters 107 (2010) 4245 Table 4 Incremental bidding. Days before auction close All (1) .346*** [.099] .260*** [.030] .472*** [.139] .124 70,216 Bid once (2) [.107] .187 .372*** [.034] .617*** [.148] .141 46,715

Dependent variable Estimation method Column Charity variables 10%-SHARE 100%-SHARE MID-SHARE Sample average 2 R N

Incremental bidder present? (Yes = 1) Probit (1) .021 [.023] .052*** [.008] .049 [.035] .323 25,898

Bids per bidder

Incremental bidder wins? (Yes = 1) Probit (3) .004 [.016] .038*** [.006] .037 [.026] .146 25,898

OLS (2) .036 [.051] .090*** [.018] .134 [.097] .031 70,216

Notes: *** = signicant at p .001; ** = signicant at p .01; * = signicant at p .05; = signicant at p .10. Standard errors, clustered on the auction identier, are in brackets. The models include (unreported) controls for: Day of week for auction close (six dummies), time of day for auction close (four dummies), seller rating, seller percentage positive rating, length of auction in days (four dummies), bidder feedback rating (quadratic), and 47 category-level dummies. Coefcient estimates for these unreported variables are available on request.

Notes: *** = signicant at p .001; ** = signicant at p .01; * = signicant at p .05; = signicant at p .10. Standard errors, clustered on the auction identier, are in brackets. Results in columns 1 and 3 are marginal effects. See notes for Table 2 for additional details.

4.2. Discrete measures of sniping For each bidder, we create an indicator variable equal to 1 if his or her highest bid was placed in the nal 60 s of the auction, and the variable takes a value of 0 otherwise. We also create an indicator for whether a bidder's highest bid was in the nal 30 s. We use a probit model to estimate the probability that bidders' indicators for lastminute bids (and, separately, last half-minute bids) are equal to 1. The explanatory variables mirror the structure of Eq. (1), including the set of category dummies (c), our descriptors of charity status (CHARITY), and other controls for the auction, seller, and bidder attributes. The results, presented in Table 3, demonstrate that charity auction bidders snipe signicantly less frequently than bidders in non-charity auctions. We focus on the choices of bidders who bid only once in an auction. Across charity auctions, we observe a signicant reduction in sniping behavior when a large share of revenue is donated to charity and thus the public goods value of an auction is as strong as possible. The results in column 1 indicate that last-minute bids are about 9% less common in auctions with 100% SHARE = 1 (a change of .0082 on the sample average of .0947). Bids in the last 30 s decline by a similar proportion (10%) in charity auctions with all revenue donated. In unreported regressions we examine the highest bids of all bidders and nd coefcients that are very similar and exhibit substantially larger t-statistics. 4.3. Timing shifts and incremental bidding For a shift in bid timing to affect revenue, a sophisticated bidder's earlier bidding must result in more aggressive bidding by nave bidders. We investigate this possibility and present the results in Table 4. We retain the full set of product, auction, seller, and bidder controls in Eq. (1) in addition to the vector CHARITY for the empirical analysis. In columns 1 and 2 we show that charity auctions with 100%SHARE = 1 are signicantly more likely to include a bidder who bids multiple times. The 5.2 percentage point increase in column 1 represents a 16% increase over the average for all auctions in the sample. In terms of number of bids per bidder, we nd that bidders in full charity auctions submit, on average, an extra .09 bids. Finally, in column 3 we examine whether bidders who bid multiple times are more likely to win charity auctions. Indeed, if the shifted bid timing of sophisticated bidders did not result in more frequent sales to bidders who were induced to bid aggressively, then it would likely be improper to infer that the public goods nature of auction revenue is causing timing shifts. The results in column 3 show that bidders in auctions with 100%-SHARE = 1 are 26% more likely to be won by a bidder who has bid more than once.

5. Conclusions Our results, taken together, are consistent with the theory that consumers view the charity component of an auction as a public good, and they will adjust their actions to increase the size of other bidders' payments. The difference in bid timing results between 100%-share and 10%-share auctions is also consistent with this view. Bidding earlier in 100%-share auctions may be more likely to generate an extra dollar in revenue for a charity than bidding earlier in a 10%-share auction, where the nal price would have to rise by $10 to achieve the same result.

Table 3 Discrete measures of sniping. Dependent variable Column Charity variables 10%-SHARE 100%-SHARE MID-SHARE Sample average N Bidders' only bid occurs in last 60 s (Yes = 1) (1) .0065 [.0094] .0082** [.0033] .0187 [.0128] .0947 46,715 Bidders' only bid occurs in last 30 s (Yes = 1) (2) .0022 [.0091] .0077* [.0030] .0207 [.0111] .0780 46,715

Notes: *** = signicant at p .001; ** = signicant at p .01; * = signicant at p .05; = signicant at p .10. Results are given as marginal effects. Standard errors, clustered on the auction identier, are in brackets. See notes for Table 2 for additional details.

D.W. Elfenbein, B. McManus / Economics Letters 107 (2010) 4245

45

References
Andreoni, J., 1989. Giving with impure altruism: applications to charity and Ricardian equivalence. Journal of Political Economy 97, 14471458. Ariely, D., Ockenfels, A., Roth, A.E., 2005. An experimental analysis of ending rules in Internet auctions. RAND Journal of Economics 36, 890907. Bajari, P., Hortacsu, A., 2003. The winner's curse, reserve prices and endogenous entry: empirical insights from eBay auctions. RAND Journal of Economics 34, 329355. Elfenbein, D.W., McManus, B., forthcoming. A greater price for a greater good? Evidence that consumers pay more for charity-linked products. American Economic Journal: Economic Policy.

Ely, J., Hossain, T., 2009. Sniping and squatting in auction markets. American Economic Journal: Microeconomics 1, 6894. Engers, M., McManus, B., 2007. Charity auctions. International Economic Review 48, 953994. Ockenfels, A., Roth, A.E., 2006. Late and multiple bidding in second-price auctions: theory and evidence concerning different ending rules for an auction. Games and Economic Behavior 55, 297320. Roth, A.E., Ockenfels, A., 2002. Last-minute bidding and the rules for ending secondprice auctions: evidence from eBay and Amazon auctions on the Internet. American Economic Review 92, 10931103.