Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
========================================================================

# Silicon Valley Everywhere? Place-Based Innovation Policy and the Direction of Entrepreneurial Entry

Authors: Freda Fang, Mohaddeseh Heydari Nejad  
Discussant: None  
Video: https://www.youtube.com/watch?v=7LDwpGoPNFc&t=23775s


## Talk (06:36:15 – 07:15:55)

[06:36:15] >> doesn't go back >> it doesn't go to my first that was the first problem okay hi everyone I'm uh this is a joint work with Freda She's  
[06:36:29] a PhD student at University of Toronto and I'm going to present this paper on uh placebased innovation policy and uh direction of uh entrepreneurial entries.  
[06:36:40] So we all know that the allocation of founders across industries uh shapes the direction of innovation and who becomes founder is important. We we're going to see how what founders choose to build also matters uh from a policy  
[06:36:55] perspective. Placebased policy increasingly targets the specific industries in each region to coordinate uh founders and all the resources into building clusters of innovation. H for example, you can think  
[06:37:08] about technology hubs, science parks, startup accelerators or uh targeted tax credits toward uh certain industries.  
[06:37:17] But founders cannot uh fully evaluate every possible industries uh to uh consider when they want to think about what what type of company to start. You can think about this as like a limited  
[06:37:30] attention story like uh limited attention and uh any type of information friction that uh creates a cost of acquiring more information to evaluate each option to enter. You can think about the irrational the rational  
[06:37:43] inattention literature or uh the cost of acquiring more information about each uh option to uh make a informed choice.  
[06:37:52] So any intervention any targeted intervention can shift the industry choice by uh founders through two main channels. So the f the first channel is the payoff channel uh that changes what an option is worth. any of these uh  
[06:38:05] interventions can bring capital mentors and investors and attract them into this industry and make the option more attractive for founders.  
[06:38:15] What we're going to see is something that we call a consideration channel and this change how much attention each option gets holding even the expected payoff uh fixed. So you can think about like an intervention that doesn't even u  
[06:38:27] attract like any input or any um investor or capital into the targeted industry but it can still stimulate entry into that uh industry in uh the targeted region uh by u making founders  
[06:38:42] evaluate that option more seriously. So the channel through which this affect the policy affects matters for uh targeting and for the design of a uh optimal policy.  
[06:38:53] So to be more specific the decision that we study is consider an electrical engineer a founder who is deciding what kind of company to start and uh their background is a good fit for building a company in software in energy in climate  
[06:39:07] in space and uh now think about the thickness of this line as how much attention they put to evaluate the option entrepreneurial options in that area. So this this electrical engineer might might actually be a better fit for  
[06:39:22] uh starting a frontier company in climate, energy or space. But they in their local environment it might be that they all hear about software companies and software entrepreneurship and they might end up they're more likely to  
[06:39:34] start a software company. Now suppose that in their local environment there is an initiative that promotes climate and entrepreneurial opportunities in building a climate company.  
[06:39:46] So as you can see this intervention can make the probably makes the climate uh more attractive in general to start a company in that industry. It can also thicken the line that founders are more  
[06:39:59] exposed to uh realizing what climate company means and then they can consider that and evaluate that option as uh something that might be a better fit for them to build.  
[06:40:13] So in this paper we are uh studying how innovation policies shape entrepreneur industry choice and we we're trying to understand whether uh the consideration channel something beyond this payoff uh channel uh is at work.  
[06:40:26] So the setting that we use is a industry specific launches of creative destruction lab. This is a science-based global mentorship accelerator that launched uh programs across uh 10 different cities and uh in each of those  
[06:40:40] cities targets a specific industry to help uh startups uh and provide uh general mentorship to them.  
[06:40:49] So we link the CDL creative destruction lab CDL data to crunch base to get the universe of uh firms founded in those cities and across different years and across different industries and we also  
[06:41:03] uh use uh real lab data to get more uh detailed information about the um background of founders.  
[06:41:11] So we basically we study how founders outside this program are affected by the launch of these accelerators in that region.  
[06:41:20] Then consistent with the with our uh findings, we're going to build a model of industry choice where founders have limited consideration and they have limited attention to uh choose among their uh industry uh among the industries. So we use the serial  
[06:41:35] founders as a attention invariant benchmark. This is uh consistent with we're following literature on advertising and uh that use informed agents as a benchmark to identify the  
[06:41:48] uninformed response. Yes. motivation sound as if the founders will already be in a place and then this program comes then you start up whatever is there also channel of people moving  
[06:42:01] into an area because they noticed area that is developing more stuff on climate change selecting into this places >> so selection of uh the accelerator into place  
[06:42:15] >> other founders >> uh so potential founders or like selecting into the region or into the industry into the region >> where the program is.  
[06:42:26] >> Uh so uh >> to see they they program people want to work in climate there.  
[06:42:34] >> Yes. So that's why we're going to limit our sample of a study to all those founders who never even apply to these programs. So and only look at the founders in that region who do not even seek or anticipate any payoff spillover from this intervention.  
[06:42:49] How do you not like set risk set of people who could become founders but is a measure did you found per >> so that it could be all coming from selection of new people into  
[06:43:03] >> uh not which is fine but I don't think that the model would actually be so appropriate >> yeah so our multi uh so it's a discrete choice model and the f the outside option which is not founding at all is  
[06:43:16] not captured in our model uh but actually that uh we can identify the relative choice among the available industries to those potential founders and uh conditional on becoming a founder. We can identify the  
[06:43:30] reallocation across industries but we cannot uh study the extensive margin because we don't actually see the potential the pool of potential founders who are choosing to become an entrepreneur and then after becoming an entrepreneur there's this choice of what  
[06:43:42] to build. So uh we're going to uh limit our sample to the founders and of course that's another margin that we cannot study because we don't have the uh that margin of who becomes an inventor uh simultaneously with their industry  
[06:43:57] choice. So you're more thinking >> simultaneous >> simultaneous particular company now I'm going to do startup that let me find something that I can do usually we think of people doing  
[06:44:11] >> yeah so we think >> yeah for for these for these channels like the effect would be uh stronger I guess if uh there's like an extensive margin of bringing people from outside of entrepreneurship to become an  
[06:44:23] entrepreneur in that industry and uh our estimates are going to be like an uh downward bias underestimation of that because we're only looking at even conditional on those people becoming an entrepreneur  
[06:44:37] are they choosing like this uh targeted this targeted uh industry over alternative industries. So the entry into these targeted industries uh might be even larger because people are coming  
[06:44:50] for from outside option. Yes, but that's actually a limitation because we don't we can't actually say anything about the extensive margin and we're going to see that for the counterfactual that matters because for  
[06:45:04] the counterfactual the outside option matters more uh in in our counterfactual experiment we're going to quantify the importance of this channel. uh our estimation uh like the cross industry uh evaluation uh doesn't have that  
[06:45:18] limitation of outside option because that will cancel out uh but the counterfactual uh needs like an outside option that we uh do not assume.  
[06:45:26] So what we so a preview of the finding is that uh we find that accelerator launches redirect the uh non-applicant founders uh enter more 12% more in uh  
[06:45:39] targeted uh industries. Uh and we find that this effect is particularly uh concentrated among science trained firsttime founders. We find that this effect this reallocation effect fades with local industry size. So if the  
[06:45:54] launch happens in uh targets an industry which is locally emerging and uh not that visible the effect uh is stronger uh we don't find any detectable response among uh serial founders. It doesn't  
[06:46:07] look like the serial founders are uh entering more into these uh this industry. We don't find u evidence of expansion in uh the total founding in that region. So it's not that more entrepreneurs are coming into this  
[06:46:20] region in general and we also don't find any expansion in financing or the reallocation of financing. So a part of that problem would also uh be alleviated here that there's actually no direct uh you know benefit or initiative that  
[06:46:34] makes the uh coming into uh this industry more uh attractive.  
[06:46:42] So in the uh in the model we're going to separate the effect into uh the consideration channel and we find that most of this uh is driven by a consideration channel as we limit the payoff channel. Um and we're going to  
[06:46:54] see u how uh we can uh separate this uh channel. So the finding is that on average an average founder uh the probability that they uh the odds of choosing a targeted industry over an  
[06:47:08] alternative uh industry raises by uh around uh three times and the payoff channel is on average zero for startups and uh it varies a lot with founders characteristic and as we limit our  
[06:47:21] sample to non uh applicant of this program uh this is reasonable uh that the payoff channel is is not at work.  
[06:47:29] So in the counterfactual we find that this actually attributes to 30% of all the first-time founder science trained uh entrepreneurs in those regions. If uh it wasn't for this channel like we had less of those uh type of uh founders  
[06:47:44] uh there so we're going to contribute to yes >> just one clarification. I may be too confused on how to think about the placebased aspect of the program because when we talk about placebased the first thing that comes to my mind is like  
[06:47:58] entrepreneurs choosing locations and whether this reallocates economic activity across space and whether that's efficient etc. But you have focused a lot on industry and you even said like placebased policies shape the the choice  
[06:48:12] of industry of entrepreneurs. So is it like an industrial policy or like >> Yeah, that that's actually a fair point.  
[06:48:19] Could be like an industrial policy, but how it interacts with the already established uh size of uh the relative advantage of that industry in that region. That's why we think that the target the optimal targeting depends on  
[06:48:32] where you're uh launching this type of uh program. For example, uh same policy uh targeting different places uh should consider the local uh environment and um like depending on the channel that is at  
[06:48:46] work. Uh sometimes like you can benefit from the elomeration and uh target the already established uh industry in one region. In some region if the information if the information is scarce uh promoting an emerging uh industry  
[06:49:00] could uh create more value. So we're going to contribute to uh three bodies of uh literature and so allocation of entrepreneurial talent across industries. We're going to uh  
[06:49:13] introduce uh policy induced attention to a sector that can shape the allocation of founders. Uh we're going to u study this in a accelerator setting. uh other papers have mostly studied the effect of the treatment of uh these programs on  
[06:49:26] the participating firms. Um we're going to study the uh effect on uh the local entrepreneur entrepreneurial activity and uh we're going to apply the limited consideration and information frictions uh literature into an entrepreneurial  
[06:49:41] setting and uh see uh if we can explain uh these patterns by a limited attention uh model.  
[06:49:50] So the setting is the creative destruction lab. It's a nonprofit mentorship accelerator founded uh 2012 at uh at University of Toronto and it only targets science-based industries uh for uh in that region. It doesn't uh  
[06:50:05] provide any uh funding even to the participants.  
[06:50:10] So it launches uh different industries like AI, health, climate, quantum space and they all should be like science based but they uh kind of focus on uh different types of uh ventures. Each of these lo these uh programs should be  
[06:50:23] hosted by a business school in a in the targeted city and one launch in our uh setting is basically a city industry year event.  
[06:50:33] CDL for example opens a climate stream in Vancouver in 2020 and uh for example a ocean stream in Halifax in 2019.  
[06:50:44] Uh so this is the map of the CDL sites and locations. Uh this is by 2025. Uh and as you can see uh it targets different cities. In each city it opens different programs and in each uh city  
[06:50:57] there might be uh more than one uh stream but they have they launch any of those stream uh different uh time at different times.  
[06:51:10] So basically so the main concern is that what if CDL expansion is not random and uh so the process of uh so we're going to give you like a little uh background of how CDL expands into different cities and how to how they choose a stream. Uh  
[06:51:25] the first thing that comes is the location. So it's not like a city industry uh targeting. It mostly happens through some connections to the founders uh through like relationship between uh  
[06:51:39] the founder the main founders of uh creative destruction lab and some people that they know in the business of school. They mostly like reach out saying that I want to promote my um school or um it also uh works the other  
[06:51:52] way. CDL team reach out to people that they know in the business schools uh to uh start uh a CDL site.  
[06:52:01] after the CDL location is uh kind of decided the stream is mostly matched based on the um specialization of that university and also uh something about the ecosystem  
[06:52:14] and the local advantage in that region and from the time beginning of this process it takes roughly between one to three years they to uh launch a program and start accepting applications uh for  
[06:52:28] uh a cohort. So this is how actually they expand and I'm going to show you how we uh uh control for this uh selection. So our empirical and YCDL provides that this um setting to study  
[06:52:41] uh the consideration channel and limit the payoff channel uh to uh local founders. uh we basically need some industry specific targets and CDL is a great setting because they open uh  
[06:52:54] industry targeted uh programs in each city and they don't provide any funding even to their uh particip participating firms. they only provide like four hours of mentorship every two months and um  
[06:53:07] every participating firm uh gets uh like four of these type of uh mentorship sessions and uh the non-licant that we study receive uh no at least direct benefit in our uh model. We allow for  
[06:53:21] these launches to generate indirect payoff spillovers to founders in that region. Um but we expect that to be u relatively small.  
[06:53:31] >> Yes. kind of get a sense of how like four hours every two months and no funding. It seems >> like not a very big thing compared to like the se Seattle and Toronto have so much going on like would we expect there  
[06:53:45] to be big effects? >> Yeah, we don't expect that to be like a big uh payoff relevant um effect for especially for non uh participating firms. They actually provide a very specialized type of mentorship to their  
[06:53:58] participating firms. Even this 4 hour is the personalized uh mentorship that they receive. There's also like every two months there's like a big public meeting like this for all the founders to gather together and see all the mentors. So for  
[06:54:12] for the participating firms it's creating value like in my job market paper I do study of like what is going on inside this uh accelerator and what kind of uh value they're providing to their participating firms and there are  
[06:54:26] like causal benefits of like different time types to uh to firms but it doesn't provide any immediate uh payoff. So like for example the thing that we find is that after one three or three years of  
[06:54:39] graduating from CDL we we find evidence of uh you know raising more money or uh doing like on average uh better.  
[06:54:50] Uh so we also uh so another interesting um feature of the setting is that there is a variation in prior local salience of the u industry that they target. Like for example in some some cities they're  
[06:55:03] targeting like a already established uh industry like for example AI.  
[06:55:09] So in our settings to uh take care of the selection of CDL expansion. All of our designs are going to have a CD CD year fixed effect to uh control for CD selection and uh we control for any industry uh global industry trend that  
[06:55:23] something becomes uh more interesting and the next uh host uh might actually more be more likely to uh select that industry. We also control for uh like the most important one which is the persistent local specialization of that city. For example, if you think about  
[06:55:38] Halifax, Halifax has a ocean. Like they there's no way they can, you know, start a an ocean in Toronto.  
[06:55:45] So uh remaining identific identification threat would be a la this concurrent thing that's happening at the same time as CDL launch that doesn't attract any investor doesn't attract any capital but it attracts uh a specific time of  
[06:56:00] founders like first time science stream founders into the this targeted industry and not other uh like science-based uh industries.  
[06:56:11] So we don't think that happens a lot because there's no direct benefit and we don't find any evidence of positive uh increase in the entry of those type of founders in that region and we also don't find any expansion in uh capital.  
[06:56:23] So if there is like an initiative at the city level they it should also attract a little bit of financing around that.  
[06:56:29] Yeah. >> Question. So you say you focus on nonparticipating firms. What is the universe of is it like every other firm in the city that is not participated?  
[06:56:39] So, >> so we're using crunchbased data. It's uh kind of close to like pitchbook if you know that data. It's um like a relatively good uh database of VC funding and all the firms that receive  
[06:56:53] any VC funding which are um kind of like a best other than pitch.  
[06:56:59] Pitchbook is a slightly better but it's kind of a coverage of the f uh firm formation in those cities. Yeah. So it's it will be basically all the firms that receive uh VC funding which are the  
[06:57:10] startups that in that region. So basically this is our setting. We have uh CDL data uh we use crunch base to uh get the all firms in that region and we use rellio to uh create founder  
[06:57:25] characteristics. So the first design so the first reduced form I'm going to show you is an event study that we look at um how uh how uh the dependent variable here is a  
[06:57:38] number of firms in CDC in industry s uh at time t and we're going to use uh ppml uh regression we control for the a high dimension of uh fixed effect and uh we  
[06:57:52] find that in in treated cities and also when we add when we match never treated cities. We find an increase in uh founding activity, entrepreneurs founding activity in these targeted industries relative to the untreated in  
[06:58:06] industries. So we have a city year fixed effect in this regression. So it's basically a relative a comparison between a trend of entry into these targeted industry versus the trend of uh  
[06:58:17] entry in uh untreated uh industries. Is there some like double counting going on here which is that I was a founder that was about to start a health firm but now because of the policy I'm going  
[06:58:30] to start a climate firm so just like you know it looks like one less health one more climate right like it's not actually a gap of two it's a gap of one you know what I mean like it's a subocations essentially  
[06:58:43] >> yeah so if that is the reallocation that's happening that's what we're going to try that's what we want to uh you know uh capture uh a potential founder a potential founder like the trend is uh so we have  
[06:58:55] like another um event study that we show that the entry into non-treated industries are relatively dropping and uh entry in uh after these launch entry into the treated industry are uh increasing and in that uh design where  
[06:59:09] uh the control group is a matched city that this is not happening in it uh so we find that both of them are kind of uh significant so we have like a relatively significant increase in entry in uh the  
[06:59:22] targeted industries on relev relative to the untreated industries.  
[06:59:28] >> But why is just number of firms in indust in the debate?  
[06:59:32] >> Yes. >> Which going back to earlier point could just come like new guys coming into that city because now there's the same.  
[06:59:41] >> Yes. So the control group here is like the founding behavior before this launch in the same city uh industry year and uh  
[06:59:52] same industry in uh another city which is not uh treated but yeah so like if there is another thing that is happening that's drive uh that's uh bringing this  
[07:00:05] specific type of founders into this industry we cannot rule that out. So there should be something happening at the same time. Sorry >> saying this is just effective. You don't need to do that out. Just there's a bit of attention with the  
[07:00:18] model that you highlight. There's somebody already there now choosing between two type of industries but could be just you guys doing climate moving into that city.  
[07:00:28] >> Yeah. That's why we're going to build a um discrete choice model where at the founder at the team level we are looking at teams choosing between uh all the 17 available industries and in that discrete choice model we can go uh  
[07:00:41] beyond this uh you know uh this event study where the dependent variable is the total number of firms and then we can look at like the probability that each team uh enters a treated industry versus uh an untreated industry. So  
[07:00:56] that's why in our model we're we're actually building a team level uh discrete choice model.  
[07:01:03] >> I think the comment goes more in the direction I guess connecting it to the model that you would have two layers.  
[07:01:08] You need to have two layers in the model. First some entrepreneurs will decide which location to go to and then one condition on the location which industry to go to.  
[07:01:17] >> Yeah. Yeah. So like this is the assumption in the model that they don't choose their location if that's and actually in our uh data I think more than um 50% actually a lot of the  
[07:01:29] founders are local um but I guess maybe we can actually exclude anyone who moves actually across different locations and then do the >> I don't really want to do that. I'm  
[07:01:42] saying I think the empirics why it's just that the model there's a bit of a tension between the empirics and the model.  
[07:01:49] >> Yeah. So like this one is not still the model that can distinguish between a payoff and consideration. And uh I'm going to show uh this as an evidence that there is a reallocation going on.  
[07:01:59] It seems that there are like entries into uh treated industries and uh relative to untreated industries. And then we're going to build a model that can separate that using some uh identifying assumptions.  
[07:02:12] Uh yes, some local institutions.  
[07:02:22] like one day gathering of people in that would be like one of the challenges like there should be some  
[07:02:42] >> so like other types of institutions unrelated to CDL we don't have data or we haven't looked at that like if there's another type of institution built around that. But for the CDL, those meetings are not like open to  
[07:02:55] public. Those are only for participating firms. That only happens during the program.  
[07:03:04] >> Yeah, we we haven't tracked them as like what they do after they go. We only track like their fin the financing of the f the participating firms.  
[07:03:13] Uh so this is the event study version of the previous um design and uh we show that there is actually an increase in uh the number of firms entering into the treated cities versus untreated cities and the effect  
[07:03:27] is still there if we exclude the applicants.  
[07:03:32] So to look at who are these uh so I guess I'm going to move from this but here we we're uh looking at whether the participating f whether these new entrance who are not  
[07:03:45] related to uh this program are more likely to be connected more likely to be schoolmates with the participating founders in that program. And here we're comparing the we're comparing the  
[07:03:58] founders's relationship or overlap uh with uh the new stream that comes in town with uh the their connection to the already operating operating streams in that city. So if the baseline is they're  
[07:04:12] like on average connected to CDL participating firms, we see that these new entrance that come at the uh time of launch are more likely to be schoolmate with those founders who got accepted into the program. And it might be coming  
[07:04:24] from you know this is just a descriptive uh result. It might be coming from the fact that they have both same expertise or or they have same uh background or same uh kind of uh experience. We're  
[07:04:37] create controlling for any overlap in their degree or field of study or if they have any employment uh history in that industry. And we still see that after controlling for uh those uh similarities, we still see some positive  
[07:04:51] correlation between uh the overlap of these entrance being a schoolmate with a new uh stream uh relative to the old already operating streams.  
[07:05:07] So to study how who are these movers, we're going to define a type for our team and We're going to proxy uh make a proxy for experience for founding  
[07:05:20] experience. So if there's no member on a team that has a founding experience, we're going to label them as a first-time founder. And a science trained measure is at least one of the founding team members should have a  
[07:05:32] science or engineering degree. And u we use this experience as a proxy for prior entrepreneurial search. So these are the founders. These are the teams that have that they have like the capacity. They  
[07:05:46] they've already done an entrepreneurial search and they are kind of more familiar with the uh landscape in that region. But we don't actually assume that they know all their option and they evaluate all their option. We only assume that these serial founders do not  
[07:06:00] change the attention that they give to different industries as a result of this uh intervention.  
[07:06:07] So science training in our uh is a proxy for relevance of this um intervention because this intervention is only um targeting science-based ventures. So it's only relevant for those type of teams that have at least like one  
[07:06:21] science uh trained uh founder. So now here we're looking at the same design for uh different types that we define and we want to see if this reallocation behavior of entering more  
[07:06:34] into a targeted industry relative to an untreated industry is a stronger in any of these types and we expect that to be uh stronger for the relevant uh founders. uh and um especially if if  
[07:06:48] there's any kind of information going on uh it should be uh stronger in first- time founders compared to the serial founders.  
[07:06:57] So here uh again we have a ppml uh regression and we see that all all the effect is coming from science trained first-time founders reallocating and choose the they're more likely to choose uh the treated industry relative to  
[07:07:11] untreated industry and most of the industries are um the all other industries some of them are also science uh industries so them choosing the targeted industry by CDL uh is coming  
[07:07:25] from uh these uh founders ers who have like a more general uh background of uh science training uh but they don't have any entrepreneurship uh experience.  
[07:07:37] So then here we're building a model that uh distinguish between the consideration channel and the payoff channel for each team. So we're building a team level model and uh team uh each team is choosing among all the available  
[07:07:51] industries and conditional on uh founding. So we don't have like the outside option but bringing the outside option in this uh model doesn't change any estimation. It might affect some of the counterfactual experiments uh cuz uh we're not uh we don't have that margin  
[07:08:06] of people going out of uh entrepreneurship.  
[07:08:10] So the first part of the model is attention weight. Each industry carries a positive weight that uh captures how strongly it enters uh the consideration of uh a potential founder.  
[07:08:22] So here the serial founders have a bas their own baseline attention allocated to each indust industry and firsttime founders have another benchmark attention that can vary across um their  
[07:08:36] characteristics and their founding experience but it can also increase uh through uh an intervention through the intervention. So this uh DCST is an active program going on in that um  
[07:08:49] targeted in that city industry and this London function captures the consideration channel and this is how uh the event or the program can increase the weight that these founders put on uh  
[07:09:01] each uh industry. So conditional on this attention they make a discrete choice and choose pick one of the industries uh among the available industries and the payoff channel here we're going to allow for a  
[07:09:15] payoff spillover that uh increases uh the payoff of entering into that targeted industry for all founders uh depending on their characteristics and also depending on the AT law which is  
[07:09:27] the local uh relative local size of that targeted industry.  
[07:09:32] So here the choice probability is a weighted uh multinnomial logit uh choice and uh the relative uh probability of choosing one industry over the other industry um doesn't depend on uh the  
[07:09:46] outside option >> yeses but I guess it doesn't have it could be a competition crowd but it might like to is allowed to be negative >> towel is allowed to be negative exactly and this is that's what actually we're  
[07:10:00] going to see uh so the identification assumption And I uh just mentioned that this is um coming from a launch uh does not change serial founders consideration weight. So they're attention invariant.  
[07:10:10] It's not that they're fully informed. Uh they just don't change their search uh behavior, the evaluation behavior. Uh and the second assumption is that conditional and all observables the response to the payoff response is uh  
[07:10:22] doesn't depend on uh experience. So we're going to estimate that through a standard maximum likelihood and the estimation results. Uh so we consider a to function and a lambda function linear  
[07:10:36] functions of their characteristics and the lambda consideration is only operating for uh the first time founders. So what we study for the what we estimate for the uh for these uh two channels on average for an average  
[07:10:49] founder the payoff channel is uh not significant is on average is not distinguishable from zero. But we see that different characteristics can actually change this response to payoff and the consideration channel is uh on  
[07:11:03] average positive and for science degree uh found founders with science degree it's uh stronger and it also fades with the with the salience of the targeted industry in that region. So it seems  
[07:11:16] that this channel is also stronger when the local uh in the targeted industry is emerging.  
[07:11:23] So some other characteristics are also u they have like a mirrorship and are um statistically significant which is interesting. If you think about like P teams with more PhD founders on the team  
[07:11:35] or more migrant founders on the team they they're more likely to be specialized and they've already uh you know done like a good search and they've specialized in a narrower setting that this uh intervention is less likely to  
[07:11:49] uh change their direction through giving them more information. But it can only move them if it offers an extra payoff to those type of founders. And this is also like this finding is also consistent with the uh predictions of the rational in attention model where  
[07:12:04] the attention is allocated where the uh payoff is already high. So a attention uh constrained founder would allocate more attention where they think that the payoff is is higher.  
[07:12:18] Uh so here we're holding all the characteristics at the average that we observe in the data. So we don't want to have like all different characteristics among the founder. So for comparable founders  
[07:12:29] that are first time science trained uh the consideration uh channel is uh significant and the total response is also significant. So if we make the non-science first- timerrs also uh comparable to these uh first time  
[07:12:44] founders we still we see that the total response of those first time founders is also positive but distinguishing between the channels for those for those uh we have like wide uh confidence interval for that. So the quickly the  
[07:12:57] counterfactual is just shutting down the consideration channel. So we put lambda function to zero and we recalculate all the choice probabilities into uh the available industries to founders and uh we then normalize that across the uh  
[07:13:12] indust industries. So here we only look at the firms that uh are treated and they enter. So conditional all those firms uh that are not related to CDL and they actually enter after these  
[07:13:25] treatments we see that there will be like among the first time serial founder serial first time science trained founders about 30% of them would uh be uh would go into another industry if uh  
[07:13:39] London was equal to zero and this is about like 11% of the uh total entry into the uh targeted uh industry and This is uh larger for more established  
[07:13:51] um industries and this is partly because uh more of the uh activities are actually in those uh established firm established industries.  
[07:14:03] So just like a rewriting of the same uh calculation the difference between the probability of choosing a targeted industry under counterfactual where where there is no consideration and the actual uh we're going to weight that  
[07:14:17] calculation by the capital that they actually raise and uh the capital attached to this part of the uh entry that's coming through this consideration channel is about uh $46 million. So it's not like the um expansion in financing,  
[07:14:32] it's the financing attached to those uh entrance that they are already in uh the treated industry.  
[07:14:40] So here we're uh doing a calculation of the change in the inclusive value of consideration. So here we're basically looking at what should be the increase in utility of each option that they already have to make the founder  
[07:14:54] indifference between receiving this information through the consideration channel or not. So here uh we are uh we find that around like 7% increase in the val the utility of each option for uh  
[07:15:07] founder is equivalent to giving up this uh consideration channel.  
[07:15:12] Uh so I think I'm on time. So here we're studying uh if uh we're studying the impact of the entry of uh accelerator program on uh the redirection of entrepreneurs across targeted industry  
[07:15:25] and we look at uh something we call a consideration channel and uh that uh founders are putting more attention toward uh building an company in this targeted industry  
[07:15:38] and we find out like around 30% of the science train first-time entry into the launch industry is coming from this channel and it actually fades with the salience of the local uh environment for  
[07:15:49] that targeted industry. Thank you. >> Very much we're at time I think. So so we're going to have a break uh and after