1 Statement of Originality
This document is written by Student Mats Mackaij who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
2
Abstract
In the after-math of the recession, bank lending dropped because banks became more reluctant or unable to lend capital to businesses. This contributed to an increase in demand of alternative ways for raising capital, one of which is crowdfunding. In this thesis I try to find evidence of the effect of price discrimination in crowdfunding. On basis of the intuition obtained from previous literature, I use a dataset of 32.000 Kickstarter projects to see if price discrimination is effective in crowdfunding. My hypothesis is that increasing the number of reward levels or increasing the range of the rewards, will increase the amount of money pledged to a project and will help the project to successfully reach its funding goal. I use a linear regression model on the amount pledged and a logit model on success, controlled for other variables, to see if I can find evidence to support my hypothesis. The result of the analysis shows that both measures of price
discrimination have a positive effect on the amount pledged as well as on success. In addition the results show that the two measures turn out to be substitutes. The insights of this paper lead to a better understanding of the dynamics of crowdfunding and to the effectiveness of price
3
Table of Contents
1. Introduction ... 4
2. Price discrimination in Crowdfunding ... 7
3. Hypothesis ... 11
4. Data and Descriptive Statistics ... 12
5. Empirical research ... 17
Model 1: Linear regression on pledged amount ... 17
Model 2: Logit regression on success ... 17
6. Results ... 19
Regression results model 1 ... 19
Regression results model 2 ... 21
Results control variables model 1 and 2 ... 22
7. Conclusion and Discussion ... 24
8. Appendix ... 26
Appendix 1: Test for heteroscedasticity and normal distributed errors ... 26
Appendix 2: Dummy Category and Dummy Location ... 27
4
1. Introduction
In the after-math of the recession, bank lending dropped because banks became more reluctant or unable to lend capital to businesses. This caused a gap between demand and supply of capital for SME (small and medium-sized enterprises) lending. Furthermore, the 5 F’s -founders, family, friends, fans and fools – where used to accommodate primary financing of startups. Harrison (2013) finds that post-2008 this kind of financing has declined as well as result of tightened household budgets. Additionally, Harrison sees a decrease in investment from venture capital and private equity in startups. All these factors contributed to an increase in the demand of alternative ways for raising capital as a starting business. One of these alternatives, which has shown a massive growth rate, is crowdfunding.
The most widely used definition of crowdfunding is “an open call, essentially through the Internet, for the provision of financial resources either in form of donation or in exchange for some form of reward and/or voting rights in order to support initiatives for specific purposes” (Belleflamme, Lambert, & Sweinbacher, 2010). According to the Massolution Annual
Crowdfunding Industry Reports, crowdfunding platforms raised a total of $16.2 billion in 2014, which is 266% of the $6.1 billion raised in 2013. The total amount raised by crowdfunding platforms is expected to increase by a factor 2,12 to $34.4 billion in 2015 (Massolution, 2015). With fewer restrictions than banks, crowdfunding poses a solution to a lot of businesses who are looking for a loan and who are denied one by the banks or are unwilling to pay a high interest on their loan (Bruton, Khavul, Siegel, & Wright, 2015).
Because crowdfunding is relatively new, the dynamics of crowdfunding are partly unknown. This form of funding is different from traditional forms and due to the characteristics of crowdfunding, otherwise not applicable economic theories become applicable to this
alternative way of funding, for example the theory of price discrimination. In this thesis I will be looking at the mechanisms of this fast growing phenomenon, where the focus will be on price discrimination within crowdfunding. I will investigate if there is evidence of price discrimination in the most popular form of crowdfunding, called reward based crowdfunding. I start by
discussing the definition of price discrimination and by reviewing literature on price
discrimination in other forms of finance. After that the focus of the literature review will shift towards price discrimination in crowdfunding. This paper will then use a dataset from Kickstarter, currently the biggest crowdfunding website, to see if any evidence of price
discrimination in crowdfunding can be found. A summary of conclusions follows and ideas for further research are raised.
5
As mentioned before the focus of this thesis lies with reward based crowdfunding. Roughly there are five types of crowdfunding: debt crowdfunding, equity crowdfunding, donation crowdfunding, pre-order crowdfunding and reward crowdfunding (Harrison, 2013). In all types of crowdfunding, there are two types of participants. First of all there is the project owner, who comes up with an idea for a project for which he needs funding. If he receives enough funding he will carry out is idea. Beside the project owner, there is the investor. He likes the idea of a project and is willing to invest money. The big difference between the types of crowdfunding is the compensation of the investor for its investment. In debt crowdfunding a company is in need for capital and in return will give the investors interest on their investment. This happens without the interference of a bank (Bruton et al., 2015). “Equity crowdfunding is a form of financing in which entrepreneurs make an open call to sell a specified amount of equity- or bond-like shares in a company on the Internet, hoping to attract a large group of
investors”(Ahlers, Cummings, Günther, & Schweizer, 2013). The investor is compensated for his investment in the form of equity in the company. Equity crowdfunding is more heavily regulated than the other two types because it has to comply with the rules of the SEC for securities. In donation crowdfunding the investor receives no compensation for his investment so the investment becomes a donation. Usually the investor does however receive a formal thank you or an individual mention on the website. Harrisson (2013) uses pre-order as a separate type, where the investor receives the product that the project owners strives to produce. In this paper I use the term reward crowdfunding for all of the last three types. Reward crowdfunding offers a nominal token in exchange for the funding. Because this nominal token can be anything, from the product itself to a formal thank you, donation and pre-order can be grouped under reward crowdfunding. Most reward crowdfunding platforms work with an all-or-nothing setup. The project owner sets creates a page with information about the project and sets a goal for the desired funding. If the goal is reached or overshot in the limited time period of the investment round, the project owner receives the money for his project. If however the goal is not reached, the project owner gets nothing and the money is returned to the individual investors.
Reward crowdfunding is used for all sorts of projects. Some websites are only intended for a niche, like Sell A Band which is a music only website. At the same time, Kickstarter accepts any kind of project as long as something is created with exception for charity projects or projects which offer monetary rewards. Examples of the kind of projects on Kickstarter are events, products, software, art and gadgets. The project has no limit on the amount of funding but it must reach its initial goal to receive any funding at all. If the project fails to reach its goal, all investors get their money back and the project will not be deemed successful.
Charity organizations are also making use of this new way of receiving funds.
6
the web, which is faster and cheaper. Echoing the move of charities to crowdfunding, scientists and researchers are trying to use the power of the crowd to fund their research. These scientific projects can be found on science-centric websites but also on the major general websites like Kickstarter and Indiegogo. Most of the projects are charity-based because the research usually doesn’t end up in producing a sellable product (Wheat, Wang, Byrnes, & Ranganathan, 2013).
In this thesis I try to find evidence of the effect of price discrimination in crowdfunding. On basis of the intuition obtained from previous literature, I use a dataset of 32.000 Kickstarter projects to see if price discrimination is effective in crowdfunding. My hypothesis is that
increasing the number of reward levels or increasing the range of the rewards, will increase the amount of money pledged to a project and will help the project to successfully reach its funding goal. I use a linear regression model on the amount pledged and a logit model on success, controlled for other variables, to see if I can find evidence to support my hypothesis. The result of the analysis shows that both measures of price discrimination have a positive effect on the amount pledged as well as on success. In addition the results show that the two measures turn out to be substitutes.
The implications of these results are twofold. First of all, project owners in crowdfunding should try to create a wide range of rewards or more reward levels, since this will yield them a higher amount pledged. Secondly, it shows that crowdfunding can outperform other ways of financing, because a project owner is able to effectively price discriminate by which he can surpass the amount of funding he would receive without price discrimination.
As for the scientific relevance of this research, an addition to existing literature on price discrimination in crowdfunding is made. All papers on price discrimination in crowdfunding are purely theoretical, whereas this research supports these models with empirical evidence. This paper also serves as an addition and update to the empirical evidence on Kickstarter, since most empirical evidence on Kickstarter is from an earlier time period. The insights of this paper lead to a better understanding of the dynamics of crowdfunding and to the effectiveness of price discrimination in new forms of finance.
The rest of this thesis is structured as follows. The following chapter reviews the existing literature on price discrimination and in particular on price discrimination in crowdfunding. After that I outline the data and discuss the descriptive statistics. Subsequently the results of the regression models are discussed and the paper is ends with a conclusion and ideas for further research.
7
2. Price discrimination in Crowdfunding
The definition of price discrimination most commonly used is: “The same commodity is sold at different prices to different consumers” (Phlips, 1983; Varian, 1985). Three conditions must be fulfilled for a company to use price discrimination as a tool to increase their own welfare. First of all, the company must have market power. Secondly the company must be able to sort costumers so it can charge different costumers different prices. Finally, resale of the product cannot be possible for price discrimination to work (Varian, 1985). Pigou was the first to divide price discrimination in three degrees (Pigou, 1920), but Tirole (1988) updated this division as follows based on the amount of information available to the company about the consumers:
First-degree or personalized pricing involves the seller charging a different price for each unit of the good in such a way that the price charged for each unit is equal to the maximum willingness to pay for that unit. In this case the firm has complete information about the preferences of each individual.
Second-degree price discrimination, or menu pricing, occurs when prices differ
depending on the number of units of the good bought, but not across consumers. In this case the firm cannot observe any information about preferences and is limited to using a form of self-selection to increase the producer surplus.
Third-degree price discrimination or group means that different purchasers are charged different prices, but each purchaser pays a constant amount for each unit of the good bought. In this case some information about consumers is available by which the firm is able to sort the consumers in groups and charge different prices to each group.
Reward crowdfunding platforms, like Kickstarter or Indiegogo, exhibit characteristics where second-degree price discrimination could exist (Cuellar & Brunamonti, 2014; Lehner, Grabmann, & Ennsgraber, 2015). Since all projects are unique, the project owner market power, ranging from some to a lot, depending on the uniqueness of the project. Because of the complete lack of information about the preferences of potential investors, a project owner has to rely on a form of self-selection to discriminate between different investors. Usually the project is in the early stages of the process and has no clear view on their customer base yet. This self-selection is also called the theory of screening, in which an monopolist maximizes her profit by using a menu of options. Every consumer will buy the option which is best for him/her. This way, the monopolist can discriminate without having information about the preferences of the consumers (Jeon & Menicucci, 2005).
Crowdfunding platforms allow for different price levels, offering slightly different rewards based on the same project. The option of different price levels for the same project
8
allows the owner to create a menu of options from which the investors can choose. This can be done in a lot of ways, for example by offering a signed version and a non-signed version of the product, creating a special edition of the project, or adding an individual mention on the projects website. Being the first to have a product can be a way to create enhanced value as well
(Cholakova & Clarysse, 2015). Each investor has a subjective valuation of the options and will invest in one if his marginal utility is equal or higher than the price of the option. This way the project owner is able to sell essentially the same product for different prices. By having more options, more consumers can be served which increases the sales of the investor. According to Richard Schmalensee (1981), an increased output, i.e. higher sales is a necessary condition for crowdfunding to work. Otherwise the producer would be better off without price discrimination.
A frequently used example of menu pricing comes from Clerides (1999). In his paper about the book market he explains why book publishers publish a hardcover book at first and later publish a paperback at a discount. He also analyzes why consumers would buy the
hardcover at a higher price, which is almost identical to the paperback, except for the cover. He comes to the conclusion that there must be consumer heterogeneity that enables book
publishers to increase their profit by selling two versions of the book. The producer does not know each consumer’s preference but he does know that there is consumer heterogeneity. By offering a menu of choices, the producer enables self-selection of the consumers into different groups: the early adapters and the regular customers.
This self-selection is also encouraged by selling pre-orders of a product. Many
crowdfunding projects offer the product itself in return for the investment, by which they have created a preordering system. The investors essentially pre-order the product when funding the project and will receive the product after the project is successful. These early investments are used to finance the startup and fixed costs of the production.
Belleflamme, Lambert and Schwienbacher (2014) study price discrimination in crowdfunding. They developed a two-period model that links crowdfunding with price discrimination and preordering. They state that this system allows for second degree price discrimination. The model proposed consists of two periods, the crowdfunding period and the regular sale period, in which the successfully crowdfunded project is available on the regular market. The producer is able to offer a menu of options, the preordered version and the regular version when the product is sold to the wider public. The producer is not able to identify the different types of consumers so it uses crowdfunding with preordering as a self-selection method. “Compared to external funding, crowdfunding has the advantage of offering an enhanced
experience to some consumers and, thereby, of allowing the firm to practice second-degree price discrimination and extract a larger share of the consumer surplus”, says Belleflamme et al (2014)
9
The model has a few implications. First, according to their model, crowdfunders always pay a higher price than regular customers.. Second, in order for price discrimination to be possible, the project must offer an enhanced experience over buying the normal product once finished
because of the higher price paid by the crowdfunders. In an explanatory study of Gerber, Hui, & Kuo (2012), evidence is found for this statement. They attribute this to the feeling of the
investors of being among the first owners as well as being part of a special and privileged community. If this feeling is created, crowdfunders will pay a higher price than a regular costumer would. Third, for high capital requirements price discrimination will be less effective. If the capital requirements are high, the project will need many investors. To attract these investors the project will need to lower the preorder prices which in turn leads to lower profits from price discrimination.
Nocke, Peitz, & Rosar (2011) also use a two-period model that is able to link price discrimination to crowdfunding when preordering is used. Although not specifically about crowdfunding, the link between theory and its application on crowdfunding is easily made. According to Nocke et al, when there is heterogeneity in the uncertainty about the future valuation of a good, price discrimination is possible. Consumers with highly expected valuation of a good will want to pre-order the product because they expect a higher price later on, while consumers with low expected valuation will wait for the price to drop ones available on the regular market. The novelty of most crowdfunding projects creates high uncertainty about the future valuation, allowing for price discrimination according to Nocke et al. Kalish (1985) also finds that heterogeneity in the uncertainty of future valuation leads to early adopters who are willing to pay a higher price.
In contrast to the two-period model, Varian (2012) uses a one-period model. The model considers the finance of public goods where private benefits are rewarded for increasing contributions to the public good. According to Varian this model is, among others, applicable to crowdfunding and Kickstarter. The model has many similarities with crowdfunding websites, where the public good is the intrinsic value of the product and the private benefits are the extra forms of utility offered by crowdfunding. These private benefits can for example be a signed version of the product, a meet-and-greet with the creators or a special edition of the product. Varian compares this outcome with the warm glow effect by Andreoni (1990). According to Andreoni, people give money to charity, not only to support the charity itself but also because the donator receives utility from giving money and helping other people. According to Hardy (2013) this warm glow effect may be an incentive to participate by itself or at least increase the value of contributing via crowdfunding. Project owners can enlarge this effect by listing
contributors on the project’s page, sending thank you notes or giving the investors access to a forum, all of which are at no costs to the project owner.
10
It is not possible for a project owner to use third-degree price discrimination. He could try to make a distinction between different groups of investors, for example by giving out student or senior discounts. However, investors invest via the website without an identity check so the project owner has no way of knowing to which group the investor belongs, making third degree price discrimination impossible. First-degree price discrimination according to the definition above, is not possible on crowdfunding platforms because there is no complete information about the preferences of each individual. However, an estimation of first-degree price discrimination in the sense that every consumer is charged a price where its marginal utility is equal to its marginal costs is theoretically possible in crowdfunding. If a project owner is able to create, for example, 1000 different reward levels, this extreme form of menu pricing would tend towards self-selected first-degree price discrimination.
In addition to the literature on second-degree price discrimination, Hardy (2013) links crowdfunding and Kickstarter-like sites in particular, to first-degree price discrimination or perfect price discrimination. Due to the Pay What You Want (PWYW) model on Kickstarter, investors will be able to invest the amount by which their marginal utility is equal to their marginal costs. Of course it is impossible for the project owner to offer a range of products with ever increasing utility to satisfy all levels of investors but in theory due to the PWYW model, a model resembling perfect price discrimination could be possible. This model leads to increased efficiency due to differentiated prices (Spann & Tellis 2006). Kim, Natter & Spann (2009) find that a PWYW model causes price differentiation which can lead to higher sales where loyalty and altruism can be of great influence. Also buyers who would normally be priced out of the market can get served is this model (Bakos, 1998). According to Hardy, when there is some level of insecurity about whether or not the project will reach its goal, consumers will invest even more just to keep the project from failing which would give them zero utility. In a paper by
Kuppuswamy & Bayus (2014) on the dynamics of project backers in Kickstarter, they find the same peak in last minute investments to help reach the projects goal.
11
3. Hypothesis
Based on the literature review and general characteristics of reward crowdfunding I formulate expectations for the empirical analysis. Both Belleflame et al (2014) and Varian (2012) agree that in order to use price discrimination, additional value must be created in order to attract different levels of investors. In this research the assumption is made that every level of reward has a different subjective value, which leads different investors to invest in different levels so that its marginal utility is equal or higher than its marginal costs. These assumptions should hold since no investor would pay a higher price for a product when the exact same product is sold at a lower price as well. This means that a project owner is able to set different prices for the almost the same product, i.e. price discriminate, if he is able to create a menu of options.
To investigate if price discrimination is effective in crowdfunding I propose two measurable characteristics of crowdfunding which can be seen as measures of price discrimination.
First of all, the number of reward levels, is a clear measure of menu pricing. By offering multiple reward level of different values to consumers the project owners creates a menu of prices and is able to charge different prices to different investors for the same commodity. In theory, if a project owner is able to offer an infinite number of levels all with corresponding rewards, a slightly altered form of perfect price discrimination might even be possible according to Hardy (2013), although the project owner would still have to rely on the self-selection of investors because of the lack of information. In either case, project owners will want to offer more reward levels in order to increase their ability to price discriminate between investors. The expectation is that more reward levels lead to more money pledged to the project and a higher chance of reaching the funding goal.
Second, the range of the reward levels is a more global measure for price discrimination. It doesn’t look at the different levels but rather looks at the range of the reward levels, i.e. the lowest reward level subtracted from the highest level. The range can tell us something about the amount of second degree price discrimination in which a higher range would generally imply that more investors are able to invest so that their marginal costs are more or less equal to their marginal utility. Project owners will want to have a broader range of reward levels in order to attract a wider group of investors. The expectation is that a wider reward range leads to more money pledged to the project and a higher chance of reaching the funding goal
12
4. Data and Descriptive Statistics
I use data that were collected from Kickstarter by TheKickbackMachine, a website created by Dan Misener. Kickstarter is currently the most popular crowdfunding website, with first places in most Top Ten rankings on the internet. It has some simple statistics available on their website but doesn’t have a publicly available database. Moreover failed projects are excluded from any searches on the website which makes it hard to include failed projects in your research. The purpose of TheKickBackMachine was to make data from Kickstarter publicly available so aspiring projects owners could learn from previous projects.
The data was extracted directly from Kickstarter.com, from 2012 until 2013. The original database contains 31.196 projects, but because this dataset contains raw data, a few projects have incomplete information. After eliminating 230 projects with missing information on any one of the variables I used, the dataset I use for this research contains 30.966 projects of which 48,9% were successful. A project is deemed successful if the goal set by the project owner on funding is reached or surpassed within the funding period. These successful projects received a total of $284 million in pledges with a mean of $18.735 per successful project. The relevant variables in the database are listed below in table 1 and the summary statistics are displayed in table 2. To control for location and category a dummy variable was created. The projects are divided into 10 groups with approximately the amount of projects, keeping the same kind of category together as much as possible. Additionally all projects are categorized based on the location where the project was created. The U.S. is divided into 10 groups with the same amount of projects with one additional group for the rest of the world, totaling the dummy for location with 11 approximately equally sized groups.
Variable Description
name Name of the project
G Funding goal of the project
P Amount of money pledged to the project
S Success: goal is reached before deadline (1) of not (0)
prG Percentage of the goal pledged
avgP Average amount pledged per backer (pledged / backers_count)
backers Amount of backers of the project
lvl Number of reward levels of the project
R_list List of reward levels of the project
min Lowest reward level of the project
max Highest reward level of the project
range Range of reward levels (Maximum reward level - Minimum reward level)
Cat_main Main category of the project, 12 categories possible
Cat_sub Sub-category of the project, 51 categories possible
13
Loc City, State or City, Country of the project
dLoc Dummy for location (1-11)
dur Duration of the investment round in days
vid The project page contains a video (1) or no video (0)
fb The project has a connected Facebook page (1) or not (0)
twitter The project has a connected Twitter account (1) or not (0)
For the empirical analysis, two models are used with two different dependent variables. The first explained variable is ln(𝑃
𝐺). This is the natural logarithm of the amount pledged divided by the goal. The variable pledged is the total amount in USD of investments in a project and the variable goal is the funding goal of the project in USD. I use the logarithm of both pledged and goal to make them more manageable because of the skewness of both variables, as well as the more intuitive explanation (percentage change instead of real change) of the regression results it facilitates. Dividing by goal the effect gives the effect of pledged without the effect of goal on pledged. The second dependent variable is dummy variable for success. This dummy variable has an output of 1 for a project which have reached its funding goal within the set time period and has an output of 0 for project which have not.
In the hypothesis section I used two characteristics of crowdfunding to formulate my expectation, the number of reward levels and the range of these reward levels. In the dataset both characteristics are available, namely lvl and range. To test my hypothesis, in addition to the amount of reward levels and the range of the reward levels, I use the lowest reward level as well. The three explanatory variables, lvl, range and min, are discussed below.
In my dataset the number of reward levels (lvl) varies from 1 to 95 with an average of 9,24 in the total dataset, 10,13 for successful projects and 8,39 for unsuccessful projects. As seen in histogram 1, the distribution is right-skewed around the mean with a very lean tail, where the 95th percentile is 19 reward levels.
Table 1: Variable description
14
If the amount pledged increases with the reward levels this can be caused by two factors, the amount of backers has increased or the average amount pledged per backer has increased.
The four scatterplots below give an indication as to which factors are causing the increase in the amount pledged.
As seen in figure 2, the average amount of P per reward level increases in the amount of levels. Figure 3 suggests that the avgP per reward level increases with lvl, while figure 4 suggests that the average amount of backers per reward level increases with lvl as well. This indicates that the increase of the amount pledged with the reward levels is caused by an increase in the
avpP as well as the amount of backers. In figure 5, this co-movement is displayed.
Kickstarter has set a maximum on the height of a reward level, which is $10.000, €7.000 or £5.000. Because all monetary amounts in my dataset are denoted in USD, the reward range (range) varies from $0 (only 1 reward level) to $10.000. The mean is $2.187 overall, $2.011 for successful projects and $2.356 for unsuccessful projects. There are several logical peaks in the distribution, around $0, $1.000, $5.000 and $10.000, with high correlation to the maximum
Figure 2: Scatterplot ln(P) (mean) and lvl Figure 3: Scatterplot avgP (mean) and lvl
15
reward level of the projects as seen in figure 6. This is due to the fact that the minimum reward level is usually low, as seen below.
In the regressions I am using the natural logarithm of the range (lnrange) because the distribution is too scattered. Moreover, the dispersion is large which shrinks the smaller ranges into insignificance compared to the ranges around $10.000 (see figure 6). By using the natural logarithm of the range, the effect of the different ranges in the left side of the distribution become more visible. Finally, the interpretation of the regression coefficient becomes more generally applicable to all levels of crowdfunding projects.
Because the dependent variable in the first three regression is a natural logarithm as well, the regression output will be in the form of elasticities, i.e. a percentage change in the independent variable has a percentage change of the dependent variable as a result. When a one percentage change in the independent variable has a less than one percentage change of the dependent variable as a result this relation is called inelastic. If the relationship is exactly one for one it is called unit-elastic and if the relationship is bigger than one it is called elastic.
To check the full effect of the range, I included the lowest reward level of a project (min) in the regression. The reward scheme of the project usually starts with a few low tiers which function as donation levels or levels with a small reward, e.g. a thank-you note, autograph or signed t-shirt, which are not the product itself. In the range, the first level isn’t of significant meaning when the maximum level is set at $10.000 because it is very hard to spot the difference between a range of $9.999 or $9.995. To include this potentially important starting level, min is included in regression (2), (3), (5) and (6). In my dataset the lowest reward level varies from $1 to $10.000 with an overall mean of $9,44, $7,10 for successful projects and $11,68 for
unsuccessful projects. The 99th percentile lies at 50, which indicates the high concentration at the lower values. In fact, 80% of all projects have a starting reward level of either $1, $5 or $10. To accord with the variable range, the natural logarithm of min is used, called lnmin.
16
Finally seven control variables are used in the regressions: lnG, dur, vid, fb, twitter, dCat and dLoc. These variables are included to keep these factors from influencing the coefficients of the independent variables on the dependent variables. lnG is the natural logarithm of the goal.
Dur is the duration of the project in days, which has a mean of 32,7 in the dataset. The duration
can be chosen by the project owner but is limited to 60 days. A project owner can choose to have a video (vid) on his page. On average 84% of the projects in my dataset make use of this option. Additionally, the project owner can link a Facebook page (fb) or Twitter (twitter) account to the project. Respectivily 73% and 2,5% have connected these accounts to their project. dCat and
dLoc are dummy variables for category and location.
Variable All Success = 1 Success = 0
G 21.722 9.696 33.251 (289.130) (38.843) (402.531) P 10.308 18.735 2.229 (90.368) (128.081) (11.623) prG 136,9 266,2 13,0 (3.489) (4.984) (16,90) avgP 68,26 80,14 56,88 (105,9) (87,7) (119,7) backers 139,1 252,4 29,5 (1.142) (1.614) (155,62) lvl 9,2 10,1 8,4 (5,57) (6,04) (4,93) min 9,44 7,10 11,68 (97,2) (12,2) (135,4) max 2.196 2.018 2.367 (3.044) (2.813) (3.242) range 2.187 2.011 2.356 (3.044) (2.814) (3.241) dur 32,7 31,4 34,0 (11,0) (10,5) (11,3) vid 0,840 0,889 0,794 (0,366) (0,315) (0,405) fb 0,730 0,743 0,718 (0,444) (0,437) (0,450) twitter 0,025 0,021 0,030 (0,157) (0,142) (0,171) observations 30.966 15.157 15.809
17
5. Empirical research
In this section, I explain the empirical models that I use to test the benefit of price
discrimination. I create two models, one in which the amount pledged over goal is the dependent variable and the other in which success is the dependent variable. In both models I use lvl, range and min as independent variables, plus two interaction variables. Both models include lnG, dur,
vid, fb, twitter, dCat and dLoc as control variables. Each model will constitute of three
regressions.
Model 1: Linear regression on pledged amount
I denote 𝑦 as the logarithm of the amount pledged over goal. Further I denote: 𝑥1 as lvl, i.e. the number of reward levels
𝑥2 as range, i.e. the range of reward levels 𝑥3 as min, i.e. the lowest reward level 𝑥𝑖 as control variables where 𝑖 = 6𝑡𝑜12 Regression 1: 𝑦 = 𝛽0+ 𝛽1𝑋1+ 𝛽𝑖𝑋𝑖+ 𝜀
Regression 2: 𝑦 = 𝛽0+ 𝛽1𝑋1+ 𝛽2𝑋2+ 𝛽3𝑋3+ 𝛽𝑖𝑋𝑖+ 𝜀
Regression 3: 𝑦 = 𝛽0+ 𝛽1𝑋1+ 𝛽2𝑋2+ 𝛽3𝑋3+ 𝛽4𝑋1𝑋2+ 𝛽5𝑋1𝑋3+ 𝛽𝑖𝑋𝑖+ 𝜀
Model 2: Logit regression on success
I denote 𝑦𝑖 as success, i.e. {1𝑖𝑓𝑔𝑜𝑎𝑙𝑖𝑠𝑟𝑒𝑎𝑐ℎ𝑒𝑑𝑏𝑒𝑓𝑜𝑟𝑒𝑑𝑒𝑎𝑑𝑙𝑖𝑛𝑒 0𝑖𝑓𝑔𝑜𝑎𝑙𝑖𝑠𝑛𝑜𝑡𝑟𝑒𝑎𝑐ℎ𝑒𝑑𝑏𝑒𝑓𝑜𝑟𝑒𝑑𝑒𝑎𝑑𝑙𝑖𝑛𝑒 𝑥1 as lvl, i.e. the number of reward levels
𝑥2 as range, i.e. the range of reward levels 𝑥3 as min, i.e. the lowest reward level 𝑥𝑖 as control variables where 𝑖 = 6𝑡𝑜12
Further I denote the conditional probability to observe 𝑦 = 1 given the independent variables and control variables by: 𝜋(𝑥1, 𝑥2, … , 𝑥𝑖). The model would look like this:
𝜋(𝑥1, 𝑥2, … , 𝑥𝑖) = 𝛽0+ 𝛽1𝑋1+ 𝛽2𝑋2+ ⋯ + 𝛽𝑖𝑋𝑖
Since the dependent variable is binary and the sum of all independent variables and control variables are not, I use a logit transformation: (𝑙𝑜𝑔𝑖𝑡𝑝 = log 𝑝
1−𝑝) by which I have created the following logit regression models:
18 Regression 1: 𝑙𝑜𝑔𝑖𝑡𝜋(𝑥1, 𝑥𝑖) = 𝛽0+ 𝛽1𝑋1+ 𝛽𝑖𝑋𝑖+ 𝜀
Regression 2: 𝑙𝑜𝑔𝑖𝑡𝜋(𝑥1, 𝑥2, 𝑥3, 𝑥𝑖) = 𝛽0+ 𝛽1𝑋1+ 𝛽2𝑋2+ 𝛽3𝑋3+ 𝛽𝑖𝑋𝑖+ 𝜀
Regression 3: 𝑙𝑜𝑔𝑖𝑡𝜋(𝑥1, 𝑥2, 𝑥3, 𝑥4, 𝑥5, 𝑥𝑖) = 𝛽0+ 𝛽1𝑋1+ 𝛽2𝑋2+ 𝛽3𝑋3+ 𝛽4𝑋1𝑋2+ 𝛽5𝑋1𝑋3+ 𝛽𝑖𝑋𝑖+ 𝜀
The transformation of the regression model changes the interpretation of the model as well. Before the transformation, the right side predicts the change in 𝑦 but for a binary value this is difficult to interpret. The logistic regression model has a different interpretation where the right side predicts the change in the log odds on success. To create understandable results, the
marginal effect is used for the coefficients. The marginal effect shows the change in probability of the dependent variable being 1 when the independent variable increases by one unit. In my case the marginal effect shows the change in probability of the project being successful in percentage points when the independent variable increases by one unit.
19
6. Results
In table 3 the regression results of both models are displayed. In all regression I reject the hypothesis of homoscedasticity so robust standard errors are used. In regression (1), (2) and (3) I look at the effect of the independent variables on the natural logarithm of the amount pledged over goal, controlled for the height of the goal. In regression (4), (5) and (6) I look at the effect of the independent variables on the dummy variable for success.
Model OLS ln(L/G) Logit Success Variables (1) (2) (3) (4) (5) (6) lvl 0,109*** 0,106*** 0,161*** 0,028*** 0,027*** 0,052*** (0,003) (0,003) (0,011) (0,001) (0,001) (0,004) lnRange 0,115*** 0,173*** 0,027*** 0,051*** (0,010) (0,015) (0,003) (0,004) lnMin 0,151*** 0,051* 0,041*** 0,016** (0,011) (0,024) (0,003) (0,006) lvlrange -0,009*** -0,004*** (0,001) (0,000) lvlmin 0,012*** 0,003*** (0,002) (0,001) lnG -0,705*** -0,786*** -0,776*** -0,146*** -0,169*** -0,165*** (0,010) (0,012) (0,012) (0,003) (0,003) (0,003) D -0,009*** -0,008*** -0,008*** -0,003*** -0,003*** -0,003*** (0,001) (0,001) (0,001) (0,003) (0,000) (0,000) vid 0,990*** 0,983*** 0,955*** 0,230*** 0,234*** 0,224*** (0,035) (0,036) (0,036) (0,009) (0,009) (0,009) fb 0,089*** 0,070** 0,073** 0,004 0,002 0,003 (0,026) (0,026) (0,026) (0,007) (0,007) (0,007) twitter -0,363*** -0,325*** -0,325*** -0,105*** -0,096*** -0,096*** (0,075) (0,075) (0,074) (0,020) (0,021) (0,021)
dCat controlled for category dLoc controlled for location
Constant 3,191*** 2,856*** 2,508*** 0,485*** 0,489*** 0,489*** (0,090) (0,092) (0,126) (0,003) (0,003) (0,003)
Observations 30.966 30.966 30.966 30.966 30.966 30.966
Note 1: Coefficients of logit are displayed as marginal effect
Note 2: Heteroscedasticity-consistent standard errors used in all regressions Table 3: Regression results
Regression results model 1
In theory, a higher level of price discrimination, i.e. more reward levels/wider range, would lead to more sales or in the case of crowdfunding, more funding.
20
In all three regressions, my first measure of price discrimination, lvl has a significant and positive effect on the amount pledged. The coefficients of 0,109, 0,106 and 0,161 respectively, signify that adding 1 reward level has an estimated increase of 10,9%, 10,6% and 16,1% in the amount pledged as a result. This result poses strong evidence for the existence of effective price discrimination in crowdfunding. The data shows that by increasing the number of options, which can be seen as partaking in menu pricing, the amount of money pledged increases. When
offering more investment levels, projects are able to attract a heterogeneous group of investors. A more heterogeneous group means more investors are able to maximize their subjective return by investing in the project. This is exactly the effect which was expected after studying previous literature.
The second measure of price discrimination, range, is included in the form of a natural logarithm in (2) and (3). In both regressions the range has a significant and positive effect on the amount pledged. The coefficients of 0,115 and 0,173 respectively mean that a 1% increase in the reward range raises the amount pledged by 0,12%-0,17%. These regression results for the reward range help to build a stronger case for price discrimination. The regression suggests that increasing the range in which the project owners offers rewards, increases the amount of money pledged by investors. Since a project only sells one product, whether or not in different versions, and is able to draw in more funding by widening the range for which the product is sold, this clearly indicates that price discrimination in crowdfunding is effective. Similarly to the reward levels, when offering a wider range, projects are able to attract a more heterogeneous group of investors. With a wider spread of options, a broader variety of investors which can maximize their utility by investing, is attracted
As mentioned when describing the dataset, the variable min is included as logarithm (lnmin) to see if the minimum reward level has any influence on the amount pledged. lnmin is included in (2) and (3), which resulted in significant coefficients of 0,151 in (2) and 0,051 in (3). The coefficients mean that an increase of the lowest reward level by 1% has a 0,051-0.151% increase in the amount pledged as a result. Since 80% of the values are 1,5 or 10, an increase in the lowest reward level will have a small effect in general. This do however means that a higher lowest level is better. This could be because most low levels usually yield an intangible reward such as a formal thank you while higher reward levels actually do yield a tangible reward. If most investors give low value to a thank you, this could explain why a project with higher values for the lowest level receives more pledges. Another explanation could come from sociology where a ‘giving standard’ is observed when donating money (Wiepking, 2007). Whether
somebody is rich or poor, when donating, every group seems find the same amount appropriate for a donation. When the lowest reward level, which is generally a donation, is below this ‘giving standard’, project owners can obtain a higher pledge by raising this level without losing backers.
21
When regressing lnmin on backers, no significant result is found which means that increasing the minimum level doesn’t decrease the amount of backers, hence supporting the sociologic theory.
In regression (3), two interactions are included to check whether the effect of independent variables is constant or not. Lvlrange which is lvl * lnrange has a negatively significant coefficient of -0,009. It means that the effect of lvl decreases with lnrange. When filling in the mean of lnrange, the negative factor of this interaction becomes -0,06. This causes the overall effect of the lvl to decrease from 16,1% to 10,1% per reward level. Filling in the mean range plus one standard deviation decreases this effect even further to 8,6%. I would expect that in a narrow range, one reward level extra would not be really effective because there are other levels of more or less the same value already present. This would mean that the two mechanisms of price discrimination are complementary. When the two mechanisms are complementary, one mechanisms does not negatively affect the other. However according to the regression the two measures for price discrimination do negatively affect each other, meaning that the two mechanisms are substitutes. Increasing one of the two mechanisms increases the amount pledged. Increasing both mechanism at the same time however does not cause an increase in the amount pledged equal to the two separate mechanisms. It seems that the range between the reward levels is a variable. When increasing both the range and the number of levels, the range between the levels remains more or less the same whereas increasing one of the two will increase or decrease the range between the number of levels. To establish a broader understanding of the effect of this interaction, further research with respect to the range
between levels is necessary. The second interaction is lvlmin, which is lvl * lnmin. The coefficient in (3) is significant and has a value of 0,012, which means that the effect of either variable is bigger when the other variable is increased as well.
Regression results model 2
In addition to the first model on the amount pledged, this model checks if price discrimination is can increase the chance of successfully funding a project in crowdfunding. The dependent variable is success, where the regression coefficients stand for the marginal effect. This marginal effect shows the change in probability of the project being successful in percentage points when the independent variable increases by one unit. Corresponding with model 1, the expectation is that more reward levels and a higher reward range lead to a higher probability of success of the project.
In all three regressions, the first measure of price discrimination, lvl has a significant and positive effect on the probability of success. The probability of success increases by 2,7-5,2 percentage point (p.p.) by adding one reward level. This is in accordance with the previous
22
results which state that a project owner can effectively price discriminate by increasing the amount of reward levels.
The second measure of price discrimination, lnrange is included in regression (5) and (6). In both regressions the range has a significant and positive effect on the probability of success. The probability of success increases by 2,7-5,1 p.p. when the range is increased by 1%. This is in accordance with the previous results which state that a project owner can effectively price discriminate by increasing the reward range.
The variable lnmin is included in regression (5) and (6). In both regressions the height of the lowest reward level has a positive and significant effect on the probability of success. This probability increases by 1,6-4,1 p.p. when increasing the lowest range by 1%. This is in accordance with the results from the first model.
In regression (6), two interactions are included to check whether the effect of independent variables is constant or not. Lvlrange has a negatively significant coefficient of -0,004. It means that the effect of lvl decreases with lnrange. When filling in the mean of lnrange, the negative factor of this interaction becomes -0,025 which causes the overall effect of the lvl to decrease from 5,2 p.p. to 2,7 p.p. per reward level. Filling in the mean range plus one standard deviation decreases this effect even further to 2,1 p.p. This confirms the previous conclusion that the two mechanisms are substitutes. The second interaction, lvlmin, has a positively significant coefficient of 0,003 which is in accordance with the results from model 1.
Results control variables model 1 and 2
Finally I will briefly discuss the results of the control variables across all regressions. First of all the height of the goal has a negative effect on both amount pledged and chance of success. A reason for this could be because a higher goal is harder to reach so if the goal is too high investors will refrain from investing at all. This is essentially the same effect as the effect of herding in discussed by Kuppuswamy and Bayus (2013). They conclude that investors tend to look at other investors to get information about quality, where a project with little funding is seen as a project of bad quality. This is also why failing projects fail amply. Moreover, the all-or-nothing rule of Kickstarter creates an incentive to invest a little bit extra when close to reaching the goal, as discussed by Hardy (2013), Kuppuswamy et al (2013) and Mollick (2013).
Furthermore, the duration has a negative effect on both amount pledged and probability of success. Kickstarter states, based on their own research, that longer projects are more likely to fail and suggests to set the duration to 30 days. Kickstarter also states that making a video is of great importance, which corresponds with the regression results. Mollick (2014) attributes this effect to a lack of confidence displayed by setting a longer duration. By including a video, the investors can form a better image about the project. A video also shows a higher quality of
23
preparation which is one of the key factors in signaling quality of a project, according to Cardon, Sudek, & Mitteness (2009) and Chen, Yao, & Kotha (2009). Next are the social media variables Facebook and Twitter. I would expect that having social media has a positive effect in both regressions. It could increase publicity and at the same time it could help to create a kind of community in which a more enhanced experience is formed. This enhanced experience is necessary to price discriminate according to Belleflame et al (2014). In traditional funding, having connected social media accounts has shown help create connections and thereby increasing the chance of receiving funds (Sørensen & Fassiotto, 2011; Stam & Elfring, 2008). Agrawal, Catalini, & Goldfarb (2011) see social media as an equivalent of ‘friends and family’ money, which could give a project a head start. According to my regressions, having Facebook has a positive effect on the amount pledged, but it has no significant effect on the probability of success. It could be that a lot of projects are well below the goal so a higher amount pledged doesn’t have effect on the probability of success. Twitter on the other hand has a large negative effect in both models. This could be because mostly young people use Twitter by which Twitter could be a dummy for age or experience. Not having Twitter would mean a higher age and usually more experience, thus generally a better project. Also, the project is more out in the open on Twitter and a few negative reactions could deter other investors. When a project doesn’t have Twitter, this negative reactions are not clearly visible. The worst category for a project is
publishing (art, children fiction and non-fiction in particular). Projects in the categories Fashion, Design, Games and Technology have the biggest positive effect on the amount pledged while creating a project Music (electro, classic, indie rock, jazz, pop and rock in particular) increases the probability of success the most. As for location, the state of New York is the best option and the South East of the U.S. is the worst. More elaborate statistics about dcategory and dlocation can be found in Appendix 2
24
7. Conclusion and Discussion
This paper studies the mechanisms of the fast growing phenomenon called crowdfunding, where the focus lies on price discrimination within reward based crowdfunding. Based on existing literature it follows that price discrimination should be effective in crowdfunding. This
hypothesis is confirmed by the empirical research in this paper. I find that two measures of price discrimination, the amount of reward levels and the range of these levels, both have a positive effect on the amount pledged and the success of reaching the funding goal of the project. An increase of one reward level leads to an increase in the amount pledged of up to 16,1% and raises the probability of success by up to 5,2 p.p.. An increase in the reward range of 1% leads to a 1,7% increase in amount pledged, as well as an increase of 5,1 p.p. in the probability of
successfully reaching the funding goal. These two measures are however not complementary but substitutes. Even so, I conclude that price discrimination is indeed effective in crowdfunding. Furthermore, the height of lowest reward level also has a positive effect on the amount pledged and the success of the project. In addition, the height of the goal has a negative effect as well as the duration of the funding period. Having a video has a positive effect while the effect of having social media is ambiguous.
From a scientific point of view, this paper provides empirical evidence which consolidates the existing theoretical models on price discrimination in crowdfunding. Furthermore, the insights of this paper lead to a better understanding of the dynamics of crowdfunding and to the effectiveness of price discrimination in new forms of finance.
Moreover, some practical implications of this paper can be mentioned. Firstly, project owners in crowdfunding can benefit from increasing the number of rewards levels because price discrimination leads to a higher amount pledged to a project as well as a higher chance of the project being successfully funded. Investors can also benefit from introducing more reward levels because they are able to better maximize their utility. Note that these reward levels should offer a slightly different reward. If the investors are unable to differentiate between two levels, every investor will choose the lowest reward level of the two and no price discrimination is possible. Also, diversifying the product to allow for more options can be costly for the project owner. As Kickstarter concludes from their own research, having too many options can be confusing in which case it has a negative effect. After adding the squared terms of lvl in the regression, the negative coefficient indicates that the effect of adding another reward level will decline and can become negative, however further research is required to confirm this effect and to find the optimal menu size. The same applies to the range of the reward levels, where project owners can benefit from increasing this range. It is important to note that the two measures of
25
price discrimination are substitutes, and that this knowledge should be applied when choosing the reward levels and range.
Secondly, this research gives some indication as to the enhanced performance it can give over other ways of financing in which price discrimination may be harder. This means that for certain projects, crowdfunding might not just be a necessary solution, because of a declined bank loan for example, but a better solution as well. Further research on price discrimination in alternative means of financing are required to confirm this primary indicator.
Like all research, this paper has a few limitations which need to be mentioned. First, although Kickstarter is one of the major crowdfunding sites, care should be taken when
extrapolating the results from this paper to other types of crowdfunding, as they might differ in characteristics. Second, in the regressions not all variables effecting the amount pledged or success were included. Some variables like the quality of the idea for example are hard or nearly impossible to capture in a number. These omitted variables could change the outcome of the regressions which is why the quantitative change is to be interpreted with caution.
26
8. Appendix
Appendix 1: Test for heteroscedasticity and normal distributed errors
Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Shapiro-Wilk W test for normal errors
Regression (1) chi2 1946,33 z-value 17,550
Prob > chi2 0,0000 Prob > z-value 0,0000
Regression (2) chi2 2025,89 z-value 17,591
Prob > chi2 0,0000 Prob > z-value 0,0000
Regression (3) chi2 2064,75 z-value 17,570
Prob > chi2 0,0000 Prob > z-value 0,0000
Regression (4) z-value 17,310 Prob > z-value 0,0000 Regression (5) z-value 17,362 Prob > z-value 0,0000 Regression (6) z-value 17,564 Prob > z-value 0,0000
27
Appendix 2: Dummy Category and Dummy Location
Appendix 2a: Effect of Category on ln(P/G)
28
Appendix 2c: Effect of Location on ln(P/G) and Success
Large positive effect Small negative effect Large negative effect No significant effect
29
9. References
Agrawal, A. K., Catalini, C., & Goldfarb, A. (2011). The geography of crowdfunding.
Ahlers, G. K., Cumming, D. J., Günther, C., & Schweizer, D. (2013). Equity crowdfunding. Available at SSRN 2362340.
Andreoni, J. (1990). Impure altruism and donations to public goods: a theory of warm-glow giving. The economic journal, 464-477.
Bakos, Y. (1998). The emerging role of electronic marketplaces on the Internet.Communications of the ACM, 41(8), 35-42.
Belleflamme, P., Lambert, T., & Schwienbacher, A. (2010, June). Crowdfunding: An industrial organization perspective. In Prepared for the workshop Digital Business Models: Understanding Strategies’, held in Paris on June (pp. 25-26).
Belleflamme, P., Lambert, T., & Schwienbacher, A. (2014). Crowdfunding: Tapping the right crowd. Journal of Business Venturing, 29(5), 585-609.
Bruton, G., Khavul, S., Siegel, D., & Wright, M. (2015). New Financial Alternatives in Seeding Entrepreneurship: Microfinance, Crowdfunding, and Peer‐to‐Peer Innovations. Entrepreneurship Theory and Practice, 39(1), 9-26.
Cardon, M. S., Sudek, R., & Mitteness, C. (2009). The impact of perceived entrepreneurial passion on angel investing. Frontiers of entrepreneurship research, 29(2), 1.
Chen, X. P., Yao, X., & Kotha, S. (2009). Entrepreneur passion and preparedness in business plan presentations: a persuasion analysis of venture capitalists' funding decisions. Academy of Management Journal, 52(1), 199-214.
Cholakova, M., & Clarysse, B. (2015). Does the Possibility to Make Equity Investments in Crowdfunding Projects Crowd Out Reward‐Based Investments?.Entrepreneurship Theory and Practice, 39(1), 145-172.
Clerides, S. K. (1999). Product Selection as Price Discrimination in the Market for Books. Department of Economics. University of Cyprus.
Cuellar, S. S., & Brunamonti, M. (2014). Retail channel price discrimination.Journal of Retailing and Consumer Services, 21(3), 339-346.
Gerber, E. M., Hui, J. S., & Kuo, P. Y. (2012, February). Crowdfunding: Why people are motivated to post and fund projects on crowdfunding platforms. InProceedings of the International Workshop on Design, Influence, and Social Technologies: Techniques, Impacts and Ethics. Hardy, W. (2013). How to perfectly discriminate in a crowd? A theoretical model of
crowdfunding. Faculty of Economic Sciences, University of Warsaw Working Papers, (2013-16). Harrison, R. (2013). Crowdfunding and the revitalisation of the early stage risk capital market: catalyst or chimera?. Venture Capital, 15(4), 283-287.
Jeon, D. S., & Menicucci, D. (2005). Optimal second-degree price discrimination and arbitrage: on the role of asymmetric information among buyers. Rand Journal of Economics, 337-360.
Kalish, S. (1985). A new product adoption model with price, advertising, and uncertainty. Management science, 31(12), 1569-1585.
Kim, J. Y., Natter, M., & Spann, M. (2009). Pay what you want: A new participative pricing mechanism. Journal of Marketing, 73(1), 44-58.
30
Kuppuswamy, V., & Bayus, B. L. (2014). Crowdfunding creative ideas: The dynamics of project backers in Kickstarter. UNC Kenan-Flagler Research Paper, (2013-15).
Lehner, O. M., Grabmann, E., & Ennsgraber, C. (2015). Entrepreneurial implications of
Crowdfunding as alternative funding source for innovations.Venture Capital, (ahead-of-print), 1-19.
Massolution. (2015). 2015CF – Crowdfunding Industry Report. New York. DC: Author Misener, D. (2013). The KickBack Machine (ongoing). Retrieved from
https://www.google.com/fusiontables/data?docid=1ecu_DIOMYdWksR-xbqoGdVxTvkTfPfYwwgNXhJs#rows:id=1
Mollick, E. R. (2013). Swept away by the crowd? crowdfunding, venture capital, and the selection of entrepreneurs. Venture Capital, and the Selection of Entrepreneurs (March 25, 2013).
Mollick, E. R. (2014). The dynamics of crowdfunding: An exploratory study.Journal of Business Venturing, 29(1), 1-16.
Nocke, V., Peitz, M., & Rosar, F. (2011). Advance-purchase discounts as a price discrimination device. Journal of Economic Theory, 146(1), 141-162.
Phlips, L. (1983). The economics of price discrimination.
Pigou, A. C. (1932). The economics of welfare, 1920. McMillan&Co., London.
Schmalensee, R. (1981). Output and welfare implications of monopolistic third-degree price discrimination. The American Economic Review, 242-247.
Sørensen, J. B., & Fassiotto, M. A. (2011). Organizations as fonts of entrepreneurship. Organization Science, 22(5), 1322-1331.
Spann, M., & Tellis, G. J. (2006). Does the Internet promote better consumer decisions? The case of name-your-own-price auctions. Journal of Marketing,70(1), 65-78.
Stam, W., & Elfring, T. (2008). Entrepreneurial orientation and new venture performance: The moderating role of intra-and extraindustry social capital.Academy of Management Journal, 51(1), 97-111.
Tirole, J. (1988). The theory of industrial organization. MIT press.
Varian, H. R. (1985). Price discrimination and social welfare. The American Economic Review, 870-875.
Varian, H. R. (2012). Public goods and private gifts.
Wheat, R. E., Wang, Y., Byrnes, J. E., & Ranganathan, J. (2013). Raising money for scientific research through crowdfunding. Trends in ecology & evolution, 28(2), 71-72.
Wiepking, P. (2007). The philanthropic poor: In search of explanations for the relative generosity of lower income households. Voluntas: International Journal of Voluntary and Nonprofit Organizations, 18(4), 339-358.
Wojciechowski, A. (2009, January). Models of charity Donations and project funding in social networks. In On the Move to Meaningful Internet Systems: OTM 2009 Workshops (pp. 454-463). Springer Berlin Heidelberg.
