An estimation, description and comparison to equity investment of the cost of crowdfunding

Bachelor Thesis

An Estimation, Description and Comparison to Equity Investment of

the Cost of Crowdfunding

University of Amsterdam

Faculty of Economics and Business Supervisor: Marijn Kool

Robert Lee

Student number: 10267514 July 2014

ABSTRACT

Crowdfunding is a fast growing source of financing, particularly for start-ups. There has been, however, no attempt made to estimate what the cost of using crowdfunding as a form of finance may be. In this paper, 82 projects funded via the platform

Kickstarter are investigated and a pseudo-cost of capital estimated for each of these. This is then compared to an estimate for observed return for angel investors provided by Boeker and Wiltman (2007). A wide, non-normal distribution of this pseudo-cost of capital is found, with a median value of 32% over the lifetime of the product. Not much ground is gained in attempting to explain the drivers of this cost of capital, which suggests that there could be more intriguing factors at play than those investigated in this paper. From the analysis it appears as though crowdfunding compares favourably to orthodox forms of financing, a possible reason for its runaway growth.

Table of Contents

3. The model and its background – p. 5

3.1 A closer look at the market for funding – p. 6 3.2 Theoretical model – p. 6

3.3 The determination of the pledge – p. 8 4. Data – p. 11

5. Results and analysis – p. 14 6. Conclusion – p. 19

7. References – p. 22 8. Appendix – p. 23

1

An Estimation, Description and Comparison to Equity Investment of

the Cost of Crowdfunding

1 Introduction

Internet-based crowdfunding is a young and fast growing source of financing. On March 3rd 2014 the website Kickstarter, founded in just 2009, passed the one-billion dollar milestone for funds raised (Kickstarter, 2014a), demonstrating the growth and viability of this form of financing, as well as offering a potential wealth of data.

As this form of financing is young, research into it has yet to take shape. However, there is increasing interest into crowdfunding from market commentators such as the Financial Times (2013), government agencies (FCA, 2013) and from within research circles such as the NBER (2013).

Non-equity crowdfunding works via a system of presales. Potential funders, here called Backers, who are interested in a product yet to be made can pre-purchase the product often at a discount. The product they are purchasing is usually called the perk, and the amount for which they buy it; the pledge. The product will then be produced by the creator with the money raised from crowdfunding, and will then often be sold to the broader public at a market price. Despite the fact that projects are funded with presales, this form of financing is not free. Each presale at a discounted pledge represents a foregone sale at the full price. However, to date there has been no work which aims to estimate the cost of using Kickstarter, which can be quantified in terms of these foregone revenues. The question this paper aims to address is what the cost of non-equity crowdfunding is for the firm and how it may compare to other, orthodox forms of start-up investment.

In this paper a pseudo-cost of capital will be approximated for 82 for-profit projects for which data will be gathered from the website Kickstarter and will be compared to estimates for angel investor rates of return.

Section 2 consists of a literature review, in which research on crowdfunding to date is discussed. Section 3 is an overview of the theory of crowdfunding. Firstly, in section 3.1, a very brief description of a crowdfunding campaign will be presented, followed in section 3.2 by a derivation of the expression that this paper uses for the pseudo-cost of capital, and lastly in section 3.3 a more in depth discussion will be held on crowdfunding as a market for capital. In section 4 the method of the research

2

will be presented. Thereafter, in section 5, the findings will be analysed. Finally, in section 6, conclusions will be drawn from the analysis.

2 Literature Review

The study of crowdfunding is still relatively young, and so literature concerning it is somewhat sparse. Despite this, there are a few pieces which give a good overview of crowdfunding in general, as well as some which go into more depth on certain questions. One of the earliest pieces to look at crowdfunding is The Geography of Crowdfunding by Agrawal, Catalini and Goldfarb (2011). In this paper the authors look at the geographic distribution of funding for the Amsterdam based, and previously dominant equity-crowdfunding platform Sellaband, which focuses

exclusively on musical enterprises (Agrawal, Catalini and Goldfarb, 2011, p. 4). From this research they found that sources of funds from crowdfunding were geographically very diverse compared to traditional forms of funding, with 77.2% of funding (for which geographical information was available) coming from more than 50 km away from the project being funded , and the average distance between creator and backer being approximately 5000 km (2011, p. 24). One implication of this broad distribution is better matching between creator and backer compared to orthodox forms of

investment, which is a conclusion that Agrawal, Catalini and Goldfarb draw in another paper, Some Simple Economics of Crowdfunding (2013), whilst referring to the data from their 2011 article. In our opinion, better matching removes some of the necessity of convincing investors of the worth of one’s idea. Agrawal, Catalini and Goldfarb also note that, although capital flows from more diverse origins, it generally flows to the same places it previously has done. To explain this they suggest that capital flows into those areas with human capital and complementary assets already in place (2013, p. 5).

Agrawal, Catalini and Goldfarb go on to mention a severe skewing of the funds via crowdfunding, whereby 0.7% of the projects raised 73% of the funds in the case of Sellaband (2011, p. 7). In their 2013 paper they find a similar characteristic for Kickstarter whereby 1% of the projects raised 36% of the funds (p. 5). This less drastic skew may be, in our opinion, due to the broader range of projects offered on Kickstarter (where music is but one category among many) and the broad range of interests of potential backers, meaning that it is less likely for there to be such clear

3

winners, and whilst there are obviously those projects that attract more attention and more funding, the effect may not be as strongly felt as in the Sellaband case.

Another reason for this skew may be a herding effect, as proposed by Agrawal, Catalini and Goldfarb (2013, p. 5), as well as Kuppuswamy and Bayus (2013, p. 3) in which propensity to fund increases with the number of backers who have already funded. Linked to this is also a ‘bystander effect’(Kuppuswamy and Bayus, 2013, p. 6) whereby potential backers will hold off on backing a project until its final hours in order to see what the consensus of the crowd may be. Whilst there may be worries that herding leads to irrational decisions, Zhang and Liu (2012, p. 901) in a study on herding in microloan markets (similar to yet distinct from non-equity crowdfunding), found evidence for the rationality of herding, whereby herding behaviour acts as a signal of quality to other backers. Agrawal, Catalini and Goldfarb (2013, p. 28) also refer to a case in which potential fraud on the platform Kickstarter was spotted by crowdfunders within hours of the project being set up, after which the project was removed. The crowd thus appears to act as an effective signaling body. A curiosity worth noting is a finding from Kuppuswamy and Bayus (2013, pp. 4, 35), in which they describe a ‘bath tub’ like funding curve whereby funding is most frequent at both the early and the late stages of the campaign and flat inbetween.

Another finding from the paper of Kuppuswamy and Bayus is that

approximately 50% of successfully funded projects on Kickstarter raised between 100 and 109 percent of their funding target (and that the distribution appears to be

asymmetric, with a long and flat tail for higher percentages) (2013, p. 35). This indicates that creators can expect to receive close to their budget.

Alongside herding, Agrawal, Catalini and Goldfarb in both their 2011 and their 2013 paper draw attention to motivations for why a backer may fund a project, namely utility from participation, early access to new products, support for a new idea (philanthropy) and the formalisation of contracts, which is relevant to the case of friends and family whereby a crowdfunding platform offers the opportunity to formalise what would otherwise be an informal, and thus unenforceable, contract between two parties. In the case of equity-crowdfunding they also draw attention to ‘new investment opportunities’ (2013, pp. 14-15). Curiously, they omit the non-equity-crowdfunding motivation of a ‘good deal’ in which the pledge rate is lower than expected value of the future product. A backer’s altruistic motivations may well

4

serve to lower the cost of capital that both equity- and non-equity-crowdfunded firms face, as backers may be inclined to give them money essentially for free.

An illustration of utility derived from participation (or maybe in this case; nostalgia) is the speed and magnitude with which LeVar Burton’s project to bring back the Reading Rainbow, a previously popular children’s television programme in the USA, is as of writing being funded, having generated over $4,700,000 from 97,000 backers – not all of which, we assume, are donating strictly in order to receive a signed postcard (Kickstarter, 2014b).

Mollick, in his 2013 paper The dynamics of crowdfunding: an exploratory study focuses mostly on the failure of project creators, and also includes some remarks on the distribution of funding, finding asymmetric distributions of funding for both failed and successfully projects, with the highest frequencies near 0% and 100% respectively (p. 7), which echoes the findings from Kuppuswamy and Bayus (2013, p. 35). Mollick also provides information on the geographic distribution of funding within the USA, in which it appears that Design is the most evenly distributed category, with other categories, such as Technology, being centred on certain

locations (in the case of Technology these being San Franciso and New York) (2013. P. 10). The main finding of the paper, however, is that 75% of projects are delayed, expected delay for any given project is 1.28 months, and in the case of delayed projects; 2.4 months, and the average delay increases with the number of backers (Mollick, 2013, p. 12). He also finds that, “the vast majority of founders [creators] seem to fulfill their obligations to funders [backers].” Mollick (2013, p. 13). Alluding to Mollick’s paper, Agrawal, Catalini and Goldfarb refer to ‘over-optimism’ on behalf of both project creators and backers (2013, p. 5). However, this seems quite a bold claim to make when the only evidence available is that there are average delays of 1.28 months.

The possibility of crowdfunding being used as a source of finance is supported by research from Agrawal, Catalini and Goldfarb who looked at the use of

crowdfunding in areas with changes in house prices, and saw a substitution effect whereby as house prices increased the use of crowdfunding decreased (2013, p. 6).

As this paper aims to compare the cost of capital of crowdfunding to that from other forms of start-up investment, in this case from angel investors, estimates for the rates of return awarded to angel investors are required. Research offering concrete estimates of returns from angel investors or venture capitalists is also sparse, and the

5

best estimate is provided by Boeker and Wiltbank (2007). This source was chosen as it relates to the returns to angel investors in groups, which is the closest equity analogue to crowdfunding for which an approximation was available and also attempts to estimate realised as opposed to expected return (Boeker & Wiltbank, 2007, p. 1). Boeker and Wiltbank gathered their data from a sample of 539 investors, active over two decades and who on average have each experienced two “exits”, or the cashing in of their equity (2007, p. 1). Boeker and Wiltbank found a mean

annualised internal rate of return of 27% (2007, p. 3). However, it is worth noting that this distribution is highly skewed; 52% of all ventures ended in a loss, returning less than the money invested and 7% of exits achieved returns of over 10 times what was invested (with an average holding period of 4.9 years, or 63% annualised), some achieving even 30 times the original investment (over 6 years, or 77% annualised) (Boeker & Wiltbank, 2007, p. 7) This paints an image of a very pronounced tail and suggests that the mean of 27% be taken with a grain of salt, especially considering that the modal return is negative.

On top of the averages, Boeker and Wiltbank (2007) also suggest correlations between returns and time spent on due diligence (p. 5), possession of industry

expertise (p. 6) and time spent on participating in the start-up (p. 7). The findings are not supported with strong statistics, but rather by a visual approach (pp. 5-7).

Although not thoroughly tested these findings are in line with hypotheses that financing from angel investors has the added advantage of bringing with it expertise or benefits from supervision. However, whilst Boeker and Wiltbank (2007, p. 5) show that due diligence brings with it higher returns, returning to Mollick’s (2013, p. 13) statement that most creators make true on their promises and also drawing attention to the 52% of angel investor exits that are negative (Boeker & Wiltbank, 2007, p. 7) casts some doubt on the efficacy or worth of angel investors’ due diligence in comparison with crowdfunding.

3 The model and its background

Having discussed the literature we are now left to consider how to specify the model of the pseudo-cost of capital of crowdfunding. This will be done first in section 3.1 by taking a closer look at the market for funding. Then in section 3.2 the model itself will be developed. Finally, in section 3.3, a brief look will be had a how the pledge which

6

backers pay is arrived at and we will also take a quick look into crowdfunding as a market for capital, with the backers as supply of capital, and the project creators as demand.

3.1 A closer look at the market for funding

On a crowdfunding platform the creator puts a project up for consideration by the potential backers. The pitch will often take the form of a webpage with an overview of the product to be produced, including any prototypes. Included on this page will also be a list of pledges and their respective ‘perks’, in the simplified model the only perk available is the future product, though in reality the perks can range from a simple thanks to direct involvement in the production process, with a range of variations on goods or services inbetween. The crowdfunding process can thus be determined as a market for funding, with the creator representing the demand for capital and the backers representing the supply of capital. An interesting point to note is that the demand for capital is constant for a given project. In the long run, i.e. over multiple projects, the creator can adjust their expectations and change the pledges and perks on offer, but in the short run, i.e. for a single project, the pledges and perks cannot be edited after they have already been sold. The creator will often offer a range of perks for different pledges, the greater the pledge the greater the value of the

proposed perk.

Before continuing with the discussion of crowdfunding as a market for capital, we will introduce the model for the pseudo-cost of capital of crowfunding so that the rest of the discussion can be put into perspective.

3.2 Theoretical Model

To determine an expression for the firm’s cost of using a crowdfunding platform take a simple example in which there are no fees from using crowdfunding. The profit function of the crowdfunding firm is in this case:

Where α represents the number of backers who fund the project at the crowdfunding pledge rate K, β represents the number of buyers of the product who purchase it at price P after it has been produced and C(α + β) the cost function of production. An assumption is that both α and β are produced at the same cost.

7

In the initial funding stage the firm earns the amount αK from the backers in exchange for the promise of a product at a future date. In the second stage, after production and the incurring of costs, the product is delivered to the backers and is also put on sale to be purchased by the residual demand β at price P.

The standard firm, on the other hand, would fund production of the good with an investment I, subject to a cost of capital r and would sell the full quantity of α + β at price P. The profit function of the standard firm is thus:

To find the cost of crowdfunding set both profit functions equal:

If we assume that the firm chooses to raise via crowdfunding as much as it would otherwise borrow or sell equity for, we can set αK = I. This gives us:

Kuppuswamy and Bayus (2013, p. 25) and Mollick (2013, p. 7) show that firms typically raise near 100% of funds requested. Even if the firm were to misjudge the amount of funding required, then they would misjudge this in both cases, and so (in the case of no severe credit rationing or other liquidity restrictions in the orthodox financing case) αK = I would still hold.

Therefore, assuming that total demand, total funding raised and cost of production are all equal in both cases, the expression (P – K)/K can be used to represent a pseudo-cost of capital, and it is this expression that is evaluated in this paper.

Intuitively, this psuedo cost of capital can be explained in terms of foregone revenue. If we assume that consumers who funded the product at the pledge K would, in a case where the firm raises funds by orthodox means, otherwise buy the product at price P, then for each unit of funding raised the firm incurs the opportunity cost of the revenue that could otherwise have been earned.

8

Now we are ready to include the fee for the use of the crowdfunding platform. Kickstarter uses a fee of 5% (Kickstarter, 2014c) whereas Indiegogo uses a fee of either 4% or 9% (Indiegogo, 2014), depending upon whether the project achieves its funding goal or not. As this paper uses Kickstarter as a proxy for crowdfunding platforms it is safe to bear the 5% fee in mind during this evaluation. To account for the fee, express the cost of capital including fee as:

Where K(1 – f) is the amount of the pledge remaining after the fee has been paid to the crowdfunding platform. It follows that:

( )

This is the adjustment that is used in this paper to determine the pseudo-cost of capital taking the platform fee into account. The advantage of this expression is that it is very simple to manipulate the ‘raw figures’ to imagine the effect of smaller or greater platform fees.

The expression for rf increases rapidly in f, and approaches infinity as f tends

to 1. The derivative of rf with respect to f is increasing in both f and r:

And so the greater r, the greater the effect of the fee. This fee can have a significant impact on the cost of capital that the project is exposed to, for example a 15% cost of capital without fee would become a 21% cost of capital including the fee, which over raised funds of tens or even hundreds of thousands of dollars can make a very large difference.

3.3 The determination of the pledge

A question that remains is how the pledge amount K is determined between the creator and the backers. Separating crowdfunders from traditional funders is the fact that crowdfunders are not backing a project in exchange for a cash return, but rather in exchange for a good and the utility they would derive from it. There is thus

9

no explicit rate of return, but rather an implicit return that varies per backer according to each backer’s preferences. For a backer to fund the project, K must thus be equal to or less than that backer’s expected utility from the product. This expected utility may be a function of the expected quality of the finished good and the expectation that the project will in fact be completed.

This could be expressed as:

Where Ui represents the expected utility derived from the product for the ith backer (for simplicity assumed to be known with certainty), and P(success) the probability that the product is in fact produced.

From the firm’s perspective, to simplify for heterogeneous utility functions of consumers, the function could be expressed as:

This probability of a failed project is especially relevant as once the project has been succesfully funded – not succesfully developed – the amount is withdrawn from the backer’s account, and so backing a project represents a risk-return trade-off. The creator of the project must thus take into account the backers’ expected utilities of the product when deciding K. The creator must thus set K sufficient to generate funding such that αK = I, and this skill may take learning over repeated projects to master. This setting of K for a given P is what determines the psuedo cost of capital that the project will pay.

Whilst, as mentioned above, the demand for capital in fixed in the short run, it is not necessarily inelastic. The creator of the project can employ limits, limiting the number of backers who can pre-order at any given pledge. In doing so the project creator can craft an elastic demand for capital, offering the most favourable pledges, and thus the highest implicit rate of return, to a limited number of backers. As it is unlikely (although with philanthropic motivations still possible) that a backer would pass up a lower pledge in order to pre-order the same perk at a higher pledge, the backers, or in this case supply of capital, thus act as if they were perfect price discriminators, supplying financing from the highest to the lowest costs of capital.

However, utilising pledge limits still has its advantages for the project creator, namely the mitigation of negative effects of misjudging supply of capital. If the creator initially over-estimated the supply of capital then, at the expense of an increased cost of capital, the project creator can use this crafted elasticity of demand

10

to help ensure that a minimum backing level is met, thus reducing the risk of

underfunding. Inversely, if the creator initially under-estimated the supply of capital then they can ensure that not too many (or even no) examples of the product are sold at too low a price, by offering a pledge close to the planned price or by limiting the number of backers entirely, thus preventing project creators from cannibalising the demand for their final product. This is illustrated in the diagrams below.

Figure 1

Figure 1 shows an initial overestimation of the supply of capital and its effects on an inelastic demand (D1) as well as a more elastic demand (D2). The exact slopes of the

curves are arbitrary. Cost of capital on the y-axis is also implicit rate of return for the supply of capital (the backers). Expected supply is shown by S1 and the revised and

reduced supply by S2. Point ‘a’ gives the creator’s expected quantity of backers and

cost of capital. As the supply curve moves to the left it crosses the inelastic demand for capital at point ‘b’ and the elastic demand for capital curve at point ‘c’. At point ‘c’ underfunding is restricted at the expense of an increased cost of capital. It is important to note that point ‘c’ does not in fact give the equilibrium cost of capital in this case, but that every point to the left of ‘c’ on the line D2 is taken up by the supply

11

Figure 2

Figure 2 shows the case of an initial underestimation of the supply of capital. Here at point ‘c’ the quantity sold in the pre-order phase is limited and the final unit sold is done so at a pledge closer to the future price.

Both of these graphics are stylised in that they are continuous, as opposed to the step-diagram you would most likely encounter with real data. Due to time limitations, hypotheses concerning the micro-market (such demand for capital being downwards sloping) can unfortunately not be investigated in this paper; though it may present an avenue for future research.

4 Data

Kickstarter was used as a proxy for crowdfunding platforms in general. Kickstarter was chosen as it is the dominant crowdfunding platform (Agrawal, Catalini & Goldfarb, 2013, p. 4). Furthermore, analysis of it is relatively simple compared to other crowdfunding platforms as Kickstarter has a constant fee of 5% which all projects face (Kickstarter, 2014c), has a constant approach to when the fee is charged (upon successful funding) and when the backer is charged the pledge amount (also upon successful funding) (Kickstarter, 2014). Furthermore, a single platform was

12

chosen so as better to isolate differences in pseudo-cost of capital due to

characteristics of the supply and demand (i.e. backer and project) sides of the market for funding; the worry being that if multiple platforms were included in a rather small sample it would prove nearly impossible to infer whether variations in the cost of capital were due to behavioural differences amongst projects or to peculiarities

between platforms. The sample size was limited by the time available for this research and the fact that this form of data mining is a very time consuming process which can only be partly automated.

For-profit projects were investigated as these most closely resemble those firms that would otherwise employ orthodox forms of financing, such as debt or equity. Non-profit projects abound on Kickstarter, but crowdfunding for them most likely represents a form of donation on which a cost of capital is not applicable.

Kickstarter groups its projects into multiple categories, examples being: Design; Technology; Fashion; Publishing; Photography; Games, etc. The choice of which category a project should be listed under can sometimes seem somewhat arbitrary. For example, photography is a category in which many photography books are funded to be published, which one could argue fits just as well under the category Publishing.

The categories Design and Technology were chosen, and 41 observations gathered from each. Design was chosen as findings from Mollick (2013, p. 10)

identify it as being fairly uniformly geographically distributed across the USA, and so it would absorb any geographical variations in cost of capital. Technology was chosen as a second category in order to test internal validity in that a comparison between the two it might indicate by how much cost of capital may vary between categories or due to geographical differences (Technology being focused mostly in San Francisco and New York). The sample was restricted to the USA so as not to inadvertently pick up on differences caused by large geographical distances – such as differences in capital markets, etc.

The first stage of gathering data involved finding a full list of projects sorted by category on the website of Kickstarter. The names of the projects were then isolated from this list and randomised. Data was then gathered by working down the list, collecting information on: the date the project was funded; the estimated date that the product went to market; the total amount of funding raised; the planned/target amount of funding to be raised; the number of pledgers (α); the different perks and

13

their associated pledge amounts, as well as the price at which the product or products would later be sold.

Cost of capital was calculated for those products for which there was information about later prices available. Each perk’s respective cost of capital was weighted according to how much of the funding was raised by that perk. These weighted costs of capital were then summed to give the estimated cost of capital for the project. Information was not always available for every perk and so the cost of capital is very much an approximation, however, the tiers most frequently missing from the approximation are those which promise some small reward such as a bumper sticker or sometimes expensive, intangible rewards such as input in the design

process; both of which would not be sold as goods were the firm to raise funds via orthodox means and so do not necessarily represent foregone revenue, so the method is somewhat defensible. However, these rewards still carry costs for the firm, and so it would not be accurate to consider them as ‘free money’, therefore they were ignored from the calculation – in effect saying that these perks carried the same cost of capital as those for which data could be found. This was considered to be more valid than approximating costs for these tiers about which very little information was available, which would be done via what is essentially little more than guesswork. Sometimes, however, a simple ‘thank you’ perk was included, with which charitable backers could offer a few dollars (usually somewhere between one and five dollars) simply to support the project; this perk can in fact lead to very small negative values for cost of capital.

14

5 Results and analysis

Figure 3

The above diagram depicts a histogram and kernel density plot for the estimated pseudo-cost of capital of crowdfunding including the 5% Kickstarter fee. The mean, median and quartile values for both the cost of capital excluding the fee and the cost of capital including the fee are presented in the following table:

mean S.D. min 25% median 75% max

Cost of Capital without fee 33.17% 28.00% -0.27% 11.98% 25.00% 46.79% 130.80% Cost of Capital with Fee 40.17% 29.47% 4.98% 17.87% 31.58% 54.51% 151.66% Table 1 List of Variables

r pseudo-cost of capital without fee r_f pseudo-cost of capital with 5% fee alpha number of backers

15

Kbar weighted average Pledge per project Pbar weighted average Price per project

Technology a dummy variable equal to 1 if the firm is from the Technology subsample and 0 if from Design

Date the date at which the project was successfully funded TTD the planned time difference between successful funding and

delivery of the product to backers (Time To Delivery) targetmet percentage of funding that the firm received

Variable alpha Kbar Pbar Date TTD targetmet

Spearman's r 0.0858 0.0472 0.2138 -0.0154 -0.0145 -0.1047

Prob > |t| 0.4432 0.6737 0.0538 0.891 0.8974 0.349

5% sig level n=82

Table 2

Noting the ungainly distribution of the cost of capital, a Shapiro-Wilk test for

normality was run on the cost of capital without fee (if the cost of capital without fee is not normally distributed, then the cost of capital with the fee will not be either). As can be seen from the table, the null hypothesis of normality was firmly rejected, and with this in mind caution was taken to proceed with only non-parametric tests. It is for this reason of non-normality, and the large positive skew of the distribution, that the median is relied upon to give an indication of cost of capital both with fee and without. Note the large impact that the fee has on the cost of capital; increasing the median cost of capital by 6.58 percentage points.

To determine relationships between the data, Spearman’s correlation coefficient was used as it conformed to the requirement of a non-parametric test. Table 2 above shows the correlations between cost of capital without fee and each of the potential explanatory variables. The Spearman’s correlation coefficient here is being used to test for a relationship between the variables, with a P-value of less than 5% taken to mean a rejection of the null hypothesis of independence and an adoption of the alternative hypothesis of interdependence between the variables.

As can be seen, not one of the variables shows a statistically significant relationship with the unadjusted cost of capital, however an honourable mention should be given to the average price (Pbar), which in a larger sample may show some relationship with the cost of capital. What is worth noting is that average pledge (Kbar) is very far removed from showing any relationship with cost of capital. It

16

seems thus that the main driving force of the cost of capital is the final price that is being charged. What makes this finding especially enigmatic is when one considers the relationship between average price and average pledge:

Spearman’s r Pbar

Kbar 0.9778

Prob > |t| 0.0000

n=82

Table 3

In which there is a clear interdependence. Why should it be that price and pledge are clearly interdependent with each other, yet only price may explain cost of capital? To explain this the distributions of both the average price and average pledge were investigated. It was found that the average pledge was more closely clustered than the average price, which is to be expected as the price represents a markup on top of the pledge. This markup causes a positive stretching or distortion, which in turn causes an increase in the cost of capital via the numerator P – K; it is in fact this stretching that defines the cost of capital. It is important to note that this relationship between P and the cost of capital would be present in any analysis, as the cost of capital is specified as the very magnitude of the difference between P and K relative to K. Thus, what remains to be understood from this is simply that the average pledge is independent from the cost of capital. Higher pledges amount to a lower diversification benefit as more money is being entrusted to any one creator and thus, theoretically, a higher required return on risk, or in non-equity terms, a better deal on the backer’s

investment by way of a greater discount versus the final price. However, this is not supported by the data and thus it seems that project creators do not set out to

compensate higher pledges with a better deal, and that it is the eventual price at which the product is sold, rather than foresight, that drives cost of capital. Even then, this relationship is statistically insignificant and even were it significant, it would

presumably be very weak. There must therefore be additional, unexplained drivers of the cost of capital or motivations for the use of crowdfunding.

17

This graph uses log-transformations of both Kbar and Pbar to make the stretching easier to see given highly skewed data. The use of a log-transformation does not affect Spearman’s correlation coefficient.

Figure 4

Before continuing with that line of thought, however, we will return to Table 2 to discuss the reasons for investigating each of these variables, and briefly discuss what the empirical results say for the cost of capital.

A relationship between cost of capital and number of backers (α) was

investigated as, in line with Mollick’s (2013, p. 12) finding that an increased number of backers lead to an increased delay in product delivery, it was thought that an increased number of backers may also lead to an increase in other costs, such as employing extra staff to help with increased demand or having to renegotiate production or transportation contracts, and so the subsequent price of the product would have to be revised upwards, thus raising the cost of capital. Inversely, an increased number of backers may also allow the creator to benefit from economies of scale in the production process, thus lowering cost of capital. Whilst both of these possibilities may well present themselves in individual cases, such as in the case of the project Hanfree in which, after seeing how much money the creator had raised on the website Kickstarter, suppliers began to hike up their prices thus making the project unfeasible (Markowitz, via Agrawal, Catalini & Goldfarb, 2013, p. 17) they

18

do not appear to be systematically present, as the relationship between number of backers and cost of capital is weak and independence is not disproven.

The correlation with targetmet had a similar aim, assuming that projects may appropriately estimate the number of backers they expect to receive, targetmet instead looks at the relationship between cost of capital and the amount by which the project received more funding (thus more interest) than it expected. Here too, however, there appeared to be no systematic relation.

The correlation against TTD, or Time To Delivery, was intended to pick up on any relationship between the stated time that it would take to complete the project and the cost of capital. The shorter the TTD the closer the project may be to completion, thus the closer P(success) is to one. A shorter TTD would thus correspond with a pledge closer to the final price. No relationship was found. A less crude analysis might include a regression with dummy variables for credibility of prototypes, credibility of TTD claims, etc.

Next, the relationship with Date was tested. The idea here was to see if there has been learning in the market overall, if the overoptimism as suggested by Agrawal, Catalini and Goldfarb (2013, p. 6) has since been taken into account and pledge rates lowered for a given price, equivalent to an increase in the cost of capital. There appears, however, to be no relationship of this kind over time. The failure to reject independence is in fact rather interesting as it casts some doubt on Agrawal, Catalini and Goldfarb’s (2013, p. 6) claim of over-optimism – especially when considered alongside such phenomena as herding as signaling and due diligence by the crowd, which are inconsistent with over-optimism. Although it may also be the case that there has been very limited learning over time.

The final test to be discussed is the difference in cost of capital between technology and design, to check for internal validity of the model. The test statistic of note is the continuity corrected Pearson Chi-squared. Our chosen p-value is 5% and so if the p-value is below 5%, then the medians of the two samples are said to differ. This is not what we find, which is encouraging as it suggests that the findings might be valid for a wider environment (i.e. the other categories on Kickstarter).

19

Greater than the median Technology 0 1 Total no 19 22 41 yes 22 19 41 Total 41 41 82 Pearson chi2(1) = 0.439 Pr = 0.508 Continuity corrected: Pearson chi2(1) = 0.1951 Pr = 0.659 5% sig level Table 4

Descriptive statistics for the two sub-samples:

_{mean S.D. min 25% median 75% max}

CoC without

fee, Technology 29.52% 27.96% -0.21% 10.83% 24.38% 45.03% 139.08% CoC without

fee, Design 36.82% 27.89% -0.27% 15.06% 31.85% 58.34% 117.47%

Whilst the median in the Design subsample is higher than that in the Technology subsample, the above test shows this difference not to be significant given the spread of the data.

Table 5

6 Conclusion

We began the paper by asking what the cost of using crowdfunding may be. This question was approached by arriving at a specification for a pseudo-cost of capital based upon foregone revenue for each unit pre-sold.

From the data we have found that the average cost of capital without a fee was 25%, and the average cost of capital with a fee of 5% was 31.58%. This should be considered as a cost of capital over the life of the product, which if the product is sold over three years – as an example – would imply a median yearly cost of capital of approximately 10%, including the fee, and if sold over two years would imply a median yearly cost of capital of approximately 15%. This is lower than Boeker and Wiltbank’s estimate for angel investors’ rate of return of and annualised 27% (2007, p. 3), and may represent foregone advantages of employing angel investor expertise,

20

the ability of angel investors to pick especially profitable projects, or the differing incentives between project backers and angel investors – angel investors seeking profits, whilst project backers are simply seeking to derive utility from the perk they receive or even from participation in helping a project get under way. However, the highly skewed distribution of angel investor returns as shown by Boeker and Wiltbank (2007, p. 7) makes such a comparison difficult and, as suggested in the literature review, the extent of the advantage due, for example, to due diligence is questionable. Furthermore, one should not forget that the effect of the fee is significant in inflating this cost of capital – and a lower fee may attract even more projects towards crowdfunding platforms and lead to a more competitive market for financing

However, the evidence suggests that crowfunding is not used purely as a source of financing in the same vein as traditional means of financing. Furthermore, what is not yet determined is the value of premia associated with the use of a crowdfunding platform or the backing of an angel investor. As noted above by Boeker and

Wiltbank, it is often the case that an angel investor will offer their expertise or assistance in order to help the fledgling project (2007, p. 2). By choosing to use crowdfunding as a form of finance, the firm foregoes this opportunity. However, the use of a crowdfunding platform also brings with it certain boons: project creators are able to assess demand of the product; they are able to market the product via the crowdfunding platform; and so on.

To illustrate this point, one project creator, when asked for his motivations for using Kickstarter, responded:

There were several reasons for Kickstarter. Production funding is the obvious, however I would also throw in:

Product validation – Maybe my idea is just crazy and only appeals to me? Marketing – I now have customers in over 15 countries

Networking – I have met and continue [to be] in contact with several supporters that have seen my project.

Building a KNOWN customer base for future support and feedback. (Dan Alich, personal communication, June 10, 2014)

21

It may be that once the requirement of a return is removed, other considerations such as marketing, network building and even philanthropy become more important drivers of funding. This would also go some way to explaining those cases where the cost of capital appears to be equal to zero; cases where pledge was equal to price, which may be caused by backers who are not necessarily backing the project for any return other than the utility they expect to derive or creators who are looking to use crowdfunding as something more than simply a source of financing, for example as a tool to gauge demand or reach out to customers. The favourable cost of capital, as well as potential provided by crowdfunding platforms to reach out to customers or gauge demand may go some way to explaining the success of crowdfunding in recent years.

The research that was undertaken in this paper represents a proof of concept, a preliminary investigation into the cost of capital. The sample size was limited by time and technology constraints and the total costs of using crowdfunding were inferred from price information where available, which it often was not – from the approximately 200 firms manually investigated, around 120 were either projects which otherwise did not fit the specification of a for-profit firm, or projects for which product prices could no longer be found or where the project itself could not be found at all, seemingly never going beyond the crowdfunding stage. These projects were not investigated further as this would require qualitative research beyond the scope of this paper.

Theoretical issues with the model could be ironed out through discussion, and empirical problems could be addressed by a more thorough undertaking. One possibility for future research may be to employ the co-operation of project creators using crowdfunding and in doing so request full access to their cost information, production dates and number of units sold (at the price P after the good has been finally produced). Whilst being immensely time consuming and even more difficult to find a large sample size for, the quality of the data would be higher and so too the standard of the inferences. More thorough data would also enable us to investigate the market for funding as briefly introduced in section 3.3 in more detail.

22

References

Agrawal, A., Catalini, C., & Goldfarb, A. (2011). The Geography of Crowdfunding. The National Bureau of Economic Research. Retrieved from

http://www.nber.org/papers/w16820

Agrawal, A., Catalini, C., & Goldfarb, A. (2013). Some Simple Economics of Crowdfunding. The National Bureau of Economic Research.

Wiltbank, R., & Boeker, W. (2007). Returns to angel investors in groups.Available at

SSRN 1028592.

Financial Conduct Authority. (2014). Crowdfunding. Retrieved from http://www.fca.org.uk/consumers/financial-services-products/investments/types-of-investment/crowdfunding Indiegogo. (2014). Retrieved from

http://support.indiegogo.com/hc/en-us/articles/526406-When-Do-I-Get-My-Money-

Kickstarter. (2014a). Retrieved from https://www.kickstarter.com/1billion Kickstarter. (2014b). Retrieved from

https://www.kickstarter.com/projects/readingrai nbow/bring-reading-rainbow-back-for-every-child-everywh

Kickstarter. (2014c). Retrieved from

https://www.kickstarter.com/help/faq/kickstarter%20basics

Mollick, E. (2013). The Dynamics of Crowdfunding: An Exploratory Study. Journal of Business Venturing, 29(1), 1-16

Moules, J., (2013, November 29). Q&A: Crowdfunding. Financial Times. Retrieved from http://www.ft.com/cms/s/0/dd1a79ba-5828-11e3-a2ed

00144feabdc0.html#axzz2y8T pjk8I

Zhang, J., & Liu, P. (2012). Rational herding in microloan markets.Management

science, 58(5), 892-912.

Kuppuswamy, V., & Bayus, B. L. (2013). Crowdfunding creative ideas: The dynamics of project backers in Kickstarter. SSRN Electronic Journal.

23

Appendix

A.1

Histograms with kernel density diagrams for cost of capital with fee, design on the left, technology on the right. Populations were not found to differ (test given in article).

A.2

Scatter plot of Date against cost of capital with fee. Note shotgun-like pattern.

24 A.3

Histogram of average price (above) and average pledge (below)

A.4

A.5

Histogram of number of backers (α)

A.6

Alpha, K-bar and P-bar are neither normal nor lognormal at a 5% (and 1%) significance level (normality rejected).

Variable Obs W V z Prob>z

alpha 82 0.39542 44.87 8.346 0.00000 Kbar 82 0.40259 41.846 8.139 0.00000 Pbar 82 0.42379 40.362 8.114 0.00000 logalpha 82 0.95513 3.143 2.513 0.00599 logK 82 0.94045 4.171 3.134 0.00086 logP 82 0.93949 4.238 3.169 0.00077

25 A.7

Histogram of the percentage of funding that was achieved. Similar to findings from Mollick (2013, p. 7)

A.8

26 A.9

Variable Obs Mean Std. Dev. Min Max r 82 0.3317 0.2799 -0.0027 1.3908 r_fee 82 0.4017 0.2947 0.0498 1.5166 alpha 82 1500.42 3994.79 33.00 26457.00 Kbar 82 209.15 485.25 11.19 3350.00 Pbar 82 279.02 628.60 11.62 3995.00 targetmet 82 7.03 20.90 0.13 170.03 TTD 82 98.46 104.73 14.00 815.00 logalpha 82 6.10 1.36 3.50 10.18 logK 82 4.35 1.22 2.42 8.12 logP 82 4.61 1.24 2.45 8.29

27 An example of how information for a project was collated (A.10): Name Portable Dual Arduino (TM) Micro XPlorerBoard

URL https://www.kickstarter.com/projects/1576747460/portable-dual-arduino-micro-xplorerboard?ref=nav_search

Purchase URL http://www.richelectronics.com

Description Breadboard with Arduino incorporation capability

Type Arduino

Location Savannah, GA

Date funding ended 02-Apr-14 Date product started 30-Jun-14

Total raised 32920

Number pledgers 169

Pledge Pledge Amount Quantity Q needed Limit Price Weighted CoC Individual CoC

Thanks 10 9 9 100 0 -0.003882825 -1 Early Bird 129 62 62 100 169.95 0.109534492 0.31744186 XPlorerBoard 139 1 1 - 169.95 0.00133526 0.222661871 XPlorerBoard + name 149 7 7 100 169.95 0.006326848 0.140604027 XPlorerBoard + micros 189 22 22 - 219.95 0.029375728 0.163756614 XPlorerBoard + micros + name 199 49 49 100 219.95 0.044287933 0.105276382

Education starter kit 399 3 - - - 0 0

Family education kit 799 1 - - - 0 0

28

Medium classroom kit 4780 0 - - - 0 0

Large classroom kit 7050 1 - - - 0 0

School kit 9995 0 - - - 0 0 Number pledgers recorded: 155 150 Total: 18.70%

Note highlighted in the Individual CoC (Cost of Capital) column a decreasing Cost of Capital for an increasing Pledge. An avenue of research beyond the scope of this paper.

29 Data for input into the statistical program Stata (A.11):

Observation r r_fee Date alpha K-bar P-bar targetmet TTD Technology

1 0.935800 1.037684 04-Jul-2013 807 58.8904 114.0000 4.8281 88 0 2 0.183701 0.246001 26-Apr-2014 648 1040.1807 1231.2629 8.8861 34 0 3 0.675009 0.763168 11-Oct-2012 1110 1191.0382 1995.0000 4.6717 292 0 4 0.187266 0.249754 18-Nov-2013 131 35.4425 42.0796 1.0910 42 0 5 0.632449 0.718367 07-Dec-2012 135 49.0000 79.9900 1.4712 54 0 6 0.215856 0.279848 30-Jan-2014 150 12.3288 14.9900 2.2617 90 0 7 0.072213 0.128645 15-Dec-2012 1921 11.1918 12.0000 6.1222 46 0 8 0.084342 0.141413 17-Feb-2014 467 51.3208 55.6492 1.7837 102 0 9 0.421789 0.496620 16-Apr-2014 848 12.8470 18.2658 3.3098 75 0 10 0.594170 0.678074 12-Dec-2013 132 123.1905 196.3866 1.0359 108 0 11 0.139535 0.199510 29-Nov-2013 110 30.7143 35.0000 3.5032 152 0 12 0.666667 0.754386 23-Jul-2012 402 30.0000 50.0000 3.5091 99 0 13 0.522467 0.602597 07-Jul-2012 47 157.6389 240.0000 6.0100 54 0 14 0.691311 0.780327 07-Jul-2013 4849 13.5989 23.0000 0.1315 85 0 15 0.249750 0.315526 11-Mar-2012 344 40.0000 49.9900 1.9080 19 0 16 0.583360 0.666694 06-Mar-2014 525 132.6956 210.1048 2.3784 101 0 17 0.119737 0.178670 09-Nov-2013 288 77.2909 86.5455 1.6361 36 0 18 1.014463 1.120487 16-Mar-2014 327 16.8557 33.9552 3.3014 30 0 19 1.174703 1.289161 01-Apr-2012 2463 16.1837 35.1946 5.3745 14 0 20 0.160037 0.221092 25-May-2013 439 55.0359 63.8437 1.4577 82 0 21 0.169212 0.230750 09-Dec-2012 639 93.5442 109.3730 1.4752 96 0 22 0.592600 0.676421 22-Feb-2014 102 69.9067 111.3333 1.9194 82 0 23 0.319124 0.388551 23-Apr-2012 146 37.4627 49.4179 1.6203 83 0 24 0.377753 0.450266 12-Aug-2012 1400 139.0070 191.5173 2.2016 34 0 25 0.109091 0.167464 22-Feb-2014 223 55.0000 61.0000 1.5199 52 0

30 26 0.059190 0.114937 16-Dec-2012 42 17.8333 18.8889 3.3360 14 0 27 0.318456 0.387848 16-Jul-2012 459 70.4026 92.8228 1.1499 14 0 28 0.355294 0.426625 14-Apr-2014 81 332.0313 450.0000 2.4551 16 0 29 0.062308 0.118219 15-May-2014 507 101.2908 107.6020 2.5822 92 0 30 0.365301 0.437159 18-Mar-2012 210 31.5497 43.0749 3.0780 58 0 31 0.249409 0.315167 16-Jan-2012 274 16.9109 21.1286 1.1137 30 0 32 0.187613 0.250119 11-Apr-2014 851 24.9405 29.6197 4.8590 65 0 33 0.250306 0.316112 24-Aug-2013 8090 65.7365 82.1908 8.1911 22 0 34 -0.002693 0.049797 03-Dec-2013 454 35.4243 30.1974 3.4494 102 0 35 0.467857 0.545113 27-Feb-2014 33 21.5385 31.6154 1.0575 16 0 36 0.437750 0.513421 28-Oct-2013 955 203.3986 292.4364 5.0804 63 0 37 0.080071 0.136917 01-Oct-2012 214 44.5479 48.1149 1.0342 60 0 38 0.146030 0.206348 04-Nov-2013 340 29.7370 34.0795 7.2847 26 0 39 0.150591 0.211149 10-Apr-2014 182 138.5345 159.3966 2.2241 81 0 40 0.467471 0.544706 19-Nov-2012 249 115.1978 169.0495 2.5774 101 0 41 0.607215 0.691805 08-Jan-2014 2391 59.0884 76.0884 18.9846 112 0 42 0.445602 0.521687 27-May-2013 175 55.3333 79.9900 1.5511 49 1 43 0.136364 0.196172 28-Jun-2012 861 220.0000 250.0000 12.9436 220 1 44 0.010692 0.063887 19-May-2013 374 65.4930 66.1933 3.4408 119 1 45 0.293267 0.361333 05-Nov-2012 1089 150.0000 193.9900 170.0340 405 1 46 0.539382 0.620403 19-Apr-2014 96 815.1724 1254.8621 2.8259 41 1 47 0.067024 0.123183 13-Aug-2012 2655 48.2328 51.4655 4.5627 110 1 48 0.192537 0.255302 09-Sep-2013 127 3350.0000 3995.0000 1.0559 97 1 49 0.053093 0.108519 15-Jul-2013 250 251.6398 265.0000 2.9476 123 1 50 0.999600 1.104842 26-Sep-2011 176 25.0000 49.9900 1.0179 50 1 51 0.108318 0.166651 27-Mar-2014 139 28.7895 31.9079 1.0699 49 1 52 0.203482 0.266823 08-Oct-2012 125 335.0000 403.1664 1.1763 68 1

31 53 0.494324 0.572973 26-Aug-2013 697 93.8314 140.2145 2.3087 293 1 54 -0.000348 0.052265 14-Jan-2012 1367 27.4361 27.4266 4.0177 16 1 55 -0.002079 0.050443 03-Jan-2013 171 72.8788 72.7273 2.7772 27 1 56 0.368008 0.440009 15-Dec-2013 373 60.3605 82.5737 1.4411 182 1 57 0.502270 0.581337 20-Feb-2012 325 66.4286 99.7937 1.8998 39 1 58 0.418996 0.493680 09-Apr-2013 638 107.8665 153.0622 1.0175 235 1 59 0.000000 0.052632 19-May-2012 612 281.8103 281.8103 2.6349 103 1 60 0.485646 0.563838 27-Jun-2013 896 49.5461 73.6079 1.0843 64 1 61 0.450308 0.526640 09-Oct-2013 2310 41.3636 59.9900 1.3869 52 1 62 0.193009 0.255799 07-Jul-2013 1229 61.3899 73.2387 2.1557 815 1 63 0.454220 0.530758 16-Feb-2014 1398 317.5838 461.8369 3.4928 103 1 64 0.187500 0.250000 09-Jan-2014 148 36.3636 43.1818 2.1991 111 1 65 0.000000 0.052632 18-Nov-2013 1409 251.8507 251.8507 45.9478 132 1 66 0.319860 0.389326 25-Jul-2013 8200 18.9338 24.9900 16.7837 67 1 67 0.000000 0.052632 21-May-2013 76 11.6154 11.6154 3.0022 70 1 68 0.116721 0.175496 13-Jan-2013 1290 163.1691 182.2145 9.7903 168 1 69 0.102687 0.160723 01-Jun-2012 129 56.7048 62.5276 1.2596 59 1 70 0.589641 0.673307 19-Apr-2013 168 228.1818 362.7273 1.7610 102 1 71 0.249750 0.315526 25-Jan-2014 21412 40.0000 49.9900 24.6522 156 1 72 0.462376 0.539343 15-Nov-2013 72 1016.4894 1486.4894 2.1230 105 1 73 0.186977 0.249450 02-Apr-2014 169 154.5267 183.4197 1.2284 89 1 74 0.040290 0.095042 14-Jun-2013 145 91.2500 94.9265 1.6724 108 1 75 0.243809 0.309273 12-Jun-2013 441 84.9884 105.7093 1.0777 79 1 76 0.297430 0.365716 29-Sep-2012 154 2292.8571 2974.8214 1.2357 123 1 77 0.250362 0.316170 25-Mar-2013 26457 84.1149 105.1740 78.1378 219 1 78 0.637779 0.723978 01-Apr-2013 486 974.0000 1610.4885 1.4627 59 1 79 0.307716 0.376543 08-Jan-2014 502 38.2270 49.9900 1.0300 81 1

32

80 0.115481 0.174190 06-Jun-2013 184 35.8142 39.9500 3.3452 54 1

81 1.390773 1.516603 30-Apr-2013 10477 43.0935 103.0266 8.3447 153 1 82 0.188346 0.250890 04-May-2012 1047 144.8400 172.1200 2.5241 57 1