Auction theory of internet auctions: how can suppliers of internet auctions maximize revenues? University of Groningen

(1)

Auction theory of internet auctions: how can suppliers of internet auctions maximize revenues?

University of Groningen

Faculty of Economics and Business

Rick Kuiper (S2344610) Supervisor: Prof. Dr. A.R. Soetevent

15 July 2017

Abstract:

Internet auctions are becoming more and more popular nowadays. The research question of this paper is whether a supplier of online auctions can maximize its revenues by setting the auction length or whether there are other factors which influence the final price of an auction. This paper analyzes the effect of the auction length, exogenous time variables (day, hour) and endogenous time variables (height of first bid, number of bids) of auctions for four products which took place between November 2016 and January 2017. We found that increasing the duration is not beneficial for the supplier. However, supplying more auctions during the weekend, prime time or between 11PM and 6AM can increase revenues.

(2)

2

1. Introduction

Many economic transactions are conducted via auctions. For example, governments use auctions to sell economic items such as: treasury bills, resource rights for oil fields and assets for firms which the government wants to be privatized. Also common products such as cars, houses and paint are sold in auctions. The literature on auction theory has originated from pioneering work of Vickrey (1961), who received a Nobel prize for his work in 1996. Until the nineties, the problem of auction theory was that it was very hard to acquire data, which made it hard to do research in this field. The start of internet auctions during the mid 90s solved this problem. Because of this, auction theory was not only focused on government auctions, but moved also towards internet auctions.

This paper will analyze English internet auctions, which every interested consumer can join. In this type of auction, the price is raised until there is a winner and the winner wins the object for the price of its bid. In this type of auction a bidder will stay in the auction and bids until the price reaches his personal value, which is the value the bidder assigns to the auctioned product and is the amount he is willing to pay. At that point, the bidder is indifferent between ‘winning’ or ‘losing'. The person with the highest personal will win the auction at the price value of the second-highest bidder (Klemperer, 1999).

(3)

3 factors, such as auctions ending during the weekend, prime time (auctions between 3 and 7 PM) and auctions ending between 11 PM and 6 AM. Moreover, we will observe which effect the number of bidders and the height of the first bid have on the final price of an auction.

This paper covers successive auctions for four products. This means that an auction of product A is auctioned again if the previous auction of the product A has ended. The four products are: two concerts (afterwards called concert A and concert B), two zoo tickets and a hotel voucher for two persons for a night in a hotel. We analyze different products, because we will try to identify whether the outcomes of all auctions are comparable or that there are differences in outcomes between products. We did this, because there is some contradiction in previous literature about the effect of the number of bidders and the timing of the auction on the outcome of an auction (which will be discussed in section 2). This might be, because previous research used different data sources, time periods or different websites which might explain the differences in outcomes. Another possibility is that every product might attract different customers which can result into different bidding behaviour between products. Both concerts mainly attract females above 40 years old. The zoo tickets might attract different types of consumers such as: young and old couples or families with children. Finally, the hotel voucher attracts two people which are most of the times young or old couples. Another difference is that the two concerts have a fixed, predetermined date. The zoo tickets and hotel voucher do not have a certain date, but the voucher and tickets expire after a certain date. Consumers might buy this, because they want to smooth consumption. Because of this, we used different product types to see whether the disagreement in the literature is because of the different data sets or just because that the bidding pattern differs per product and the type of bidders it attracts. The reason we investigate two concerts has to deal with the fact that concert B takes place at the 24th of December, which is our observation period. Concert A takes place in May and is also auctioned after our observation period. Since bidders know that concert B takes place in December, they know in November that this concert is not auctioned for a long period anymore. This might result in different bidding behaviour than for concert A. For this reason, we decided analyze two concerts. Note that we only observed the consumers who have placed a bid. It might be that there were more interested consumers during the auction, but decided not to participate (for example because the current price is higher than their personal value). So, our research is based on the consumers who placed a bid and does not take into account interested consumers which did not place a bid, because we were not able to observe the number of interested consumers who did not place a bid.

(4)

4 ‘almost common value’ auctions in their papers, because not all bidders are symmetric and do not have the same personal value in mind, even though a common value is mentioned. However, this sounds the same as pure common value models, there is a difference between these concepts. In almost common value auctions, giving bidder i a slightly higher value compared to bidder j when i wins, bidder i is willing to bid more aggressively. Bidder j has to be cautious, because he can face the winner’s curse and pay more than he initially wanted. Examples according to our data might be that the common value price is €100 for 2 concert tickets, but that a certain bidder already has 2 tickets, but wants to go with his friends and needs 2 more. Because he definitely wants 2 more tickets, he might be willing to bid slightly more than other bidders want to and is willing to bid more than €100 (which also can be observed in our data). These small asymmetries can have a large effect on the price. It can be that escalation of bidding can play a role in auctions. It might be that people really want an auctioned product and that they bid desperately high in order to ‘win’ the auction and get the product. Ariely and Simonson (2003) argue that consumers see the outcome of an auction as either a ‘win’ or a ‘loss’. The question is whether bidders act rationally or that bidders do not act rationally anymore when deciding to enter an auction. Besides, after losing an auction, it might be that the bidder will bid desperately high the next auction in order to win the product. There is no previous research which investigates whether a bidder faces escalation of commitment. This paper tries to fill this gap and analyzes whether bidders face this escalation of commitment and will bid higher in the next auction after losing the current one. We do this by observing the difference in bids of a certain bidder between the current auction and the previous auction. Besides, we observe whether the previous auction is won or lost, in order to investigate whether bidders increase their bid in the next auction after losing the previous one.

(5)

5 participate. We analyze whether these results also hold for auctions. If it does have an effect, the moment of getting salary might have two effects from the demand side. First, the number of bidders increases which increases price. Second, when people have received their salary, they are less financially constrained and can afford to bid higher.

Finally, the first bid can determine the final price of the auction. It has been demonstrated that a higher first bid will increase the final auction price (Ariely and Simonson, 2003; Bapna et al. (2008). Although, the first bid can have a relation with the number of bidders. If the first bid is high, more potential bidders will drop out, because the price is above their personal value. Previous research demonstrated that as the number of bidders increases, final prices do (Hong and Shum, 2002). Moreover, English auction studies found a positive relation (Nelson, 1997; Bailey et al., 1993). However, not all studies find a positive relation (Pinkse and Tan, 2005; Li and Zeng, 2009). Besides, in an empirical study of repeated auctions, there is also found a negative correlation (Bajari and Hortaçsu, 2003). We investigate whether the number of bidders and the height of the first bid have an effect on the outcome of the auction. Based on previous research, we can conclude that the number of bidders certainly have an effect on the winning bid, although there is discussion about whether the effect is positive or negative. The effect might depend on the type of product which is presented during the auction.

The following three research questions will be addressed:

Can the organizer of the auctions maximize his revenues by choosing the length of the auction? Which other factors do influence the height of a winning bid of an internet auction?

Do these factors influence every auction on the same way or are there difference between different product categories?

Do bidders bid more in the next auction after losing an auction?

This paper will be structured in the following manner. Section 2 discusses the main theories and concepts, reviews the auction theory literature and state hypotheses. Thereafter, section 3 discusses the econometric model and methodology. Section 4 presents the data used for the empirical research. Section 5 presents the results. Finally, section 6 provides a conclusion.

2. Literature review

(6)

6

2.1 Length of the auction

The organiser of the auction can determine the duration. With this decision he tries to maximize his revenue. Haruvy and Leszczyc (2010) investigated in a field experiment the effect by extending the auction duration from 1 to 3 days for eBay auctions and from 1 to 10 days and for a local auction site with a lower number of auctions and a more constant number of participants. Increasing the auction duration for eBay auctions does increase the number of bidders and the outcome of the auction. On the other hand, increasing the duration for the local site will decrease the outcome. Further, there is no evidence found for a higher number of bidders on the local site when extending the duration. Laboratory experiments found evidence that shorter durations do increase the revenue (Kwasnica and Katok, 2007; Katok and Kwasnica, 2008). We can use these findings to observe the effect of the auction length. However, we have a different dataset than Havuvy and Leszczyc. In our dataset a new auction which is supplied after the previous auction has ended. Besides that, do we have durations between one minute and 20 hours, instead of auctions which have a duration of 1, 3 or 10 days. However, we can investigate whether increasing the duration with one or several minutes has an effect on the outcome. Lastly, in laboratory experiments it is easier to control for external factors than in field experiments, which might imply different results in the lab than for field experiments. Another difference is that Katok and Kwasnica (2008) provided evidence for Dutch auctions and we investigate the effect of English auctions.

2.2 Timing of the auction

During the week, there are popular moments to participate in an auction and less popular moments. Most of the action during auctions happens at the beginning of the evening and during the weekend, when most households do not have to work and have time to participate in auctions. Livingston (2003) argues that auctions in the late night between 12 and 6 AM may not receive as much attention. On the other hand, auctions ending during prime time (3PM-7PM) receive much more activity. Not only the time of the auction matters for the price, also the auction day might influence the auction result. Melnik and Alm (2002) considered, besides prime time, also the influence of the weekend. They provided two specifications, one with Friday, Saturday and Sunday and one only with Saturday and Sunday. Both variables have a positive effect on the final price of the auction. However, time does not necessarily have an effect on the final price of the auction (Cabral and Hortacsu, 2010; Lucking-Reiley et al., 2007).

2.3 First bid and number of bids

(7)

7 which results in less severe competition and a lower price. The other side of the story is that a higher first bid might result in a higher auction price. The first bid should affect the participation effect by bidders, but not the decision how much it will bid (Livingston, 2005). So, it might be that the height of the first bid is correlated with the number of bids placed in an auction. The number of bids gives an indication of the how severe the competition is during the auction. Intuitively, more competition decrease prices in markets, while in auctions more competition will increase prices. Most studies use in determining which factors explain the final price of the auction one of the two variables. McDonald and Slawson (2000) indirectly used both variables to determine the effect of the first bid and the number of bids. They found a negative relation between the amount of the opening bid and the number of bids. Moreover, the final price has a positive relation with the number of bids. This also indicates a negative relation between the amount of the opening bid and the final price. In our analysis, we will both investigate the effect of the first bid and the number of bids placed.

McDonald and Slawson were able to combine the number of bids and the amount of the first bid to determine the effect on the final price. However, most research used just one of these variables. As mentioned in the introduction, theory is inconclusive about the relation between the height of the winning bid and the number of bidders. This can be explained by the fact that auctions for different type of products yield different results. This paper will take this into account and test different products. Since we have different results in the literature about the effect of the number of bidders it is hard to perform an expectation whether we expect a positive or a negative effect. However, based on McDonald and Slawson, we expect a positive relation.

Previous research about the effect of the height of the first bid on the final price is really one-sided. Wang et al. (2008) and Lucking-Reiley et al. (2007) found a positive relation.

2.4 Paycheque

(8)

8 Receiving the paycheque might influence bidding behaviour at auctions. Since there is no paper which researched this effect, this paper wants to fill this gap. When households just received their paycheque, they consume more as mentioned in the paragraph above. Since consumption increases, households might also be willing to participate in auctions or willing to bid more than they will do before receiving the paycheque. We expect that the first week after receiving the salary has a positive effect on the final price of auctions. If households have more money to spend and are willing to bid (slightly) more than they otherwise do. Moreover, households who are more constrained are not willing to participate in auctions, but when they just received their paycheque they do. We will identify the effect of the first two days after receiving the paycheque, the first four days and the first week. We do this to research whether the effect of receiving salary is different for the mentioned time periods.

2.5 Learning effect

Since we observe multiple auctions, the learning effect can change the strategy of the bidders. Jeitschko (1997) found that revelation of winning bids in previous auctions will update the believe of bidders which opponents they face in an auction. This update will affect the bidding pattern of bidders, which means that they bid, on average, lower than when they do not know this information. However, Jeitschko used two auctions and just three bidders. So, after the first auction has been played, the two ‘losing bidders’ of the first auction compete in the last auction and do exactly know the willingness to payoff of each other. Our auctions are also repeated, but the bidders will never know who their opponents are. So, the learning effect consists in our setting only in knowing the previous winning bids. Further, the bidder never knows whether bidders of previous auctions will participate in the current one or that ‘new bidders’, who did not participate in the previous auctions, will join. Despite that our setting is different than Jeitschko’s, we might find that final auction prices decrease after more auctions have taken place, just as Jeitschko found, because bidders can update their beliefs by knowing what the winning bids of the previous auctions were.

2.6 Loser’s curse

(9)

9 bidder, it might think what he wants to do with the ‘won product’. However, he does not currently own the product, but is at that moment the highest bidder. If he visits the website of the auction a few hours later and observes that he is not the highest bidder and thus will ‘lose’ the auction, he might increase his bid to reclaim the endowment. However, we observe auctions and products which are supplied more than once a day. So, do we also find this excessive bidding or does this only hold for very rare products, which are auctioned once?

Another possibility is that after losing an auction, the ‘losing consumer’ will bid more aggressively during the next auction for the same product in order to ‘win’ the auction and thus ‘winning’ the product. In some auctions we observe that winners of an auction will also participate in the next auction. Do these winners behave differently than losers and stick to the value of the previous won auction? Or do these winners escalate in order to win? In order to test this, we observe whether the winner of an auction did bid in the previous five auctions and whether we can find differences between these previous auctions. Moreover, we will also identify whether winning or losing one of the previous auctions will influence the bidding behaviour of the auction.

2.7 Hypotheses

Based on previous mentioned literature and theory which has been discussed, a couple of hypotheses have been formulated. These are based on previous research on auctions (especially on Ebay). On the basis of these hypotheses this research will be based. Based on previous mentioned literature and economic theory the following hypotheses have been formulated:

- Hypothesis 1: increasing the duration of an auction will have a positive effect on the final price.

- Hypothesis 2a: an auction ending during prime time (between 3PM and 7PM) will have a positive effect on the final price of the auction.

- Hypothesis 2b: an auction ending in the weekend will have a positive effect on the final price of the auction.

- Hypothesis 2c: an auction ending between 11PM and 6AM will have a negative effect on the final price of the auction.

- Hypothesis 3: the height of the first bid will have a positive effect on the final price of the auction.

- Hypothesis 4: the number of bids placed in an auction have a positive effect on the final price. - Hypothesis 5: the paycheque receipt has a positive effect on the final price.

(10)

10

3. Methodology

This section discusses the methodology that has been applied in order to test the expectations mentioned in the previous section. It will explain the econometric model which has been used and discuss the estimation techniques which have been used.

3.1 Model specification

In order to test which factors have an impact on the final auction price and to test the expectations, an auction function will be estimated. An auction function will explain which factors do influence the end price of an auction. Moreover, we use auctions for different types of products in order to observe whether the factors on the end price hold for auctions in common or just for a certain type of products (section 2.3 gives an example of this). The function will have the following specification:

LNFINALPRICEt = β1 + β2*AUCTIONLENGTHt + β3*PRIMEt + β4*WEEKENDt + β5*LATEt +

β6*FIRSTBIDt + β7*NUMBERBIDSt + β8*PAYCHEQUE2t + β9*PAYCHEQUE4t +

β10*PAYCHEQUE7t + β11*AUCTIONNUMBERt + β12*AUCTIONNUMBER²t + t (1) t = auction number = 1,2, ..., T

Where LNFINALPRICEt is the log of the winning bid in the auction t. We decided to take the log,

because we observe auctions for different products. To compare the outcomes, it is easier to have changes in percentages rather than in Euros, because the final prices and market values are different for every product. We also checked whether the error term is more normally distributed with a log specification. We performed a Jarque-Bera test and found for both the linear specification and the log specification a p-value of 0.000. This indicates that both specifications are normally distributed. AUCTIONLENGTHt indicates the length of the auction t. With this we will test whether the supplier

of the auctions can maximize its revenues by setting the duration of the auction. PRIMEt is a binary

variable which equals 1 if the end of the auction is between 3PM and 7PM in auction t, zero otherwise and WEEKENDt indicates an auction ending during the weekend. LATEt is a binary variable which

indicates whether the ending of the auction t is in between 11PM and 6AM. FIRSTBIDt indicates the

amount of the first bid in auction t and NUMBERBIDSt the number of bids placed in auction t.

PAYCHEQUE2t indicates whether the end of the auction is in the first two days after receiving the

(11)

11 also perform a robustness analysis where we do not take into account the period 15 December until 15 January to see whether results change. Finally, we included the auction number and the square of the auction number in order to see whether there is a linear or quadratic relation in the final prices of time and to see whether we can find a learning effect (discussed in section 2.5). In addition, day and part of the day fixed effects will be included in the model specification. The model will be estimated with OLS. For this model, we use clustered standard errors, in order to adjust for possible serial correlation and heteroskedasticity. We have bidders which participate in more than one auction, therefore we cluster standard errors by bidder in order to adjust for potential heteroskedasticity and serial correlation.

A final problem might be multicollinearity, because the number of bids and the amount of the first bid are related. If the correlation is more than 0.80 or less than -0.80 we have multicollinearity and have to adjust our model. So, for every product type we have to observe the correlation between these two variables. Fortunately, no model has multicollinearity problems, but we will discuss this in section 5.

3.2. Hypothesis testing

In order to test the relation between the end price and the end duration of an auction, a significance test for β2 will be performed. If the final price is positively influence by increasing the duration, it gives the following hypothesis:

Hypothesis 1: H0: β2 = 0 | H1: β2 > 0

For the 5 other hypotheses, the following significance tests will be performed:

Hypothesis 2a: H0: β3 = 0 | H1: β3 > 0 Hypothesis 2b: H0: β4 = 0 | H1: β4 > 0 Hypothesis 2c: H0: β5 = 0 | H1: β5 < 0 Hypothesis 3: H0: β6 = 0 | H1: β6 > 0 Hypothesis 4: H0: β7 = 0 | H1: β7 > 0 Hypothesis 5: H0: β8 β9/β10 = 0 | H1: β8 β9/β10 > 0 3.3. Second specification

In order to test hypothesis 6, we have to create another model specification. In order to take into account whether a bidder has lost or won an auction and whether it will increase its bid during the next auction(s), the next specification is created:

SPREADi,t,t-x = β1 + β2* PRIMEi,t + β3* WEEKENDi,t + β4*LATEi,t + β5*LOSTi,t-x,

+ β6*AUCTIONNUMBERi,t + β7*AUCTIONNUMBER

2

i,t + i,t (2)

i = bidder = 1,2, ..., N

t = auction = 1,2,...,T

(12)

12 Where SPREADi,t,t-x is an index for the spread, which compares a bid made by person i in auction t

with a bid in auction t-x. We will observe the spreads between auction t up to auction t-5. The index is build up by taking the bid of person i in auction t, divide it by the bid of bidder i in auction t-x and multiplied by 100 to construct the index. So, if the index is 100 the bid of person i in the current auction is the same as the bid of person i in the previous observed auction. The variables PRIME, WEEKEND and LATE have the same function as in equation (1) and indicate whether the current auction ends between 3PM and 7PM, during the weekend or between 11PM and 6AM. We did this to investigate whether the exogenous time variables have an influence on the spread. Further, we added auction number and the square of the auction number for the same reason as we did in equation (1). The model will be estimated with OLS.

In order to test the sixth hypothesis, the variable LOSTi,t-x is included. This is a binary variable which

indicates whether auction t-x, which is indexed to auction t, is lost by person i. Our dataset has auction winners who also have won one of the previous five auctions for the same product. Further, the dataset also has winners who have lost previous auctions. Therefore, we can investigate whether winning or losing will influence the decision to bid differently or that bidders stick to a predetermined amount (which equals the personal value) and that the outcome of the previous auction does not influence bidding in the current auction. In order to test the relation between the spread and winning the previous auction, we test the significance for β5. We perform the following significance test in order to test hypothesis 6:

H0: β5 = 0 | H1: β5 > 0

4. Data

This sections provides a description of the data and how these data have been recorded. Moreover, it explains which products we decided to include in our analysis.

4.1 Data collection

(13)

13 of products. However, relating outcomes to different types of bidders is just an assumption, because we do not have any information about the bidders. This paper covers two concerts, because concert B is held at the 24th of December which is in our observation period. As discussed in the introduction, this might change bidding behaviour, because in November bidders know that this concert is not supplied a very long period anymore.

Our dataset consists out of successive auctions. This means that when auction t for concert A has ended, immediately after the end of auction t, a new auction t+1 will be supplied for concert A. This also holds for all other products. Because of this, we can analyze the behaviour of bidders in successive auctions and compare the placed bids of person i in auction t up to auction t-x in order to analyze whether we find evidence for escalation of commitment behaviour. The next sections will provide more information about these auctions and the descriptive statistics.

4.1 Data concert A

This concert is the most popular product on the site with a supply of 6,719 auctions and 14,533 unique bidders during these auctions. In this auction the winner wins two tickets and the tickets have a market value of €97,90. However, as we can see in graph 1, there are only a few persons who bid close to this amount. Further, table 1 presents the descriptive statistics of concert A. We have a wide range of final prices (lowest €10 and highest €100), the height of the first bid and the number of bids placed. In figure 2 we see the percentage of auction supplied per date over the total number of auctions. So, the first day of the month supplies approximately 10% of all auctions. The red line indicates 3.23%. This would be the percentage if the auctions were equally distributed over all days of the month. If there is on a certain date more supplied Relating this to days after the paycheque receipt, we see that at the 27th and the 28th of the month the supply is higher than the average 3.23%.

Table 1: Descriptive statistics concert A

In table 2 the descriptive statistics for the second specification are presented. We see that the index is more than 200 for all auctionst-x indicating that on average consumers place in auction t a bid which is

more than two times the bid placed in auction t-x. Based on the data, we see that on average bidders increase their bids with 105% to 154% compared to the previous auctions. Moreover, we see that all the dummy variables lost-t have values close to 1. This indicates that a small number of bidders

Observations Mean Std. Dev Min. Max.

(14)

14 0 5 0 0 1 0 0 0 F re q u e n cy 0 20 40 60 80 100 Final price 0 2 4 6 8 10 p e rce n ta g e 1 6 11 16 21 26 31

day of the month

continue bidding after winning an auction. Another possibility is that just a few bidders win an auction.

Table 2: Descriptive statistics concert A for the second specification

4.2 Data concert B

This concert has a supply of 762 auctions in the collected period and 2,167 unique bidders during these auctions. The winner wins 2 tickets for the concert and the market value of the tickets is €107.90. However, also seen for concert A, few consumers bid close to this amount which can be seen in figure 3 (the final prices ranges from €29 to €110). Further, table 3 presents the descriptive statistics of concert B. We observe that auctions for this concert have, on average, two times the duration of concert A. We see that these auctions are only supplied between the 22th and the 29th of the month. Therefore, the supply is high during the first days after receiving the paycheque. The red line indicates, just as for concert A, 3.23% (see figure 4).

Table 4 presents the descriptive statistics for the second specification. We decided not to use the specification for Spreadi,t,t-5, because of a too small number of observations (less than 50). Comparing

this to the specification of concert A we observe higher spreads. This indicates that bidders bid in auction t between 136% and 232% more than in auction t-x. Moreover, we observe lower values for

Spreadi,t,t-1 12,939 204.794 531.982 1.521 6,300 Spread i,t,t-2 4,758 229.083 581.030 1.615 6,800 Spread i,t,t-3 2,773 239.703 584.741 1.723 6,100 Spread i,t,t-4 1,878 253.930 645.421 1.615 6,400 Spread i,t,t-5 1,233 212.008 469.981 1.729 5,100 Lost-1 12,939 0.966 0.181 0 1 Lost-2 4,758 0.982 0.134 0 1 Lost-3 2,773 0.989 0.105 0 1 Lost-4 1,878 0.995 0.069 0 1 Lost-5 1,233 0.999 0.028 0 1

(15)

15 0 50 1 0 0 1 5 0 2 0 0 2 5 0 F re q u e n cy 20 40 60 80 100 120 Final price 0 4 8 12 16 20 p e rce n ta g e 1 6 11 16 21 26 31

day of the month

lost. This indicates that for this concert there are more bidders which continue bidding after winning an auction.

Table 3: Descriptive statistics concert B

Table 4: Descriptive statistics concert B for the second specification

4.3 Data zoo tickets

The auction for zoo tickets is supplied 855 times in the collected period and has 3,555 unique bidders during these auctions. The winner wins 2 entrance tickets for the zoo. The market value of these depends on the visiting day, but varies between €40 and €50. In figure 5 the spread of final prices is presented (the final prices ranges between €7 and €50). We see that the spread in final prices for the zoo voucher is higher than for both concerts (higher std. dev) Further, table 5 provides descriptive statistics. We see that the supply during the first days after receiving the paycheque is higher than the

Ln final price 762 54.210 6.224 29 110 First bid 762 9.996 13.897 1 50 Number of bids 762 11.438 5.7774 1 31 Duration 762 14.285 53.642 2 944 Prime 762 0.232 0.423 0 1 Weekend 762 0.301 0.459 0 1 Late 762 0.073 0.261 0 1

Spreadi,t,t-1 1,539 236.720 626.229 1.613 6,600 Spread i,t,t-2 546 243.463 547.107 1.724 5,000 Spread i,t,t-3 327 303.852 725.281 1.923 4,700 Spread i,t,t-4 218 323.171 761.928 4 5,100 Lost-1 1,539 0.969 0.174 0 1 Lost-2 546 0.940 0.239 0 1 Lost-3 327 0.938 0.251 0 1 Lost-4 218 0.917 0.276 0 1

(16)

16 0 2 4 6 8 p e rce n ta g e 1 6 11 16 21 26 31

day of the month

red line, indicating that during the first days after the paycheque receipt there is a higher supply than when the supply is equally distributed over the month (see figure 6).

Table 6 presents the descriptive statistics for the second specification. We do not use the specification for Spreadi,t,t-5 because the number of observations is smaller than 50. Further, we observe that the

spreads of bids placed in auction t and auction t-x are smaller than for both concerts. Lastly, we see that the coefficients for lost are comparable to the coefficients of concert A. Compared to the coefficients of concert B, we see that concert B has more winners which continue bidding after winning an auction.

Table 5: Descriptive statistics zoo voucher

Table 6: Descriptive statistics concert B for the second specification

Ln final price 855 20.629 3.830 9 50 First bid 855 3.726 4.939 1 25 Number of bids 855 9.159 3.952 1 23 Duration 855 98.011 115.698 5 1,158 Prime 855 0.225 0.420 0 1 Weekend 855 0.269 0.444 0 1 Late 855 0.077 0.267 0 1

Spreadi,t,t-1 1,119 174.445 289.114 3.571 2,200 Spread i,t,t-2 390 204.809 322.930 3.846 2,000 Spread i,t,t-3 210 194.863 331.962 3.846 2,400 Spread i,t,t-4 136 234.950 352.473 5.556 2,000 Lost-1 1,119 0.971 0.169 0 1 Lost-2 390 0.972 0.166 0 1 Lost-3 210 0.981 0.137 0 1 Lost-4 136 0.993 0.086 0 1

Figure 5: normal distribution final price zoo voucher

0 50 1 0 0 1 5 0 2 0 0 2 5 0 F re q u e n cy 10 20 30 40 50 Final price

(17)

17 0 20 40 60 80 F re q u e n cy 20 30 40 50 60 Final price 0 20 40 60 p e rce n ta g e 1 6 11 16 21 26 31

day of the month 4.4 Data hotel voucher

The auction for the hotel voucher has 351 observations and has 1,776 unique bidders. The winner gets a voucher and with this voucher the winner can sleep at a hotel room for 1 night with 2 persons. There is no mentioned market value, since the market value depends on the day of arrival and the choice of hotel. Table 5 presents the descriptive statistics and in figure 7 we observe the spread between all winning bids. We observe that the standard deviation for the final price is higher than for both concerts, but lower than for the zoo voucher. In figure 8 we see the number of auctions per date. We see that the supply of auctions is very high at the 14th and the 15th of the month and very small during the rest of the month. We also see a very small percentage of auctions supplied at the end of the month. We do not run the second specification for the hotel voucher, since we have less than 50 observations for all variables of Spreadi,t,t-x

Table 7: Descriptive statistics hotel voucher

5. Results

We will run six regressions for every category. First, we will only use the exogenous variables (columns 1 to 3) to test the effect of the time variables. Afterwards, we will add the endogenous

Final price 351 31.306 5.578 18 61 First bid 351 5.303 7.500 1 32 Number of bids 351 9.560 4.064 2 25 Duration 351 122.897 227.045 2 1,258 Prime 351 0.254 0.436 0 1 Weekend 351 0.097 0.297 0 1 Late 351 0.071 0.258 0 1 Paycheque2 351 0.040 0.196 0 1 Paycheque4 351 0.049 0.215 0 1 Paycheque7 351 0.063 0.243 0 1

(18)

18 variables amount of bids and height of the first bid (column 4 to 6). Moreover, we test the second specification for concert A, B and for the zoo tickets in order to see whether we can observe some excessive bidding in our auctions.

5.1 Concert A

(19)

19 extra bid during an auction, the price increases approximately by 0.07%. Lastly, if the first bid is one euro higher, the final price increases by approximately 0.076%. Most results are in line with our hypothesis, except for the effect of the paycheque receipt. The first two days have a negative effect and it becomes positive after seven days. This result might be explained by the fact that assumption for the December paycheque is not right. Therefore we also run a robustness analysis which does not take into account auctions between 15 December and 15 January.

The results for the second specification are presented in table 9 with the spread between the bid in auction t and the bid in auction –t by individual i as dependent variable. We see that in all specifications the exogenous time variables are insignificant and thus do not influence the spread. In the regression with Spreadt,t-1 auction number and auction number² are significant at the 10%

confidence interval. The coefficient is positive for auction number, indicating that the spread increases over time when there are more auctions finished. Further, Auction number² has a coefficient of 0.0002, which is very small. However, it does influence the spread, especially when many auctions have taken place. Losing an auction will increase the bid only if the loss has happened in the previous auction or 2 auctions ago. So, only the win or loss of the last two auctions are taken into account when bidding for the current auction. We observe a very low R². Less than 1% of the variation of the model is explained. This will also be the case in the second specification for the other products. However, this does not mean that our model is bad. We have very big outliers in both the left and right tail of our spread variable which may cause this low R².

Table 10 presents our robustness analysis. We found in our normal specification a negative relation on the final price for the first two days after receiving the paycheque, no relation for the first four days and a positive relation for the first 7 days after receiving the paycheque. Doing a robustness analysis and drop observations between 15 December and 15 January, we might find other results. What we immediately notice is that we have 2,433 observations left and we observe a much lower R². The effect of the paycheque receipt is now only significant and positive for the first 7 days compared to the normal analysis of table 8. Further, we see that prime time, duration and auction number² do not influence the final price anymore. Weekend has now only a positive relation which is significant when we include the first 7 days after the paycheque receipt. Lastly, the effect of the number of bids and the height of the first bid are more than two as high as in table 8. Other results are comparable to the results found in table 8. In conclusion, we find for the paycheque receipt results which are in line with our hypothesis. However, some results are now insignificant and we have a much lower R².

5.2 Concert B

(20)

20

Table 8: Empirical results for concert A

Note: *** significant at .01; ** significant at .05;* significant at 0.1. Standard errors are within the parentheses. The dependent variable is the log of the final price of the auction. Hour and day dummies are used. Moreover, standard errors are clustered on bidder id.

Table 9: Empirical results for the second specification for concert A

Note: *** significant at .01; ** significant at .05;* significant at 0.1. Standard errors are within the parentheses

(1) (2) (3) (4) (5) (6) Constant 4.109*** 4.126*** 4.118*** 4.109*** 4.110*** 4.102*** (0.011) (0.010) (0.010) (0.011) (0.010) (0.010) Prime 7.336*** 7.387*** 7.269*** 7.202*** 7.256*** 7.136*** (0.009) (0.009) (0.009) (0.009) (0.009) (0.009) Weekend 3.623*** 3.571*** 3.737*** 3.580*** 3.517*** 3.693*** (0.005) (0.005) (0.005) (0.005) (0.005) (0.004) Late -15.446** -15.540** -15.676** -15.105*** -15.177** -15.336** (0.066) (0.067) (0.066) (0.065) (0.065) (0.064) Paycheque2 -2.126*** -2.110*** (0.007) (0.007) Paycheque4 0.034 -0.026 (0.005) (0.004) Paycheque7 1.620*** 1.599*** (0.005) (0.005) Duration 0.010*** 0.011*** 0.011* 0.010 0.011* 0.011* (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number -0.009*** -0.009*** -0.009*** -0.009*** -0.009*** -0.009*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number² 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of bids 0.069*** 0.072*** 0.068*** (0.000) (0.000) (0.000)

Height of first bid 0.077***

(0.000) 0.076*** (0.000) 0.076*** (0.000) N 6,719 6,719 6,719 6,719 6,719 6,719 R² 0.681 0.677 0.677 0.681 0.680 0.681

Spreadt,t-1 Spreadt,t-2 Spreadt,t-3 Spreadt,t-4 Spreadt,t-5

(21)

21

Table 10: Robustness analysis paycheque for concert A

Note: *** significant at .01; ** significant at .05;* significant at 0.1. Standard errors are within the parentheses. The dependent variable is the log of the final price of the auction. Hour and day dummies are used. Moreover, standard errors are clustered on bidder id.

reason as we did for concert A. Lastly, we multiplied the coefficients with 100, so that we immediately can observe the percentage change of the variables. When an auction ends between 3 and 7PM the final price is 14.24% higher in the first two columns and 19.71% higher in column three. Compared to concert A, we see that this effect is bigger for concert B. In the third model, we find a positive and significant coefficient for weekend. This indicates that in auctions ending during the weekend, will increase the final price with almost 30%. We did not find a significant relation for the first two models. In concert A we found in all our models a positive relation. However, the effect was much smaller than what we found in column 3. An auction ending between 11PM and 6AM has a lower final price between 14.81% and 16.16%. These numbers are comparable to our findings of concert A. The paycheque receipt has only a significant coefficient when we take into account the first 7 days, only we observe that the coefficient is very high. A possible explanation might be that the paycheque interval is roughly the same as our observation period since we have only observations from the 22th until the 29th of the month. This is a drawback of this auction in determining the effects of the pay date. We did not find a significant relation when using the first two or first four days after receiving the paycheque. (1) (2) (3) (4) (5) (6) Constant 4.114*** 4.104*** 4.352*** 4.261*** 4.348*** 4.290*** (0.191) (0.179) (0.161) (0.183) (0.172) (0.157) Prime 0.248 0.360 0.543 0.373 0.472 0.653 (0.043) (0.043) (0.043) (0.042) (0.042) (0.041) Weekend 0.350 0.605 2.363*** 0.466 0.717 2.428*** (0.009) (0.010) (0.009) (0.009) (0.010) (0.008) Late -12.556* -12.492* -12.501* -12.630* -12.571* -12.582** (0.070) (0.070) (0.070) (0.068) (0.067) (0.008) Paycheque2 -0.008 -0.643 (0.106) (0.010) Paycheque4 0.467 0.453 (0.008) (0.008) Paycheque7 1.910** 1.905** (0.008) (0.008) Duration -0.006 -0.006 -0.006 -0.005 -0.005 -0.005 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number -0.014** -0.017*** -0.015*** -0.017** -0.017*** -0.014*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number² 0.000 0.000 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of bids 0.174*** 0.175*** 0.175*** (0.000) (0.000) (0.000)

(22)

22 Further, we observed the effect of the duration. If the auction lasts one minute longer, the final price will increase with 0.05%. This effect is somewhat bigger than for concert A, but it still not quite beneficial for the supplier to increase the length of an auction. As discussed for concert A, it might be an idea to shorten the duration and supply more auctions if there are many consumers willing to place a bid. Moreover, revenues might increase if more auctions are supplied during prime time and the weekend and less auctions are supplied between 11PM and 6AM. Lastly, we included the auction number and the square of it. The auction number is significant in columns 1,2,3 and 6. So, the price in auction t is higher than in auction t-1. This provides evidence for the findings for a learning effect. The square of the auction number is insignificant and does not influence the final price of an auction for concert B.

In the last three models, the endogenous variables are added to the first three models. As we did for concert A, we first observed the correlation between the number of bids placed and the height of the first bid. We found a correlation of -0.447, which does not give multicollinearity problems. The findings of these models are comparable with the first three models. The only difference is that the variables auction number and late are not significant in column 4 and 5 and are significant in column 1 and 2. The number of bids is not significant and does not influence the final price and when the height of the first bid is €1 higher, the final price increases by 0.062%. Comparing the results of concert A and B, we observe some small differences between the influence of variables on the final price. We did run a robustness analysis for concert B. However, the robustness is somewhat different as we did for concert A. Concert B has only observations in November. Because of this, we cannot drop observations between 15 December and 15 January. For the robustness analysis we changed the paycheque receipt to the 25th of November in order to see whether the effect of changing the date of getting salary does change the results. Changing the paycheque receipt date to the 25th does not change our results.

We also regressed the second specification for concert B. The results are presented in table 12. There are no significant variables in any model. This means that the spread between bids in auction t and t-x of person i is not influenced by any of the variables in our models. Also losing the previous auction or losing two auctions ago does not influence the spread, whereas it did for the auctions of concert A.

5.3 Zoo tickets

(23)

23

Table 11: Empirical results for concert B

Note: *** significant at .01; ** significant at .05;* significant at 0.1. Standard errors are within the parentheses. The dependent variable is the log of the final price of the auction. Hour and day dummies are used. Moreover, standard errors are clustered on bidderid.

Table 12: Empirical results for the second specification for concert B

(1) (2) (3) (4) (5) (6) Constant 3.880*** 3.878*** 3.389*** 3.870*** 3.863*** 3.398*** (0.065) (0.086) (0.228) (0.671) (0.089) (0.230) Prime 14.242** 14.242** 19.706*** 14.633** 14.633** 19.822*** (0.062) (0.062) (0.065) (0.063) (0.063) (0.066) Weekend -0.424 -0.269 29.934** -0.657 0.017 28.761*** (0.027) (0.026) (0.135) (0.276) (0.090) (0.136) Late -14.809* -14.809* -16.164* -13.867 -13.868 -15.199* (0.090) (0.090) (0.090) (0.090) (0.090) (0.090) Paycheque2 -0.155 -0.675 (0.443) (0.044) Paycheque4 0.000 0.000 (0.440) (0.451) Paycheque7 87.648** 83.435** (0.375) (0.380) Duration 0.052*** 0.052*** 0.052*** 0.052*** 0.052*** 0.052*** (0.002) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number -0.043* -0.043* -0.139* -0.052 -0.041 -0.132*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number² 0.000 0.000 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of bids -0.016 -0.016 -0.016 (0.000) (0.000) (0.000)

Height of first bid 0.062**

(0.000) 0.062** (0.000) 0.059** (0.000) N 762 762 762 762 762 762 R² 0.291 0.291 0.296 0.297 0.297 0.302

Spreadt,t-1 Spreadt,t-2 Spreadt,t-3 Spreadt,t-4

(24)

24 10%. Lastly, increasing the duration with 1 minute will increase the final price with 0.024%. So, it is not quite beneficial to increase the duration. All other variables are insignificant. In the last three columns we added the number of bids and the height of the first bid. First, we observed the possible multicollinearity problem. We did not find multicollinearity problems, since the correlation between the number of bids placed and the height of the first bid is -0.455. We did not find different results in the last three models than we did in the first three. Besides that, we found that when the height of the first bid increases with €1, the final price increase with approximately 1%. Moreover, an extra bid will increase the final price with approximately 1.5%.

Further, the second specification is only regressed for the spread between auction t up to t-4. We did not regress the spread between auction t and auction t-5, because of a lack of observations. The results are shown in table 14. All variables are insignificant and thus do not influence the spread between bids in auction t and t-x. In conclusion, we do not find evidence for escalation of bidding, whereas we found it for concert A when the previous auction and/or two auctions ago were lost.

Table 13: Empirical results for the zoo tickets

(1) (2) (3) (4) (5) (6) Constant 2.789*** 3.784*** 2.767*** 2.651*** 2.651*** 2.642*** (0.052) (0.049) (0.050) (0.049) (0.047) (0.047) Prime 4.667 4.705 6.927 4.471 4.419 6.334 (0.049) (0.049) (0.049) (0.046) (0.045) (0.045) Weekend -1.675 0.192 0.516 -2.010 -0.593 -0.332 (0.024) (0.024) (0.024) (0.024) (0.023) (0.023) Late 4.341 4.340 7.452 5.102 0.051 7.715 (0.067) (0.066) (0.065) (0.062) (0.061) (0.060) Paycheque2 7.588*** 5.117** (0.025) (0.025) Paycheque4 10.054*** 8.553*** (0.017) (0.017) Paycheque7 8.927*** 7.582*** (0.014) (0.014) Duration 0.024*** 0.024*** 0.012 0.018*** 0.019*** 0.022*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number 0.028** 0.019* 0.000 0.028*** 0.020** 0.014 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number² 0.000 0.000 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of bids 1.459*** 1.379*** 1.345*** (0.002) (0.002) (0.002)

(25)

25

Table 14: Empirical results for the second specification for the zoo tickets

Table 15: Robustness analysis paycheque zoo tickets

Spreadt,t-1 Spreadt,t-2 Spreadt,t-3 Spreadt,t-4

Constant 143.283** 216.315** 167.387 209.660 (58.377) (109.175) (188.669) (381.194) Prime -9.869 12.192 -12.950 -38.097 (21.640) (41.310) (60.622) (84.834) Weekend -11.927 61.594 -6.466 64.059 (21.093) (40.372) (53.508) (70.566) Late -28.616 -60.696 60.529 130.677 (29.628) (52.734) (78.496) (93.462) Lost-t 62.136 -51.858 119.613 77.689 (51.368) (99.783) (170.740) (364.924) Auction number -0.144 0.175 -0.330 -0.367 (0.135) (0.257) (0.360) (0.498) Auction number² 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) N 1,119 390 210 136 R² 0.004 0.011 0.015 0.032 (1) (2) (3) (4) (5) (6) Constant 2.767*** 2.726*** 2.761*** 2.691*** 2.655*** 2.640*** (0.079) (0.077) (0.078) (0.089) (0.088) (0.088) Prime 17.922* 17.732* 18.432* 19.643** 19.186** 19.443** (0.096) (0.095) (0.096) (0.088) (0.088) (0.089) Weekend 8.562*** 12.010*** 7.626** 8.709*** 11.273*** 12.022*** (0.033) (0.033) (0.023) (0.033) (0.034) (0.033) Late 12.435 11.910 12.494 10.087 9.455 9.741 (0.107) (0.106) (0.106) (0.107) (0.107) (0.107) Paycheque2 11.320*** 9.425*** (0.025) (0.028) Paycheque4 4.579* 3.956 (0.026) (0.026) Paycheque7 12.317 0.784 (0.023) (0.022) Duration 0.025** 0.025** 0.023** 0.025** 0.025** 0.024** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number -0.030 -0.026 -0.022 -0.035 -0.032 -0.023 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number² 0.000** 0.000** 0.000** 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of bids 1.110*** 1.184*** 1.218*** (0.003) (0.027) (0.003)

(26)

26 Lastly, we did a robustness analysis in order to see whether the effect of the paycheque receipt does change when we drop the observations between 15 December and 15 January. The results are presented in table 15. We have 408 observations left and we observe a higher R² than in table 13. Now, the variable prime time is significant and increase the final price between 18% and 20%. Another difference with table 13 is that the effect of the weekend is significant and increase the price with 7.5% to 12%. We also observe that not all paycheque variables are significant anymore. The last difference is between the auction number and auction number² variables. In table 13 we see that there is an linear relation between the auction prices over time, which is positive. In table 15 we see a small, positive quadratic effect. However, prices do increase when more auctions have been supplied. This finding also contradicts the learning effect.

Table 16: Empirical results for the hotel voucher

5.4 Hotel voucher

The last investigated product is a hotel voucher. The results of our models are presented in table 16. In our first three models, we see that auctions ending during prime time have a final price which is

(1) (2) (3) (4) (5) (6) Constant 3.409*** 3.408*** 3.403*** 3.346*** 3.343*** 3.339*** (0.087) (0.087) (0.087) (0.091) (0.091) (0.091) Prime 18.917** 18.899** 19.232*** 19.628*** 19.605*** 19.894*** (0.073) (0.073) (0.073) (0.071) (0.071) (0.072) Weekend -3.716 -3.565 -3.711 -0.333 -0.183 -0.334 (0.064) (0.064) (0.063) (0.065) (0.064) (0.072) Late 15.525** 19.325** 15.549** 19.628*** 19.121** 15.814** (0.072) (0.074) (0.072) (0.073) (0.073) (0.073) Paycheque2 -1.722 -1.719 (0.053) (0.053) Paycheque4 -1.830 -1.938 (0.046) (0.047) Paycheque7 -3.352 -3.278 (0.039) (0.039) Duration 0.000 0.000 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number -0.088** -0.088** -0.088** -0.074* -0.074* -0.074* (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number² 0.000** 0.000** 0.000** 0.000** 0.000** 0.000** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of bids 0.001 0.001 0.001 (0.002) (0.002) (0.002)

(27)

27 approximately 19% higher. Further, we find that auction ending during 11PM and 6AM have a final price which is between 15.5% and 19.3% higher, which is in contrast with our hypothesis and findings for both concerts. Lastly, we found the a quadratic final price function over time (coefficient of Auction number² is 0.000001 which is 0.0001%). Solving this, we find that in the first 880 auctions the price will decrease and increase afterwards. Since we only have 350 auctions, the price will decrease every auction which provides evidence for the learning effect. In the last three columns, the findings are comparable to those of the first three. Further, we found that a €1 increase in the first bid, increases the final price by 0.4%. A second specification is not performed, because of a lack of observations. Therefore, we cannot conclude whether bidders increase their bid after losing an auction. We performed a robustness analysis for the paycheque receipt. We dropped the observations between 15 December and 15 January in order to see whether this will change the results found in table 16. The robustness results are presented in table 17. We see that 30 observations drop out and we have a slightly higher R². Compared to table 16, we see that the effect of prime time has increased from

Table 17: Robustness analysis paycheque zoo tickets

(1) (2) (3) (4) (5) (6) Constant 3.252*** 3.254*** 3.253*** 3.179*** 3.181*** 3.176*** (0.072) (0.072) (0.076) (0.077) (0.077) (0.080) Prime 30.486*** 30.560*** 30.732*** 29.349*** 29.431*** 29.559*** (0.050) (0.050) (0.050) (0.048) (0.048) (0.048) Weekend -0.328 -0.770 -0.739 3.301 2.887 3.022 (0.069) (0.067) (0.069) (0.067) (0.067) (0.069) Late 25.735*** 25.739*** 25.748*** 24.179*** 24.176*** 24.088*** (0.049) (0.049) (0.049) (0.051) (0.051) (0.051) Paycheque2 3.108 2.918 (0.054) (0.053) Paycheque4 2.622 2.189 (0.050) (0.051) Paycheque7 0.162 -0.496 (0.043) (0.443) Duration 0.002 0.002 0.002 0.003 0.003 0.003 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number -0.106*** -0.106*** -0.108*** -0.090** -0.091*** -0.091*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auction number² 0.000*** 0.000*** 0.000*** 0.000** 0.000*** 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of bids 0.207 0.209 0.227 (0.002) (0.002) (0.002)

(28)

28 approximately 19% to 30%. Besides, we observe an increase of the effect of auction supplied between 11PM and 6AM. We also see an increase in the effect of auction number. However, solving the quadratic equation of auction number, auction number², we find that the in the first 1060 auctions (column 1 and 2) the price decrease and increase afterwards. We only have 320 auctions, so the price decreases in auction t+1. This is provides evidence for the learning effect, which is also found in table 16. Lastly, other results are comparable to the findings of table 16.

6. Conclusion

This paper has investigated whether increasing the duration of an auction will increase revenues for a supplier or whether there are other factors which influence the final price of an auction. Data have been recorded from a website which supplies auctions from the 8th of November 2016 till 31st of January 2017. Four products were recorded: two types of concerts, zoo tickets and a hotel voucher. The effect of increasing the duration by one minute is positive and significant for both concerts and for the zoo voucher. However, the effect is so small that it is not beneficial for the supplier to increase the duration with one or several minutes. To increase revenues, to supplier can better shorten duration and supply extra auctions in periods when there are many bidders at the site or when bidders are willing to place higher bids.

This paper finds that not all variables do influence the results of auctions on the same way for different products. The only significant result which is found for all our products is a positive effect between 0.1% and 1.1% for the height of the first bid on the final price of an auction. Some results were significant for some products and insignificant for others. The number of bids during an auction only influences the final price of both concerts. There is a positive effect on the final price between 0.7% and 1.5%

We found for example that auctions during prime time increase the final price for concert A, B and the hotel voucher (respectively +7%, 14% and 19%). However, auctions during prime time do not influence the final price of zoo tickets in the initial specification. We found a positive effect of approximately 19% in our robustness analysis. Auctions ending during the weekend have a positive effect on the final price of concert A and for the zoo tickets. For concert B depends the significance of the effect of auctions during the weekend on the paycheque variable. So, we find that there is no effect and an increase in final prices around 30%. The results of auctions ending between 11PM and 6AM are a little bit mixed. We find for both concerts a decrease in final price of around 15%. However, we found an increase between 15% and 30% for the hotel voucher. Whereas the zoo tickets auctions are not influenced.

(29)

29 that the last week of December is different than other weeks in the year, we also performed a robustness analysis without observations between 15 December and 15 January. For concert A we found a negative effect for the first two days after receiving the salary, no effect for the first 4 days and a positive effect for the first week after receiving the salary. In the robustness analysis we found only a positive effect in the first week after the paycheque receipt. For concert B, we only find a positive effect in the first week after receiving the salary. This effect is approximately 85%, which indicates that in the first week after the paycheque receipt the final price is 85% higher. An explanation for this high number, might be that we only had observations between the 22nd and the 29th of the month, which may cause these results. The first 2,4 and 7 days have a positive effect on the final price for the zoo tickets. With the robustness analysis we find no significant relation for the first week after the paycheque receipt anymore.

In section 2.5 we discussed the learning effect. For concert A, we found evidence for the learning effect in the first 900 auctions. Afterwards, the price will increase every auction. For concert B, we also found evidence for a lower price when more auctions have taken place. This results was also found for the hotel voucher. In contrast, prices do increase every auction for the zoo tickets, so we find the opposite result of the learning effect for the zoo tickets.

Lastly, we observed whether escalation of bidding does play a role. We did this by creating an second specification where we take into account the spread between the bid in the current auction and bid in the previous auctions by person i. Moreover, we also take into account whether the previous auction was lost. For concert A we saw that bidders will bid more when they have lost the previous auction and/or two auctions ago. For concert B and the zoo tickets we did not find evidence for escalation of bidding. Lastly, for the hotel voucher we did not have enough observations to run a model. For this reason, we cannot conclude whether escalation of bidding does play a role for the hotel voucher auctions.

(30)

30 outcomes between product types are caused by different types of bidders which might be attracted, further research is needed which includes demographic information about the bidders. In order to increase revenues it is not beneficial to increase the duration of the auctions. To increase revenues, the supplier can shorten the duration and supply more auctions when many bidders are visiting the auction site or when bidders are willing to bid high. Further, the supplier should supply more auctions in time periods which have a positive effect on the final price and should supply less auctions in time periods which have a negative effect on the final price. To increase revenue, the supplier should observe these patterns per product and respond to these patterns.

References:

Alford, B. L., Gilley, O. W., Wood, C. M., & Obilo, O. (2015). “No sale” items in auctions: do they really matter? Marketing Letters, 1-14.

Ariely, D., & Simonson, I. (2003). Buying, bidding, playing, or competing? Value assessment and decision dynamics in online auctions. Journal of Consumer psychology, 13(1-2), 113-123.

Bailey, D., Brorsen, B.W.,& Fawson, B.W. (1993). Buyer Concentration and Feeder Cattle Auctions.

Review of Agricultural Economics 15, 103-119.

Bajari, P., & Hortacsu, A. (2003). The winner's curse, reserve prices, and endogenous entry: Empirical insights from eBay auctions. RAND Journal of Economics, 329-355.

Bapna, R., Jank, W., & Shmueli, G. (2008). Price formation and its dynamics in online auctions. Decision Support Systems, 44(3), 641-656.

Carmon, Z., & Ariely, D. (2000). Focusing on the forgone: How value can appear so different to buyers and sellers. Journal of consumer research, 27(3), 360-370.

Cox, R. G. (2005). Optimal reservation prices and superior information in auctions with common-value elements: Evidence from field data. Ekonomia, 8(2), 142–167.

Hastings, J., & Washington, E. (2010). The first of the month effect: Consumer behavior and store responses. American Economic Journal: Economic Policy, 2(2), 142-162.

Hong, H., & Shum, M. (2002). Increasing competition and the winner's curse: Evidence from procurement. The Review of Economic Studies, 69(4), 871-898.

Cabral, L., & Hortacsu, A. (2010). The dynamics of seller reputation: Evidence from eBay. The Journal

of Industrial Economics, 58(1), 54-78.

Haruvy, E., & Leszczyc, P. T. P. (2010). The impact of online auction duration. Decision Analysis, 7(1), 99-106.

Huffman, D., & Barenstein, M. (2005). A monthly struggle for self-control? Hyperbolic discounting, mental accounting, and the fall in consumption between paydays. Institute for the Study of Labor

(IZA) Discussion Paper, 1430.

(31)

31 Katok, E., & Kwasnica, A. M. (2008). Time is money: The effect of clock speed on seller’s revenue in Dutch auctions. Experimental Economics, 11(4), 344-357.

Klemperer, P. (1998). Auctions with almost common values: ‘The Wallet Game' and its applications. European Economic Review, 42(3), 757-769.

Klemperer, P. (1999). Auction theory: A guide to the literature. Journal of economic surveys, 13(3), 227-286.

Krishna, V. (2002). Auction theory. San Diego. Cal.: Academic Press.

Kwasnica, A. M., & Katok, E. (2007). The effect of timing on jump bidding in ascending auctions. Production and Operations Management, 16(4), 483-494.

Levin, D., & Kagel, J. H. (2005). Almost common values auctions revisited. European Economic

Review, 49(5), 1125-1136.

Li, T., & Zheng, X. (2009). Entry and competition effects in first-price auctions: theory and evidence from procurement auctions. The Review of Economic Studies, 76(4), 1397-1429.

Livingston, J. A. (2003). What attracts a bidder to a particular internet auction? In Organizing the New

Industrial Economy (pp. 165-187). Emerald Group Publishing Limited.

Livingston, J. A. (2005). How valuable is a good reputation? A sample selection model of internet auctions. Review of Economics and Statistics, 87(3), 453-465.

Lucking-Reiley, D., Bryan, D., Prasad, N., & Reeves, D. (2007). Pennies from eBay: The determinants of price in online auctions. The journal of industrial economics, 55(2), 223-233.

Mastrobuoni, G., & Weinberg, M. (2009). Heterogeneity in intra-monthly consumption patterns, self-control, and savings at retirement. American Economic Journal: Economic Policy, 1(2), 163-189. McDonald, C. G., & Slawson, V. C. (2002). Reputation in an internet auction market. Economic

Inquiry, 40(4), 633-650.

Melnik, M. I., & Alm, J. (2002). Does a seller’s ecommerce reputation matter? Evidence from eBay auctions. The journal of industrial economics, 50(3), 337-349.

Nelson, J.P. (1997). Market Structure and Incomplete Information: Price Formation in a Real-World Repeated English Auction. Journal of Economic Behavior and Organization, 27, 421-437.

Pinkse, J., & Tan, G. (2005). The affiliation effect in first-price auctions. Econometrica, 73(1), 263-277. Shapiro, J. M. (2005). Is there a daily discount rate? Evidence from the food stamp nutrition

cycle. Journal of public Economics, 89(2), 303-325.

Stephens, M. (2003), 3rd of tha month: Do Social Security Recipients Smooth Consumption Between Checks? American Economic Review, 93, 406-422.

(32)

32 Strahilevitz, M., & Loewenstein, G. (2001). The differential roles of gender, previous ownership, and

time on attachment to possessions. Working paper, University of Arizona.

Vickrey, W. (1961). Counterspeculation, auctions, and competitive sealed tenders. The Journal of

finance, 16(1), 8-37.