Sample
We collected data from FundedByMe, the largest crowdfunding platform in Sweden (Ingram and Teigland, 2013).Footnote 3 Our sample includes all investors in 31 campaigns,Footnote 4 which include successful equity crowd-funding campaigns posted on FundedByMe from the start of equity crowdfunding by FundedByMe in 2012 to the end of March 2015. We collected firm information about the campaign from the FundedByMe website. We were able to extract the name of investors and the time of investment from the activity log of campaigns. We excluded investments made from team members who enjoy private information (we know their names from the campaign) to keep a consistent sample in line with our focal investigation. The final sample includes 2537 investments by 1979 unique investors.
We took several steps to code genders based on first names. We first algorithmically used the API of genderize.io (a similar procedure used in Greenberg and Mollick 2014) by providing several country and language inputs such as Swedish, German, and Finnish. The algorithm returns the gender and a probability that a specific name-gender attribution (male or female) was correct; in the case it cannot decide, the algorithm returns none. In a second step, one of the authors speaking the Swedish language double-checked the accuracy of the codes and completed the missing variables, with additional help from the profile picture of the users, LinkedIn and Google Search (mostly in ambiguous cases such as unisex names). Most of the investors used their real name instead of pseudonyms as FundedByMe encourages this practice (in our sample, only approximately 3% of investors used pseudonyms).
Analysis of gender-related risk taking
Dependent variables
There are several observed characteristics of firms that we hypothesize to show the risk profile of future cash flows. First, younger firms are riskier because nascent firms suffer from liabilities of newness and smallness (Stinchcombe 1965). They have short track records and have had less time to accumulate tangible resources, which increases the risk of investment. Firm age is the numbers of years since the firm’s establishment. On average, investors invest in firms that are 2.6 years old.
Second, the technology category is another proxy for risk. Technology firms are involved in developing and commercializing innovative projects with a high uncertainty in outcome (Hall and Lerner 2012). Technology firm is a dummy variable denoting 1 for firms operating in the technology category such as mobile apps. Technology firms comprise 60% of investments and 58% of the total firms.
The third variable of risk is the Equity offering (%) of the campaign. Leland and Pyle (1977) suggest how firms opportunistically choose to raise equity when managers know that their shares are overvalued, and investors perceive equity offerings as a negative signal, taking this former information into consideration. Likewise, an owner’s decision to offer a lower amount of equity can indicate less adverse selection risk in that a bad outcome is less likely to be perceived by owners (Ahlers et al. 2015). Furthermore, greater equity offerings can dilute entrepreneurs’ incentive to commit to their firms. Altogether, more Equity offering suggests that a firm is riskier. On average, investments are in firms that offer 12.5% of their equity.
Independent variable
Female is a dummy variable equal to one for female investors and zero for male investors. Female investors commit about 20% of investments.
Control variables
Several variables might influence the investment decisions of investors. We control for the number of prior investors (No. prior investors) and the frequency of investments in prior days (Investment rate). Number of prior investors is on average 85 (with a maximum of 365 investors). Investment frequency is calculated as the number of prior investors divided by days that have passed since the start of the campaign. These values show the traction of a campaign and how successful it has been so far. The campaigns receive on average 0.8 investments per day (with a maximum of 5.5). The percentage of days passed (Share of days passed) presents the number of days passed since the launch of the campaign over its planned duration. The passage of time can provide more information, such as the percentage of funding and a forward-looking estimation on whether the campaign will reach its funding goal. The investment takes place on average after 37.5% of the campaign duration has passed. There are also time-invariant control variables. We group a number of variables that (a) are correlated positively with the unobservable direct quality of the firm and (b) are difficult and costlier for a low-quality type of firm to imitate compared to a high-quality firm. External certificate is a dummy variable and takes value one if a firm has filed for patent, received governmental seed investment, or introduced a lead investor (VC or angel); otherwise, it is zero. Patent shows the capability of R&D staff and technical capabilities of the firm and is viewed as a valuable positive signal decreasing information asymmetry for investors (Hsu and Ziedonis 2013). Affiliation with prestigious external stakeholders (e.g., (reputable) VCs) increases the legitimacy of the new firm, as new firms can borrow the reputation and legitimacy of those firms (Stuart et al. 1999). On average, 52.7% of investments are in firms with external certificates. Furthermore, we include the natural logarithm of valuation in Swedish Krona (SEK). The valuation of firms varies between 2.1 and 69.9 million SEK, with a mean value of 26.9 million SEK.Footnote 5 Given that entrepreneurs have a richer endowment of social capital from their home country (Dahl and Sorenson 2012), we control for location, which is set to one if the firm is located in Sweden and otherwise is set to zero. As female investors might be more likely to invest in projects with a higher share of female team members, we insert a dummy Woman on Team that is set to one if there is at least one female member on the project team and otherwise is set to zero. Finally, first-time investors on the platform might be systematically different from investors with a prior history of investment using this platform in that these investors might include friends and family. Therefore, Investment experience takes a value of one for investors with a prior history of funding a firm on the platform and otherwise is zero for first-time investors. Investments made by repeated investors comprise 13% of investments. To capture possible temporal trends, we insert week-day (6 dummies) and year fixed-effect (2 dummies) in all models.
Model specification
We use ordinary least square (OLS) regression when dependent variable is Firm age and Equity offering and we employ logit regression when dependent variable is Technology firm. Specifically, we use the following specification:
$$ Y={\beta}_0+{\beta}_{1.} Female+\alpha\ Controls+\varepsilon $$
(1)
Y defines separately the following risk proxies: Firm age, Technology firm, and Equity offering. The standard errors are robust and clustered around investors to control for non-independence of observations for investors across firms.
Analysis
Table 1 reports the descriptive statistics of variables in addition to correlation matrix, and Table 2 presents the results of regressions models.
Table 1 Summary statistics and pairwise correlation (N = 2537)
Table 2 Regression analysis
We perform formal tests of variance inflated factor (VIF) and conditional index (Belsley et al. 1980), and these tests do not suggest severe issues of multi-collinearity.
Model I presents the estimates of OLS model predicting Firm age. The coefficient of Female is positive and statistically significant (p < 0.01), suggesting that female investors are more likely to invest in older firms. In terms of magnitude, female investors, compared with male investors, are associated with firms that are on average less than one year older (0.76). Model II presents the coefficients of logit regression predicting Technology firm. Female investors are 35.7% less likely to invest in firms categorized as technology intensive (p < 0.01). Finally, model III shows estimates of the OLS model predicting Equity offering. Female investors are less likely to invest in firms with higher equity offering (p < 0.05). The coefficient implies that female investors, compared with male investors, are associated with firms that on average offer −0.48% less equity. Altogether, these results provide supporting evidence in favor of H1.
We also employ Tobit specification on model I and model III because the dependent variables are non-negative. The results remain unchanged.
Herding and gender in equity crowdfunding
Dependent variable
Number of female (male) investors in each day is the count of the current incremental number of female (male) investors. These variables are logged.
Independent variable
Female share of prior investors is the proportion of investors who are female to the total count of all investors until the previous day of the campaign.
Control variables
We include time-varying control variables such as No. of prior investors and Frequency of prior investors. The number of prior investors is the total count of prior investors until the previous day, and the Frequency of prior investors represents the total number of prior investors divided by the number of days that have passed until the previous day. We also included week-day fixed effects. Table 3 reports descriptive statistics, including the mean and standard deviation in addition to the correlation of variables.
Table 3 Summary statistics and pairwise correlation (N = 1639)
Model specification
We identify herding using the following specification. y
jt
represents the incremental number of female investors at each day t investing in firm j (for brevity of argument, we focus only on female investors as a dependent variable in the following description). Let Y
j, t − 1 be the lagged total proportion of female investors and X
jt
be other observable time-varying attributes related to the funding progress of firm j.
$$ {y}_{jt}=\alpha {Y}_{j,t-1}+{\beta}_1{X}_{jt}+{\beta}_2{Z}_j+{u}_j+{v}_{jt} $$
(2)
It is unlikely that we will capture every source of heterogeneity across firms given our available data. For instance, firms could have products that appeal to female investors, such as fashionable women’s clothes, which might attract female investors, yet our data does not include a variable denoting the product-category variable. Therefore, the unobserved firm attributes represented as u
j
could consist of, for instance, fashionable clothes for women. As a result, u
j
might be correlated with both the proportion of female investors (i.e., Y
j,t − 1) and the current incremental female investors attracted (i.e., y
jt
). This would cause endogeneity problems in estimating the effect of Y
j,t − 1 on y
jt
(i.e., coefficient of α in the Eq. 2). Therefore, we need to control for unobserved firm heterogeneity with firm-fixed effects to capture the unobserved correlation of preferences among female investors facing the same firm. We assume that u
j
is time invariant because firm attributes are unlikely to radically vary from the launch of the campaign to the end of the campaign. Given the strict multicollinearity between observable time-invariant firm attributes Z
j
with unobserved firm-fixed effect u
j
, the effect of Z
j
cannot be separately estimated. We argued that given website design features of FundedByMe, such as featuring the well-funded firms and the inclusion of these firms in the newsletter emails subscribed by investors, it is likely that well-funded firms become salient to subsequent investors and give rise to irrational herding. By drawing on cross-sectional variation in the publicly observable firm attributes, we can distinguish whether investors are replicating others’ decisions and ignore how others have arrived at such decisions. Consistent with Zhang and Liu’s (2012) operationalization of this idea, we include the interaction term between lagged proportion of female investors and publicly observable firm attributes (Z
j
) such as External certificate.
$$ {y}_{jt}=\alpha {Y}_{j,t-1}+{\beta}_1{X}_{jt}+{\beta}_2{Z}_j+{\beta}_3{Y}_{j,t-1}{Z}_j+{u}_j+{v}_{jt} $$
(3)
As a result of rational observational learning, a subsequent female investor would make more positive incremental quality inference after observing the momentum associated with higher-proportion of male investors (i.e., male-based herding momentum) about a firm without external certificate. We expect the male-based herding momentum will be accentuated by unfavorable firm characteristics and dampened by favorable firm characteristics in case of rational herding. Thus, the moderating effect of male-based herding on External certificate is relevant in determining whether female investors are rational observational learners or not. As such, in Eq. 3, if β
3 has the same (opposite) sign as external certificate effect (i.e., positive) for subsequent female investors, subsequent female investors are irrational observational learners of proportion of male (female) investors.
Analysis
We present descriptive statistics in Table 3 and the results from fixed-effect regression analysis in Table 4. Models I and II are models with the dependent variable set to the number of female investors, and models III and IV are related to the number of male investors as dependent variables.
Table 4 Firm fixed-effect panel data regression
There are two noteworthy results in models I and II, which predict the number of female investors in a given day. First, the coefficient of Female share of prior investors is negative (p < 0.1) in model I. Second, the interaction term between share of Female share of prior investors and External certificate is negative (p < 0.01) in model II (in this model, External certificate is dropped due to the strict collinearity with project fixed effects). Combined, these results suggest that women are less (more) likely to follow women (men) and that this effect is stronger when there is an external certificate of the project. The amplified effect of gender-related herding in the presence of favorable characteristics indicate the women are not assigning gender-related herding momentum to the quality of the project and are thus ignoring the reasons behind male investors’ decisions.
We perform the same set of analyses in models III and IV, which predict the number of male investors in a given day. The coefficient of Share of prior female investors is positive but not significant at conventional significance levels. In model IV, we also do not find a moderating effect of External certificate for Share of female prior investors. Overall, these results suggest differing patterns of gender-related herding for male investors compared with female investors.