Reservation price setting experiment
in the Dutch Flower auction
This thesis is presented for the degree of Master in Economics of
the University of Amsterdam.
Faculty of economics and business
October 2014
Student: Lianne Klijn
Studentnumber: 5806992
Evaluated by: Aljaž Ule
2
Abstract
In a ten week market experiment at a Dutch flower auction, sellers of cut Gerbera flowers got the opportunity to choose their own reservation price for their products. We use the data from this experiment in this thesis to find out if sellers will choose to use a
reservation price and which reservation price they will pick. We find that sellers hesitate at the beginning, but that all of them choose to use a reservation price at the end of the experiment. Also we find that sellers with a bad reputation choose a higher reservation price at the beginning of the experiment, compared to sellers with a bad reputation. In the
last weeks of the experiment a clear equilibrium is shown, in which every seller chooses a reservation price of eight cents. No evidence is found that the implementation of the
3
1. Introduction
Auctions are generally used in situations where someone wants to sell a product, but doesn’t know what price buyers are willing to pay. Auction theory is a considerable part of game theory since 1969. The study on how people act in auction environments and what the properties are of different types of auctions, used to be mostly theoretical. The introduction of electronic auctions and more specifically internet auction has caused opportunities to empirically study auctions. This has led to a great amount of auction experiments.
This thesis is about the use of a reservation price in a Dutch auction. A Dutch auction is auction mechanism in which a high price is gradually lowered until a bidder decides to stop the auction and buy the product at the current price. A reservation price is a minimum price in an auction, under which it is not possible to bid.
At the Dutch flower auction, FloraHolland, an experiment was performed in which sellers of Gerbera flowers had the opportunity to choose a reservation price for the products they wanted to trade. The motivation for this experiment is that flower sellers are not content with the current reservation price of four cents per flower. Especially in the low season, most selling prices end up at exactly four cents. With this experiment FloraHolland wants to see if liberating the choice for a reservation price can improve flower selling prices and the sellers’ satisfaction.
A lot of economic experiments have been performed to see what the effect is of a reservation price on auctions. However, reservation price experiments specifically performed in Dutch auctions are rare. Economic theory about the use of a reservation price in a Dutch auction exists, but varies. Different theories exist about the optimal reservation price and it is unclear what the effect of this price on the bidding behavior will be. This experiment is an unique opportunity to test the effect of a reservation price in a Dutch auction in a market situation.
The general plan of this thesis is to give a brief introduction in section two, about what auctions are and what is known about them so far. In this section we will also describe what the theoretical optimal bidding strategies are in different types of auctions. Also it will be explained what a reservation price is, and what the general effect is of the use of a reservation price in an auction.
In section three, we will write about FloraHolland, the Dutch flower auction where the experiment took place. We will explain the auction process that is used to trade cut flowers. In the fourth section we will continue with answering the question why FloraHolland decided to perform this reservation price experiment and how the experiment was implemented.
In section five we will tell what the five hypothesis are, which will be tested using the experimental data. The results will be shown in section six. Section seven is the conclusion in which results will be summarized. And section eight is used to discuss the experiment and its results.
4
2. Auction theory and literature
2.1 Auction types
The word auction comes from the Latin word augere, which means to increase. The earliest use of auctions was reported in Babylonia 500 B.C. as a method to trade women of marriageable age (Milgrom & Weber, 1982). Even though auctions have been used since, it has taken them a long time to show up in economic literature. William Vickrey was the first to write about the game-theoretical aspects of auctions in 1961. This was the beginning of a new branch of game theory, called auction theory. The main
characteristic of an auction is that a good is sold to the bidder who submits the highest bid (Osborne, 2001). Usually there is one seller, who has an item to sell. This seller faces a number of n potential buyers. Auction theory deals with the question how the seller and the potential buyers act in an auction and studies the properties of different types of auctions.
The paper of Vickrey is extensive and covers already most of the basics of what is currently known about auction theory. He describes the four basic types of auctions that are still used today. The first type is the ordinary or progressive auction. Nowadays this auction is referred to as ascending-bid or English auction. In this auction the price is successively raised until only one bidder remains. This last bidder obtains the object for the price at which the auction is stopped. This auction can be performed in different ways. The bidders can call out bids in turns, or submit them electronically. Alternatively, prices can be gradually raised and bidders can choose to leave the auction when the price exceeds their maximum preferred bid. This last type is also called a Japanese auction (Klemperer, 1999).
The second type is the descending or Dutch auction. This type obtained its name from its use to trade flowers in the Netherlands. In this auction, an auctioneer starts the auction at a high price and gradually lowers it until a bidder decides to stop the auction and to buy the product at the current price (Klemperer, 1999).
Other types are sealed-bid auctions. A sealed-bid auction means that every bidder submits her bid simultaneously or within a certain period of time and this bid is
unknown to the other bidders. The bidder who submitted in the highest bid wins the auction. In the first-price sealed- bid auction, the winner pays her own bid. In the second-price sealed -bid auction, the winner pays the bid of the second-highest bidder. The second-price sealed-bid auction is also called the Vickrey auction, because William Vickrey was the one writing about this auction and about the fact that this auction is comparable to the English auction. We will explain this auction in further detail later.
2.2 Bidding in auctions
To analyze bidding strategies in an auction, it is important to have assumptions about the value of the bidders of the product. Two value-models are commonly used. The
5
private-value model is used when every bidder has her own value, but this value is only known to her. So the bidder only considers her own value while bidding and this value is not affected by the valuation of the other bidders. If a bidder has an idea about her valuation of an object, but this valuation is affected by the valuation of other bidders for the object, we speak about interdependent valuation. This model can for example be used for products that can be re-sold (Krishna, 2010, p. 3). The common-value model is a special form of interdependent value. This model can be used when the value of the product being offered for sale, is the same for all bidders, but no bidder knows exactly what this value is. Every bidder makes her own estimate, using private signals (Paarsch, 1992). This is, for example, the most commonly used model for auctions of oil- or
mineral leases (Kagel & Levin, 1986).
As mentioned before, auction theory deals with the question how rational bidders act in an auction. An auction can be described as a game. In an auction defined as a game with the private value model, every player i attatches a value Vi to the object that is being sold. If she obtains this opject and pays price p, her payoff will be Vi - P. A player can only obtain the object if she wins the auction by submitting the highst bid. The payoff of player i is equal to zero, if any other bid is higher than hers, and an other player j obtains the object.
For every player in an English or second-price sealed-bid auction, bidding her own value is weakly dominant to all other possible bids. Imagine player 1 has the highest valuation of all n players V1> V2> V3>…..Vn. If player 1 bids her own value B1 = V1, her
payoff can either be zero, if any other players bids higher than B1. Alternatively, if no
other player bids higer, her payoff equals V1-Bj, where Bj is the second highest bid, the
bid of player j. If player 1 bids higher than her valuation V1 > V1, her payoff is either zero,
if Bj>B1. Or it is V1-Bj>0, if B1>V1>Bj. Or it is negative V1-Bj<0, if B1>Bj>Vj. An overview of
the payoffs of the players is given in Table 1Table 1
B1<Bj (j≠1) & Bj < Vj B1<Bj (j≠1)& Bj > Vj B1>Bj (j≠1) & Bj>V1 B1>Bj (j≠1) & Bj<V1 B1 < V1 0, Vj-B1>0 0, Vj-B1= Z - V1-Bj>0, 0 B1 = V1 - 0, Vj-B1<0 - V1-Bj>0, 0 B1 > V1 - 0, Vj–B1<0 V1 – Bj < 0, 0 V1-Bj>0, 0
Table 1 Payoff player j and payoff player 1, given V1 > Vj in English auction
Given that player 1 bids her own value, for any other player j, who has a lower valuation than player 1, their payoff will be zero, if they put in a bid lower than B1. If
they put in a higher bid than B1, their payoff will be negative Vj-B1<0, because B1=V1>Vj.
Of course submitting any bit lower than her own valuation also yields a payoff of zero Submitting a bid uneuqaul to her own valuation Bi≠Vi always yields the same or a
lower payoff than submitting a bid equal to her own valuation Bi=Vi. For this reason,
bidding one’s own value weakly dominates any other bid.
In the English and second- price sealed-bid auction it for every player it is a weakly dominant strategy to bid her own value or to continue bidding until the auction
6
reaches her own value. In theory, the object will always be won by the player with the highest valuation and the price that will be paid is the second highest bid, which is equal to the second highest valuation (Vickrey, 1961) (Riley & William, 1981).
When a player is bidding in a Dutch auction, she needs to consider on the one hand, the possible gain of a lower bid and on the other hand the lower possibility of winning of a lower bid. This same balance needs to be found when a bidder is bidding in a first-price sealed-bid auction (Vickrey, 1961). The Dutch auction and the first-price sealed-bid auction are strategically equivalent (Milgrom & Weber, 1982).
Determining a bidding strategy in a Dutch or first-price sealed-bid auction is slightly more complicated. For every player, bidding higher than her own value Vi yields a payoff equal or lower than zero. In the case of winning the auction, the player pays her own bid. Bidding exactly Vi weakly dominates any higher bids, because this always leads
to a payoff of zero. The payoffs are shown in Table 2Table 2 2. Bidding slightly lower than vi weakly dominates bidding exactly vi, because this can lead to a payoff of zero, or
a payoff higher than zero. But if V1> V2> V3>…..Vn and B2<B1<V2<V1, player 2 can increase her payoff by increasing her bid so that B1<B2<V2<V1. For this reason, in a dutch first-price auction, there is always one action of one player weakly dominated. The only equilibrium in which no action is dominated, is if every player bid is equal to the the next highest value. For this reason in every distingquised equilibrium of each private value game, the player with the highest valuation wins the auction and pays a price equal to the second highest valuation (Osborne, 2001).
B1<Bj (j≠1) & Bj < Vj B1<Bj (j≠1) & Bj > Vj B1>Bj (j≠1) B1 < V1 0, Vj-Bj>0 0, Vj-Bj<0 V1 - B1 > 0 B1 = V1 - 0, Vj-Bj<0 V1 - B1 = 0 B1 > V1 - 0, Vj-Bj<0 V1 - B1 < 0
Table 2 Payoff player j and payoff player 1, given V1>Vj in Dutch auction
William Vickrey (1961) computed the equilibria explained above in the four different auctions, and found out that the expected revenue was exactly the same. This finding was called the Revenue Equivalence Theorem. In 1981 both Myerson (1981) and Riley & William (1981) added the assumptions of risk neutrallity and independent private values to this theorem.
In a common value auction, a bidder has to anticipate that if she wins the auction, it is probably that she overestimated the value of the object being sold.
This can happen when a player receives a more optimistic signal than the other bidders. Failing to take into account the signals the other bidders received, can cause a negative payoff when the price to pay is higher than what the product is worth. This phenomenon is called the winner’s curse. (Thaler, 1988). This concept was first introduced by Capen, Clapp, & Cambell (1971), three engineers who were wondering why the gas and oil industry in the Gulf of Mexico was almost never making as the return on investment as intended.
7
2.3 Reservation price
Even though the Revenue Equivalence Theorem predicts the same expected pay-off in every auction type, many papers have been written about how to design the optimal auction. One of the topics used to influence the expected revenue is the use of a reservation price.
A reservation price, is a minimum price in an auction under which a product will not be sold. Many economists have written about the optimal reservation price to use in auctions. Most theories describe the optimal reservation price in second price or
English auctions. As the bidding strategy in these auctions is more straight forward compared to the strategy in a Dutch auction.
In 1981 John Riley and William Samuelson wrote a paper about optimal auctions. In this paper they announce that seller’s profit is maximized if he sets a well-chosen reservation price. This reservation price is independent of the number of buyers and is slightly higher than the seller’s personal value. If the chosen reservation price V* =V0+δ
is slightly higher than sellers own value V0, the profit of the seller will be affected in two
cases. First, when the value of all buyers lie between V0 and V*, because in this case the
product won’t be sold. Second, when the value of one buyer is higher than V*and of at least on one other buyer the value is between V0 en V*. In this case the product will be
sold for a price of v* instead of a price between V0 and V*. If δ is sufficiently small, the
gain of the reservation price outweighs the cost (Riley & William, 1981, p. 385). In the same year the same result was found by Roger Myerson (1981).
Dan Levin and James Smith relax the independent private value model, that was used in the papers of Riley & William (1981) and Myerson (1981). They show that when bidder’s information is correlated, which is often the case, the optimal reservation price converges to the true value of the seller, as the numbers of bidders grow (Levin & Smith, 1996).
A new insight about the use of a reservation price is introduced by Cai, Riley, & Ye (2007). They argue that a reservation price can be used to signal information about the value of the object being sold. In their model they assume that a seller of an object in an auction has private information about the product. For example a seller of a work of art can know more about characteristics that affect the value of this item, like the quality, rarity and the history. If a seller has positive information about the value of his product he would want to signal this to the buyers, to increase their valuation of te product. Therefore a high type seller has an incentive to signal this to the buyers by setting a high reservation price (Cai, Riley, & Ye, 2007).
There are two different rules of using a reservation price in an auction. The first is a public reservation price. Which is set by the seller and is known to all bidders before they submit their bid. The second rule is a secret reservation price. This means that the seller chooses a reservation price, but this price is not known to the bidders before they submit their bids. The most common argument in favor of using a secret reservation
8
price is that a high public reservation price can possibly discourage bidders form entering the auction (Katkar & Lucking-Reiley, 2000, p. 6).
Theoretical work in game theory about the effect of hiding or showing the reservation price is limited. Elyakime, Laffont, Loisel and Vuong first show in their paper that in a first-price sealed-bid auction, the optimal reservation price equals the seller’s own value. Furthermore they show that hiding the reservation price never leads to an increase of the average selling price and that using a public reservation price leads to an expected pay-off at least as good as the expected pay-off of a secret reservation price (Elyakime, Laffont, Loisel, & Vuong, 1994).
One year later, an example is provided by Vincent (1995), who shows that hiding the reservation price can increase the seller’s revenue. This can happen in a second-price auction using the interdependent value model. This means that a buyer’s valuation depends on the valuation of other buyers (Bergmann, Shi, & Valimaki, 2009).
Li and Tan (2000) show that in an independent private value first-price auction, under the assumption of risk averse bidders hiding the reservation price may increase the seller’s revenue. In a second-price auction, they find that the seller should be indifferent between showing or hiding the reservation price (Li & Tan, 2000).
Inspired by the theoretical work done on the topic of the optimal use of
reservation prices, many experiments are performed. Especially the emergence of on-line auctions generated many opportunities to empirically test the effect of different auction rules.
Lucking-Reiley, Bryan, Prasad and Reeves find in their eBay experiment that the use of a reservation price has a positive effect on the selling price (Lucking-Reiley, Bryan, Prasad, & Reeves, 2007). However, they don’t investigate the effect on the probability of selling the product. Bajari & Hortaçsu (2003) performed an experiment on an eBay second-price auction and found that the optimal reservation price is equal to the seller’s own value, but that sellers generally choose to set a low reservation price. They also find a difference between a secret and a public reservation price. Their results show that sellers using the secret reservation price rule, set lower reservation prices compared to sellers who make their reservation price public. Also they find that the significant lower number of auctions result in a sale when a secret reservation price is used (Bajari & Hortaçsu, 2003, p. 333).
Experiments performed at a first-price auction are more limited. One takes place in the timber auction in France. Elyakime and others find out that switching from a secret to a public reservation price, increases the expected selling price of timber with 3,14% (Elyakime, Laffont, Loisel, & Vuong, 1994).
2.4 Games
The experiment that will be analyzed in this thesis can be seen as a price setting game. Multiple sellers of the same product can all choose to set their own price. The process of price setting in any market can be presented as a game. The most basic form of a game
9
used in game theory is the Prisoner’s dilemma. This game was invented by Merill Flood and Melvin Dresher, who were RAND scientist. RAND was a think-thank founded by the air force to perform strategic studies on the intercontinental nuclear war (Poundstone, 1993).
This game describes a situation in which two players can both choose between two actions, co-operate or defect. The total pay-off of the game is highest when both players choose to cooperate see table 3. But as generally know, this game has a unique Nash equilibrium in which both players choose the action defect (Osborne, 2001).
Cooperate Defect
Cooperate 2, 2 0, 3
Defect 3, 0 1, 1
Table 3
However, when a Prisoner’s Dilemma is infinitely repeated, the game also has a Nash equilibrium in which both players choose Cooperate. Because in the long run, the benefits from having a cooperative relationship are higher than the short term gain form defecting. The other Nash equilibrium strategy is that if the other player plays defect, the first player will play defect from then on (Osborne, 2001).
To determine which strategy is optimal for the player, the discounted pay-off of both strategies need to be compared. And the best strategy depends on the discount factor δ.
3. Flower auction
3.1 FloraHolland
This thesis uses data from FloraHolland, a flower auction in the Netherlands.
FloraHolland is a cooperative owned by 4000 Dutch and 600 non- Dutch growers of plants and cut flowers. Around 7000 growers use the services from FloraHolland to sell their products. In the year 2012 FloraHolland achieved a turnover of Euro 4.2 billion. On average 32 million flowers and 2 million plants are auctioned per day, via 38 auction clocks. In the Netherlands, FloraHolland has five establishments in five different locations, three export locations, and two which focus on the regional market (Annual report FloraHolland, 2012). The experiment which we will describe takes place in the three export locations Aalsmeer, Naaldwijk and Rijnsburg.
3.2 Auction mechanism
To auction plants and flowers, the Dutch auction mechanism is used at the flower auction. A clock is showed to all potential buyers/ bidders, every position of the clock equals a certain price. The hand of the clock falls down from an initial high price till a
10
bidder stops the clock by pushing a button. The bidder who pushes first, achieves the object and pays the price at which she stopped the clock.
At the locations Aalsmeer and Rijnsburg the winning bidder gets audio contact with the auctioneer, and tells him how many units of the product she wants to buy. In Naaldwijk, this process is slightly different. Instead of pushing the button and decide how many units to buy after winning the auction, buyers have to push a button with the number of units on it. So in Naaldwijk, bidders decide how many products to buy, before knowing if they won the auction. Figure 1 shows a picture of an auction clock at
FloraHolland.
Figure 1 The auction clock source: Floraholland.nl
Every group of products (e.g. roses or tulips) has its own minimum price also called the reservation price. If no bidder pushes the button before the clock reaches this reservation price, the product is not sold and will be destroyed. The process of the auction clock reaching the reservation price and the product being destroyed is called withdrawal. There is no option for growers to receive the product back and sell it at a different time or in a different place. This withdrawal process is needed to make sure only daily fresh products are offered at the auction and there is no option to disturb the market process. Disturbing the market in this case means that after the reservation price is reached during the auction process, it is not possible for a buyer to personally contact a grower and offer him to buy his products for a price lower than the
reservation price at the auction. The reservation prices used in the auction are determined years ago and haven’t been changed since. Every flower auction in the Netherlands uses the same reservation prices, to prevent for price competition between the different auction. These prices are not shown during the auction process, but they are known to all bidders out of experience (personal communication).
Products are auctioned five days a week. Every day the same auction schedule is used. This schedule shows in which order product groups will be auctioned. A product
11
group is a group with the same type of flowers, for example roses, tulips or gerbera’s. This schedule differs per location. This means that there is a possibility to switch location. If it is not possible to buy the product in one place, a buyer can try later at a different location. The products are auctioned as separate lots, which are defined as a homogeneous product from one buyer. One lot consists of multiple units. One unit is the minimum quantity of stems that can be bought. The price in the auction clock is the price per stem. The order in which lots from which grower are auctioned is every day randomly decided by a computer (personal communication at FloraHolland).
A flower grower or seller can choose to bring his product to any of the five auction locations, which for him is most convenient, or which he believes will yield the highest revenue. There are multiple reasons why a grower brings his products to the auction. First of all, the convenience, the grower can bring as many flowers on whatever day he likes to the auction. As soon as the flowers are at the auction, he doesn’t have to worry about anything, the whole selling process is taken care of by the auction. Second, the auction is a place where many buyers are together, all demand is in the same place and the seller doesn’t have to go looking for the demand himself. Last of all, the seller has a security of receiving his money. All grower that bring their products to the auction, receive their revenue the same day. This means that a grower doesn’t have to wait for his money and that he doesn’t have a risk of buyers delaying their payment or refusing to pay at all.
A buyer can choose to participate in an auction at any of the five locations of FloraHolland, but he needs to buy a subscription for every location separate. The buyer can choose to be personally present in one of the auction rooms, and push the button there, or he can use a KOA (buying from a distance) subscription, which means he can participate in the auction using the internet from any place in the world. If buyers have multiple KOA subscription, they can participate in multiple auctions at the same time. In this paper I will refer to buyers as bidders of the auction. Buyers are generally not the end consumers, but wholesalers. The reason for this are the entry barriers for buying at the auction. One barrier is that only large amount of flowers can be bought. Additionally, the subscription to buy at the auction is costly. In 2013 the subscription fee was 15.000 euro. The advantages for a buyer to buy his flower at the auction is that he can buy different type- and amount of flowers every day. This way it is easy to anticipate on current demand.
The typical bidder is highly experienced in the auction process. Even though the process sounds simple, one needs extensive knowledge about the flower market and needs to be able to handle a lot of information at the same time. A typical bidder does this work for many years, but a new bidder will start after three to four months of training.
In almost every market the selling price of a product depends on many variable. The same is the case in for products sold at the flower auction. Many factors influence the demand for flowers, and because of the fact that a flower is a fresh product, it is relatively difficult to rapidly adapt the flower supply.
12
In 1197, Gjolberg and Steen used data from the Flower auction about the demand of three different types of flowers to determine calendar patterns in flower demand. They showed that the demand for flowers is highly volatile during the year. More specifically, demand generally increases during winter and decreases during summer (Gjolberg & Steen, 1997).
In 2007 van den Berg and van der Klaauw also used flower auction data. They used these data to evaluate the impact of reservation prices on the expected revenue in the auction. They found that plants cannot be seen as entirely homogeneous products, because the identity of the grower is an important characteristic. To determine the optimal reserve prices for pot plants, they used the model from (Van den Berg, 2007). They found that reservation prices at the auction are too low, and that increasing the reservation price will lead to a higher average selling price (van den Berg & van der Klaauw, 2007).
4. The experiment
4.1 Why a secret reservation price at the flower auction?
Once a grower delivers his flowers to the flower auction, there is no turning back. The flowers will be auctioned that same day. If the product is not sold, it will be destroyed. The grower will be charged the cost of destroying the products. In certain months of the year, the demand for flowers is low, this is mostly the case in summer. In this time of the year, buyers generally offer no more than the reservation price of four cents at the auction. This price is usually not enough to cover the costs of growing the flower. The Netherlands Competition Authority doesn’t allow the flower auction to raise the general reservation price. As FloraHolland is by far the biggest flower auction in Holland (there is only one small other flower auction, which has a turnover of 87,5 million in 2013), it almost has monopoly power in the flower trading industry. Raising the reservation price would mean they use their monopoly power to increase selling prices. Growers however, argue that if the reservation price can be higher and is not shown to the bidders, bidders cannot wait for the auction clock to reach the reservation price anymore, bidders will have to push the button at a higher price (personal
communication). For this reason it is decided to let every grower pick a reservation price himself, and make it possible to change this price for every product he brings to the auction. This way competition is created between the growers, and it cannot be claimed that FloraHolland is using her monopoly power to increase prices.
4.2 The experiment
FloraHolland decided to perform an experiment to test what would be the effect of implementing a flexible reservation price. To test this, it is decided that for a period of sixteen weeks growers of the flower Gerbera can set their own reservation price. This means that for every lot of gerbera’s a grower brings to the auction, he can choose any reservation price as long as the price is higher than the regular reservation price of four
13
cents per stem. If a grower decides not to set a reservation price, the regular price of four cents will be used during the auction process (personal communication). This experiment is a natural experiment, because it takes place in a real market environment. It is not possible to make it a fully controlled experiment, because the players are real buyers and sellers in the flower industry. For this reason it is not possible to for
example randomly assign certain buyers and sellers to the experiment and it is also not possible to obligate buyers or sellers to take part in the experiment.
The gerbera type used in this experiment is the large flower gerbera sold in carton boxes. There are two main reasons why the product group gerbera is chosen. First of all, this group is a relatively large group, with many products sold every day and with many different sellers and buyers. Second, this product group often reaches low prices close or equal to the reservation price during the auction process.
The rules of experiment are simple. Every seller who brings his flowers to the auction needs to fill out an electronic delivery form. On this form he writes which flowers and how many he will bring to the auction that day. During the experiment an extra box is added to this form. In this box the seller can add his chosen reservation price. To write down the reservation price, a pricing code is used. 105 means a reservation price of 5 cents, 106 means a reservation price of 6 etc.
When the auction process starts the reservation price is shown only to the auctioneer, and not to the buyers. To make sure that the reservation price remains unknown to the buyers, the auctioneers are instructed not to tell the reservation price, or to sell the same product again after the clock reached the reservation price. The auctioneers are all informed about these rules, which are discussed during multiple meetings.
The auction process works as usual, only the clock will stop at the chosen
reservation price instead of the usual price of four cents. This means that for the bidders there is an extra element of uncertainty, not only do they know when the other bidders will press the button, also they don’t know when the clock will reach the reservation price. It is not possible for anyone outside the auctioneer to know what is the
reservation price. Only when a product is not sold and the clock stops at the reservation price, buyers know that that was the reservation price given. For sellers it is also
possible to follow the auction process. This means that they are able to observe the reservation prices of other sellers. The experiment takes place at all three auction locations, to make sure that buyers are not able to switch and avoid taking part in the experiment.
4.3 Implementation of the experiment
The experiment started in week 37 of the year 2013 and lasted till the end of the year. All sellers of gerbera flowers are informed about the experiment in two ways. The auctioneers of the flower auction visited most of the growers personally to explain the experimental rules. Additionally, all growers received an e-mail with the same
14
questions of a survey. In this survey, which was done after the start of the experiment, 69 percent of the sellers stated that they were well informed about the rules of the experiment. The other 31 percent say they were not well informed. Almost all of the sellers, 94 percent, said that it was clear to them how they could add their own
reservation price to their products, and a 100 percent said it was easy to understand the minimum price codes.
The buyers were informed about the experiment in a slightly different way. An e-mail with all the information was sent to all of them, slideshow presentations,
presenting all the information, were shown at the auction during the weeks prior to the experiment, and right before the start of the experiment all auctioneers explained the experiment again to all buyers over the audio system to which all buyers are connected.
During the experimental period, 461 unique buyers bought gerberas at the auction. From these buyers, only 143 are considered active buyers, a buyer who on average buys flowers once a day. Thirty active buyers were asked questions about their experience. From the survey 83 percent stated that it was clear when the experiment would start, however only 73 percent of the buyers said that the rules of the experiment were completely clear to them. 70 percent of the buyers answered “no” to the question, are you content with the experiment. To find out what was the reason for their
dissatisfaction, the buyers that were not content with the experiment were asked for an explanation, which was given by only 14 buyers. The reasons are shown in table 4.
4.4 The dataset
The dataset used to analyze this experiment consist of auction transactions made during the period week 30 till week 52 in 2013. The experiment only takes place from week 37-52. In this period, growers of large flower gerberas are able to choose a reservation price for their products. Also the data from the same period in the years 2009-2012 are collected to use as control data.
Many different types of Large Flower Gerbera’s are traded at the auction, and new types are introduced each year. These types are mainly distinguished based on the color of the flower. Because of the large quantity of types only 10 types are included in the data-set. These ten types are selected based on two criteria. The first criteria is that the flowers are the most commonly sold. The second criteria is that the flowers where sold during the whole period of the last five years, because the prices of newly introduced
Reason for dissatisfaction Number of buyers
Confusion during the auction process, because the experiment was not clear
7 The reservation price should be
showed to the buyers
3 The price should be determined by the
market, not by the seller
4 Table 4 reasons for dissatisfaction
15
flower types can be more volatile. In total the selection used for the data set covers almost 40 percent of the total amount of large flower gerberas sold during the registered period.
As mentioned before, the dataset consists of auction transactions. Every
transaction includes the following information: it tells what product is sold, who is the grower, who is the buyer, what is the date, day and time of the transaction, at which location did the transaction take place, what is the amount offered, what is the amount sold, what is the reservation price and what is the selling price. In case a product was not sold, this transaction is still included in the data-set, only the amount sold and the selling price are equal to zero.
In the data of 2013, 36490 transactions are reported, 27 unique sellers, bring their flowers to the auction. From these, 4 sellers are deleted from the data set because they only show up a limited number of times. Because their limited presence at the auction, the possibility is high that these sellers did not get enough information about the
experiment or did not have many motivation to take part in the experiment. 561 unique buyers buy Gerberas at the auction in this period.
Because it is not mandatory for sellers to choose a reservation price, not every seller chooses one. Especially in the beginning less than 40 percent of the transaction have a reservation price. This number increases during the experimental period to almost a 100 percent in the last view weeks. The number of transaction including a reservation price are shown per week in graph 1.
Graph 1 0 20 40 60 80 100 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 Per ce n tage Week
16
5. Hypotheses
5.1 Choosing the reservation price
Hypothesis 1
As written before, a reservation price can be used to signal private information. According to Cai, Riley, & Ye (2007), sellers with positive private information about their product, can use the reservation price to signal this information.
At the flower auction it is tried to provide bidders with as much information as possible about the quality of the flowers. Every seller has to specify a number of characteristics of the flower and has to give a quality code for his product. Before the start of the auction process a selection of flowers is checked to see if the quality code indeed corresponds with the quality of the flowers.
For a potential buyers it is difficult to know the exact quality of the flowers. First of all, because the amount of information given is high. For a buyer it is impossible to read all the information about all the potential flowers he wants to buy. Next to this, there is a possibility to personally inspect the flowers before the start of the auction process. But as the auction process start at 6.00 a.m., a buyers should start his inspection very early in the morning, which is not a very appealing time for most buyers. Also, most buyers join the auction via internet, and are nog physically present at the same place as the flowers.
Because many factors influence the quality of the flower (amount of light, amount of water, temperature, type of soil, timing of harvest), the seller will always have a better idea of the quality of the product than the buyers. For this reason we expect that sellers of high quality products will choose a higher reservation price.
H1: Sellers of the high type will choose a higher reservation price then
sellers of the low type.
The difficulty of this test is to find out which sellers are of the high type and which sellers are of the low type. Because of the lack of information on the quality of the flowers, the results of previous years are used. Considering that sellers who sell high quality flowers yield higher selling prices than sellers of low quality flowers, the selling prices of the years 2009 – 2012 are used to create to groups of sellers.
To test what is the effect of the name of the seller on the selling price, the following model is tested:
𝑃𝑟𝑖𝑐𝑒 𝑝𝑒𝑟 𝑠𝑡𝑒𝑚 = 𝛽0 + 𝛽1 ∗ 𝐷𝑢𝑚𝑚𝑦 𝑆𝑒𝑙𝑙𝑒𝑟 𝑛𝑎𝑚𝑒 + 𝛽2 ∗ 𝐷𝑢𝑚𝑚𝑦 𝑃𝑟𝑜𝑑𝑢𝑐𝑡 𝑛𝑎𝑚𝑒 From testing this model, it is clear that both the name of the seller and the type of product has a significant effect on the selling price. These results are as expected, as van den Berg & van der Klaauw (2007) found a significant effect of seller name on selling price as well. These results show how much the name of a seller increases or decreases
17
the average selling price. From the 21 sellers, we give the five sellers with the highest selling prices the label “good” and the sellers with the lowest selling price the label “bad”.
To see if ,on average, the good-type sellers set a higher reservation price that the bad-type sellers, we look at the reservation prices these sellers choose for their
products during the first five weeks of the experiment. We only use the first weeks, because we especially in the beginning high quality flower sellers feel more confident asking for a higher price compared to low quality flower sellers. Also we expect that later during the experiment reservation prices of all sellers will end up at the same price in an equilibrium.
Hypothesis 2
In the experiments of Elyakime, Laffont, Loisel, & Vuong (1994) and Bajari & Hortaçsu, (2003) a hidden reservation price leads to a lower probability of sale. The same can happen in the experiment at the flower auction. If sellers don’t know what the
reservation price is, there is a possibility that they wait too long to press the button. If this happens the flowers will be destroyed. Because destroying flowers is costly for the seller. He might want bidders to know what his reservation price is. For this reason a seller can choose to, once he found the right price, keep his reservation price the same over a long period of time. This way bidders can learn what his reservation price is, and the probability of withdrawal can be decreased. For this reason we expect that
reservation prices will stabilize over time during the experiment.
As described in section two, a general price setting infinitely repeated game has two equilibria. The first is that all players choose to cooperate and all choose a high price. The second that at least one deviates by choosing a low price. This action will be followed by all other players choosing a low price as well. Because in this experiment the number of price setters is high, a sustainable cooperation is difficult. For this reason the fourth hypothesis is that all reservation prices will end up at the original price of four cents, which is the lowest price possible.
This information leads to two hypothesis. The first is that reservation prices will stabilize in time. The second is that all reservation prices will end up at the lowest price possible, 4 cents.
𝐻2: 𝑀𝑜𝑠𝑡 𝑟𝑒𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛 𝑝𝑟𝑖𝑐𝑒𝑠 𝑤𝑖𝑙𝑙 𝑒𝑛𝑑 𝑢𝑝 𝑎𝑡 𝑓𝑜𝑢𝑟 𝑐𝑒𝑛𝑡𝑠
5.2 Increase of selling prices and withdrawal
Hypothesis 3
The theortictal work from van den Berg & van der Klaauw( 2007) shows that the use of a reservation price, can increase the sellingprice of products sold at the flower auction. Therefore the fourth hypothesis is:
18
To test if an increase in the selling price takes place during the period of the experiment, the data set is converted into a time series data. For every year from 2009-2013 we have the average selling price per week for the weeks 30-52. This means 115 observations in total. In the graph below it is possible to see that there is a clear pattern in selling prices ever year.
Graph 2
With these data it is possible to generate an autoregressive model with a break at the moment of the start of the experiment.
𝑆𝑒𝑙𝑙𝑖𝑛𝑔 𝑝𝑟𝑖𝑐𝑒(𝑡) = 𝛽0 + 𝛽1𝑆𝑒𝑙𝑙𝑖𝑛𝑔𝑝𝑟𝑖𝑐𝑒 𝑡 − 1 + 𝛾0𝐷𝑡(𝜏) + 𝛾1[𝐷𝑡(𝜏)]𝑆𝑒𝑙𝑙𝑖𝑛𝑔𝑝𝑟𝑖𝑐𝑒𝑡 − 1
𝐷𝑡(𝜏) is the binary variable for the break date. To test if there is a break in the data, a Chow test will be performed. The following hypothesis will be tested.
𝐻0: 𝛾0 = 𝛾1 = 0
If there a break excists in the data of the selling price, this hypothesis will be rejected.
Hypothesis 4
According to the theory of Riley & William (1981)and Myerson (1981), setting a reserve price increases the probability of the object not being sold. This is why we
0 0.1 0.2 0.3 0.4 0.5 0.6 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 Se lli n g p ri ce in E ru o 's Week
Average selling price per week, year 2009-2013
2009 2010 2011 2012 2013
19
expect that during the experiment the number of transactions that do not lead to a sale will increase. The fifth hypothesis for this reason is:
𝐻4: 𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑜𝑓 𝑤𝑖𝑡𝑑𝑟𝑎𝑤𝑎𝑙 𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒𝑠 𝑑𝑢𝑟𝑖𝑛𝑔 𝑡ℎ𝑒 𝑒𝑥𝑝𝑒𝑟𝑖𝑚𝑒𝑛𝑡
The percentage of withdrawel over the years 2009-2013 is shown in the figure below. It is clear to see that a yearly trend excists for the percentage of withdrawal. Also it is notable that in week 44, a sudden increase is shown in the year 2013.
Graph 3
Again the time series data are used to test this hypothesis. The same type of model is used to see if a break occurs in the percentage of withdrawal at the start of the
experiment. The weekly percentage of withdraw of the total amount of flowers brought to the auction is used.
𝑊𝑖𝑡ℎ𝑑𝑟𝑎𝑤𝑎𝑙(𝑡) = 𝛽0+𝛽1𝑊𝑖𝑡ℎ𝑑𝑟𝑎𝑤𝑎𝑙 𝑡 − 1 + 𝛾0𝐷𝑡(𝜏) + 𝛾1[𝐷𝑡(𝜏)]𝑊𝑖𝑡ℎ𝑑𝑟𝑎𝑤𝑎𝑙𝑡 − 1
Again a Chow-test will be performed to see if: 𝐻0: 𝛾0 = 𝛾1 = 0 0 0.05 0.1 0.15 0.2 0.25 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 Per ce n tage Week
Percentage of withdrawal per week, year 2009- 2013
2009 2010 2011 2012 2013
20
6. Results
6.1 Choosing the reservation price
Hypothesis 1
H1: Sellers of the high type will choose a higher reservation price then sellers of
the low type.
To see if high type sellers choose higher reservation prices compared to low type sellers in the first five weeks of the experiment, we conducted a graph. In graph 5
Graph 4
We can see in graph 4 that sellers from the bad group choose a higher reservation price on average. It is also interesting to know if these sellers are also more eager to use a reservation price to begin with. In the first weeks of the experiment, not all sellers are using a reservation price to sell their products. Graph 1 shows that in the first weeks, only about forty percent of all transactions include a reservation price. Looking at the difference of group good and group bad during the first five weeks of the experiment we see that a reservation price is included in 11% of the transactions of the good sellers. This number is higher in the bad group, where 40% of the
transactions include a reservation price. We test if significantly less sellers of group good choose a reservation price compared to the sellers of group bad with the following regression:
𝐷𝑢𝑚𝑚𝑦 𝑟𝑒𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛 𝑝𝑟𝑖𝑐𝑒 = 𝛽0 + 𝛽1 𝐷𝑢𝑚𝑚𝑦𝐺𝑟𝑜𝑢𝑝 𝑔𝑜𝑜𝑑
We find 𝛽1 a equal to -028, which is significantly different from zero. This means that seller in group good are less likely to choose a reservation price in the first few weeks.
0 20 40 60 80 100 104 105 106 107 108 N o rm al ize d n u m b e r o f tim e s c h o o sen Reservation price
Reservation prices schoosen by Good and Bad sellers in first five weeks
Bad Good
21
Hypothesis 2
𝐻2: 𝑀𝑜𝑠𝑡 𝑟𝑒𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛 𝑝𝑟𝑖𝑐𝑒𝑠 𝑤𝑖𝑙𝑙 𝑒𝑛𝑑 𝑢𝑝 𝑎𝑡 𝑓𝑜𝑢𝑟 𝑐𝑒𝑛𝑡𝑠
In graph 5, which is shown below, you can clearly see that the reservation price does stabilize in time. In the first eight weeks of the experiment, different prices are chosen. From of week 45 a clear increase of the reservation price of eight cents can be seen. Hypothesis two is rejected, because the reservation price doesn’t end up at four cents, but at eight cents
Graph 5
Because it is an interesting result that all reservation prices end up at eight cents, we asked all sellers how they chose their reservation price. The answers to this question are shown in table 5.
reason Number of sellers A feeling 5
Cost price of production 2
Trial and error 1
Reasonable price 3
Double of the original price 1
Cost price of production minus four cents
2
Random 2
Table 5 reasons to choose a particular reservation price
0 10 20 30 40 50 60 70 80 90 100 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 Per ce n tage Week
Percentage of choosen reservation prices per week
104 105 106 107 108 109 110 Reservation price codes
22
6.2 Increase of selling prices and withdrawal
Hypothesis 3
𝐻3: 𝑠𝑒𝑙𝑙𝑖𝑛𝑔 𝑝𝑟𝑖𝑐𝑒𝑠 𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒 𝑑𝑢𝑟𝑖𝑛𝑔 𝑡ℎ𝑒 𝑒𝑥𝑝𝑒𝑟𝑖𝑚𝑒𝑛𝑡
The F-test performed to determine if a break exists in the average selling price per week shows a P-value of 0.3432. Because the P-value is > 0,05, the H3 is rejected. And the start of the experiment does not cause a break in the regression coefficient for average selling prices.
Hypothesis 4
𝐻4: 𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑜𝑓 𝑤𝑖𝑡𝑑𝑟𝑎𝑤𝑎𝑙 𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒𝑠 𝑑𝑢𝑟𝑖𝑛𝑔 𝑡ℎ𝑒 𝑒𝑥𝑝𝑒𝑟𝑖𝑚𝑒𝑛𝑡
The F-test performed to determine if a break exists in the average selling price per week shows a P-value of 0.9706. Because the P-value is > 0,05, the H4 is rejected. And the start of the experiment does not cause a break in the regression coefficient for the percentage of withdrawal.
7. Discussion
The experiment at the flower auction was a market experiment. Because of the fact that the participants were all real sellers and buyers, it was difficult to control for many factors. On example is that it was not possible to prevent contact between the different participants of the experiment. Potential contact between these participant can have affected the results of the experiment.
The relationship between Gerbera sellers has two sides. On the one side they are competitors, because they all try to sell the same products to the same buyers. On the other side they are cooperators, because their product competes with other,
comparable products like roses or tulips. For this reason it is possible that the sellers had contact with each other about what reservation price they would use for their products.
One of the most surprising results from the experiment is that all reservation prices ended up at eight cents. It is not clear from the data and from the survey why this reservation price was chosen by almost all sellers. It would be interesting to find out what led to this equilibrium and to see how this experiment would develop in the long run. An experimental period of sixteen weeks might be too short to see what the long term equilibrium of this experiment is.
In the graph 1, a clear jump is shown in week 45. From of this week a striking larger number of sellers choose to participate in the experiment by choosing a reservation price. It is hard to explain this jump, further questions in a survey can be useful.
23
The first hypothesis tested the difference between high- and low quality flower sellers. To determine which sellers were of type high and which sellers were of type low, we used the average selling prices of the previous years. Of course the selling prices show the believes of buyers about the quality, not the quality itself. A better way to determine the quality of the flowers would be to use objective quality reports from the flower auction.
The reservation price that was chosen by the flower sellers was not shown to the buyers at the auction. For this reason the buyers did not know what the reservation price was. However, many buyers are present at the auction every day and there is a possibility that after a certain period of time, most buyers learned which sellers used a reservation price and what this reservation price was. To really see what the effect on selling price and withdrawal would be in a secret reservation price experiment, it would be better to test this in a more randomized experiment. For example in a setting where reservation prices are randomly assigned to different products. Also it would be better if there was no possibility for buyers to contact sellers. Another interesting experiment would be the same experimental set up, but to make the reservation price public. This way, sellers get the possibility to use their reservation price as a signal.
8. Conclusion
In this thesis, data are used from an experiment at a flower auction. To trade flowers, a Dutch auction mechanism is used. This means that an auction clock starts at a high price and that this prices decreases until a bidder stops the auction to buy the flowers for the current price. During the experiment, all sellers were allowed to choose their own reservation price for their products.
At the start of the reservation-price experiment, not every seller participated by choosing a reservation price. Less than half of all sellers choose a reservation price in the first view weeks. We have found that flower sellers with a bad reputation are more likely to choose a higher reservation price compared to sellers with a good reputation. Moreover, sellers from the good group are less eager to use a reservation price in the first weeks of the experiment compared to sellers of the bad group. As the experiment continued, more sellers choose reservation prices, and also reservation prices started to stabilize to a price of eight cents. In the last weeks of the experiment, almost all sellers choose a reservation price and almost all choose a price of eight cents. It is not clear why the reservation price ends up at 8 cents. The reasons for choosing a reservation price differ, but most sellers say they choose their price because of a feeling, random or because they think it is a reasonable price.
The effect of the reservation price on the amount of withdrawal and the selling price is unclear. From reading the literature we expected that the amount of withdrawal and the selling price would increase. This is not the case. Both variables did not show a clear break in their time series data.
A sudden increase of withdrawal is shown in week 44 of the year 2013. Also it is shown that around this week, the number of transactions with a reservation price
24
rapidly increases. Also the number of reservation prices of 8 cents starts to increase around the same time. It seems that a sudden increase of reservation prices and a sudden increase of the height of reservation prices does lead to an increase in the number of transactions which don’t end up in a sale.
25
9. Bibliography
FloraHolland. (2012). Annual report 2012. Aalsmeer, Pieters RA
http://jaarverslag2012.floraholland.com/media/1469819/floraholland-jaarverslag-2012.pdf
Bajari , P., & Hortaçsu, A. (2003). The Winner's Curse, Reserve Prices, and Endogenous Entry: Empirical Insights from eBay. The RAND Journal of Economics, (34), 2, pp.329 - 355. Bergmann, D., Shi, X., & Valimaki, J. (2009). Information acquisition in interdependent value
auctions. Journal of European Economic Association (7)1, 61-89.
Cai, H., Riley, J., & Ye, L. (2007). Reserve price signaling. Journal of Economic Theory (135), 253-268.
Capen, E. C., Clapp, R. V., & Cambell, W. M. (1971). Competitive bidding in high-risk situations.
Journal of Petroleum Technology (23) 6, 641 - 653.
Elyakime, B., Laffont, J., Loisel, P., & Vuong, Q. (1994). First-Price Sealed-Bid Auctions With Secret Reservation Prices. Annales D'économie et de statistique (34), 115-141. Gjolberg, O., & Steen, M. (1997). Forecasting Prices at the Dutch Flower Auctions: Calendar
Patterns and Other Time regularities. Discussion Paper #D-10/1997.
Kagel, J. H., & Levin, D. (1986). The winner's curse and public information in common value auctions. The American economic review 67(5), 849 - 920.
Katkar, R., & Lucking-Reiley, D. (2000). Public Versus Secret Reserve Prices in eBay Auctions: Results from a Pokémon Field Experment. Working Paper, 1-30.
Klemperer, P. (1999). Auction Theory: A guide to the literature. Journal of economic surveys, 227- 286.
Krishna, V. (2010). Auction Theory. London: Elsevier.
Levin, D., & Smith, J. L. (1996). Optimal reservation prices in auctions. The economic Journal
(106), 1271-1283.
Li, H., & Tan, G. (2000). Hidden Reserve Prices with Risk-Averse Bidders. Working paper , 1-31. Lucking-Reiley, D., Bryan, D., Prasad, N., & Reeves, D. (2007). Pennies from eBay, the
determinants of price in online auctions. Journal of Industrial Economics (LV) 2, 223-233. Milgrom, R. P., & Weber, R. J. (1982). A theory of auctions and competitive bidding.
Econometrica, 1089 - 1122.
Myerson , R. B. (1981). Optimal auction design. Mathematics of Operations Research (6), 58-63. Osborne, M. J. (2001). An introduction to game theory. Oxford University Press.
Paarsch, H. J. (1992). Deciding between the common and private value paradigms in empirical models of auctions. Journal of Econometrics 51, 191 - 215.
Poundstone, W. (1993). Prisoner's Dilemma. United States: Random House.
Riley, J. G., & William, S. F. (1981). Optimal Auctions . The American Economic Review, 381- 392. Thaler, R. H. (1988). Anomalies, The winner's curse. Journal of Economic Perspectives (2) 1, 191 -
202.
van den Berg, G. J., & van der Klaauw, B. (2007). If winning isn't Everything, Why do they keep Score? A Structural Empirical Analysis of Dutch Flower Auctions. Tinbergen institute
Discussion paper.
Vickrey, W. (1961). Counterspeculation, auctions, and competeitive sealed tenders. Journal of
Finance, 8-36.
Vincent, D. R. (1995). Bidding of the Wall, Why Reserve Prices May be Kept Secret. Journal of
