Entrepreneurial overconfidence is usually emphasized as a possible cause of misallocation of resources into projects with little expected value. From this point of view, if overconfidence was predominant among entrepreneurs, public policies should try to correct its causes and consequences, rather than promote start- ups affected by this bias. Understanding the incidence and causes of overconfidence among entrepreneurs is, thus, crucial from a policy perspective. This study explores the existence and characteristics of overconfidence among entrepreneurs that participated in Buenos Aires Impended (BAE), a local government program that aims to promote innovative start-ups. Each year, the program evaluates, scores, and publishes the ranking of all applicants and selects those above a threshold as beneficiaries. This ranking is common knowledge to all applicants. We base our study on a survey distributed in 2012 to entrepreneurs that applied for a BAE grant from 2008 to 2011. In particular, in a first set of questions, participants were asked to compare themselves with others in the same round of the program based on two dimensions: their ability as entrepreneurs and the viability of their projects. In a second set of questions, participants were asked about their past performance and expectations for the future compared with the average of the same reference group. Finally, we explore whether information about the performance of a typical entrepreneur affects the perceptions of respondents. Specifically, the sample was divided survival rate and growth rate of net income). We compare the answers to the second set of questions of both groups to analyze the effect of this information. The answers to our survey show strong signs of overconfidence. In the first set of questions, about 80% of respondents place themselves above the median and more than 60% of the answers are within the first quintile. In the second set of questions, we find that past performance self- assessment is centered on the mean, whereas future expectations are strongly biased toward optimism. Additionally, we find an effect of the information provided to a random group of entrepreneurs. The group with information tends to be less optimistic than the group with no information. This effect is strong and significant for those that rank themselves within the first 20%. Entrepreneurial activity has been an important field for analyzing overconfidence and its effects. Nevertheless, empirical evidence is frequently derived from experimental settings rather than actual entrepreneurs [1]. We thus contribute to the literature by providing new evidence of overconfidence from actual entrepreneurs. Importantly, we exploit the fact that our respondents have a clear reference group to which we can refer for comparison purposes. In lab experiments, the comparison group is often un- clear or unknown to subjects [2]. In our case, we ask our entrepreneurs to rank themselves compared with other entrepreneurs of the same round of the program, a reference group that is clear and known to our subjects. An additional strength of our data is that entrepreneurs not only have  a clear reference group but also that they receive a strong signal of their relative ranking in that group. This information on the ranking of the program is clearly published and, given that selection is according to this ranking, is crucial to our entrepreneurs. The finding that entrepreneurs provide overconfident answers is even more remarkable given the information they have about the reference group and their relative performance within that group. To our knowledge, no studies have thus far analyzed the overconfidence of actual entrepreneurs with such a clear reference group that holds previous information about their relative position with respect to the other members of the reference group. Several studies have documented the fact that entrepreneurs tend to be overoptimistic when appraising their prospects for success [3-7]. These works typically compare the expectations (of survival, for example) of a sample of entrepreneurs with actual previous performance in that dimension (e. g., the failure rates of new businesses) from a population. We differ from this literature in that we concentrate on over placement within a well-defined reference group. The paper by shows how entrepreneurial optimism persists to bad signals [8]. The paper examines a sample of inventors that pay a fee and fill a form to get their ideas evaluated by a non-profit organization. A main finding is that about half of inventors continued with their projects after receiving bad evaluations, including those that received a clear advice to terminate their projects. In our paper we show persistent overconfidence among entrepreneurs, who over place themselves within a well-defined group even after receiving a clear signal of their actual ranking. To interpret our findings, we consider a set of possible explanations for overconfidence. For example, emphasize that above average responses such as those typically observed in experiments can be rationalized within Bayesian updating under certain distributions of types and information [9-10]. They derive the conditions for the structure of responses to ranking and scale questionnaires that will fail to be rationalized. In ranking questions, people are asked to rank themselves on a particular attribute within a given population, with their answers laying in the percentile in which they consider themselves to be. They conclude that 50% of the population ranking themselves above the median is evidence of “apparent overconfidence”; for “true overconfidence,” 50% of the population must rank themselves above the first quartile (more generally, 2*x% of people rank themselves above the top x% of the population distribution). Our results satisfy this more strict definition of “true overconfidence”. This is remarkable, because most of the literature identify “above average” responses, but do not satisfy the criteria of “true overconfidence”. In the scale questions, participants are asked to use a scale to compare themselves with the average of a population on a given attribute. Show that if the “population average” is interpreted as the mean and people self- evaluate using the mean of their beliefs, then for the answers to be rationalized, they should average to the population mean. 2 In our data, we find the noteworthy result that respondents’ answers average to the population mean when asked about past performance, whereas the results show overconfidence when asked about future performance. As we discuss at the end of the paper, existent theories of overconfidence fail to explain the structure of the responses shown in our data. In particular, it is difficult to square the following two observations: (i) “true” overconfidence in the self-assessment of respondents’ own abilities when information about ranking within the reference group is common knowledge and (ii) biased beliefs about future prospects along with centered self-assessments about past performance. For this reason, we propose a conceptual framework, consistent with all the evidence provided by the survey and that provides a credible explanation of entrepreneurial overconfidence. 3 In this framework, entrepreneurs use attribution to explain away bad signals, boost their confidence and motivation in order to exert more effort. In this way, they use enhanced confidence as a mechanism to counteract the tendency to procrastinate. Our framework is linked to who show that motivation could be driving overconfidence if beliefs about abilities affect effort [11]. Our findings are important for policymakers that wish to boost entrepreneurial activity. Overconfidence and over optimism within entrepreneurs have cast doubt about the convenience of increasing entrepreneurial activity through policies because many start-ups have lower chances of success than those predicted by entrepreneurs [12]. However, this is not necessarily the case if overconfidence is a means to increase self-motivation. In particular, using our framework we discuss the welfare cones quinces of overconfidence. The remainder of this paper is organized as follows. In Section 2 we discuss our frame- work. In Section 3, we describe the BAE program and characteristics of our survey. In Section 4, we present the answers to the ranking and scale questions. We also analyze their correlation with the individual characteristics of the respondents and discuss the effects of information on the structure of those answers. In section 5 we discuss the different explanations of entrepreneurial overconfidence and we compare them to our framework. Finally, we conclude in Section 6.

Consider an entrepreneur uncertain about his or her own ability and uncertain about the outcome of a task. These types of uncertainties (about one’s own abilities, about project viability, and about the state of the world in the future) are important for any entrepreneur that is willing to invest in a project. We consider an initial period in which information about ability arrives and is processed to form beliefs. The agent then takes into account these beliefs to choose an effort level. Finally, all uncertainty is resolved and payoffs are received. These depend on the realized state of the world, action taken, and ability of the agent. It is important to emphasize that these are sequential decisions. First, the entrepreneur receives a set of signals such as the imperfect evaluation of the chances of the success of the start-up. The entrepreneur then builds his or her beliefs about these chances. Taking beliefs as the input and considering the costs, the entrepreneur decides whether to engage in the activity. Once this sequence is explicit, it is clear that through these beliefs the individual can manipulate his or her own actions in the future. In this sense, this is a game between the “current self” (Self 0) and “future self” (Self 1). Thus, in this framework, an individual can self-deceive if the gains of doing so are higher than the costs. The main reason why an individual (Self 0) May want to distort his or her own (Selfs) beliefs away from what objective information indicates is increasing effort. Consider that Self 1’s decision is affected by a lack of self-control or procrastination (implemented as hyperbolic discounting in a formal model). Then, Self 0 may want to bias Self 1’s beliefs about the return to effort to balance such a self-control problem. This bias in optimism about one’s own ability implies that actions could sometimes imply committing too much effort and is thus costly. However, when the self-control problem is a sufficient concern, then this bias can be advantageous. We consider that entrepreneurs can use self-serving attribution to process information and form beliefs. Individuals can explain away bad signals by attributing bad outcomes to bad luck instead of low ability. They can also attribute good outcomes to high ability instead of good luck. Thus, we can think of individuals placing less weight on bad past signals, such as a low program ranking and poor past performance, to maintain their self- confidence and motivation. At the moment of processing information, the individual faces a basic tradeoff. On the one hand, the individual must avoid ignoring information because it provides important insight about the likelihood of future success. On the other hand, the agent wants to maintain motivation and wants to avoid his or her tendency to procrastinate. Thus, the individual must weigh the value of motivation with the value of information. We think of agents as performing optimal self- deceiving strategies to determine the attribution process and the degree of self-deception. Importantly, the individual could receive multiple signals of different quality. The lower the quality, the noisier the signal is and the lower their value of information. The larger the number of signals, the larger the value of information. In other words, quantity can compensate poor quality of signals.

In what follows we describe the implications of our framework that can be tested in our data. If information is noisy and self- motivation is important, a “better than average” effect would be present in entrepreneurial beliefs, and “true overconfidence” in the sense of can arise. We prove this point in Appendix B.1 and B.3. We show evidence on this implication in section 4.1. Agents can have an unbiased (centered) belief about signals and at the same time persist in overly favorable self-assessments of attributes or future outcomes. We analyze this point more formally in Appendix B.5. We show evidence on this implication in section 4.2. Agents that present better self-assessment would then exert higher levels of effort. We discuss this point in Appendix B.6. This implication is related to the evidence provided in section 4.1.3. Even having concrete information about the relative performance within a reference group does not eliminate overconfidence among that group. In our framework, information asymmetry is not the driver of overconfidence, as in [13-14]. We show this in Appendix B.2. We present evidence on this issue in sections 4.1.1 and 4.1.2. More information tends to reduce overconfidence and in the limit, with the arrival of a higher number of signals, overconfidence recedes. We prove this observation within our model in Appendix B.4. This implication is consistent with the fact that the newer waves of entrepreneurs, with fewer signals about their projects, present higher overconfidence as discussed in sections 4.1.1 and 4.1.

The BAE Program

Our data come from a survey of applicants to the BAE government program that aims to promote the creation and development of innovative firms. Each year, the program receives about 100 applications. It then evaluates and scores each of these applications under two dimensions. First, project viability is appraised, including the project’s planning, consistency, estimations, and projections. A maximum of 40 points is assigned to each project in this dimension. Second, entrepreneurial ability is evaluated through different methods including an in-depth interview with the applicant analyzing past experience, leadership, commitment, project knowledge, and entrepreneurial attitude. A maximum of 60 points is assigned to each project in this dimension. The sum of both dimensions then determines the score for the project. Applications with more than 55 points are selected into the beneficiary group and are provided with a monetary transfer (non-refundable) at the beginning of the project. Those applicants also receive technical assistance or tutorship during which entrepreneurs obtain advice and consulting from experts from specialized organizations for six to 12 months. This selection process provides substantial information to the entrepreneur about the other participants, which is an important feature for the purpose of this study. First, each application implies a non-negligible cost for the entrepreneur. Thus, it is plausible that each potential participant would try to assess his or her chances of success in BAE’s selection process by analyzing the characteristics of the beneficiaries in previous rounds, which are publicly available. 4 Second, during the evaluation process and interviews, evaluator’s explicitly inform the entrepreneur about his or her strengths and weaknesses. Third, the tutorship provides insights into the relative quality of the project. Finally, the list of beneficiaries is published by the government each year, along with the total score and a ranking of merit based on this score. 5 Thus, beneficiaries observe their exact position in the ranking. While non-beneficiaries do not have precise information about their total score, they are aware that they remained below the threshold of 55 points (and below the median, given that more than 50% of applications become beneficiaries). We exploit this selection process and ask respondents to compare themselves with those that participated in the same round of the program. Thus, the reference group is clear in our sample and respondents have information about it as well as about their relative position in this group according to BAE. Overall, the programs between 2008 and 2011 received 412 proposals and selected 233 projects to become beneficiaries. We restrict the analysis to these rounds.

Survey of the entrepreneurs

This study is based on the survey distributed in 2012 to the more than 400 entrepreneurs that participated in the program from 2008 to 2011 regardless of whether they were beneficiaries. The survey gathers information about the project (industry, initial capital), characteristics of entrepreneurs (age, education), and outcomes such as survival, sales, profits, and employment. Self- assessment questions, both the ranking and the scale questions, are included. Subjects were informed that the survey was not conducted by the program but rather for research purposes and that their responses would remain confidential. The survey was answered by 108 firms/entrepreneurs, with 68 beneficiaries and 40 non-beneficiaries (Table 1). Table 2 shows the descriptive statistics of the sample. Entrepreneurs have a mean age of 36 years, 70% of them are men, 42% are college graduates, and 40% have a master’s degree. Projects are diverse but one-third of them are in IT-related technologies, while 16% are in manufacturing. There is no significant difference between beneficiaries and non-beneficiaries. Of the surveyed entrepreneurs, 57% had previous experience in start-ups and 33% consider that at least one of these experiences were failures. A total of 85% of the sample began with their projects, while 15% finally did not implement their ideas. Of the created projects, almost 90% were active at the time of the survey. Firms are usually small and begin with about three workers, but they tend to increase in size: after four years, their mean size is about 10 workers. The program aims to foster individual entrepreneurs, but it is not uncommon to find more than one entrepreneur. 

Table 1: Beneficiaries and non-beneficiaries in the sample.

Table 2: Descriptive statistics of the surveyed entrepreneurs.

An important issue is whether respondents comprise a particular selection. We find that the data are consistent with a random sample of entrepreneurs that participated in the program. To show this, we generated a variable that represents the ranking that BAE awarded to each project within the participants of each round of the program. This variable was rescaled to represent a uniform distribution between 1 (the highest score of the round) to 100 (the lowest score). The variable was generated for all projects presented to BAE (including those respondents that did not answer our survey) in each year as well as for an indicator of overall assessment and on each of the dimensions: (i) ability as an entrepreneur and (ii) viability of the project. We only have information on both these components of the score for 2010 and 2011 We call this variable “BAE’s ranking.” We also computed the overall ranking (i.e., the overall score is the sum of the two dimensions) for all four rounds of the program. Figure 1 shows the cumulative distribution function of BAE’s ranking for our sample. The graphs show the proportion of observations classified below a given BAE’s ranking. For the universe, this graph would be a 45-degree line as in a uniform distribution. Our sample is the same as that distribution. Indeed, Panel (c) shows that, if anything, the sample is biased against the best projects: the proportion of respondents ranked by BAE in the first four deciles is about 30%. Nevertheless, the difference is not substantial. We also analyze the balance of the other variables in the sample compared with the universe of applicants. We find no significant difference in the industry or size of the project between those who answered the survey and those who did not. Moreover, all these variables combined do not explain the probability of responding to the survey (a probity model of the probability of responding on these variables generates a pseudo R2 of less than 0.03 and a likelihood ratio test comparing the model without explanatory variables provides a p-value of 0.25). Finally, as we show below, we find that overconfidence is stronger in those years in which the selection is less problematic, which also suggests that selection is not driving our results.

Figure 1: Empirical cumulative distribution function of BAE’s ranking in the survey.

Perceptions of the ranking questions

Entrepreneurs are asked to rank themselves relative to all the entrepreneurs of the round in which they participated. The first question is related to their skills as an entrepreneur and the second refers to the viability of the project. The first question is in your opinion, how do you rank yourself compared with the other participants of Buenos Aires Impended in that year (refers to the year in which the entrepreneur submitted his or her business plan to BAE) in terms of your ABILITY AS AN EN- TREPRENEUR? (Ranking from 1 to 100) To answer this question, assume that there were 100 submissions to BAE and rank yourself among those 100 entrepreneurs. As an example, choosing 1 would mean you consider yourself to be the most skilled entrepreneur among the 100 entrepreneurs. The question about the viability of the project is identical: In your opinion, how do you rank yourself compared with the other participants of Buenos Aires Impended in that year in terms of the VIABILITY OF YOUR PROJECT? (Ranking from 1 to 100) To answer this question, assume that there were 100 submissions to BAE and rank yourself among those 100 entrepreneurs. As an example, choosing 1 would mean your project was the best among the 100 submitted projects. First, we describe the existence of overconfidence by analyzing the answers to these questions. We interpret the answers based on the median of the expectations of respondents about their relative skills. We first observe that 79.4% of entrepreneurs rank themselves above the median (their answer was a number below 50) in ability and 78.6% above the median in project viability. This excessively positive self-evaluation is concentrated at the top: 52% of entrepreneurs rank themselves in the first quintile in terms of ability as an entrepreneur and 55% do so in terms of project viability. Figure 2 shows histograms of the distribution of the answers to these questions, illustrating that most entrepreneurs rank themselves in the first 20 positions. These results are in line with the literature in that more than 50% of answers are concentrated above the median. However, in our data, the effect is strong, evidencing “true overconfidence” in the sense of to show this, Figure 3 presents the proportion of entrepreneurs that consider themselves to be above a particular percentile for the two questions (ability and viability). In each plot, waded wore ferrous lines: (i) a 45-degree line that we label “apparent overconfidence” and (ii) a line that doubles that slope to represent the condition of “true overconfidence” of If the distribution of answers is above the 45-degree line, this means that people tend to over place themselves (i.e., more than x% of respondents rank themselves in the top x% of the population distribution).

Figure 2: Distribution of the answers to the ranking questions.

Figure 3: Empirical cumulative distribution function of the answers to the ranking questions.

If the distribution of answers is above the “true overconfidence” line, this means that 2*x% of people rank themselves above the top x% of the population distribution. Thus, our data provide evidence of “true overconfidence”. In particular, more than 50% of entrepreneurs rank them in the first 20% of the ranking. This structure of responses cannot be rationalized using Bayesian arguments. This result is strong: as highlighted by data rarely show an effect as skewed to meet the strictest definition of overconfidence. When we restrict the sample to those respondents of the 2011 round (who compare themselves with the same group), this structure of responses is even more biased toward the higher positions in the ranking. For example, more than 85% rank themselves above the median and more than 60% rank themselves above the first quintile (see Table 3 and Figure 6 in the Appendix) [15]. When we further restrict the sample to non-beneficiaries (i.e., those below the median of BAE’s ranking), we find that more than 70%rank themselves above the median and about 50% rank themselves above the first quintile.

Our results are even more puzzling considering that all respondents have direct information about their relative performance as a  result of the  selection process by BAE. In particular, non-beneficiaries know that they performed worse than more than half of the applicants. We emphasize this fact in Section 5.

Table 3: Answers to the ranking questions.

Self-assessment and program ranking

We now relate the subjective self-assessment of entrepreneurs to BAE’s ranking for several reasons. First, we wish to analyze the extent to which a respondent may over place him- or herself compared with BAE’s ranking. Second, given that BAE’s ranking is available to respondents, the relationship between these two variables can be seen as an indicator of the extent to which this information is relevant to respondents’ beliefs. Additionally, we compare the results of the different rounds of BAE to analyze whether over placement tends to recede over time, exploiting the fact that newer rounds of the program capture younger firms, which have less information about their projects. We first restrict this analysis to the 2010 and 2011 rounds given the lack of a score for each dimension for the other years. Panels (a) and (b) of Figure 4 show the scatterplots for each year.

Table 4: Over placement results.

First, there is no strong correlation between program evaluation and entrepreneur self-assessment. Although the fitted line is upward sloping, indicating a relationship between both variables, the regression coefficients are relatively low. Second, the regression lines lie below the 45-degree line and fitted values are always less than 50, indicating a clear above average effect. This is true even for those that in the ranking of BAE are well below the median. Third, when comparing the fitted lines for 2010 and 2011, it is apparent that the latter is always below the mean and that it has a steeper slope.

Figure 4: Scatterplot and fitted lines: Self-assessment vs. project evaluation.

To include the observations of the 2008 and 2009 rounds and compare the self-assessment of the entrepreneur with the  ranking by the program, we construct two overall self-assessment rankings. First, we consider the worst of both answers as an overall ranking. Second, we make a predicted overall ranking as a weighted average of the ability and viability rankings, where the weights are 0.7 and 0.3, respectively. These weights arise from the regression of the overall ranking of BAE on both individual rankings (ability and viability) by BAE (see Section A in the Appendix for a detailed description). Panel (c) of Figure 4 presents the scatterplot for our second definition of overall self- assessment (for the first definition, the graph is almost identical). The analysis of these variables generates similar results as before: there is little correlation between self-assessment and BAE’s ranking and the fitted values are below 50. Again, the slope of 2011 is higher than those of previous years.

The over placed

We identify those entrepreneurs that rank themselves above the position in BAE’s ranking by an identification variable labeled “over placed.” We then apply this definition based on the different alternatives with respect to ability, project viability, and overall self-assessment. This is a binary indicator. To characterize the width of the gap between self-assessment and BAE’s ranking, we define the degree of over placement as the difference between self- assessment and BAE’s ranking. Finally, we also identify those respondents that rank them- selves in the top 20^th percentile in both dimensions by constructing a binary variable called the “first quintile.” Table 4 shows that 71% of entrepreneurs are over placed in ability, while 68% are over- placed in project viability. In other words, a large majority of entrepreneurs self-assess better than the program. Furthermore, about 70% of respondents are over placed in the overall self- assessment (68% according to the first definition and 74% according to the second definition). Finally, 43% of respondents rank themselves within the first quintile in both ability and viability (at the same time). We use this variable below to divide the sample into two groups. When we analyze the degree of over placement by using the difference between the positions in the rankings, we find that respondents rank themselves about 20 positions above the program’s ranking (22 positions in the ability dimension, 20 positions in the project viability dimension, 15 positions when considering the first definition of overall self- assessment, and 20 positions when considering the “predicted” overall self-assessment). These definitions are useful to show some of the correlations with the characteristics of respondents and thus explore whether this overly positive self-evaluation is correlated with some such characteristics. First, we regress the degree of over placement on gender, age, years of education, and year of round. We also include two potentially relevant variables: whether the entrepreneur had previous entrepreneurial experience and whether he or she considered previous start-ups to be a failure. Second, we use a logit model to analyze the factors that increase the probability of being over placed. Finally, we also include a regression of overall self-assessment on BAE’s ranking and other characteristics. Table 5 presents all these regressions. Few variables are statistically significant in these regressions and overall R2 is below 0.20 for all the OLS regressions. For example, gender and age do not seem to be relevant for over placement. Years of education, on the contrary, seem to reduce over placement, although the effect is marginally significant. In particular, one additional year of education is related to three positions below in the difference between self- reported and BAE’s ranking (see the first column of Table 5). Notably, past experience of previous start-ups seems to be uncorrelated with self-assessment when analyzing both the degree of over placement and the ranking. Nevertheless, past failures seem to be marginally correlated with more positive self- assessment: the coefficients are negative in the regressions (only being marginally significant in the first column) and positive in the logit estimation. Respondents of the 2011 round tend to rank themselves about 10 positions above comparable respondents of other rounds. Finally, the regression of overall self-assessment on BAE’s ranking is significantly different from zero, being above 0.30 in both definitions. While this coefficient only reports the correlation between ranking and beliefs, it is also consistent with the idea that the ranking of BAE serves as a signal that affects beliefs.

Table 5:  Regression results.

Hours worked and overconfidence

Table 6 presents the distribution of workers according to the number of weekly hours of work. We present the results for those in the first quintile of the self-reported ranking and for the remaining observations. We find that hours worked are significantly higher for those that consider themselves to be high in the ranking. For example, 50% of those with high self- assessment work more than 51 hours, while this proportion is 25% for the remaining observations. We also regress the number of hours worked on the self-reported ranking and find that climbing 10 positions in the ranking is correlated with working 5% longer hours. This correlation is significant even when controlling for the ranking of BAE. We interpret these results as a correlation between an indicator of effort and self-confidence: those that rank themselves higher also tend to work longer hours.

Scale questions: Past performance and future expectations

Two other survey questions inquire about (i) the past performance of the firm and (ii) the expectations about future performance. These questions are of the scale type, meaning that respondents must compare their beliefs with the population average. The tables 7 and 8 show the proportion of responses as well as the mean of answers by considering a simple numerical scale from -2 (far below average) to +2 (far above average). The first is a retrospective question that compares the average of entrepreneurs: How was the performance of your business (compared with the average of the other start-ups presented in Buenos Aires Impended in that year [refers to the year in which the entrepreneur submitted his or her business plan to BAE])? Exceptional (far above average) Very strong (above average) Average Weak (below average) Very weak (far below average) the evaluation of performance from a retrospective point of view does not show a particularly benevolent self-assessment (see Table 7). Specifically, 47% of respondents evaluated that the performance was about average, 24% evaluated themselves above average, and 28% below average.

Table 6: Hours worked and self-assessment.

When we compare beneficiaries with non-beneficiaries, we find that 36% of non-beneficiaries considered that their performance was below average, while about 20% answered that their performance was above average. Among the beneficiaries, the answers were similar: 25% answered that they performed below average, while 26%thoughtthat they performed above average. When we restrict the analysis to those that rank themselves in the top 20^th percentile in both dimensions (the “first quintile” sample as defined in Section 4.1.2), the evaluations are more positive: while 14% answered that they performed below average, 43% of respondents considered they did better than average. The two- sample tests show that these responses are significantly different from those that are not in that sample. We also show the self- evaluation of entrepreneurs of the 2011 round. In this group, about 57% considered that they were about average, with 24% above average, and 20% below average. The last rows of Table 7 show the mean of the answers when we apply the scale centered on the mean (-2 for “far below average,” -1 for “below average,” 0 for “about average,” +1 for “above average,” +2 for “far above average”).

Table 7: Self-evaluation of past performance.

Table 8: Self-evaluation of potential growth.

We also present the standard deviation and 95% confidence interval for this statistic. As Table 7 shows, the averages of the answers of all groups are around zero and the confidenc

We have shown that there is high and persistent overconfidence among the surveyed entrepreneurs, that overconfidence is higher among entrepreneurs in the last round of BAE, and that higher self-confidence is correlated with longer hours worked. We have also pointed out that this structure of responses is even more striking given that all respondents have common knowledge about their relative position in a ranking published by the program. Furthermore, we noted that answers about future expected outcomes are strongly biased to- ward optimism, while assessment about past performance is centered on the mean. Finally, we have shown that beliefs respond to new information about the performance of the average entrepreneur, but in a mild way: those that received new information tend to show slightly less overconfidence about future expected outcomes compared with the remaining sample. In this section, we discuss the extent to which these results are novel and consistent with different theories of overconfidence. First, it is worth noting that entrepreneurs do not have any incentives to exaggerate their self-assessment. These entrepreneurs have already been rated by BAE and many of them are already beneficiaries; therefore, there is no gain for them in lying about their beliefs, as this would not raise their likelihood of receiving future transfers or assistance from the program. Importantly, if that was the objective, the consistent way of lying would be to exaggerate their past good performance given that BAE has no independent information on actual sales, profits, or other firm outcomes. However, we learned that past self-assessment was strongly centered on the mean, exhibiting no indication of overconfidence. This argument also leads us to discard the “social signaling” argument for overconfidence for our data [16]. According to that explanation, individuals like to over place themselves as a signal to others because they derive utility from the perceptions of others about their own abilities (e.g., the positive opinions of others could improve the chances of financing their own firm). To be explicit, that theory does not depend on overconfident beliefs but on overconfident answers: even respondents that know that they are unskilled can answer falsely that they place themselves in the top decile just to provide a signal to others. Within this view, anyone that wants to affect others’ beliefs should provide a strong and consistent signal of above average performance. For example, when asked repeatedly about different but correlated desirable attributes, they should consistently place themselves above their own beliefs. In our data, this is not the case. First, 72% of those over placed (as defined in Section 4.1.2) answered that their past performance was about average or below average and more than 50% of those in the “first quintile” sample self-evaluated their past performance as about average or below average, as shown in Table 7. Second, 32% of entrepreneurs that considered that their past performance was below average expected above average potential growth (see Table 9). Overall, these characteristics of the data do not seem compatible with social signaling objectives. A second novel aspect of our data is that they cannot be explained by the lack of information about the reference group. Indeed, individuals do have information on the previous rounds of the program, particularly beneficiaries, through the various brochures that BAE publishes each year. Furthermore, the selection process provides entrepreneurs with relevant information about their abilities and characteristics relative to a benchmark based on the feedback received in the tutorship and evaluation by BAE. Finally, the ranking by BAE after the evaluation process is common knowledge. Thus, theories that explain overconfidence by emphasizing information asymmetry are less relevant for our data. For example, the “reference group neglect” theory of Camerer and Lovelies based on the observation that an individual could fail to internalize the fact that the reference group is not a random sample of the universe and for that reason could be biased in his or her relative perception with respect to their ferrous group, even while his or her absolute perception could be correct. The “differential regression” theory proposed by Moore and Cain builds on the observation that self- information is more useful and precise for estimating one’s own performance, while the estimates of others tend to be closer to a general prior, which could generate an above average effect in easy tasks. These two explanations of overconfidence are inconsistent with precise information about the relative ranking of individuals. Moreover, our data are also at odds with the view that individuals would bias their information by searching for signals that could improve their self-confidence [17-18]. Our point is that it is not an issue of information availability (e.g., the ranking is common knowledge), but about how this information is processed. Additionally, we have shown that new information can change self-assessment: from Section 4.3, we learned that entrepreneurs (especially the over placed or those in the first quintile) do update their answers after providing them with new information, but their update does not seem to be too responsive. Finally, one of the most important features of our data is that the self-assessment of past performance is unbiased, while average future expected outcomes are significantly above the mean. In other words, it does not seem to be a biased perception about the signal but rather a self- serving interpretation of that signal. In that sense, we tend to discard the “unskilled and unaware” explanation of Kruger and Dunning for analyzing our data: if the “unskilled and unaware” were to be given their “actual/objective” ranking, they should update their self-assessment accordingly [19-22]. In our data, this ranking is provided but overconfidence is still present. Finally, respondents do recognize their past failures, which shows that they can understand a bad signal; yet, they remain overconfident about the future, which indicates that they filter bad signals in a way that preserves their confidence about the future. Importantly, the typical interpretation of entrepreneurial overconfidence is that it could lead to suboptimal results. For example, “[i] inflated expectations of success at the point of firm creation may contribute to subsequent failure in nascent firms as overconfident entrepreneurs can [. . .] over-allocate acquired capital to high risk projects with little intrinsic chance of success” If overconfidence leads to the systematic misallocation of capital, then public policies should try to correct its causes and consequences rather than promote start-ups or patents. Our results, interpreted in light of our framework, suggest a different conclusion. If overconfidence is built through the optimal self-deceiving process for self-motivation, then it could be solving the bias generated by procrastination and increasing total output (see Appendix B.7). Overall, our explanation thus implies that overconfidence is a symptom that does not necessarily need a treatment; it is the result of agents that use attributions to boost motivation.

Our survey of participants in the BAE program shows that about 80% of entrepreneurs ranked themselves above the median and more than 50% among the first quintile in the dimensions of entrepreneurial ability and project viability, showing a clear overconfidence effect if we interpret that the answers are based on the median beliefs of a random sample of participants. Additionally, we show that when asked about their potential growth, 60% predicted above average performance, which is evidence of true overconfidence if we assume that respondents interpret the scale correctly as a comparison with the population average and if they answer based on their mean beliefs. Besides these two important points, we also emphasize several other key facts. First, respondents have strong signals about their relative performance in the relevant dimensions within the comparison group. Indeed, non-beneficiaries know that they performed worse than the median given that their applications were rejected and beneficiaries have precise information on their ranking in an overall assessment based on ability and project viability. Given that they answered our survey after this information, this makes our results even more striking. Second, we find that these effects are stronger for entrepreneurs of the last round of the program than of previous rounds (who might have more information about their abilities and projects). Third, we find strong differences between past performance and future growth self- assessment: while the former is centered on the mean (answers average out to the population average), the latter is strongly biased toward optimism (the average of answers is well above the population average). Fourth, a non-negligible proportion of those that expect exceptional growth in the future think that their past performance was below average. Analogously, some of those that place themselves among the first quintile in both ranking questions answered that their actual performance was worse than average. Fifth, when information on survival and income for the universe of self-employed in Buenos Aires is provided, respondents with the highest self-image seem to rethink and reduce their overconfidence about the future. We use these observations as our main arguments to discuss different theories of over confidence. We find that under particular but plausible assumptions, Bayesian overconfidence is incompatible with our main results. We also discuss that the utility value of beliefs, reference group neglect, differential regression, social signaling, and “unskilled and unaware” explanations are difficult to be sustained under the information and signals that our respondents have. The motivation value explanation of overconfidence (B´enabouandTirole 2002) seems to be more relevant. This setup is also in line with individuals that gather information about their absolute and relative performance but that update their beliefs slowly (less than Bayesian agent), as if they were protecting their self-image behind excuses, after a self- serving use of information driven by the attempt to motivate themselves. This leads to persistently overconfident individuals. However, this persistence tends to eventually recede as more problems are faced and evidence about one’s own performance accumulates. One of our respondents provided an informative answer that connects to the motivation value of overconfidence: “when becoming an entrepreneur one must consider oneself among the best; that is the strength that allows the entrepreneur to reach his goals”. Our findings point to the importance of motivation for start-ups. In our view, motivation is crucial for enabling entrepreneurs to interpret events and information in a way that shields them from any discouragement. From that perspective, entrepreneurial overconfidence can result from the pursuit of self-motivation rather than being a cognitive bias to be corrected.

For full article please go through this link : https://www.pubtexto.com/pdf/?entrepreneurial-overconfidence-evidence-from-a-public-program