1. An investigation of Provision Points and Stretch goals
as a remedy to the Cold Start Problem
Thomas Minshull
Department of Economics
Dalhousie University
April 2014
2. Abstract
Crowd-funding has dramatically increased in popularity in recent years. During this surge in
popularity a new tactic, the use of stretch-goals, has been implemented by a small but notable
number of crowd-funding campaigns. Using empirical analysis of data collected from a prominent
crowd-funding website over a 3 month period, this paper will attempt to quantify the effect of this
new tactic, as well as the effect of an older tactic known in the economics literature as a provi-
sion point, on contributor’s behaviour. Specifically, we will investigate wither the implementing
stretch-goals, or a provision point increases the rate of contributions within the first 30 days of a
campaign. If affective this would help to elevate the “cold start problem” identified by Ward and
Ramachandran.
ii
3. 1 Introduction
Crowd-funding has dramatically increased in popularity in recent years. During this surge in pop-
ularity a new tactic, the use of stretch-goals, has been implemented by a small but notable number
of crowd-funding campaigns. Using empirical analysis of data collected from a prominent crowd-
funding website over a period of 3 months, this paper will attempt to show that the implementation
of this new tactic or the implementation of the more established tactic the use of provision points
may elevate the “cold start problem” identified by Ward and Ramachandran.
Crowd-fundings recent rise in popularity is undoubtedly, in part, due to the passing of the
Jumpstart Our Business Startups Act (JOBS Act) of 2012. The JOBS Act was signed into law in
the United States of America on April 5 2012. And, among other things the JOBS Act directed
the Security and Exchange Commission (SEC) to construct rules implementing an exemption for
equity based crowd-funding. Equity based crowd-funding is the use of crowd-funding as a means to
exchange securities. On October 23 2013 the SEC posted rules implementing this exemption. Prior
to this exemption the sale of securities to non-accredited investors via crowd-funding was illegal in
the United States.
Although these events have increased the popularity of crowd-funding, most crowd funding in
north america, to date, does not involve the transfer of equity. This paper’s scope is limited to
non-equity crowd-funding, which is the more established form of crowd-funding in North America.
Crowd-funding, for those unfamiliar with this phenomenon, is a type of crowd sourcing where
financial resources are contributed by a large group of dispersed individuals in order that some
pre-specified project can be carried out. In exchange for their contributions contributors to a
project (here forth referred to as contributors or funders) are offered pre specified material or non-
material rewards from the individual coordinating the project (here forth referred to as the creator).
The nature of the projects that are crowd-funded very greatly. Crowd-funding is commonly used
to finance the development of new products or artistic works, but is also used to raise funds
for political, charitable, and other causes (eg crowd funding has been used to finance academic
1
4. research). Consequently, the nature of the reward offered to funders also varies widely. Each
campaign offers a variety of rewards that funder may choose to receive if they contribute more
than the specified amount associated with each reward. Typically, the rewards associated with
lesser amounts comprise simply of tokens of appreciation for supporting the project, but campaigns
typically offer more substantive rewards if larger contributions are made. Projects that aim to
develop a new product, service, or artistic work will routinely offer this product, service, or artistic
work as a reward. In such cases, crowd-funding can be seen as a type of pre-sale mechanism.
Furthermore, slight variations of the product, service, or artistic work are routinely offered as
different rewards which are available at different contribution levels.
Crowd funding may or may not take place over the internet, and may or may not take place
using a website dedicated solely to hosting crowd-funding campaigns (here forth referred to as a
crowd funding platform); however, since most prominent crowd funding campaigns are hosted on
the internet, by crowding platforms this papers scope will be limited to such campaigns.
Provision points are a mechanism where the creator of the crowd-funding campaign public states
a goal, and commits to refunding all contributions made to the campaign if this goal is not met.
If a provision point is not implemented the creator still publicly states a goal that he/she wishes
to raise; however, in such a case the creator will keep any contributions made to the campaign,
wither or not the goal has been reached (minus the commission taken by the platform, which is
taken wither or not provision points are used).
Stretch-goals are a mechanism where the creator publicly states goals in addition to the initial
goal discussed above, and commits to improving the rewards received by the contributors in pre-
specified ways when the cumulative amount raised by the campaign reaches each additional goal.
Stretch-goals can be implemented in conjunction with provision points or they can be implemented
independently. No money is refunded if a stretch-goal is not met, unless they are implemented in
conjunction with provision points and the initial goal (the provision point) is not met. Stretch-
goals create an external benefit, that is captured by contributors when additional contributors
2
5. contribute to the campaign. This creates an interesting set of incentives that influence contributor
behaviour. In this paper we will be concerned with the affect that stretch-goals have on the timing
of contributions, but this is by no means the only, nor their most dominant influence of stretch
goals on contributor behaviour.
Prior literature, by Ward and Ramachandran, has shown that the likelihood that a potential
contributor will contribute to a crowd funding campaign increases as the cumulative amount raised
by that campaign increases. This result lead Ward and Ramachandran, to pose what they deemed
the “cold start problem”[Ward and Ramachandran, 2010]. The cold start problem is summarized
by the following question: how do crowd-funding campaigns raise the initial funds required to
increase the likelihood that potential contributors will contribute to there campaign? Increasing
the likelihood that potential contributors contribute to a campaign is important, because it, in
turn, increases the cumulative amount raised which further increases the likelihood that potential
contributors will contribute to the campaign, and this causes the cumulative amount raised to
increase at an exponential rate.
Due to the contemporary nature of this topic, the economic literature on crowd-funding is
sparse. To date empirical economic papers have yet to quantify how the cold start problem is
elevated, nor have empirical papers attempted to quantify the affect of provision points or Stretch-
goals. Using data collected from a predominant crowd-funding platform (indiegogo.com) over a
period of 3 months, this paper will attempt to answer the question: does the use of stretch goals
or provision points elevate the cold start problem? We find evidence that suggests that campaigns
that implement provision points or stretch goals have a greater cumulative funding growth rate than
campaigns that do not implement provision points or stretch goals. This suggests that campaigns
that implement these tactics are less inhibited by the cold start problem. Since this topic has, yet
to be addressed in the economic literature these results, as well as the orchestration of the data
collection process maybe viewed as my contribution to the field.
Before continuing I would like to take this opportunity to both recognize and thank Ian Whitman
3
6. for both collecting and parsing the data used in this paper; without Mr. Whitman’s technical
support this paper would not have been possible. When I began this project I was unaware of
the technical challenges that would need to be tackled in order for the project to be completed. I
would like to thank Mr. Whitman for providing the hardware required to collect this data, and
for constructing the data base used to store the data, and for the hours of coding that he put into
constructing the parser, and most importantly for lending his technical expertise to this project.
2 Background
As mentioned previously, technically crowd funding need not involve the use of the internet. It
may be useful to point out that readers of this paper are likely more familiar with crowd-funding
than they may realize. Anyone who has witnessed a NPR or PBS pledge drive has observed crowd-
funding in action. The creator, NPR or PBS, publicly reaches out to its audience, the potential
funders (here forth refereed to collectively as the crowd,) and asks for contributions so that the
creator may complete the project, the provision of unique content on their respective networks. NPR
and PBS also provide contributors with rewards for donating minimum amounts to the campaign.
In these examples NPR and PBS already have a means of accessing the crowd and have no need of
a crowd funding platform, and although these campaigns do not have a provision point mechanism,
nor stretch-goals they do meet the definition of a crowd-funding campaign and otherwise conduct
themselves in a manner very similar to a typical modern crowd-funding campaign.
In contrast to the example above contemporary Crowd-funding occurs almost exclusively online,
and there has been a proliferation of crowd-funding platforms as crowd-funding has increased
in popularity. In 2003 the crowd-funding platforms ArtistShare Created the internet’s first fan
funding platform for artists, and may very well have launched north america’s first crowd-funding
platform [ArtistShare, 2014]. Since 2003, several crowd funding platforms have emerged of these the
most well known are likely Sellaband (2006), IndieGoGo (2008), and Kickstarter (2009). Unlike
ArtisitShare and Sellaband, which primarily raised money for musicians who wished to record
4
7. new albums or go on tour, crowd-funding platforms launched since 2007 have tended to host
campaigns pursuing a much wider variety of projects. Crowd-funding is now commonly used to
raise funds for entrepreneurial ventures, charitable causes, political causes, and a wide variety of
artistic projects. Since 2008, $1,009,000,000 (USD) have been pledged by 5,755,966 funders on the
kickstarter platform alone [Kickstarter, 2014].
3 Literature Review
Stretch-goals, being a relatively new phenomenon, have yet to receive adequate attention from the
Economics community. The only exception to this seems to be Hardy’s paper How to perfectly
discriminate in a crowd? A theoretical model of crowd funding. Hardy explains how stretch goals
may theoretically be used to perfectly price discriminate. Hardy also makes the interesting point
that it may be rational for funders to increase their contributions when a stretch goal is reached
in order to obtain a higher priced reward that has become more appealing due to the increase
in quality that resulted from the products improvement when the stretch goal is reached [Hardy,
2013].
In contrast to Hardy contemporary work Bagnoli and Lipman predates the modern emergence
of crowd-funding by 13 years. Despite this Bagnoli and Lipman’s theoretical work very clearly
describes the provision point used by contemporary crowd funding campaigns today. Because of the
similarity between Bagnoli and Lipman’s description and the mechanic that is used by contemporary
crowd funding campaigns the term provision point is used in the economics literature to describe
what is refereed to by the public as“fixed funding.” It would be interesting to know if crowd funding
platforms discovered this mechanic through Bagnoli and Lipman’s work, or if the use of provision
points in crowd funding campaigns emerged independently.
Bagnoli and Lipman’s paper describe a mechanism for the private provision of public goods.
In the system they propose an impartial third party collects contributions from the public for
the provision of a public good, while providing a grantee that if a pre-specified amount deemed
5
8. large enough to ensure the provision of the public good is not raised all contributors will be fully
refunded. They conclude that such a scheme would be affective at reducing the free rider problem
as it relates to the provision of public goods [Bagnoli and Lipman, 1989]. Since contributors to the
public good are guaranteed a refund if the public good cannot be supplied, the contributors who’s
expected benefit from the public good out weigh the required contribution amount can rationally
choose to contribute, since they no longer have to be concerned that the lack of contributions from
potential free riders will result in their contribution returning a negative net benefit due to the
projects failure from under capitalization. Similarly, the use of provision points in contemporary
crowd-funding reduces the risk associated with under capitalization of a project.
In 2010 Ward and Ramachandran were able to show; using archival, seasonally corrected, click
through data from Sellaband.com (Aug 2006 — Feb 2010;) that a higher cumulative amount raised
increased the likelihood that a potential contributor would choose to contribute to a campaign.
They attributed this to the notion that contributors use information conveyed by the cumulative
amount raised to update their beliefs about the quality of the reward provided by the campaign
[Ward and Ramachandran, 2010]. When a contributor contributes to a campaign the cumulative
amount raised increases, this sends a signal to potential contributors that others believe the project
(or the reward) is of high quality, and this increase the likelihood that those potential contributors
will contribute to the campaign. This increased likelihood that potential contributors will contribute
to a campaign in turn further increases the cumulative amount raised. Having empirically observed
these peer effects, Ward and Ramachandran identify what they coined the cold start problem. As
described above the cold start problem asks the question how do campaigns gain the initial funding
needed to initiate this peer effect which drives exponential growth of the cumulative amount raised.
Ward and Ramachandran, point to the importance of initial contributions from friends and family.
Finally, Agrawal et al. provides an excellent survey of the economics of crowd-funding. It
touches on a wide variate of topics related to crowd funding ranging from the difference between
equity and non-equity crowd funding to the incentives faced by the various agents involved in crowd
6
9. funding. Unfortunately, this paper fails discuss the recent emergence of stretch-goals, but despite
this this survey paper by Ajay, Catalini, and Goldfarb’s serves as an excellent introduction to the
topic of crowd-funding
4 Conceptual Framework
The economic literature on crowd-funding has indicated that peer-effects, specifically the informa-
tion conveyed by a campaigns cumulative amount raised, increases the likelihood that potential
contributors will contribute to a specific campaign. This in turn increases the cumulative amount
raised for these campaigns which will in turn stimulate yet more peer-effects. This implies that
the cumulative amount raised to date increases at an exponential rate. However, this exponential
growth will not occur if an initial source of funding is not found.
I hypothesis that stretch-goals and provision points create a set of incentives that encourage
contributors to contribute to campaigns at an earlier stage.
To see why the implementation of stretch goals may incentives contributors to contribute to
campaigns at an earlier date we must make two assumptions. First that on some level, contributors
recognize that the peer-effects mentioned above exist, and that there contributions therefore will
entice more potential contributors to contribute to the campaign. And secondly, that Ward and
Ramachandran interpretation is correct and that potential contributors use information conveyed
by the total contribution to date to update their beliefs about the quality of the reward provided
by the campaign
If we know consider the case where stretch goals are not implemented, we see that the potential
contributor faces a time inconsistency problem. That is to say that at any given moment prior to the
campaigns termination date, the contributor can stand to benefit by delaying his/her contribution.
Since this delay will allow the potential contributor to gain more information about the quality of
the reward. Therefore it is only rational for contributors to contribute just prior to the termination
of the campaign (which is publicly available information).
7
10. But, if we consider the case where stretch goals are implemented and peer effects are recognized
the potential contributor faces a different set of incentives. The potential contributor still benefits
from the additional information gained by delaying the contribution, but this is off set, because by
delaying the contribution the potential contributor has reduced the period in which the peer effects
created by their contribution may entices others to contribute, which in turn would have increased
the external benefit they would have received, when each stretch goal was reached. Consequently,
when stretch goals are implemented an opportunity cost to delaying is introduced. Therefore
contributors are expected to contribute to a campaign at an earlier date.
The mechanism by which provision points encourage earlier contributions to crowd funding
campaigns is more straight forward. Because provision points are a form of guarantee that the
project will not be under capitalized they directly increase the expected value of an early stage
crowd funding campaign by reducing the expected costs associated with such a campaign. This
should increase the likelihood that contributors will contribute to such campaigns during this early
stage.
To empirically test wither the implementation of stretch goals or provision points will alleviate
the cold start problem I will utilized two sets of two simple linear regression model, with clustered
standard errors at the campaign level on data gathered during the first 30 days since a campaign
was launched.
ln(Amount Raised) = β0 + β1 Days + β2 Provision Point + β3( Days × Provision Point) + u (1)
The first model will regress the natural log of amount raised on Days, an indicator variable that
indicates if a provision point has been implemented or not, and on an interaction term for Days and
this provision point indicator variable. Days represents the number of days from the start of the
campaign. Days was calculated by subtracting the Start date of the campaign (which is provided
in the html code) from the date stamp provided by the data scraper at the time the observation
was taken.
8
11. This simple model will be used to determine if provision points are indeed effective at increasing
the amount raised by crowd funding campaigns within the early stage of the campaign. The base
case is the case were no provision point was implemented. β0 indicates the initial funding that the
average campaign without a provision point receives, to find the average amount such a campaign
raises simple raise e to the power β0. The coefficient β2 indicates the initial funding that the
average campaign with a provision point receives, in addition to the amount accounted for by
β0. If β1 (the percentage increase in the cumulative amount raised per day for campaigns when
provision points are not implemented) is greater than, or statistically indistinguishable from β3
(the percentage increase in the cumulative amount raised per day for campaigns where provision
points are implemented) at a given significance level we will be unable to reject the null hypothesis
that provision points do not elevate the cold start problem.
ln(Amount Raised) = β0 + β1 Days + β2 Stretch Goal + β3( Days × Stretch Goal) + u (2)
The second model is identical to the first model except every instance of the indicator variable
indicating if a provision point was implemented is replaced by an indicator variable indicating if
stretch goals were implemented. The base case is the case where no stretch goals were implemented.
β0 and β2 are the intercept terms that convey the amount of initial funding campaigns without
and with stretch goals receive, respectively. β1 is the percent change in the cumulative amount
raised per day for campaigns without stretch goals, and β3 represents the percent change in the
cumulative amount raised per day for campaigns with stretch goals. If β1 is greater than, or
statistically indistinguishable from β3 at a given significance level we will not be be able to reject
the null hypothesis that stretch goals do not elevate the cold start problem.
Finally, these models will be rerun, but campaigns unable to raise 5% of their goal by the
end of the campaign will be dropped from the analysis. The justification for this is as follows.
Creating a low quality crowd funding campaign is relatively costless, consequently a great number
of campaigns exist that appear to be created by creators on the off chance that they may be
9
12. successful. By dropping campaigns that are unable to raise 5% of their goal, we are hopping to
introduce a minimum quality standard that will reduce the noise induced by such campaigns.
5 Sources of Evidence
Data has been collected from the crowd funding platform indiegogo.com in accordance to their
terms and conditions on November 15 of 2013. My college Ian Whitman, and myself began this
process in November of last year. To gather this data we created a web scraping bot using a java
library jsoup. This bot navigated the indiegogo.com website and collected each link that appeared
in the “new this week” section of the indiegogo.com website. New links were appended to this list
on a near daily basis. The bot would then load each of these links and copy the html script for each
campaign page, and store it on our database. These pages were not scrapped in parallel, and a
delay was added between each scrap to insure that we were in compliance with the sites terms and
conditions. Care had to be taken to insure that this process would not interfere with indiegogo’s
daily operations.
In February of 2014 we completed the parsing program. This parsing program systematically
searched the html scripts for a wide variety of information that we were interested in collecting. The
majority of this information could be identified in the html code using the html tags and attributes.
The one notable exception is the dummy variable that indicated weather or not a campaign utilizes
stretch goals. Unfortunately because stretch goals are not part of the standard template used by
indiegogo no html attributes exist that coould be associated with the implementation of stretch
goals. To create this dummy variable we parsed all the text that appears on the campaigns webpage,
and we set the variable to 1 if the terms “stretch,” “stretch goal,” “stretchgoal” or some variant
of these terms appears within this text. This may have lead us to over count the use of stretch
goals in these campaigns. As a consequence of this, additional noise may have been added and our
results may under represent the affect of stretch goals.
As of February 26 2014, we are collecting 18 variable for each campaign. These variables
10
13. are, the campaign Name, the link to the campaign’s webpage, the date and time of the html
code was collected, the number of days remaining, the cumulative amount raised at the time
of the observation, the campaign’s goal, the currency the campaigns goal and contributions are
denominated in, the type of funding (wither or not provision points were implemented), the number
of contributors, the number of updates made by the creators, the number of comments made by the
contributors, the category (indiegogo.com list 25 separate categories, all but “Verified Non-profit
were represented in our data set), the location, the campaigns launch date, the campaigns end date,
an indicator variables that indicate if the goal had been reached at the time the observation was
taken, and the dummy variable indicating if stretch goals were implemented in the campaign.
From November 14 2013 to February 26 2014 we were able to collect data on 69,873 different
crowd funding campaigns. Five different currencies AUD, USD, EUR, GBP, CAD were represented.
However, in order to simplify our analysis we considered only the subset of campaigns that were
delineated in USD. This reduced our data set to 24,192 campaigns. The goals for these campaigns
ranged from $500 to $2 billion, the median goal is $6000 and the mean goal is $58,900. By our
measure of the 24192 campaigns only 768 (3.17%) of campaigns implemented stretch-goals, and
1,224 (5.06%) implemented provision points. Recall that the actual number of campaigns that
implement stretch goals is fewer than 768 due to the over counting discussed above.
It is also important to note that the other major crowd funding platform in north america
kickstarter.com mandates that all crowd funding campaigns hosted on their site implement provision
points. This is likely creating a selection bias that causes creators that do not wish to implement
provision points to be over represented in our data. This impart explains why we observe so few
campaigns implementing provision points. This could be a cause for concern if there was reason to
believe that the implementation of provision points was correlated with other variable, besides the
amount raised.
This data is structured as unbalanced panel data, where each panel contains observations on a
given campaign. The data contains 134,035 distinct observations. We attempted to take observa-
11
14. tions daily, but due to logistical problems, equipment breakdowns, and other technical difficulties
there were several occasions were the collection of data was not conducted according to schedule.
6 Analysis
To determine if the implementation of provision points elevate the cold start problem we regress
the cumulative amount raised on the number or days since the campaign is launched, a dummy
variable that takes on the value one if a provision point mechanism was implemented and zero
otherwise, and the term representing the interaction between this dummy variable and the number
of days since the campaign was launched, using data from the first 30 days of the campaigns. A
similar model is used to test if the implementations of stretch goals elevates the cold start problem,
but in this case the dummy variable indicating if stretch goals were implemented was used.
After running these simple models we found the residuals were positively related to the number
of days from start, and the residuals were distributed in a heteroskedastic manner, as shown in
Figure 1. Furthermore the magnitude of many of the positive residuals was clearly larger than the
magnitude of any of the negative residuals indicating that the residuals were skewed.
Applying a logarithmic transform to the cumulative amount raised reduces the complicating
factors observed in Figure 1. As show in Figure 2, once the logarithmic transform is applied the
residual are less skewed, the relationship between the residuals and the number of days from start
is reduced, and the residuals appear to be homoskedastic.
Consequentially, in our regression models we used the logarithmic transform of the cumulative
amount raised as our dependent variable. These results are summarized in Table 1.
In the first regression the coefficients associated with the base case, and the coefficient associated
with the initial funding level when provision points were implemented were found to be statistically
significant at the 0.001 significance level. The magnitude of the coefficient associated with the initial
funding for the base case where provision points were not implemented indicates that the average
campaign without a provision point starts with $149.61 of initial funding. The coefficient associated
12
15. (a) (b)
Figure 1: (a)Residual plot for the regression plotting Residuals on Days From Start of Campaign for,
(Amount Raised) = β0+β1 Days+β2 Provision Point+β3( Days× Provision Point)+u. (b) Residual
plot for the regression plotting Residuals on Days From Start of Campaign for, (Amount Raised) =
β0 + β1 Days + β2 Stretch Goal + β3( Days × Stretch Goal) + u
(a) (b)
Figure 2: (a) Residual plot for the regression plotting Residuals on Days From Start of Campaign
for, ln(Amount Raised) = β0 + β1 Days + β2 Provision Point + β3( Days × Provision Point) + u.
(b) Residual plot for the regression plotting Residuals on Days From Start of Campaign for,
ln(Amount Raised) = β0 + β1 Days + β2 Stretch Goal + β3( Days × Stretch Goal) + u
13
16. with the initial funding level of campaigns with provision points indicates that this increases by
32% or by $47.87 when provision points are used. The magnitude of the coefficient for the number
of days remaining indicates that the cumulative amount raised increases on average by 3.09% per
day. These results are summarized in the Provision Points column in Table 1. We cannot reject
the null hypothesis that provision points do not elevate the cold start problem since the remaining
coefficient was not statistically significant.
As the Stretch Goals column in Table 1 shows, all the coefficients in the second model are
statistically significant at the 0.01 significance level. The base case is the case where no stretch
goals have been implemented. The coefficient associated with the intercept for this base case
indicates that the average base case campaign has $147.23 dollars in initial contributions. The
coefficient for Days from start indicates that his increases by 3.03% per day. The coefficient for the
stretch goal dummy variable indicates that campaigns that implement stretch goals start with an
initial funding level that is 77.9% higher than those that do not implement stretch goals, and the
coefficient for the interaction term for this dummy variable and the number of days since the start
of the campaign indicates that the cumulative amount raised by campaigns that implement stretch
goals increases by an additional 0.988% per day.
When the first model is rerun but with the campaigns that are unable to reach 5% of their
goal by the end of their campaign dropped from the data set we find that all the terms that
were statistically significant are still significant at the 0.001 significance level, but in addition to
this the coefficient for the interaction between the implementation of provision points and the
number of days remaining is now significant at the 0.05 significance level, as noted in the Provision
Points column in Table 2. The magnitudes for these coefficients can be interpreted as follows.
The intercept term for the base case where provision points are not implemented indicates that the
average campaign without a provision point mechanism now starts with $315.45, and the coefficient
for the indicator variable indicates that campaign with provision points start on average with an
additional $174.44. The coefficient for the number of days since the beginning of the campaign
14
17. Amount Raised in first 30 days of campaign
Provision Points Stretch Goals
Days From Start Of Campaign
0.0309∗∗∗ 0.0303∗∗∗
(0.000647) (0.000667)
Provision Point Implemented
0.320∗∗∗
(0.0476)
PP time
-0.00166
(0.00285)
Stretch Goals Implemented
0.779∗∗∗
(0.0529)
SG time
0.00988∗∗
(0.00316)
Constant
5.008∗∗∗ 4.992∗∗∗
(0.0112) (0.0111)
Observations 134035 134035
Adjusted R2 0.017 0.025
Standard errors in parentheses
∗
p < 0.05,∗∗
p < 0.01,∗∗∗
p < 0.001
Table 1: The Provision Points column contains the Coefficients and the associated standard errors
for the regression model Equation 1 regressed on the all observations denominated in USD. The
Stretch Goals column contains the Coefficients and the associated standard errors for the regression
model Equation 2 regressed on the all observations denominated in USD.
15
18. Amount Raised in first 30 days of campaign that raised more than 5% of goal
Provision Points Stretch Goals
Days From Start Of Campaign
0.0407∗∗∗ 0.0467∗∗∗
(0.000673) (0.000672)
Provision Point Implemented
0.553∗∗∗
(0.0521)
PP time
0.00634
(0.00310)
Stretch Goals Implemented
0.660∗∗∗
(0.0517)
SG time
0.00725∗∗
(0.00302)
Constant
5.754∗∗∗ 5.750∗∗∗
(0.0113) (0.0113)
Observations 74045 74045
Adjusted R2 0.073 0.077
Standard errors in parentheses
∗
p < 0.05,∗∗
p < 0.01,∗∗∗
p < 0.001
Table 2: The Provision Points column contains the Coefficients and the associated standard errors
for the regression model Equation 1 regressed on the all observations denominated in USD, and
where the campaign was able to obtain more than 5% of its stated goal by the conclusion of the
campaign. The Stretch Goals column contains the Coefficients and the associated standard errors
for the regression model Equation 2 regressed on the all observations denominated in USD, and
where the campaign was able to obtain more than 5% of its stated goal by the conclusion of the
campaign.
16
19. indicates that the cumulative amount raised by campaigns that do not implement provision points
increases by 4.7% per day, while the coefficient for the interaction between the number of days since
the beginning of the campaign and the indicator variable indicates the cumulative amount raised
by campaigns with provision points increase by an additional 0.634% per day.
Finally when the second model is rerun on the data where the campaigns that are unable to
reach 5% of their goal by the end of the campaign are dropped we see that all the coefficients, which
are listed in the Stretch Goals column in Table 2 are statistically significant at the 0.05 significance
level. The initial amount raised by campaigns that did not implement stretch goals has increased to
$314.19 from $147.23. Campaign that do implement stretch goals start on average with 66% more
initial funding than campaigns without stretch goals. The cumulative amount raised for campaigns
without stretch goals increases at an average rate of 4.67% per day, while the cumulative amount
raised for campaigns with stretch goals on average increases by an additional 0.725% per day.
In the final three regression statistically significant coefficients were observed that would allow
us to reject the claim that provision points or stretch goals do not elevate the cold start problem.
And in some sense this is correct, campaign with provision points or stretch goals appear to be
attracting contribution at a faster rate, and are suffering form the cold start problem to a lesser
extent than campaigns that do not implement these tactics. However, every regression model we
have run in this paper has indicated that campaigns that implement these tactics start with a
greater amount of initial funding. Since we know from previous literature that increased levels of
funding induce peer effects that increase the rate of contribution, we cannot eliminate the possibility
that it is this initial disparity in initial funding levels that is causing the difference in the rate of
growth of the cumulative amount raised when provision points or stretch goals are implemented.
7 Conclusion
We were able to find evidence that campaigns that implement stretch goals, or provision points
are less inhibited by the cold start problem, though we were not able to determine if this was
17
20. primarily due to the implementation of these techniques or the increased initial funding levels that
was associated with the implementation of these techniques. As more data is collected we hope to
be able to include a term that accounts for the implementation of both provision points and stretch
goals. We did not test for this in this paper because our data set contained so few campaigns that
met this criteria that the results would not have been statistically significant. However, as more
data is gathered there is no reason why this should not be included in future analysis.
When drawing conclusions from this study one must keep in mind that external validity is a
concern since all of the data was collected for a single crowd funding platform. Unfortunately
we were unable to collect data from the other major north american non-equity crowd funding
platform Kickstarter.com, since the use of bots is prohibited by their website’s terms and conditions.
Concerns regarding external validity could be accounted for if permission to gather data from
kickstarter.com was obtained.
Due to inexperience and time constraints I have only scratched the surface of what I believe
to be a valuable data set. The current data set only consist of data from a three month period;
however, moving forward we hope to continue to collect data from indiegogo.com. It is our hope
that we will be able to compile a larger, more representative data set that includes observation from
different points through out the year. This data set could be used to address a variety of question
about non-equity crowd funding such as, what factors make a campaign successful, or what factors
are associated with greater initial funding levels.
Although our results thus far have been some what inconclusive, we are confident that this data
source will, in time, yield important results.
18
21. References
Ajay K Agrawal, Christian Catalini, and Avi Goldfarb. Some simple economics of crowdfunding.
Technical report, National Bureau of Economic Research, 2013.
ArtistShare. About us, 2014. URL https://www.artistshare.com/v4/About.
Mark Bagnoli and Barton L Lipman. Provision of public goods: Fully implementing the core
through private contributions. The Review of Economic Studies, 56(4):583–601, 1989.
Wojciech Hardy. How to perfectly discriminate in a crowd? a theoretical model of crowdfunding.
Technical report, 2013.
Kickstarter. Kickstarter stats, 2014. URL https://www.kickstarter.com/help/stats?ref=footer.
Chris Ward and Vandana Ramachandran. Crowdfunding the next hit: Microfunding online ex-
perience goods. In Workshop on Computational Social Science and the Wisdom of Crowds at
NIPS2010, 2010.
19