Spite and overbidding in second price all-pay auctions A theoretical and experimental investigation Oliver Kirchkamp∗ Wladislaw Mill† February 6, 2016 An extensive literature provides evidence for overbidding in auctions. Risk aversion, joy of winning, anticipated regret etc. do not always offer sufficient explanations. We show that the second-price all-pay auction presents a very interesting environment to study spite and overbidding. In equilibrium we should expect spiteful participants to increase their bids for small valuations and decrease their bids for high valuations. We measure spite and bidding behaviour for our participants and find that, indeed, the more spiteful participants also follow a bidding pattern which is theoretically in line with more spitefulness. Keywords: Auction, Overbidding, Spite 1. Introduction Auctions are an important part of everyday life. Not only do we find explicit auction institutions in markets. Auctions are also often a good model of non-market interaction. For example, fights between animals (Riley, 1980; Smith, 1974)1 , competition between firms (Fudenberg and Tirole, 1986; Ghemawat and Nalebuff, 1985; Oprea et al., 2013), the voluntary provision of public goods (Bilodeau et al., 2004), legal expenditures in litigation environments (Baye et al., 2005), the settlement of strikes, fiscal and political stabilization, the timing of exploratory oil drilling, and many more (see Hrisch and Kirchkamp, 2010, p. 1) are applications of (all-pay-)auctions. ∗ Department of Economics, Friedrich-Schiller-University Jena, Germany Department of Economics, Friedrich-Schiller-University Jena, Germany and International MaxPlanck-Research School 1 Smith (1974) uses a war-of-attrition game with common valuations to model fights between animals. † 1 While risk neutral Bayesian Nash equilibria can be derived for many auction types, participants in lab-experiments do not always follow these equilibria. Many authors find overbidding (or over-dissipation): actual bids are higher than equilibrium bids in all-pay auctions,2 in rent-seeking contests,3 in winner-pay auctions.4 A wide literature discusses possible adjustments of equilibrium behaviour to explain overbidding.5 However, some papers indicate that explanations like risk-aversion, joy of winning, anticipated regret etc. are not in line with all aspects of observed behaviour.6 Some authors suggest spite as a possible explanation.7 Here we present the second-price all-pay auction as an interesting environment to study bidding behaviour. We show that in the Bayesian Nash equilibrium of this auction spiteful subjects should overbid up to a certain (high) threshold. In our experiment we measure separately spitefullness and bids in auctions. Bidding behaviour confirms the theoretical prediction: More spite leads to higher bids. We think that we can make three contributions to the literature: 1. We extend the theoretical model of spiteful behaviour to first-price and secondprice all pay auctions. 2. With the help of experiments we investigate bidding behaviour in the second-price all-pay auction. 3. Most importantly, we can show that spite is an important determinant of overbidding. The remainder of the paper is structured the follows: In Section 2 we briefly summarise the relevant literature. In Section 3 we present the model and the theoretical predictions. Section 4 will explain the design of the experiment. In Section 5 we show the results of the experiment. Section 6 concludes. 2. Literature In this paper we study consider second-price all-pay auctions with sealed-bids and private information.8 Furthermore, we restrict our attention to the case where the ex-post 2 (Noussair and Silver, 2006; Ernst and Thni, 2013; Goeree et al., 2002; Ong and Chen, 2013; Lugovskyy et al., 2010). 3 Potters et al. (1998) 4 Morgan et al. (2003); Andreoni et al. (2007); Barut et al. (2002). 5 Filiz-Ozbay and Ozbay (2007, 2010); Cooper and Fang (2008); Andreoni et al. (2007); Cox et al. (1985, 1988); Fibich et al. (2006); Kagel and Levin (1993); Kirchkamp and Reiss (2008); EngelbrechtWiggans and Katok (2009); Kirchkamp et al. (2008); Armantier and Treich (2009). 6 Kagel and Levin (1993); Kirchkamp et al. (2008); Engelbrecht-Wiggans and Katok (2009); Andreoni et al. (2007). 7 Morgan et al. (2003); Nishimura et al. (2010); Andreoni et al. (2007); Bartling and Netzer (2014); Nishimura et al. (2010); Kimbrough and Reiss (2012). 8 Equilibria for all-pay auctions with common values are provided by Hendricks et al. (1988) andKovenock et al. (1996). Sacco and Schmutzler (2008) provide mixed strategy equilibria for 2 allocation (once all bids are known) is not stochastic.9 In addition, we will consider first and second price all-pay auctions only.10 We also assume here that the playing subjects are already specified.11 2.1. Factors other than spite that influence bids One of our aims is to better understand the influence spitefulness has on bidding behaviour. Besides spitefulness a large number of other factors, internal and external to the bidders, have already been studied. Among the external factors, Katok and Kwasnica (2008) show that the speed of the ticking clock in Dutch auctions affects bids. Cox and James (2012) show that the structure of the presented games have an influence on the bidding behavior of the participants. Kirchkamp et al. (2009) show that outside options also influence bids. Among the factors more internal to bidders Nakajima (2011) explains the deviation of the Dutch auction from the first price auction with the Allais paradox. Gth et al. (2003), Dittrich et al. (2012), and Ockenfels and Selten (2005) relate bidding behaviour to learning effects. Kagel et al. (1987) and Hyndman et al. (2012) show that the provision of information influences overbidding. Chen et al. (2013) provide empirical evidence that even the menstrual cycle has an influence on the bidder’s decision. 2.2. Literature on overbidding In many experiments overbidding has been found and explained with different motives, ranging from risk aversion, over spite, to anticipated regret. Filiz-Ozbay and Ozbay (2007, 2010) explain overbidding with the emotion of anticipated regret in winner-pay auctions. They assume that players anticipate their regret after a wrong choice. Cooper and Fang (2008) are discussing whether joy of winning is a possible explanation for overbidding. Andreoni et al. (2007) provide evidence against against joy of winning. Cox et al. (1985, 1988) suggest risk-aversion to explain overbidding. Fibich et al. (2006) introduce risk averse players in all-pay auctions to explain overbidding. Kagel and Levin (1993), Kirchkamp et al. (2008), and Engelbrecht-Wiggans and Katok (2009) argue that risk aversion might by itself not enough to explain overbidding. common value auctions where the prize is influenced by the own behavior. Feess et al. (2002) show a pure equilibrium strategy in case of handicapped players. Klose and Kovenock (2015) show equilibria for the case of externalities which depend on the bidders’ identities. Bertoletti (2006) show equilibria for common value all-pay auctions with reserve price. Dechenaux and Mancini (2008) and Baye et al. (2005) are model ligation systems with all-pay auctions. The case of affiliated valiations is studied by Krishna and Morgan (1997). 9 The survey by Dechenaux et al. (2015) also covers rent-seeking games where the ex-post allocation is stochastic and where also bidders who did not submit the highest bid have a chance to win the auction. 10 Intermediate situations between the first and second price all-pay auction are studied by Albano (2001). 11 Bos (2012) considers the situation where the number of bidders is uncertain. 3 Anderson et al. (1998) are introducing bounded rationality as a reason for overbidding in all-pay auctions. Similarly, Kirchkamp and Reiss (2008) claim that overbidding may result from bidding heuristics. Armantier and Treich (2009) argue that bidders are unable to assess winning probabilities correctly. Andreoni et al. (2007) suggest that spite may be on of the determinants of overbidding. Also Bartling and Netzer (2014) discuss that “spiteful preferences are an important determinant of overbidding in the second-price auction” (Bartling and Netzer, 2014, p.23). To the best of our knowledge, only Morgan et al. (2003) and Nishimura et al. (2010) introduce spite into theoretical bidding behavior. Different from our paper, Nishimura et al. (2010) and Morgan et al. (2003) study spite in common-valuations-auctions and winner-pay auctions, respectively. Nevertheless, our theoretical part can be seen as an extension of Morgan et al. (2003). 2.3. Literature on spite Besides papers suggesting spite being a determinant of overbidding, there are papers showing data suggesting that spite is not uncommon in general. D’Almeida et al. (2014) claim to show that children can behave spiteful in economic games. Chun et al. (2011) observe that Koreans and Chinese behave less spiteful than Japanese. Fehr et al. (2008) show experimentally that spiteful behavior is rather wide spread in the least developed parts of India. Kimbrough and Reiss (2012) observe consistent spiteful behavior in a second-price winner-pay auction, and Marcus et al. (2014) developed a psychological test for spitefulness. In the next section we will present a model that explicitly incorporates spite in the context of an all-pay auction. 3. Model In the following we will derive the risk neutral Bayesian Nash equilibrium for spiteful bidders in the first-price all-pay auction and in the second-price all-pay auction. 3.1. First-price all-pay auction Two bidders i and j have private valuations for a prize with with values vk , k ∈ {i, j}. Both submit a bid bk following a monotonic bidding function βk (vk ) with first derivative and inverse βk−1 (bk ) = vk . Valuations are distributed according to F , β ′ (x) ≡ dβ(x) dx i.e. v ∼ F (0, v), and f (x) = dFdx(x) . The prize is allocated to the player with the highest bid. If vi = vj , the prize is distributed randomly. We assume that both bidders have the same utility function u(x). Following a standard argument in equilibrium βk (0) = 0. is the utility function and we assume bidders to be risk neutral. To integrate spite into the model we assume that the losing participant gains an additional disutility of α-times the payoff of the winning player. Assume without loss of 4 generality that bj > bi , i.e. bidder j is the winner. Then α · (vj − βj (vj )) reflects bidder i’s spite.12 The absence of spite is equivalent α = 0. We assume that α ∈ [0, 1), i.e. we do not consider α < 0 which could represent sympathy or profit sharing. The payoff of player i is as follows: ΦI-AP (βi , vi ) S u(vi − βi (vi )) if βi > βj (i wins) = if βi = βj (a tie) u(−βi (vi ) − α(vj − βj (vj ))) if βi < βj (j wins) u( vi −β2i (vi ) ) (1) We follow the standard approach and assume that bidder i with valuation vi makes a bid b. The expected utility of this bidder is given as follows: E(b, v) = ∫ βj−1 (b) |0 u(v − b)f (vj )dvj + {z } bidder i wins and obtains the prize ∫ | v βj−1 (b) u(−b − α(vj − βj (vj )))f (vj )dvj {z (2) } bidder i loses and pays his own bid and additionally experiences spite Rearranging the FOC leads to: (u(v − b) − u(−b − α(βj−1 (b) − βj (βj−1 (b))))f (βj−1 (b)) β ′ (βj−1 (b)) = ∫ β −1 (b) ∫ j u(v − b)′ f (vj )dvj + βv−1 (b) u(−b − α(vj − βj (vj )))′ f (vj )dvj 0 j Assuming a symmetric equilibrium bidding function, we obtain the following condition: ((v − b) − (−b − α(v − b)))f (v) = (1 + α)vf (v) − αf (v)b (3) 1 Solving the ODE (3) with the initial value b(0) = 0 we obtain the symmetric equilibrium bidding function bI-AP : S β ′ (v) = bI-AP (v) S =e −αF (v) ∫ v (α + 1)sf (s)e αF (s) 0 ( ) 1 ∫ v αF (s) α+1 v − αF (v) ds = e ds α e 0 (4) For α = 0, Equation (4) becomes the familiar equilibrium bidding function for all-pay auctions without spite: ∫ v bI-AP := bI-AP = sf (s) ds (5) α=0 0 For uniformly distributed valuations, F (x) = x, we have ( (v) bI-AP S α+1 e−αv − 1 = v+ α α ) . (6) The left part of Figure 1 illustrates in particular the following: 12 Note that spite is similar but not equivalent to the negative aspect of inequality aversion. The latter would consider relative gain α · (vj − βj (vj ) − βi (v) ). 5 α=2 v=1 0.8 1 0.9 0.6 0.5 0.6 0.8 b b 0 0.4 0.4 0.7 0.6 −0.5 0.5 0.2 0.2 0.4 0.0 0.0 −1 0.2 0.4 0.6 0.8 0.3 0.2 00.1 1.0 0.2 v 0.4 0.6 0.8 1.0 α Figure 1: Equilibrium bids in first-price all-pay auctions Equilibrium bids in first-price all-pay auctions for different valuations v and different levels of spite α with uniform distributions of valuations (see Equation 6). Proposition 1. The bidding function in the first-price all-pay auction is increasing in the bidder’s valuation: dbI-AP S ≥0 dv Proof ( ∫ v dbI-AP 1 S = αF (v) −αf (v) (α + 1)sf (s)eαF (s) ds + (α + 1)vf (v)eαF (v) dv e 0 ) (α + 1)f (v) ∫ v αF (s) = e ds ≥ 0 eαF (v) 0 The right part of Figure 1 illustrates the following: Proposition 2. Bids in the first-price all-pay auction are increasing in spite: dbI-AP S ≥0 dα The proof is shown in Appendix A. In particular bids in the first-price all-pay auction with spite (α > 0) are larger or − bI-AP ≥ 0. equal than bids without spite (α = 0), i.e. bI-AP S 6 Proposition 3. The difference between equilibrium bids with spite and without spite is increasing in bidder’s valuation: d(bI-AP − bI-AP ) S ≥0 dv The proof is shown in Appendix B. Proposition 3 is actually quite intuitive. Consider two bidders competing for an object in the first-price auction. For a low valuation the probability of winning is low, too. In this situation bidders will suffer from spite almost always. As a result the overall value of the auction is small and bids will be small, too. If, on the other hand, valuations are high bidders will want to avoid the disutility from losing by increasing their bids. As a result we see that when spite increases also bids increase. 3.2. Second-price all-pay auction The model of the second-price all-pay auction is quite similar to the already mentioned model of the first-price all-pay auction. Consider again two risk-neutral bidders, i and j with utility function u(x). Valuations vi and vj are distributed according to F , i.e. v ∼ F (0, v), and f (x) = dFdx(x) . Both bidders submit a bid βk (vk ) with k ∈ {i, j}. In the second-price all-pay auction both players pay the second highest bid. The prize is allocated to the bidder with the highest bid. We assume that for the candidate equilibrium we have βk (0) = 0 and that β ′ (x) = dβ(x) dx and βk−1 (bk ) = vk exist. As before we model spite as a disutility for the losing bidder i of α · (vj − βi (vi )) where α describes the amount of spite and (vj − βi (vi ) the gain of the winning bidder j. We again consider α ∈ [0, 1). The payoff of player i is as follows: ΦII-AP (βi , vi ) S u(vi − βj (vj )) if βi > βj (i wins) = if βi = βj (a tie) u(−βi (vi ) − α(vj − βi (vi ))) if βi < βj (j wins) u( vi −β2i (vi ) ) (7) Again we assume that bidder i with valuation vi makes a bid b. The expected utility of this bidder is given as follows: E(b, v) = ∫ | βj−1 (b) 0 u(v − βj (vj ))f (vj )dvj + {z } bidder i wins and obtains the prize and pays the loser’s bid ∫ | v βj−1 (b) u(−b − α(vj − b))f (vj )dvj {z bidder i loses and pays his own bid and additionally experiences spite Rearranging the FOC yieds: β ′ (βj−1 (b)) (u(v − b) − u(−b − α(βj−1 (b) − b))f (βj−1 (b)) = ∫ (1 − α) βv−1 (b) u(−b − α(vj − b))′ f (vj )dvj j 7 } (8) 6 2.0 5 4 1.0 b b 1.5 α=1 0.75 0.5 0.25 3 2 0 v=1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 −0.5 0.5 1 0 0.0 0.2 0.4 0.6 0.8 1.0 0.2 0.4 v 0.6 0.8 1.0 1.2 α Figure 2: Equilibrium bids in second-price all-pay auctions Equilibrium bids in second-price all-pay auctions for different valuations v and different levels of spite α with uniform distributions of valuations (see Equation 11). Assuming a symmetric equilibrium we obtain β ′ (v) = v + α(v − b))f (v) v(1 + α)f (v) α(b))f (v) = − (1 − α)(1 − F (v) (1 − α)(1 − F (v) (1 − α)(1 − F (v) (9) Solving the differential equation (9) with initial value b(0) = 0 gives us the symmetric equilibrium bidding function bII-AP : S ( ) ∫v α ∫ v α 1 α+1 (1 − F (s)) α−1 ds α+1 0 v− = (1−F (v)) 1−α sf (s)(1−F (s)) α−1 ds = α 1−α 1−α 0 (1 − F (v)) α−1 (10) For α = 0, Equation 10 becomes the familiar equilibrium bidding function for secondprice all-pay auctions without spite: bII-AP (v) S ∫ II-AP b := bII-AP α=0 = 0 v sf (s)(1 − F (s))−1 ds For uniformly distributed valuations, F (x) = x, we have (v) = bII-AP S ( ) ) α (α + 1) ( (1 − α) (1 − v) 1−α − 1 + vα α(2α − 1) (11) Figure 2 illustrates this case. The left graph in the Figure shows that, as in first-price all-pay auctions, bids are monotonically increasing in valuations. The right part shows that, different from first-price all-pay auctions, bids are only increasing in spite for small 8 valuations. If the valuation is sufficiently large, equilibrium bids decrease when spite increases. In the Figure we see that for small valuations bids are, as in the first-price case, higher with spiteful bidders. For large valuations bids rather decrease with spite. To simplify the notiation we assume in the following that valuations v ∈ [0, 1]. It is easy to see the following: Proposition 4. The bidding function in the second-price all-pay auction is increasing in bidder’s valuation: dbII-AP S ≥0 dv Proof ( 2α−1 1 α dbII-AP 1+α α ∫v S 1−α α−1 sf (s)(1 − F (s)) α−1 = (1 − F (v)) f (v) v(1 − F (v)) − dv 1−α 1−α 0 | {z } ( q≥0 = q v(1 − F (v)) α α−1 − v(1 − F (v)) ≥ 0 α α−1 ∫ v + 0 (1 − F (s)) α α−1 ) ) Interestingly, there is no equivalent for Proposition 2 for second-price auctions. Bids are not necessarily increasing if bidders are more spiteful. On the contrary: As illustrated in Figure 2, if valuations are sufficiently large, more spite can actually lead to a decrease in bids. 3.3. Predictions The predictions we have obtained in Section 3.1 for the first-price all-pay auction are that spiteful bidders will bid more, in particular when their valuation is high. This is interesting, but this might also be difficult to disentangle from bidding more due to, say, risk aversion. In this sense the predictions we have obtained in Section 3.2 for the second-price all-pay auction are more specific. Spiteful bidders will bid more, in particular for intermediate valuations. In our experiment we will, therefore, use the second-price auction as a workhorse. We have the following hypotheses: Hypothesis 1 Increased spite will lead to higher bids Hypothesis 1 is the main hypothesis of this paper and it is based on the theoretical prediction, that spiteful agents will overbid compared to non-spiteful agents. We will show in the result section that, indeed spite leads to higher bids. 9 Hypothesis 2.1 Overbidding will be increasing in valuation for a give level of spite as long as the valuation is not too large. Hypothesis 2.2 For very high valuations and a given level of spite we will find underbidding. Hypothesis 2 are mainly to check for the specific predictions of the model. In the result section we will find support for hypothesis 2.1 but we will not find any support for Hypothesis 2.2. 4. Design of the experiment In the experiment, we first measure spite. We will discuss the different measures of spite in Section 4.1. Then we observe bids in the second-price all-pay auction which we will discuss in Section 4.2. 4.1. Measures of Spite We use three different measures for spite. We will discuss the measure by Marcus et al. (2014) in Section 4.1.1. The measure by Kimbrough and Reiss (2012) will be discussed in Section 4.1.2 and our own measure in Section 4.1.3. 4.1.1. Spite according to Marcus et al. The questionnaire by Marcus et al. (2014) is based on 17 not-payoff relevant questions at the end of the experiment. Examples of the 17 questions are the following:13 • If I am checking out at a store and I feel like the person in line behind me is rushing me, then I will sometimes slow down and take extra time to pay. • I would rather no one get extra credit in a class if it meant that others would receive more credit than me. Participants were asked to indicate their agreement on a scale between 1 and 5. The distribution of spite in this measure is shown in the left part of Figure 3. 4.1.2. Spite according to Kimbrough and Reiss The second measure is a modification of the design by Kimbrough and Reiss (2012) who want to observe spiteful behavior in an second-price auction. First participants were asked to bid in ten auctions with one opponent.14 Then participants were informed about the outcome of the ten auctions, i.e. who had won the auction and the winner’s bid. Thereafter, participants had the opportunity to increase 13 14 All questions are shown in Appendix D.1. See Appendix D for details. 10 1.0 0.8 0.6 0.4 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Female Male 0.0 0.2 0.8 0.4 0.0 0.2 0.6 1.0 0.8 0.4 0.6 Own measure 0.2 0.0 Empirical CDF Kimbrough-Reiss 1.0 Spite-Score 0 20 40 60 80 100 0 10 20 30 Measure for Spite Figure 3: Distribution of Measures for Spite or to keep their own bid. This increase could not be more than the difference between the winner’s and the loser’s bid. Hence, bidders could not change the allocation, only the payoff of the winner. Participants who had lost would still lose the auction (and hence, have a payoff of 0) but they could increase the price the winner had to pay. At the end participants were asked to indicate their willingness to pay for this bid adaptation. Participants who had increased their losing bid were considered spiteful. The spitemeasure is a continuous measure between 0% (no adjustment) and 100% (100% if the loser increases the own bid up to the winner’s bid). A distribution of spite based on this measure is shown in the middle of Figure 3. 4.1.3. Our own measure for spite Our own spite measure asks participants to decide six times among 9 possible allocations similar to the SVO-Slider measure by Murphy et al. (2011); Murphy and Ackerman (2014). Figure 4 shows all six sets.15 For each set participants had to chose their preferred allocation. In all six sets the allocation with the highest payoff for the other player maximises also the own payoff. Any deviation from this allocation reduces the payoff of the other player and never increases the own payoff. Such a deviation can, hence, be seen as a sign of spite. This deviation is costless in sets IA1, RG1 and PS1, it is costly in IA2, RG2, and PS2. While one reason for these deviations can be spite, there are other explanations. Deviations in sets IA1 and IA2 can be a sign of “inequality aversion”. Deviations in sets RG1 and RG2 can be a sign of “concerns for relative gain”. 15 Details of the allocations are shown in Appendix D.2. 11 100 IA1 Other payoff 90 RG1 IA2 RG2 80 PS1 70 PS2 60 50 50 60 70 80 90 100 Own payoff Figure 4: The six allocation sets for our own slider measure For each of the six sets players choose one allocation. For each set we consider the pareto efficient allocation not spiteful. Less efficient allocations will be considered more spiteful. 12 Notable only 9 participants were behaving extremely spitefully and only one participant was willing to pay for this behavior. A distribution of the combined spite measure is shown in the right graph in Figure (3). 4.1.4. Other controls Additionally we controlled for risk-aversion by implementing the task by Holt and Laury (2002), we controlled for social value orientation and inequality aversion by using the slider measure by Murphy et al. (2011); Murphy and Ackerman (2014) and we controlled for rivalry, as this could be linked to spite, by implementing the questionnaire of Back et al. (2013). 4.2. Design of the auction After having measured spite16 participants play the second-price all-pay auction. We explained in great detail (with video-instructions for each behavioral measure) the rules of the second-price all-pay auction. Participants played the auction for 15 rounds with stranger matching within matching groups of 6. Each round participants were asked to indicate their biding function for valuations between 0 and 100. Participants indicated their bid for valuations of 0, 10, 20,. . . , 90, 100. Bids for intermediate valuations were linearly interpolated. To give more feedback each pair of bidders plays not only one but ten auctions for a random pair of valuations. For each of the ten auctions participants were told how much their valuation their own bid, and their the opponent’s bid was. Participants were told the outcome of the auction and how much they had won or lost. Figure 5a shows an example of the bidding interface. Figure (5b) shows an example of the feedback interface. 4.3. Payment of one random task Participants were paid at the end of the experiment for one random task, i.e. either one lottery from the risk-measure or one allocation from the SVO-Slider-Measure or the Spite-Measure or the adaptation of Kimbrough and Reiss (2012) or one of the allpay auctions.17 To ensure the average payoff of each behavioral task to be similar we adjusted the payoff ratio of ECU (experimental currency unit) to Euros in each task. Moreover participants were given a higher endowment if the all-pay auction was paid, as the payoff auction could lead to negative payoff. 16 The implementation of Kimbrough and Reiss (2012) and our all-pay auction were counterbalanced as both parts are auctions and we want to control for order effects here. 17 In case of the all-pay auction only one of the 10 × 15 auctions about which we gave feedback was played out. 13 (a) Interface of the bidding stage. (b) Interface of the feedback stage. Imputing the bidding function for the possible valuations between 0 and 100. The bidding function will be drawn after the input of the respective bids. Mapping the 10 random valuations and the respective bids on the bidding function. Additionally subjects could see the opponents bid, whether they won and the amount they won/lost. Figure 5: Auction Interface 5. Results We conducted the experiments in June 2015 in the laboratory of the school of economics of the University of Jena (Germany). We recruited 138 participants in 8 sessions using the online recruiting platform ORSEE (Greiner, 2004). We implemented the experiment using z-Tree (Fischbacher, 2007). Instructions were presented as a 25 minutes video followed by test questions for the all-pay auction and for the auction of the spite-measure by Kimbrough and Reiss (2012). The entire experiment lasted for about 100 minutes. Participants earned on average 15.83e, which is slightly above the minimum wage in Germany. We had 45% male and 55% female participants with a median age of 24. Participants were on average in their third year of studying and about 12% were students of business or economics. 5.1. Measures of Spite Figure 6 shows the joint distribution of the three measures for spite. There is no visible correlation. For the three instruments we find a Cronbach α of 0.0371 (CI=[0.0172,0.0908]). The three instruments seem to measure different things. Having said that, we find substantial consistency within each scale. For the questionnaire (Marcus et al., 2014) we find a Cronbach α of 0.868 (CI=[0.818,0.933]). For our own measure we find a Cronbach α of 0.555 (CI=[0.41,0.733]). Neither the questionnaire nor our own measure seems to be strictly one-dimensional. For the questionnaire we find that the first element of a principal component analysis explains 33.3% of the variance, (CI=[25.2,40.6]). For our own measure we find that the first element of a principal component analysis explains 87.2% of the variance, (CI=[81.7,91.4]). • The adaptation of Kimbrough and Reiss (2012) could be mainly measuring retal- 14 80 60 40 20 0 1.0 2.0 3.0 4.0 Spite-Score 100 4.0 80 3.5 Spite-Score Kimbrough-Reiss Kimbrough-Reiss 100 60 40 3.0 2.5 2.0 20 1.5 0 1.0 0 10 20 30 Own measure 0 10 20 30 Own measure Figure 6: Joint Distribution of Measures for Spite iation or punishment, as the amount of lost auctions is a strong predictor of the amount of spite—this means, that participant are increasing the bid to harm the opponent to “punish him for winning”. • The questionnaire is measuring the general motivation towards spiteful behavior. It is, however, not necessary describing how people will in fact behave. • Our own measure is potentially in general measuring the attitude towards spite in an behavioral measure. As there is, in general, no way to disentangle which of the three spite-measures is “really” measuring spite we will look at the combined measures. 5.2. Aggregated Behavior Figure 7 shows the average bidding behavior across all 15 rounds compared to predicted behavior. The left graph in the Figure shows median bids, the right graph shows median overbidding, i.e. the median difference between the bid and the bid of a non-spiteful bidder. Clearly, median bids are not fit spiteful bids too well. Still, spiteful participants indeed bid more than non-spiteful participants, in particular if their valuations are low. To check for Hypothesis 1, i.e.. increased spite will lead to higher bids, we will use a mixed-effects model as spite, risk, social value orientation (SVO) etc. are fixed effects but the individuals and the matching-group are random effects. We compare four different models which differ only in the controls C1 , . . . , C4 . 15 b − bII-AP b 100 Spite=0 Spite=0.3 Spite=0.6 Spite>Median Spite<Median 50 0 20 40 60 80 100 50 Spite=0 Spite=0.3 Spite=0.6 Spite>Median Spite<Median 0 0 20 v 40 60 80 100 v Figure 7: Median bids and overbidding in the auction The left graph shows median bids. The right graph shows median overbidding. Bidi,t,j,v C1 C2 C3 =β0 + β1 Spitei + νi,j + ηj + ϵi,j,k,l + CM =0 =β2 t + β3 v =C2 + β4 1Gender=♀ + β5 Riski + β6 rivalryi + β7 SVOi + β8 IAi C4 =C3 + β9 Spitei × v (12) (13) (14) (15) (16) where νi,j is a random effect for bidder i in group j, ηj is a random effect for group j, and ϵi,j,k,l is the residual. Estimation results are shown in Table (1). For all our specifications spite has a positive and significant influence on bids behavior. This is in line with Hypothesis 1. Risk and (obviously) valuation also have a significant and positive influence on bids. Furthermore, female bidders bid more than men. The decrease in bidding over the rounds could be interpreted as a learning effect of overbidding. Result 1 Spite has a significant and positive effect on bids. To investigate Hypothesis 2, increased overbidding in valuation and underbidding for high valuations, we look at the right part of Figure 7. The Figure shows overbidding, i.e. the difference between actual bids minus the theoretical bids for a non-spiteful agent. We can see that observed bids seem to be in line with Hypothesis 2. To formally investigate Hypothesis 2 we again use a mixed effects model. To model the dependency of oberbidding on the valuation we use linear splines where knots of the spline function 16 (1) Spite Male Risk Rivalry SVO IA Period Spite × Valuation Valuation Constant Observations Log Likelihood Akaike Inf. Crit. Bayesian Inf. Crit. Notes: 3.51∗ (1.70) 53.68∗∗∗ (3.43) 23,760 −123,402.10 246,816.20 246,864.60 (2) 3.52∗ (1.70) (3) 4.37∗ −0.40∗∗∗ (0.05) (2.00) (6.27) 6.09∗ (2.98) −1.04 (3.14) 0.40 (0.25) −1.93 (2.55) −0.40∗∗∗ (0.05) 0.64∗∗∗ (0.01) 27.42∗∗∗ (3.15) 23,760 −120,479.30 240,972.60 241,029.10 0.64∗∗∗ (0.01) 27.54∗∗∗ (6.67) 23,760 −120,462.00 240,948.10 241,045.00 −19.05∗∗∗ (4) 4.19∗ (2.01) (6.27) 6.09∗ (2.98) −1.04 (3.14) 0.40 (0.25) −1.93 (2.55) −0.40∗∗∗ (0.05) 0.004 (0.004) 0.64∗∗∗ (0.01) 27.54∗∗∗ (6.67) 23,760 −120,466.20 240,958.40 241,063.40 −19.05∗∗∗ p : ∗ ∗ ∗ < .001 ∗ ∗ < .01∗ < .05 Table 1: Estimation of Equation 12. Spite is the sum of the three (normalised) spite measures. IA is the sum of the (normalised) inequality aversion score obtain from the slider measure and the (normalised) score obtained from inequality allocation of our own spite measure. are at valuations 25, 50 and 75. We compare three different models: ′ Bidi,t,j,v − bII-AP =β0 + β1 Spitei + νi,j + ηj + ϵi,j,k,l + β3 t + CM (17) ′ C1 =β4 1v∈[0,25] + β5 1v∈[25,50] + β6 1v∈[50,75] + β7 1v∈[75,100] (18) ′ ′ C2 =C1 + Spitei · (β8 1v∈[0,25] + β9 1v∈[25,50] + β10 1v∈[50,75] + β11 1v∈[75,100] ) (19) C3′ =C2′ + β12 IAi + β13 1Gender=♀ + β14 Riski + β15 rivalryi + β16 SVOi (20) where νi,j is a random effect for bidder i in group j, ηj is a random effect for group j, and ϵi,j,k,l is the residual. Estimation results are shown in Table (2). Figure 8 shows estimation results for the fitted spline. First we can see that generally behavior seems to be in line with theory: Overbidding first increases and then decreases. Furthermore, the interaction of spite and valuation seems to be most pronounced for smaller valuations, i.e. between 25 and 50. Indeed, this is similar to what we have seen also for equilibrium bids in Figure 2. As before we find risk to be a significantly positive determinant for overbidding as gender is a highly significant factor of overbidding. Note, that neither rivalry, social value orientation nor inequality aversion seem to have a significant impact. 17 C1′ Male Risk Rivalry SVO IA Spite V ∈ [0, 25] V ∈ [25, 50] V ∈ [50, 75] V ∈ [75, 100] Period Spite·V ∈ [0, 25] Spite·V ∈ [25, 50] Spite·V ∈ [50, 75] Spite·V ∈ [75, 100] Constant Observations Log Likelihood Akaike Inf. Crit. Bayesian Inf. Crit. Notes: 3.52∗ (1.70) 13.59∗∗∗ (1.14) 14.24∗∗∗ (0.95) −16.50∗∗∗ (1.02) −95.69∗∗∗ (1.03) −0.40∗∗∗ (0.05) 27.82∗∗∗ (3.20) 23,760 −120,829.30 241,678.60 241,759.30 C2′ C3′ 2.77 (1.75) 13.59∗∗∗ (1.14) 14.24∗∗∗ (0.95) −16.50∗∗∗ (1.02) −95.69∗∗∗ (1.03) −0.40∗∗∗ (0.05) 0.66 (0.63) 1.66∗∗∗ (0.52) 0.53 (0.56) 0.60 (0.57) 27.82∗∗∗ (3.20) 23,760 −120,823.10 241,674.30 241,787.40 −19.05∗∗∗ (6.27) 6.09∗ (2.98) −1.04 (3.14) 0.40 (0.25) −1.93 (2.55) 3.62 (2.03) 13.59∗∗∗ (1.14) 14.24∗∗∗ (0.95) −16.50∗∗∗ (1.02) −95.69∗∗∗ (1.03) −0.40∗∗∗ (0.05) 0.66 (0.63) 1.66∗∗∗ (0.52) 0.53 (0.56) 0.60 (0.57) 27.94∗∗∗ (6.69) 23,760 −120,805.90 241,649.80 241,803.20 p : ∗ ∗ ∗ < .001 ∗ ∗ < .01∗ < .05 Table 2: Estimation results for Equation 17 (overbidding) The table shows estimation results for the different controls C1′ , C2′ , and C3′ . Splines have knots at valuations 25, 50, and 75. Spite is the sum of the three spite measures. IA is the sum of the inequality aversion score obtain from the slider measure and the score obtained from inequality allocation of our own spite measure. 18 0 20 40 60 80 100 C10 C20 C30 b − bII-AP 50 0 Spite: -50 -5 0 5 0 20 40 60 80 100 0 20 40 60 80 100 v Figure 8: Estimation results for the spline from Equation 17 (overbidding) The Figure show splines for different controls C1′ , C2′ , and C3′ and for different (normalised) levels of spite. Result 2a Spite significantly contributes to overbidding. Result 2b The interaction of spite and valuations is, in line with theory, only significant for moderate valuations. Result 2c Gender and risk-attitude influence overbidding significantly. 6. Discussion and Conclusion In this paper we want to contribute with the help of theory and with experiments to a better understanding of bidding behaviour. We propose, in particular, that spite could be a relevant factor to explain bids. As a workhorse we are using the second-price all-pay auction. In the theoretical part we have seen that spitefulness should have a substantial influence on bids in this auction. For the participants in our experiment we measure spite using a questionnaire, an allocation task and an experimental design similar to Kimbrough and Reiss (2012). In line with theory more spiteful bidders make relatively higher bids for small valuations and relatively smaller bids for large valuations (compared to the non-spite equilibrium) . Besides this rather robust finding of our experiment, some limitations should be noted. First, we can explain only a part of the deviation from non-spite equilbria. Other forces, such as risk-aversion, joy of winning, anticipated regret or bounded rationality, might play a role, too. We have also seen that spite seems to be hard to define. The three spite measures we are using seem to measure different things. It is not so clear how spite is connected to 19 anger, retaliation, inequality aversion or even concerns about justice? Our approach, to simply sum up the normalised values of each measure is rather pragmatic. It might be interesting to further disentangle the contribution of each of the three measures to the joint effect. References Albano, G. L. (2001). A class of all-pay auctions with affiliated information. Discussion Papers (REL - Recherches Economiques de Louvain) 2001012, Universit catholique de Louvain, Institut de Recherches Economiques et Sociales (IRES). Anderson, S., Goeree, J. K., and Holt, C. A. (1998). Rent seeking with bounded rationality: An analysis of the all-pay auction. Journal of Political Economy, 106(4):828–853. Andreoni, J., Che, Y.-K., and Kim, J. (2007). Asymmetric information about rivals’ types in standard auctions: An experiment. Games and Economic Behavior, 59(2):240–259. Armantier, O. and Treich, N. (2009). Subjective probabilities in games: An application to the overbidding puzzle. International Economic Review, 50(4):1079–1102. Back, M. D., Kfner, A. C. P., Dufner, M., Gerlach, T. M., Rauthmann, J. F., and Denissen, J. J. A. (2013). Narcissistic admiration and rivalry: Disentangling the bright and dark sides of narcissism. Journal of Personality and Social Psychology, 105(10):1013 – 1037. Bartling, B. and Netzer, N. (2014). An externality-robust auction: Theory and experimental evidence. Cesifo working paper series, CESifo, Center for Economic Studies & Ifo Institute for economic research. Barut, Y., Kovenock, D., and Noussair, C. (2002). A comparison of multiple-unit allpay and winner-pay auctions under incomplete information. International Economic Review, 43(3):675–708. Baye, M. R., Kovenock, D., and de Vries, C. G. (2005). Comparative analysis of litigation systems: An auction-theoretic approach. The Economic Journal, 115(505):583–601. Bertoletti, P. (2006). On the reserve price in all-pay auctions with complete information and lobbying games. Mpra paper, University Library of Munich, Germany. Bilodeau, M., Childs, J., and Mestelman, S. (2004). Volunteering a public service: an experimental investigation. Journal of Public Economics, 88(12):2839 – 2855. Bos, O. (2012). Wars of attrition and all-pay auctions with stochastic competition. Journal of Mathematical Economics, 48(2):83–91. 20 Chen, Y., Katuscak, P., and Ozdenoren, E. (2013). Why can’t a woman bid more like a man? Games and Economic Behavior, 77(1):181–213. Chun, Y., Kim, J., and Saijo, T. (2011). The spite dilemma experiment in korea. Seoul Journal of Economics, 24(1):87–98. Cooper, D. J. and Fang, H. (2008). Understanding overbidding in second price auctions: An experimental study. The Economic Journal, 118(532):1572–1595. Cox, J., Smith, V., and Walker, J. (1988). Theory and individual behavior of first-price auctions. Journal of Risk and Uncertainty, 1(1):61–99. Cox, J. C. and James, D. (2012). Clocks and trees: Isomorphic dutch auctions and centipede games. Econometrica, 80(2):883–903. Cox, J. C., Smith, V. L., and Walker, J. M. (1985). Experimental development of sealed-bid auction theory: Calibrating controls for risk aversion. American Economic Review, 75(2):160–65. D’Almeida, A. F., Teixeira, R. C., and Chalub, F. A. C. C. (2014). Much or more? experiments of rationality and spite with school children. North American Journal of Psychology, 16(1):163–178. Dechenaux, E., Kovenock, D., and Sheremeta, R. (2015). A survey of experimental research on contests, all-pay auctions and tournaments. Experimental Economics, pages 609–669. Dechenaux, E. and Mancini, M. (2008). Auction-theoretic approach to modeling legal systems: An experimental analysis. Applied Economics Research Bulletin, 2:142–177. Dittrich, D. A., Gth, W., Kocher, M. G., and Pezanis-Christou, P. (2012). Loss aversion and learning to bid. Economica, 79(314):226–257. Engelbrecht-Wiggans, R. and Katok, E. (2009). A direct test of risk aversion and regret in first price sealed-bid auctions. Decision Analysis, 6(2):75–86. Ernst, C. and Thni, C. (2013). Bimodal Bidding in Experimental All-Pay Auctions. Games, 4(4):608–623. Feess, E., Muehlheusser, G., and Walzl, M. (2002). When bidding more is not enough: All-pay auctions with handicaps. Bonn econ discussion papers, University of Bonn, Germany. Fehr, E., Hoff, K., and Kshetramade, M. (2008). Spite and development. American Economic Review, 98(2):494–99. Fibich, G., Gavious, A., and Sela, A. (2006). All-pay auctions with risk-averse players. International Journal of Game Theory, 34(4):583–599. 21 Filiz-Ozbay, E. and Ozbay, E. Y. (2007). Auctions with anticipated regret: Theory and experiment. American Economic Review, 97(4):1407–1418. Filiz-Ozbay, E. and Ozbay, E. Y. (2010). Anticipated loser regret in third price auctions. Economics Letters, 107(2):217–219. Fischbacher, U. (2007). z-tree: Zurich toolbox for ready-made economic experiments. Experimental Economics, 10(2):171–178. Fudenberg, D. and Tirole, J. (1986). A theory of exit in duopoly. Econometrica, 54(4):943–60. Ghemawat, P. and Nalebuff, B. (1985). Exit. RAND Journal of Economics, 16(2):184– 194. Goeree, J. K., Holt, C. A., and Palfrey, T. R. (2002). Quantal response equilibrium and overbidding in private-value auctions. Journal of Economic Theory, 104(1):247–272. Greiner, B. (2004). The Online Recruitment System ORSEE 2.0 - A Guide for the Organization of Experiments in Economics. Working Paper Series in Economics 10, University of Cologne, Department of Economics. Gth, W., Ivanova-Stenzel, R., Knigstein, M., and Strobel, M. (2003). Learning to bid an experimental study of bid function adjustments in auctions and fair division games. The Economic Journal, 113(487):477–494. Hendricks, K., Weiss, A. M., and Wilson, C. A. (1988). The war of attrition in continuous time with complete information. International Economic Review, 29(4):663–80. Holt, C. A. and Laury, S. K. (2002). Risk aversion and incentive effects. American Economic Review, 92(5):1644–1655. Hrisch, H. and Kirchkamp, O. (2010). Less fighting than expected. Public Choice, 144(1):347–367. Hyndman, K., Ozbay, E. Y., and Sujarittanonta, P. (2012). Rent seeking with regretful agents: Theory and experiment. Journal of Economic Behavior & Organization, 84(3):866 – 878. Kagel, J. H., Harstad, R. M., and Levin, D. (1987). Information Impact and Allocation Rules in Auctions with Affiliated Private Values: A Laboratory Study. Econometrica, 55(6):1275–1304. Kagel, J. H. and Levin, D. (1993). Independent private value auctions: Bidder behaviour in first-, second- and third-price auctions with varying numbers of bidders. Economic Journal, 103(419):868–79. Katok, E. and Kwasnica, A. (2008). Time is money: The effect of clock speed on seller’s revenue in dutch auctions. Experimental Economics, 11(4):344–357. 22 Kimbrough, E. O. and Reiss, J. P. (2012). Measuring the distribution of spitefulness. Research Memorandum 040, Maastricht University, Maastricht Research School of Economics of Technology and Organization (METEOR). Kirchkamp, O., Poen, E. E., and Reiss, J. P. (2009). Outside options: Another reason to choose the first-price auction. European Economic Review, 53(2):153–169. Kirchkamp, O. and Reiss, J. P. (2008). Heterogeneous bids in auctions with rational and markdown bidders - theory and experiment. Jena Economic Research Papers 2008-066, Friedrich-Schiller-University Jena, Max-Planck-Institute of Economics. Kirchkamp, O., Reiss, J. P., and Sadrieh, A. (2008). A pure variation of risk in privatevalue auctions. Research Memorandum 050, Maastricht University, Maastricht Research School of Economics of Technology and Organization (METEOR). Klose, B. and Kovenock, D. (2015). The all-pay auction with complete information and identity-dependent externalities. Economic Theory, 59(1):1–19. Kovenock, D., Baye, M. R., and de Vries, C. G. (1996). The all-pay auction with complete information. Economic Theory, 8(2):291–305. Krishna, V. (2002). Auction Theory. Academic Press. Krishna, V. and Morgan, J. (1997). An analysis of the war of attrition and the all-pay auction. Journal of Economic Theory, 72(2):343 – 362. Lugovskyy, V., Puzzello, D., and Tucker, S. (2010). An experimental investigation of overdissipation in the all pay auction. European Economic Review, 54(8):974–997. Marcus, D. K., Zeigler-Hill, V., Mercer, S. H., and Norris, A. L. (2014). The psychology of spite and the measurement of spitefulness. Psychological Assessment, 26(2):563–574. Morgan, J., Steiglitz, K., and Reis, G. (2003). The Spite Motive and Equilibrium Behavior in Auctions. Contributions in Economic Analysis & Policy, 2(1). Murphy, R. O. and Ackerman, K. A. (2014). Social value orientation: Theoretical and measurement issues in the study of social preferences. Personality and Social Psychology Review, 18(1):13–41. Murphy, R. O., Ackerman, K. A., and Handgraaf, M. J. J. (2011). Measuring social value orientation. Judgment and Decision Making, 6(8):771–781. Nakajima, D. (2011). First-price auctions, dutch auctions, and buy-it-now-prices with allais paradox bidders. Theoretical Economics, 6(3):473–498. Nishimura, N., Cason, T. N., Ikeda, Y., and Saijo, T. (2010). Spite and counter-spite in auctions. 23 Noussair, C. and Silver, J. (2006). Behavior in all-pay auctions with incomplete information. Games and Economic Behavior, 55(1):189 – 206. Ockenfels, A. and Selten, R. (2005). Impulse balance equilibrium and feedback in first price auctions. Games and Economic Behavior, 51(1):155 – 170. Ong, D. and Chen, Z. (2013). Tiger women: An all-pay auction experiment on the gender heuristic of the desire to win. Mimeo. Oprea, R., Wilson, B. J., and Zillante, A. (2013). War of attrition: evidence from a laboratory experiment on market exit. Economic Inquiry, 51(4):2018–2027. Potters, J., de Vries, C. G., and van Winden, F. (1998). An experimental examination of rational rent-seeking. European Journal of Political Economy, 14(4):783 – 800. Riley, J. G. (1980). Strong evolutionary equilibrium and the war of attrition. Journal of Theoretical Biology, 82(3):383 – 400. Sacco, D. and Schmutzler, A. (2008). All-pay auctions with negative prize externalities: Theory and experimental evidence. SOI - Working Papers 0806, Socioeconomic Institute - University of Zurich. Smith, J. M. (1974). The theory of games and the evolution of animal conflicts. Journal of Theoretical Biology, 47(1):209 – 221. A. Proof of Proposition 2 Proof We want to show the following: dbI-AP 1 S = αF (v) dα e ∫ v = 0 (∫ 0 v sf (s)eαF (s) ((α + 1)(F (s) − F (v)) + 1) ds ) sf (s) eαF (s)−F (v) ((α + 1)(F (s) − F (v)) + 1) ds | {z } : =Q(α,F (v),F (s)) ≥ 0 Let us rewrite F (v) = w and F (s) = z and therefore F (z)−1 = s and F (w)−1 = v are the case. Hence, Q(α, w, z) := eα(z−w) ((α + 1)(z − w) + 1). When we now consider the derivative of Q(α, w, z) in α we see that: d(Q(α, w, z)) = (−z + w) eα (z−w) (−α z + α w − z + w − 2) dα = (z − w) eα (z−w) (α + 1) (z − w) +2 | {z ≤0 } | {z } | | = ≤0 24 ≤2 {z ≥−1 {z ≥0 } } Also considering the derivative of Q(α, w, z) in w we see that: ( ) d(Q(α, w, z)) = −1 + (−z + w) α2 + (−2 − z + w) α eα (z−w) dw α (z−w) = −e 1 + α (α + 1) (z − w) +2 | {z } | {z } ≤2 ≥−1 | {z } ≥0 = ≤0 Our goal is to show that the following equation holds (which would be Proposition 2) ∫ F (w)−1 0 F (z)−1 Q(α, w, z)dz ≥ 0 Using the derivatives of Q(α, w, z) in w and in α we get: ∫ F (w)−1 0 ∫ F (z)−1 Q(α, w, z)dz ≥ F (1)−1 −1 ∫ F (w)−1 ∫ 0 ∫ 0 F (1)−1 ∫ F (1)−1 F (z)−1 Q(1, 1, z)dz 0 F (z)−1 ez−1 ((2)(z − 1) + 1) = F (z) Q(1, 1, z)dz = 0 F (z)−1 Q(1, w, z)dz ≥ F (1)−1 F (z)−1 ez−1 ((2z − 1)) dz = | 0 {z } =Q(1,1,z) Considering this we see that this is positive even for the lowest possible values. ∫ 1 0 Q(1, 1, z)dz = 3 · e−1 − 1 ≈ 0.1 ≥ 0 Now we can also show that the function Q(α, w, z) is increasing in z ( ) d(Q(α, w, z)) = − −1 + (−z + w) α2 + (−2 − z + w) α eα (z−w) dz = α (z−w) e 1 + α (α + 1) (z − w) +2 | {z } | {z } ≤2 ≥−1 {z } | ≥0 = ≥0 Now as we know that F (z)−1 is an increasing function and as we know moreover that ∫ ∫ d(Q(α,w,z)) ≥ 0 and also 01 Q(α, w, z)dz ≥ 01 Q(1, 1, z)dz ≥ 0 we conclude the following: dz ∫ 0 F (w)−1 F (z)−1 Q(α, w, z)dz ≥ ∫ F (1)−1 0 25 F (z)−1 ez−1 ((2)(z − 1) + 1) ≥ 0 B. Proof of Proposition 3 Proof To prove that the deviation is increasing in valuation in the first-price auction we can use the result of 1 ) ( ∫ v d(bI-AP − bI-AP 1 S rational ) = αF (v) −αf (v) (α + 1)sf (s)eαF (s) ds + (α + 1)vf (v)eαF (v) − vf (v) dv e 0 (α + 1) = f (v) αF (v) e Now we need to show that ( ↔ ↔ ↔ ( (α+1) v eαF (v) (α+1) v eαF (v) v 0 ( (α + 1) e| {z } ds − v ≥ f (v) v−v eαF (v) ≥1 ) αF (s) ) − v is positive. −v ) −1 −F (v) −F (v)e−F (v) ( ↔ W −F (v)e−F (v) ↔ ∫ W (−F (v)e−F (v) ) −F (v) ≥0 ) ≥ −(α + 1)e−αF (v) ≥ −F (v)(α + 1)e−αF (v) ≥ −F (v)(α + 1)e−F (v)(α+1) ≥ −F (v)(α + 1) −1 ≤α Whereas W(.) represents the Lambert W-function. W (−F (v)e−F (v) ) − 1 is zero. Therefore, the deviation is increasing in According to Maple −F (v) valuation in the first-price auction for α ≥ 0. C. Revenue We have seen that the introduction of spite could explain overbidding in all-pay auctions. In addition, it would be interesting to see some results on revenue ranking in case of spiteful bidders. In this paper we are only looking at two players and therefore the revenue of the seller is just two times the ex-ante expected payment, which is the bid multiplied with the probability to pay the bid: Expected payment: m(v) = Bid[v] · Prob[Paying Bid] Ex-ante expected payment: E[m(v)] := Expected revenue for one player= 12 · Expected revenue for the seller Now we can look at the revenues of the all-pay auctions with spite: Proposition 5. The revenue in the first-price all-pay auctions with two players is given by: ∫ 1 −αF (s) (e − e−α ) ) = 2 RS1AP = 2E(m1AP (α + 1)sf (s)eαF (s) ds (21) S α 0 26 The revenue in the second-price all-pay auctions with two players is given by: ∫ RS2AP = 2E(m2AP ) S 1 =2 2 0 α+1 sf (s)(1 − F (s))ds 2−α (22) Proof ∫ RS1AP = 2E(m1AP ) S = ∫ = 1 2 0 1 2 bI-AP f (v)dv S f (v)e−αF (v) ∫ ∫ 1 0 z}|{ = ∫ 1 2 ∫ 1 2 0 ∫ RS2AP = 2E(m2AP ) S 1 = 2 0 ∫ 1 = 2 0 ∫ 1 ∫ α+1 1−α ∫ 1 s 1 = 2 0 ∫ 1 ( 0 = 2 1 2 0 ∫ 0 v 1 e−αF (v) (α + 1)sf (s)eαF (s) ds −α s (e−αF (s) − e−α ) (α + 1)sf (s)eαF (s) ds α 1 α sf (s)(1 − F (s)) α−1 ds(1 − F (v)) 1−α (2(1 − F (v)))f (v)dv 1 (1 − F (v)) 1−α 2f (v)dv (1 − F (v)) = 2 ∫ f (v)e−αF (v) dv(α + 1)sf (s)eαF (s) ds bII-AP (2(1 − F (v)))f (v)dv S = 2 0 1 s 0 = ∫ 2 if α̸=0 (α + 1)sf (s)eαF (s) dsdv 0 0 = v 2−α 1−α 1 α+1 sf (s)(1 − F (s)) α−1 ds 1−α 1 1 − α 1 α + 1 2 sf (s)(1 − F (s)) α−1 ds α−2 s 1−α 2−α (1 − F (s)) 1−α 2 ) 1 1−α α+1 sf (s)(1 − F (s)) α−1 ds 2−α 1−α α+1 sf (s)(1 − F (s))ds 2−α Using the revenue equivalence principle18 , which says that the expected revenue is the same through all standard auctions (meaning the winner obtains the object for sure) with risk-neutral and rational bidders, we get the first interesting result concerning revenue: 18 For a simple explanation see:Krishna (2002) p. 28. 27 Proposition 6. The expected revenue of the first-price and the second-price all-pay auction in case of spite-experiencing bidders is higher than the expected revenue in case of rational bidders. E(m1AP ) ≥ E(mrational ) S 2AP E(mS ) ≥ E(mrational ) (23) (24) Proof We first compare the revenues of spiteful bidders in first-price all-pay auction and standard-rational revenue RS1AP − Rrational = − mrational ) = 2 2E(m1AP S ∫ 1 α+1∫ 1 sf (s)(1 − F (s))ds sf (s)(1 − e−α eαF (s) )ds − 2 α 0 0 ∫ 1 = 2 sf (s)( α + 1 α + 1 α(F (s)−1) − e − 1 + F (s))ds α α sf (s)( 1 α + 1 α(F (s)−1) − e + F (s))ds α α 0 ∫ 1 = 2 0 It is quite straightforward that this difference is positive as we know that bI-AP ≥ bI-AP . S To see this: → → ≥ bI-AP ≥ f (v) · bI-AP bI-AP S f (v) · bI-AP S ∫1 0 → → f (v) · bI-AP S E(m1AP ) S 2E(m1AP ) S 1AP RS ≥ ∫ 0 1 f (v) · bI-AP ≥ E(mrational ) ≥ 2E(mrational ) ≥ Rrational We now compare the revenues of spiteful bidders in second-price all-pay auction and standard-rational revenue ∫ 1 α+1 sf (s)(1 − F (s))ds sf (s)(1 − F (s))ds − 2 2−α 0 0 ( ) ∫ 1 α+1 = 2 sf (s)(1 − F (s)) 2 − 1 ds 2−α 0 E(m2AP − mrational ) = 2 S ∫ 1 2 As we are interested just whether the difference is positive we check only: 2 α+1 −1 ≥ 2−α 2AP 0 ↔ 2α + 2 ≥ 2 − α ↔ 3α ≥ 0. Therefore E(mS − mrational ) is positive for all α ≥ 0. ∀α ≥ 0 ⇒ RS2AP ≥ Rrational 28 We have seen that the revenues are higher if bidders are experiencing spite relative to the rational case. Therefore, it seems plausible that the revenue equivalence principle does not hold anymore when bidders experience spite and thus we expect a revenue ranking on standard-auction formats for spiteful bidders. For that reason we are going to look at the expected revenues of the four formats (first and second winner-pay and all-pay auctions respectively) and we will try to identify an ordering of those revenues. Let us first look at the revenues of the all-pay auctions. Proposition 7. The expected revenue of the second-price all-pay auction in case of spite-experiencing bidders is higher than the expected revenue of the first-price all-pay auction with spiteful bidders. E(m1AP ) ≤ E(m2AP ) S S (25) Proof ∫ ∫ 1 α+1 (e−αF (s) − e−α ) (α + 1)sf (s)eαF (s) ds − 2 sf (s)(1 − F (s))ds α 2 − α 0 0 ( ) ∫ 1 1 1eαF (s) 2 = (α + 1) sf (s) − − (1 − F (s)) ds α αeα 2−α 0 ) = E(mS1AP − m2AP S 1 ∫ 1 = (α + 1) 0 ∫ 1 = (α + 1) 0 ∫ = (α + 1) 0 1 ) ( (2)eα − 3αeα − 2eαF (s) + eαF (s) α + 2αF (s)eα sf (s) ds (2 − α)(α)eα ( ) (α − 2)eαF (s) + 2(1 + (F (s) − 32 )α)eα sf (s) ds (2 − α)(α)eα ) ( (α − 2)eα(F (s)−1) + 2(1 + (F (s) − 32 )α) ds sf (s) (2 − α)(α) | {z } : =n(F (s),α) If we could show that n(F (s), α) ≤ 0 it would be obvious that the difference is negative. Therefore, we show that this is negative by proofing that it is not positive. n(F (s), α) ↔ (α−2) 2 → (α−2) −1+ 1 α 2 e 2 ↔ W( (α−2) ) 2 → F (s) (α − 2)eα(F (s)−1) + 2(1 + (F (s) − 32 )α) >0→ >0 (2 − α)(α) 3 > −e−α(F (s)−1) (1 + (F (s) − )α) 2 3 3 > −e(1+(F (s)− 2 )α) (1 + (F (s) − )α) 2 3 > −(1 + (F (s) − )α) 2 (α−2) 3 α − 1 − W( 2 ) > 2 α Whereas W(.) represents the Lambert W-function. 3 α−1−W( As we know due to numeric programs that 2 α can conclude the difference is negative, as F (s) ≯ 1. 29 (α−2) ) 2 is one except a null set we Consequently, we know that a seller could make more profit with the second-price than with the first-price all-pay auction if participants would bid according to our model. To get a ranking we are going to look now at the expected revenues of the winner-pay vs. all-pay auction: Proposition 8. The expected revenue of the first-price all-pay auction in case of spiteexperiencing bidders is higher than the expected revenue of the first-price winner-pay auction with spiteful bidders. ) ≥ E(m1S ) E(m1AP S (26) The expected revenue of the second-price all-pay auction in case of spite-experiencing bidders is higher than the expected revenue of the first-price winner-pay auction with spiteful bidders. E(m2AP ) ≥ E(m1S ) (27) S Proof RS1 − RS1AP = 2E(m1S − m1AP ) S ∫ 1 −αF (s) (e − e−α ) 1+α∫ 1 α = 2 (sf (s)F (s) − 2 (α + 1)sf (s)eαF (s) ds 1−α 0 α 0 ) ( ∫ 1 F (s)α − F (s) 1 − eα(F (s)−1) − = 2(1 + α) sf (s) ds 1−α α 0 | {z } :=g(F (s),α) Let us now set g(F (s), α) : = g(k, α): dg(k, α) = dk dg(k,α) dk dk αkα k −1 + eα(k−1) 1−α ( = α eα(k−1) − k α−2 ) Now we want to prove that the second derivative of g(k, α) is negative. Hence, we assume the opposite. ( α eα(k−1) − k α−2 ) → eα(k−1) k 2−α → eα(k−1) k > eα(k−1) k 2−α → eα(k) kα → kα → k 30 >0 >1 >1 > eα α > W(eα α) W(eα α) > = 1a.s. α Whereas W(.) represents the Lambert W-function. Thus, we know that max(k) = 1 and hence k ≯ 1 and hence we know that ⇒ ⇒ dg(k,α) dk dk ≤ 0. dg(k,α) dk dk ≤ 0 → min( dg(k, α) dg(k, α) )= |k=1 = 0 dk dk dg(k, α) ≥ 0 → max(g(k, α)) = g(k, α)|k=1 = 0 dk g(k, α) ≤ 0 Therefore : ∫ 1 2(1 + α) 0 ( ) F (s)α − F (s) 1 − eα(F (s)−1) sf (s) ds ≤ 0 − 1−α α {z | } =g(s,α)≤0 ⇒ E(m1S − m1AP )≤0⇒ S RS1AP ≥ RS1 The second part of the proposition is just straightforward, as we know due to (10) that the revenue of the second-price all-pay is higher than the revenue of the first-price all-pay auction we can easily conclude that E(m2AP ) ≥ E(m1S ) ⇒ S RS2AP ≥ RS1 So far we have the following ranking of expected revenues: Proposition 9. The expected revenues of the standard games (without second-price and third-price winner-pay) with two spiteful bidders can be ranked generally in the following manner: E(m(v)2AP ) ≥ E(m(v)1AP ) ≥ E(m(v)1S ) ≥ E(m(v)rational ) (28) S S Summarizing our results we have seen that in theory spiteful bidders would bid more than rational bidders. Now, we have seen that this behavior would lead to more revenue for the seller. Also, we have seen that the revenue equivalence principle is not applicable with spiteful bidders, and the result is that all-pay auctions are yielding higher revenue than the first-price winner-pay auction. Moreover, we have seen that the second-price all-pay auctions yield higher revenue than the first-price all-pay auction. Now, for the ranking to be complete we have to look at the second-price winner-pay auction. For the ranking of the second-price winner-pay auction in relation to the first-price all-pay auction we obtain the following result: Proposition 10. The expected revenue of the second-price winner-pay auction in case of spite-experiencing bidders is higher than the expected revenue of the first-price all-pay auction with spiteful bidders. E(m1AP ) ≤ E(m2S ) S 31 (29) Proof E(mS1AP − ∫ m2S ) (e−αF (s) − e−α ) (α + 1)sf (s)eαF (s) ds − α 0 1 1+α∫ 1 (sf (s)(1 − F (s) − (1 − F (s)) α ))ds 1−α 0 1 ∫ 1 α(F (s)−1) α 1 − F (s) − (1 − F (s)) 1−e = (1 + α) sf (s) − ds α 1−α 0 1 = ≤ (1 + α) ∫ 0 ∫ = (1 + α) 0 0 (1 + α) 1−α (1 + α) = 1−α 1 1 0 ( ) 1 sf (s) (1 − F (s)) α − α(1 − F (s)) ds (∫ 1 0 1 α sf (s)(1 − F (s)) ds − α ∫ 1 0 sf (s)(1 − F (s))ds ) [ ∫ 1 1 1 (1 + α) −α 1 +1 −α α = s(1 − F (s)) |0 − (1 − F (s)) α +1 ds 1−α 1+α 1+α 0 ( −α s(1 − (1 + α) = 1−α [∫ −1 1 F (s))2 |0 2 1 0 (1 − F (s)) − 1 +1 α ∫ 0 1 21 (1 − F (s)) 2 )] ds ∫ 1 α 1 ds − (1 − F (s))2 ds 1+α 2 0 ] (1 + α) ∫ 1 2α(1 − F (s)) α +1 − (1 + α)(1 − F (s))2 ds = 1−α (1 + α)2 0 = = (1) 2(1 − α) 1 [∫ 0 1 1 2α(1 − F (s)) α +1 − (1 + α)(1 − F (s))2 ds ] ∫ ( ) α−1 1 (1) 1 (1 − F (s)) α +1 2α − (1 + α)(1 − F (s)) α ds ≤ 0 2(1 − α) 0 | {z } :=M (F (s),α)≤0 ( 1 − F (s) − α + αF (s) − 1 + F (s) + (1 − F (s)) α sf (s) ds 1−α = (1 + α) = 1 1 ∫ 1 − F (s) − (1 − F (s)) α sf (s) 1 − F (s) − ds 1−α 1 ∫ 1 1 − 1 − α(F (s) − 1) 1 − F (s) − (1 − F (s)) α sf (s) − ds α 1−α 1 To see that 2α − (1 + α)(1 − F (s)) α−1 α ) ≤ 0 we just derive it on F (s) and hence we 32 get: −1 α − 1 dM (F (s), α) = − (1 + α) (1 − F (s)) α (−1) dF (s) α 1 − α2 1 = − α } (1 − F (s)) α1 | {z | ≥0 ≤ 0 {z ≥0 } Thus, the function is decreasing in F (s). If we look at the maximum we see that it is still negative: M (0, α) = 2α − 1 + α = α − 1 ≤ 0 Therefore, we can conclude that: ) ≤ E(m2S ) E(m1AP S Unfortunately we cannot replicate the same, or even a similar, property onto the second-price all-pay auction as can be seen in the following graph: ) and E(mIIS ) with uniform distributions of valuations. Figure 9: Difference of E(mII-AP S 33 Figure 10: Revenue Ranking with uniform distributions of valuations We can see in the Figure 10 that the gap in revenue between the second-price winnerpay and all-pay auction is first negative (the winner-pay auctions yields higher revenue first) and then turns positive (the all-pay auctions yields higher revenue to the seller). Therefore, the revenues do not have a clear ordering as can be seen in the Figure 9. We can extend Proposition 9, but only in uniform distribution and not in general: Proposition 11. The expected revenues of the standard games with two spiteful bidders and uniform distributions can be ranked as follows: for α ∈ [0, 0.125] and F (v) = v E(m(v)2S ) ≥ E(m(v)2AP ) ≥ E(m(v)1AP ) ≥ E(m(v)1S ) ≥ E(m(v)rational ) S S for α ∈ [0.125, 1) and F (v) = v E(m(v)2AP ) ≥ E(m(v)2S ) ≥ E(m(v)1AP ) ≥ E(m(v)1S ) ≥ E(m(v)rational ) S S Proof This proof is basically simple analysis but nevertheless for reasons of completeness the reasoning is presented: E(m(v)2AP ) S − E(m(v)2S ) ∫ = (1 + α) 0 1 sf (s)(1 − F (s)) 2 1 (1 − F (s)) α −1 − + ds 2−α 1−α (1 − α) =F (v)=v −1/6 34 1 a (8 a − 1) (2 a + 1) (−2 + a) This is again positive when when α ∈ [ 18 , 1). All other orderings were already proofed. Consequently the proof is complete. 35 D. Spite Measure D.1. Spite Measure -Questionnaire The questions of the questionnaire according to Marcus et al. (2014) included the following questions (translated into English): • I would be willing to take a punch if it meant that someone I did not like would receive two punches. • I would be willing to pay more for some goods and services if other people I did not like had to pay even more. • If I was one of the last students in a classroom taking an exam and I noticed that the instructor looked impatient, I would be sure to take my time finishing the exam just to irritate him or her. • If my neighbor complained about the appearance of my front yard, I would be tempted to make it look worse just to annoy him or her. • It might be worth risking my reputation in order to spread gossip about someone I did not like. • If I am going to my car in a crowded parking lot and it appears that another driver wants my parking space, then I will make sure to take my time pulling out of the parking space. • I hope that elected officials are successful in their efforts to improve my community even if I opposed their election. (reverse scored) • If my neighbor complained that I was playing my music too loud, then I might turn up the music even louder just to irritate him or her, even if meant I could get fined. 36 • I would be happy receiving extra credit in a class even if other students received more points than me. (reverse scored) • Part of me enjoys seeing the people I do not like fail even if their failure hurts me in some way. • If I am checking out at a store and I feel like the person in line behind me is rushing me, then I will sometimes slow down and take extra time to pay. • It is sometimes worth a little suffering on my part to see others receive the punishment they deserve. • I would take on extra work at my job if it meant that one of my co-workers who I did not like would also have to do extra work. • If I had the opportunity, then I would gladly pay a small sum of money to see a classmate who I do not like fail his or her final exam. • There have been times when I was willing to suffer some small harm so that I could punish someone else who deserved it. • I would rather no one get extra credit in a class if it meant that others would receive more credit than me. • If I opposed the election of an official, then I would be glad to see him or her fail even if their failure hurt my community. D.2. Spite Measure—Own Measure Our own spite measure is measuring spite similar to the social value orientation measures of Murphy et al. (2011); Murphy and Ackerman (2014). The slider measure of Murphy et al. (2011); Murphy and Ackerman (2014) are presenting the participants with 6 (or 15 if inequality aversion is also measured) allocational sets. For each set the participants need to decide between 9 allocations determining the own payoff and the payoff of the “opponent”. Similarly our spite measure asks participants to decide six times between 9 possible allocation. An overview of the six allocations can be seen in Figure 4. The table below provides all the details. The leftmost allocation is always the nonspiteful allocation and the rightmost allocation is always the maximally spiteful allocation. In the experiment each set was shown on a separate screen and two sets were presented in reverse order. You IA1 Other You IA2 Other You RG1 Other You RG2 Other You PS1 Other You PS2 Other 70 ◦ 100 70 ◦ 100 100 ◦ 100 100 ◦ 100 100 ◦ 85 100 ◦ 85 70 ◦ 98 68 ◦ 96 100 ◦ 98 98 ◦ 96 100 ◦ 83 98 ◦ 81 70 ◦ 96 65 ◦ 92 100 ◦ 96 96 ◦ 92 100 ◦ 81 96 ◦ 76 70 ◦ 94 62 ◦ 89 100 ◦ 94 94 ◦ 89 100 ◦ 79 94 ◦ 72 70 ◦ 92 60 ◦ 85 100 ◦ 92 92 ◦ 85 100 ◦ 78 92 ◦ 68 70 ◦ 91 58 ◦ 81 100 ◦ 91 91 ◦ 81 100 ◦ 76 91 ◦ 63 70 ◦ 89 55 ◦ 78 100 ◦ 89 89 ◦ 78 100 ◦ 74 89 ◦ 59 70 ◦ 87 52 ◦ 74 100 ◦ 87 87 ◦ 74 100 ◦ 72 87 ◦ 54 70 ◦ 85 50 ◦ 70 100 ◦ 85 85 ◦ 70 100 ◦ 70 85 ◦ 50 Each of the six tasks are supposed to measure one feature of spite. The sets IA1 and IA2 are measuring spite when it is behaviorally in line with inequality aversion. A decision maker with positive concerns for social efficiency would choose the allocation with the highest payoff for the other player since this choice also maximises the own payoff. A spiteful person but also an inequality averse person would choose possibly a different allocation. In IA1 being spiteful has no cost. Decision makers get 70 ECU for sure and can basically reduce the payoff of the opponent. In IA2 spitefulness has a cost. In both IA1 and IA2 it may be that the motivation of the decision maker of not maximising the payoff of the other player could be to either harm the other (spite) or to decrease the overall inequality. RG1 and RG2 are measuring spite when spite is behaviorally in line with relative gain. Again, a decision maker with positive concerns for social efficiency would choose 37 Submeasures IA IA-WP RG RG-WP PS PS-WP ∑ No Spite in % Spite in % Average Spite 83.00 93.00 93.00 99.00 93.00 99.00 17.00 7.00 7.00 1.00 7.00 1.00 1.53 0.07 0.26 0.01 0.30 0.01 80.00 20.00 2.09 Table 3: The allocation of choices considered (non-)spiteful in the six allocational tasks of our own spite measure. (a) Interface of the auction spite measure. (b) Interface of the feedback of each auction. Imputing the bidding function for the possible valuations between 500 and 1000. The bidding function will be drawn after the input of the respective bids. Mapping the 10 random valuations and the respective bids on the bidding function. Additionally subjects could see the opponents bid (if the opponent won) and whether they won or lost. Figure 11: Auction Interface the allocation with the highest payoff for the other player since this choice also maximises the own payoff. In RG1 a person who deviates from this choice is considered spiteful as this person decreases the payoff of the opponent. However, this behaviour would also be in line with the behavior of an agent who wants to have relative better payoff compared to the opponent. RG2 is a variant of RG1 where the spiteful choice is costly. In PS1 and PS2 the efficient outcome implies already a positive relative standing of the decision maker who can only decrease the payoff of the other player. Here we find it less plausible to attribute a deviating choice as driven by relative standing. Instead, we take the last two sets as extreme spite. PS2 is a variant of PS1 where the spiteful choice is costly. The allocation of the overall spite in this measure can be seen in Figure ??. The decisions of the individual set can be seen in Table ??. 38 Figure 12: Interface of the bid adaptation. To reduce the demand effect participants were allowed to increase their losing but also the winning bid. The screen was split in two parts so that they made the decisions on the winning part on the right part of the screen and the losing decisions on the left. D.3. Spite Measure -Auction Adaptation The most complicated part of this experiment was actually the adaptation of Kimbrough and Reiss (2012). In the original paper by Kimbrough and Reiss (2012) participants were playing in groups of 3, 16 rounds of an auction (a second-price winner-pay auction which is the design of f.e. eBay). Participants were supposed to bid for an object for which they were provided a personal induced value v ∼ U [500, 1000]. After this participants were doing some effort task. Afterwards participants were informed about whether they won or lost the auction and thereafter they were allowed to increase their bid. However, they also had a possibility to buy the object, they were competing for, at a random price p ∼ U [300, 500] if they lost. We had a similar design to Kimbrough and Reiss (2012) but we changed some aspects of their design. We excluded the fixed price and we also excluded the real effort task. Moreover, we played this part only for one round. Additionally, we measured the participants willingness to pay for the bid adaptation. Therefore, the measure consisted basically of four stages: Stage 1 Participants (I{1,2} ) submitting initial bids (BI{1,2} ) according to the possible valuation. This part was accessed by a strategy-like-approach were we presented the participants with the possible valuation between 500 and 1000 in steps of 50 and asked them to indicate their respective bid (see Figure 11a) Stage 2 10 random valuations with v ∼ U [500, 1000] were drawn for each subject and the respective bids from the bidding function of stage 1 were mapped. Each valuation and bid of each pair represent an auction. Participants were now, informed 39 about the highest bid and bidder in each of the 10 auctions (see Figure 11b). Stage 3 Participants were asked to submit final bids (BF{1,2} ) where BF{1,2} ≥ BI{1,2} . However, the outcome of the auction could not be changed with the final bids. Hence, the initial bids determined who won which auction, but the final bids determined how much the winner need to pay. The participants saw the interface as can be seen in Figure 12.19 Stage 4 Participants were randomly matched with a new partner and were asked to state how much they are willing to pay for the final bid to be implemented. The willingness to pay of each pair was compared and the final bids of the person who stated a higher willingness to pay were implemented. However, those participants whose final bid was implemented needed to pay the willingness to pay of their partner in the stage 4. Hence,the willingness to pay was design-wise also a second price auction which is supposed to be reveling the true willingness to pay as the dominant strategy is to pay the true valuation. E. Linear Bidding As we have seen, that average bidding behavior can be best described by a linear function we want to estimate the optimal linear bidding behavior for spiteful agents. Let us assume that we have two players i and j how are competing for a object which each player values with vk ∈ [0, 1] with k ∈ {i, j}. The valuation is supposed to be drawn from a distribution with density function F (v) and distribution function f (v). We assume also that both are bidding for the object with the bidding function βk (vk ) with k ∈ {i, j}. Additionally u(x) is representing the utility function which is equal for both players. We also assume that both players have the same spitefulness α ∈ [0, 1) with α = 0 being the rational player. The utility of the second price all-pay auction is given by the following equation: ΦII-AP (βi , vi ) S u(vi − βj (vj )) if i wins, i.e. βi > βj = if the bid ties, i.e. βi = βj u(−βi (vi ) − α(vj − βi (vi ))) if j wins, i.e. βi < βj u( vi −β2i (vi ) ) (30) Now, we want to make the following assumptions: • Subjects are not risk averse etc u(x) = x • We assume the uniform distribution F (v) = v. 19 An indicator, that there may be a demand effect existing (or participants did not fully understood this part) is that 41% of the participants increased their bid also in the winning case. It may also be, that participants wanted to ensure that they won or they experienced joy of winning—but in any case behavior can be driven by these motivations only if the task is not completely understood. 40 • Both subjects are bidding linear: βi (vi ) = mi · vi + ti βj (vj ) = mj · vj + tj • We postulate the following restrictions on the bidding function: 0 ≥ βk (vk ) ≤ 1.5. This implies: t k + mk · v ≥ 0 t k + mk · v ≤ b t k + mk · v ≤ b t k + mk · v ≥ 0 ↔ ↔ ↔ ↔ tk ≥ 0 tk ≤ 1.5 mk ≤ 1.5 − tk mk ≥ −tk As we want to find the bidding function which maximizes the utility independent of the actual valuation for the object, we maximize the ex-ante payoff. Which is given by: 41 42 bidder i loses and pays his own bid and additionally experiences spite and pays the loser’s bid mj ) m +t −t ( mi ·vi +ti −tj ∫1 ∫ j mji i ∫ mj (vi − mj · vj − tj )dvj + mi ·vi +ti −tj (−(mi · vi + ti ) − α(vj − (mi · vi + ti )))dvj dvi + tj −ti 0 mj mi ∫ tj −ti (∫ ) 1 m (−(mi · vi + ti ) − α(vj − (mi · vi + ti )))dvj dvi + 0 i 0 (∫ ) 1 1 ∫ m (v − m · v − t )dv dvi j +tj −ti i j j j j 0 mi ( m ·v +t −t ) mj +tj −ti i i i j ∫ ∫ ∫ mj 1 m i (vi − mj · vj − tj )dvj + mi ·vi +ti −tj (−(mi · vi + ti ) − α(vj − (mi · vi + ti )))dvj dvi + 0 0 mj (∫ ) ∫ 1 1 mj +tj −ti (vi − mj · vj − tj )dvj dvi 0 Φi (bi ) mi ( m ·v +t −t ) ∫1 ∫1 ∫ i imj i j (vi − mj · vj − tj )dvj + mi ·vi +ti −tj (−(mi · vi + ti ) − α(vj − (mi · vi + ti )))dvj dvi + tjm−ti 0 mj i tj −ti ( ) ∫ ∫ 1 0 mi (−(m · v + t ) − α(v − (m · v + t )))dv i i i j i i i j dvi 0 ) ( m ·v +t −t ∫1 ∫ 1 ∫ i imj i j (vi − mj · vj − tj )dvj + mi ·vi +ti −tj (−(mi · vi + ti ) − α(vj − (mi · vi + ti )))dvj dvi 0 0 This can be rephrased in the following way: ∫ and and and tj −ti mi tj −ti mi when 0 ≤ when 0 > and tj −ti mi tj −ti mi when 0 > when 0 ≤ mj +tj −ti mi mj +tj −ti mi mj +tj −ti mi mj +tj −ti mi mi ·vi +ti −tj ∫ ∫ 1 min(max( ,0),1) 1 mj E(b) = (vi − mj · vj − tj )dvj + (−(mi · vi + ti ) − α(vj − (mi · vi + ti )))dvj mi ·vi +ti −tj dvi 0 0 min(max( ,0),1) mj {z } | {z } | bidder i wins and obtains the prize >1 >1 ≤1 ≤1 The general approach will be to find an optimal mi and an optimal ti which maximize the utility. Opt As we are only interested in symmetric equilibrium we will search for a mi Sym which Opt i (bi ) fulfills: δΦδm |mi =mj ,ti =tj = 0. And for a ti Sym which fulfills: δΦδti (bi i ) |mi =mj ,ti =tj = 0. i Opt Opt ti Sym will depend on α and mi . The same goes for mi Sym , which is depend also Opt Opt on α and ti . Therefore we will find the optimal bidding function bi Sym with mi Sym Opt and ti Sym . However, we will see that the optimal bidding function is not within the defined area of ti . First we generally derive the ex-ante utility for mi ( ) ( )2 ( ) ( )( )) ( −tj +ti mi +2 mj 2 + −6 ti +6 tj mj +6 −tj +ti (−α−1)mj 2 + (−3 α−3)tj +3 ti +2 α mj +3 (α+1)tj +(α−1)ti −α 1/6 mi 3 ( )2 ( (( ) ) ( ) ) (2 α−1)ti +(α+1)tj −2 α mi +2 ti −2 tj mj −α −tj +ti mi mj +tj −ti (−2+(α+1)mi )mj 2 + −1/6 mj 2 mi 3 δΦi (bi ) ( ) ( (( ) ) ( )) 3 (−4 α+2)mj +2 α mi 4 + (3 α−3)mj 2 + −6 ti +3 tj α+3 ti−3 tj +2 mj +3 α −tj +ti mi δmi 1/6 2m 3 m j i ((( ) ) ( ))( )2 ( )3 +2 ti +1/2 tj α−1/2 ti +1/2 tj mj −1/2 α −tj +ti −tj +ti mi +2 mj −tj +ti 1/6 mj 2 mi 3 (( ) ) ( ) 1/6 (3 α−3)mj 2 + −4 mi −6 ti +3 tj α+2 mi +3 ti −3 tj +2 mj +2 α mi+3/2 ti −3/2 tj mj 2 when 0 ≤ tj −ti mi and when 0 > tj −ti mi and when 0 ≤ tj −ti mi and when 0 > tj −ti mi and To get the optimal mi we derive the utility with respect to mi and assume symmetry = 0 and solve for mi will give us the optimal mi . The symmetric optimal mi is described by the following function: δΦi (bi ) | δmi mj =mi ,tj =ti mi = − 3αti−2α−2 α+1 mi = − 3αti−2α−2 α+1 δΦi (bi ) |mj =mi ,tj =ti δmi mi = − 3αti−2α−2 α+1 m = − 3αti−2α−2 i α+1 43 when 0 ≤ tj −ti mi and mj +tj −ti mi ≤1 when 0 > tj −ti mi and mj +tj −ti mi ≤1 when 0 ≤ tj −ti mi and mj +tj −ti mi >1 when 0 > tj −ti mi and mj +tj −ti mi >1 m m m m Now we generally derive the ex-ante utility for ti ) ( (−2 α+2)ti −mj −2 tj +α mi +mj −2 ti +2 tj 1/2 mi 2 ( )( (( ) ) ( ) ) mj +tj −ti (2 α−1)ti +tj −α mi −tj +ti mj −α −tj +ti mi (mi −1)mj 2 + mj 2 mi 2 −1/2 δΦi (bi ) ) ) ( )) 2 ( ) ( (( mi (−2 α+1)mj +α mi 3 + (2 α−2)mj 2 + −4 ti +2 tj α+2 ti−2 tj +1 mj +2 α −tj +ti δti 1/2 ( mj 2 mi 2 )(( ) ( )) ( )2 −2 −tj +ti −1/2 ti +1/2 tj +α ti mj −1/2 α −tj +ti mi −mj −tj +ti 1/2 mj 2 mi 2 (( ) ) ( ) 2 1/2 (2 α−2)mj + −2 mi −4 ti +2 tj α+mi +2 ti −2 tj +1 mj +α mi +2 ti −2 tj mj 2 when 0 ≤ tj −ti mi and mj +tj −ti mi ≤1 when 0 > tj −ti mi and mj +tj −ti mi ≤1 when 0 ≤ tj −ti mi and mj +tj −ti mi >1 when 0 > tj −ti mi and mj +tj −ti mi >1 To get the optimal ti we derive the utility with respect to ti and assume symmetry: = 0 and solve for ti will give us the optimal ti . The symmetric optimal ti is described by the following function: δΦi (bi ) |mj =mi ,tj =ti δti ti = ti = α−mi+1 2α when 0 ≤ tj −ti mi and mj +tj −ti mi ≤1 α−mi+1 2α when 0 > tj −ti mi and mj +tj −ti mi ≤1 α−mi+1 2α when 0 ≤ tj −ti mi and mj +tj −ti mi >1 α−mi+1 2α when 0 > tj −ti mi and mj +tj −ti mi >1 δΦi (bi ) |mj =mi ,tj =ti δti ti = t = i OptSym This results in the following optimal solutions for ti (α, mi ti = ti = δΦi (bi ) |mj =mi ,tj =ti δti ti = t = i ): α2 −1 2α2 −α when 0 ≤ tj −ti mi and mj +tj −ti mi ≤1 α2 −1 2α2 −α when 0 > tj −ti mi and mj +tj −ti mi ≤1 α2 −1 2α2 −α when 0 ≤ tj −ti mi and mj +tj −ti mi >1 α2 −1 2α2 −α when 0 > tj −ti mi and mj +tj −ti mi >1 44 OptSym Using the optimal ti (α, mi OptSym ) we obtain mi (α, ti mi = mi = α+1 2α−1 when 0 ≤ tj −ti mi and mj +tj −ti mi ≤1 α+1 2α−1 when 0 > tj −ti mi and mj +tj −ti mi ≤1 α+1 2α−1 when 0 ≤ tj −ti mi and mj +tj −ti mi >1 α+1 2α−1 when 0 > tj −ti mi and mj +tj −ti mi >1 δΦi (bi ) |mj =mi ,tj =ti δmi mi = m = i ): Therefore the optimal b sofar seems to be: OptSym bi (vi ) α+1 α2 − 1 = · vi + 2 2α − 1 2α − α To know whether we have a local maximum at the bi (vi )OptSym we need to check whether the hesse matrix is positive definit. Hesse = 1/3 mi −1 + 1/3 mαi 2 − 2/3 mαi − mαi + 1/2 mi −1 + 1/2 mαi 2 − mαi + 1/2 mi −1 + 1/2 −2 α mi 2 α mi + mi −1 + α mi 2 Looking at the eigenvalues of the hesse matrix we get: √ 1/6 13)(2 α mi −α−mi ) mi 2 √ (4+ 13)(2 α mi −α−mi ) mi 2 (−4+ −1/6 1/6 = √ 2 13)(2 α2 +α−1) (−4+ −1/6 (1+α)4 √ (4+ 13)(2 α2 +α−1) 2 (1+α)4 Obviously the eigenvalues are all negative hence, the Hesse-Matrix is negative definit. Therefore, the optimal solution is a local maxima. So we see that the symmetric equilibrium is maximizing the payoff. However, for all α ≤ 0.5 ti is above 1.5 which is the upper bound and for α > 0.5 ti is negative. Therefore, the optimal ti cannot be fulfilled. Furthermore, the bidding function bi (vi )OptSym is above the upper bidding bound (1.5) for α < 0.5. However, the bidding function for α > 0.5 is at least partly in the allowed area, which could describe a stepwise bidding function in these parts. PROBLEMS: Using numerical optimization we find that mi = ti = 0 is optimal for all alphas if both bidder behave the same. However, if player j sticks to this function it is best for player i to change the function. Therefore, we do not have a equilibrium. 45
© Copyright 2024 Paperzz