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

# Risk Aversion and Barriers to Firm Growth: Experimental Evidence from Small Retailers

Authors:   
Discussant: None  
Video: https://www.youtube.com/watch?v=mC4ESywpIDc&t=3645s


## Talk (01:00:45 – 01:52:03)

[01:00:45] someone come to me had a deal. So, >> thanks so much. Um, I'm Grady. I'm from  
[01:01:09] the University of Chicago and I'm delighted to present my paper, Risk Aversion and Barriers to Firm Growth with Experimental Evidence from Small Retailers.  
[01:01:17] So this work is motivated by the fact that most production developing countries occurs among very small firms with fewer than 10 employees and so they make up more than 99% of developing country enterprises and account for more than 90% of employment.  
[01:01:32] And the literature has found that the growth rate of such firms is very limited and that in particular small firms in developing countries tend to grow much more slowly versus rich countrys.  
[01:01:43] This slow growth is often attributed in part to low levels of firm level upgrading in developing countries which encompasses both innovation in terms of inventing new products or technologies but also simpler uh decisions like adopting products or technologies that  
[01:01:58] were already invented elsewhere and introducing them into markets for the first time. And this poses a bit of a puzzle because it seems like it should be easier for firms that are operating inside of the frontier to upgrade and yet they do so at a lower rate.  
[01:02:11] And so in this paper I ask if small firms engage in too little of the risk-taking that is needed to grow. So the core idea of the paper is that the owners of small firms receive most of the output from their enterprise which may make them heavily exposed to gains  
[01:02:25] or losses in the absence of complete insurance markets. And so this may cause firm owners consumption preferences to enter their production decisions which create scope for firms to be risk averse. And so I ask if risk aversion  
[01:02:39] prevents firms from pursuing profitable but risky investment opportunities.  
[01:02:45] I study this question in the context of retailers decision to adopt and sell a new consumer product. This is one of the most common firm upgrading decisions that is available to enterprises in developing countries and it's one of broader macroeconomic importance since  
[01:03:00] the low adoption of new products among retailers would constrain incentives for manufacturers to introduce new products or technologies in the first place by limiting the demand that is available to them. And I'll argue that it's a setting that where the investment is risky  
[01:03:14] because firms don't know if consumers will demand the product at the time when they invest in inventory for the first time.  
[01:03:21] So, I specifically focus on Kenyon shops decision to stock new motorcycle helmets. Effective helmets were historically too expensive for a typical consumer to afford, but a local factory was constructed 2 years before my study began that dramatically lowered the cost  
[01:03:36] of the products. I document that retailers adoption of the helmets was very limited where with a screening exercise of about 1,500 suitable retailers I recorded that under 3% of them had ever tried stocking  
[01:03:49] helmets. And yet this is in contrast evidence from my previous research that when consumers are offered helmets by researchers they seem to value them suggesting some market inefficiency.  
[01:04:00] And descriptive evidence suggests that the ingredients for risk aversion to affect firm stocking decisions may be present. And so namely firms belief suggests that helmets are viewed as a profitable but risky opportunity.  
[01:04:12] And so in particular 51% of shops report positive expected helmet profits when I survey them at baseline. And yet only 6% of them accept stock in the absence of any form of intervention.  
[01:04:25] And I record that uncertainty about sales is strongly negatively predictive of subsequent stocking decisions which suggests some role for risk aversion or volatility.  
[01:04:36] And so motivated by this, this paper asks if risk aversion prevents small retailers from stocking a profitable new product when they're uncertain about demand.  
[01:04:44] I begin by formulating a model of um that shows when firms are uncertain about demand for a product. Riskneutral firms will invest more because the upside is high, which ensures that profitable products are quickly adopted.  
[01:04:58] However, risk aversion creates disutility from uncertainty that can flip this relationship and undermine the process of learning about new products.  
[01:05:06] Then I test the model's predictions using two field experiments. First, in an insurance experiment, I ask if risk aversion affects stocking and it's designed to induce a mean-preserving contraction of profits to produce a theoretically clean test of risk  
[01:05:19] aversion. Then in a second learning experiment, I ask if risk aversion prevents the discovery of profitable new products by offering a temporary buyback option to encourage experimentation and then tracing out the lumber run effects.  
[01:05:33] And to preview the design of both of the experiments will take seriously potential confounding forces from credit constraints, biased beliefs, any signaling effects of offering interventions or learning by doing.  
[01:05:46] To preview the results in the first part of the talk, I'll show you evidence that risk aversion affects product stocking decisions. My method will be to offer an insurance contract that would be strictly dominated if firms are risk neutral and yet in practice leads to a  
[01:05:59] doubling of the shortrun helmet stocking rate.  
[01:06:03] Then in the second part of the paper, I'll look into the interaction of risk aversion with new product adoption over several results. First, I'll show that inducing firms to try stocking helmets leads them to permanently adopt them at a much higher rate. So my method will be  
[01:06:17] to offer return policy in a first phase in an intervention and then offer helmet stock without returns in phase two where the key test is to examine stocking decisions in phase two. I'll find that this leads to a seven percentage point  
[01:06:30] or 70% expansion in helmet stocking in this phase of the intervention and that it doubles the share of shops that decide to permanently enter the market for helmets by the end of the study.  
[01:06:42] Then next I'll show evidence that firms understand the learning value of experimentation by ensuring their ability to resupply a from manufacturer after the study ends.  
[01:06:52] And I'll find that when the long run potential for profits goes up, firms are about twice as likely to try experimenting with helmet sales today.  
[01:07:01] And then lastly, I'll show evidence that returns matter because they cause riskaverse firms with uncertain beliefs to try stocking who then resolve uncertainty with experience in the market. So I'll first show that the effects of returns are concentrated among firms with optimistic but  
[01:07:15] uncertain beliefs in phase one of the experiment. And then I'll show that the treatment does not change beliefs about expected profits in phase two, but it does substantially resolve uncertainty about demand.  
[01:07:28] And then in a final part of the paper that I likely won't be able to get to today, but wanted to preview, I show that firms can also learn about demand from their neighbors. And so these information externalities will help to rationalize why competition doesn't drive risk averse firms out of business.  
[01:07:43] But I find that the spillovers are geographically very localized to within about three city blocks. And so they don't necessarily resolve the problem because social learning doesn't occur broadly enough for firms to be able to acquire the the information without  
[01:07:55] their own experimentation. So the principal contribution of this paper is to a literature studying firm growth and risk aversion. I provide among the first direct evidence that risk aversion is a barrier to firm growth preventing retailers from  
[01:08:09] discovering profitable new products. And this finding also raises questions about the common practice of modeling small firms as riskneutral in development which is relevant to tests of misallocation collusion as well as many other important topics where our  
[01:08:23] interpretation of tests tends to be through the strict lens of profit maximization.  
[01:08:29] Second, I build on literature on pharma upgrading and product diffusion where I show that risk aversion and demand uncertainty can undermine the retail adoption of goods. Um and so this helps to extend evidence that manufacturers inefficiently adopt new technologies  
[01:08:42] downstream to the retail sector. And this has this is important because it suggests reduced incentives for manufacturers to introduce new products in the first place which may contribute uh to a low set of products available to consumers.  
[01:08:56] And then lastly, I build on a literature on learning about demand where I show that risk aversion can undermine firm learning which can rationalize firms holding uncertain or incorrect equilibrium beliefs about demand. So first in the model I'll formulate an idea that uncertainty about  
[01:09:10] profitability is costly to firms since it may lead them to stock the wrong set of items and in general this will cause riskneutral firms to experiment but learn and learn truth but then in both the model and the empirical results I'll show that risk aversion can undermine  
[01:09:24] this experimentation process causing a breakdown in learning and so to preview the rest of the talk I'm next going to go over a theoretical model which will both illustrate this core trade-off between experimentation  
[01:09:38] and risk aversion and from which I'll derive a class of predictions which um the experiments will be designed to test. Then I'll go over an experiment which asks if risk aversion affects firm stocking decisions, empirical results about whether risk aversion prevents  
[01:09:52] retailers from adopting new products and then conclude and so I formulate a model of possibly riskaverse firms learning about demand.  
[01:10:01] Yes. >> Uh companies like Coca-Cola, Pepsi or even companies in the US or um you know big wholesalers, manufacturers in developing countries, they find ways to put their products in the market even in  
[01:10:16] very remote places with small retailers, right? And I I guess partly they do it by offering something like insurance, right? So why isn't that solving the problem here?  
[01:10:28] >> Yeah, that's a great question. And so when I talked to the the like CEO of the helmet manufacturer about why they didn't offer return policies themselves, they cited the fact that that um like operating with so many very small retailers makes the costs like  
[01:10:42] transportation and delivery prohibitive relative to the amount of stock um that's being offered. And then that's compounded by the fact that it's an environment where there's really weak contract enforcement. And so if they instead offered something like a trade credit with new firms, they perceive  
[01:10:56] that they'd just get stolen and it works in reverse. And then the the last issue is that because it's so difficult to track firms in the absence of things like more established addresses, they perceive just massive reputational risks where if they said, um, you need to pay  
[01:11:10] me for stock, but I'll come and and give you your money back if you don't sell it. There's a huge risk that they'd fail to find some of the shops again, then that information would spread, that they were unreliable, and it would undermine their business broadly. And so I think there are characteristics of low-income countries specifically that make it very  
[01:11:25] challenging for these markets to exist. And to kind of broaden that a little bit, I actually collected data on that.  
[01:11:29] And I found that under 15% of shops had ever been offered something like returns for any new product at all from a new manufacturer. And so there are trade credits which exist on like relational contracts with um suppliers that they've purchased from for a while, but that  
[01:11:44] doesn't tend to emerge for new products specifically.  
[01:11:47] >> Sorry, just a quick follow up. Uh Yeah, everything you said makes a lot of sense, but I I'm guessing the way Coca-Cola does it is that they have multiple product lines and then they can, you know, uh bundle things or, you know, just to encourage you to try  
[01:12:02] something new, right? If they think it's going to be profitable. Uh so there it's not just about trade credit. There's other mechanisms, you know, is it just that the this particular motorcycle manufacturer is just too small, doesn't have multiple product lines and >> Yeah, that's a good question. So I think  
[01:12:16] like that's speaks to the types of of products where we might expect this to bind for that if something's more novel and so it falls outside of the scope of an existing product line. It's much more challenging to leverage existing relationships um in order to get it to diffuse. And so if you're like  
[01:12:30] introducing a new machine for instance that seems really binding or like helmets that didn't exist. If you're like a juice seller and you have a new flavor then it's much easier to leverage those relationships. And so I view it more in kind of line with like the novelty of innovations um that that we'd  
[01:12:44] expect to spread. Um so returning to the model the goals are first to produce a test of risk aversion which is robust to important features of the economy such as for instance capital constraints and second to examine how risk aversion affects new  
[01:12:58] product adoption. And so the setup is that firms begin with beliefs about theta which is a parameter that indexes the demand function. Given their beliefs, they choose how much to invest in product of a new or in stock of the new product, saving or borrowing from a  
[01:13:13] risk-free asset and consumption levels. After investing in stock, demand is realized where profits of the new product are a function of the stocking level, a variable new which will capture stoastic fluctuations in demand period to period. And then the true value of  
[01:13:28] theta, theta kn which is fixed but unknown to the agent. So they need to learn about it. After observing profits, the firm gains a signal about demand.  
[01:13:36] Their beliefs update according to BA's rule and then the process repeats but with updated information about demand.  
[01:13:42] And so I'll refer to uncertainty about theta as what I call demand uncertainty which is uncertainty driven by lack of experience but which can be overcome by learning since the true value of theta is fixed. And this is differentiated from true periodto period volatility and  
[01:13:57] demand for all products which is captured by new and which I'll refer to as demand volat volatility for clarity of language.  
[01:14:05] The other feature of the model is that in the future firms must pay a transaction cost of gamma to continue stocking the new product and this provides a parameter that can be manipulated to affect the continuation value of learning.  
[01:14:17] Firm's objective is to maximize the present value of expected lifetime utility of consumption where they face a budget constraint that saving or borrowing plus consumption plus investment in stock at wholesale cost of W subn cannot exceed their savings on  
[01:14:32] hand plus their their profits off of stock from prior periods. I model model a minimum order size of investment in the new product of Kai which creates a discrete risk of experimentation. Um, and so this is a true feature of the the market that I study, but I noted that  
[01:14:46] it's isomeorphic to, for instance, search or transportation costs, which would similarly create some discrete jump in terms of investing something rather than nothing. And then finally, since capital constraints are salient in this setting, I I include a borrowing limit that borrowing cannot fall below  
[01:15:01] some lower bound. And so expectations are both over new due to stochastic fluctuations in demand and theta due to uncertain beliefs about demand which are mitigated by learning as a function of one's investment level.  
[01:15:14] And so therefore investing more will increase expected utility in future periods by allowing for for better future optimization. Yes.  
[01:15:27] >> But how do you think about other ways looking to see if people are wearing helmet chatting with the customer and going like a helmet. It feels like there are things to learn about.  
[01:15:40] >> Yeah, that's absolutely true. And so one thing you'll see I do in the experiment is I give shops um like an ample amount of time to come to a final decision about whether to purchase. And so I'm I'm measuring the effects of the interventions once all of that is baked in. And so find that things like talking  
[01:15:54] to consumers or like pre-selling is kind of imperfect alone. Um, and then I I have a a component of the experiment that that tests for spillovers. And essentially what I find is just that markets are fragmented enough that those spillovers matter very locally, but  
[01:16:09] there are not enough shops that are experimenting in the first place such that the opportunity exists in order to learn in that way. But that is included in a richer version of the model that's in the paper. Yes.  
[01:16:25] So that's the uncertainty about your own need for cashing on the uncertainty about demand or is  
[01:16:39] that some problemated by the borrow? >> Um that's a good question. I think that that certaintly like an inability to borrow would create like a buffer stock savings problem that that amplifies  
[01:16:53] this. Um I don't think it should kind of create the risk aversion over products uh sufficiently in the first place because it so it would if they're like irreversible investments. So it could in the extent they're irreversible investments and you're like I don't know  
[01:17:06] like I may need cash like at this specific period in time and so that could confound this returns design specifically. Um, and so in in that intervention, I'll actually only allow people to return stock at a specific point in time to make it ineffective at  
[01:17:21] countering that type of risk. And then in a second insurance policy, because it it just affects uh transfers at a specific state in time, I don't think it would would be useful for that either.  
[01:17:30] But feel free to return to it once I get to the decremental design.  
[01:17:37] >> All of your customers have motorcycles. >> That's right. So I view the the uncertainty as essentially both uncertainty over what like the residual demand curve would be. So that is both  
[01:17:51] uncertainty about whether any consumer is demanded at all as well as kind of what the the equilibrium in terms of other shops stocking will be. And so then if you kind of interpret it that way, the tests that I'll use for risk aversion are robust to potential competitive features in the market. And  
[01:18:06] so that's why I abstract away from it. That being said, of course, that types of like interactions with other firms learning from you are very important and there's there's follow-up work that I have that looks at that specifically.  
[01:18:18] >> Seems exactly the right model for someone who's thinking, should I take all my wealth and buy a Domino's franchise and and see if people like pizza or not? But if you're a retailer who's stocking 40 products and you're adding one, are you going to give us a  
[01:18:31] some numeric to help think that this is really meaningful in terms of risk aversion and curvature of utility function?  
[01:18:38] >> Yeah. And so I think that's an empirical question of whether adding one new product is is a sizable enough risk. And so what I'll hope to convince you of when when I get to the results is that even this risk of of stocking um a couple of helmets is consequential enough relative to the size of typical  
[01:18:52] retailers that it matters a lot. um and like essentially like doubles the share of shops that stock it. Part of that is that uh because of firm sizes, the like minimum order size comes out to about like one to two weeks profits for for the median firm. And so um even though  
[01:19:06] it's not that expensive of a product because they're small enough, essentially adding anything that is kind of a durable asset is is a pretty big discrete risk.  
[01:19:15] >> This a variant of like why does the market solid and is real imperfection like failure of the helmet manufacturer?  
[01:19:22] But I guess I'm If you just gave everybody one helmet like some of these retailers one helmet that you know to Ben's point that then customers could try it on and see if they like it. Like how does that compare the price of one helmet to the the flow  
[01:19:37] profits to the helmet maker you know of having the new retailer >> the just give somebody you know to to like encourage adoption.  
[01:19:50] >> No. Yeah, that's a great question. So don't observe their profits. they were not kind of forthcoming with sharing data. What I can say is that it did change their equilibrium behavior in other ways, but they did not do things like allow like a single helmet sale.  
[01:20:02] They did actually drop their minimum order size from like 10 to three, which was a pretty sizable reduction. And all of the treatment effects survived that because they did it unexpectedly in the middle of of the experiment. And so it's possible if they dropped it all the way to one that would solve it. It's kind of  
[01:20:15] not something that observe. I think kind of more more broadly in terms of external validity. At minimum, we would expect this to bind for products that are more expensive than motorcycle helmets, for instance. For instance, things like machines for for like farmers that we need to go through  
[01:20:29] retail markets where even a single unit is quite expensive. Um, and I have follow-up work with Nick Swanson and Lisa Follow that looks at cook stoves and Bundi that's exactly that case where we allow single purchases and this still seems to be pretty binding.  
[01:20:44] hat to say that the the failure the market failure wasn't something like deep about contract failure but like yeah failure of m the upsale seller is risk averse or doesn't have perfect management skills etc  
[01:20:57] >> 100% and so they might not be optimizing perfectly what I can say is that they like like I talked to them a lot and they were really really opposed to offering this and so I tend to think that there are probably actually high costs especially since they did change change other behavior and so it's not  
[01:21:10] just cheap talk But so I'm next going to jump forward a bit to the the solution for optimal investment and we'll introduce new terms that I'll then walk you through. So optimal investment is guided by term capturing capital constraints um expected marginal profits  
[01:21:24] a term which will capture risk aversion the marginal value of learning and then this will be equal to marginal stocking costs and the core friction in the model is that risk aversion deter stocking under uncertain demand which risk neutral firms would optimally stock  
[01:21:37] more. So to see this, let's first focus on the learning value term, which is capturing the fact that uncertain beliefs about demand imply that there's a chance that a very good is very profitable. And so a firm can stock to see if demand exists. If it doesn't,  
[01:21:51] exit so losses are short-lived. If it turns out consumers do man demand it, they can restock it. And so their profits increase in the future as well.  
[01:21:59] And observe that if we hold fixed the expected gains or losses in period one, then this becomes strictly more attractive to but spread out the distribution. This becomes strictly more attractive to a riskneutral firm because the continuation value of learning  
[01:22:12] increases. However, this upside is mitigated by future transaction costs gamma since there's less incentive to experiment today if one's future ability to leverage that information um is is limited either because there's uncertain  
[01:22:26] supplier relationships or because they're transaction costs to continue accessing stock.  
[01:22:32] And so this learning value term competes with risk aversion in the model which creates utility costs from uncertain demand which if sufficiently large can flip the stocking uncertainty relationship from one that is efficiently positive to one that's  
[01:22:45] negative. And so to see this observe that if U is concave then a bad realization of profits will drive up the marginal utility of consumption. And so this covariance term will be negative if and only firms are risk averse.  
[01:22:59] um as soon as you do this discovery right if your competitor next door can also um can also stock motorcycles immediately right so then there is no upside to stocking helmets is that correct  
[01:23:13] >> um yeah that'd be right and that's the focus of of a follow-up project and just note that um there's at least some upside because I'll find that just relaxing risk alone is >> I mean that might explain a lot of I mean the questions that that have been coming up which is that there's  
[01:23:28] potentially a a little bit of a downside. There's absolutely no upside.  
[01:23:31] So, we just don't start like it's like a Danny Rodri has this old paper on why there's too little entrepreneurship in developing countries and he has an explanation like this.  
[01:23:39] >> Yeah. No, absolutely. I think it's like if you were the residual claimment to all of the monopoly profits. We'd probably see more experimentation with one helmet. It may be that the risk is sufficiently large with one helmet that that kind of risk aversion is binding.  
[01:23:51] What I'll say is that what you'll see is that I just relax risk and I find that that does sizably correct the market. So even in the presence of those types of business dealing externalities the the risk aversion I target will be important in its own right.  
[01:24:04] So next arrive three classes of predictions which I'll then guide design the experiments to test. So first I'll examine if risk aversion affects stocking behavior where the typical tool to test if agents are risk averse is something like a mean preserving contraction. Uh so something like  
[01:24:19] actuarily fair insurance. The complication in this setting is that we need a change which is only one period in duration so that it does not affect the learning value so that we can separately identify risk preferences um from the continuation value of learning.  
[01:24:33] So proposition one will state that a one period mean preserving contraction increases new product investment if and only if a firm is risk averse. And so this is a prediction for or test for risk aversion that's robust to both possible capital constraints as well as  
[01:24:47] learning value. And to preview where I'm going empirically, I'll design the insurance experiment to test it.  
[01:24:54] Second, I'll ask if learning about demand is economically important. And to do so, I'll consider the model's predictions about a return policy, which you can think of as just ensuring prices next period are equal to stocking costs and a reduction in supply chain  
[01:25:07] uncertainty gamma. And so if firms face demand uncertainty, then first with regards to their experimentation with the new product, so do they ever try stocking it? It should increase when a firm is offered returns since they mitigate the risk of losses. It should  
[01:25:21] increase when future supply chain uncertainty falls because the continuation value of learning increases.  
[01:25:27] A bit more subtly. It should um be unchanged when future supply chain uncertainty falls if the firm has access to returns. And the intuition for this is that if a firm can make returns then there's essentially no downside to  
[01:25:41] stocking helmets. So we should see that any firm which is constrained by uncertainty about demand should already choose to enter the market and therefore a reduction in gamma should not have an additional effect. And the reason that this is empirically valid or valuable is  
[01:25:55] because this is relatively unique to the model that I've written down whereas other forces which we could imagine would generate the second prediction like learning by doing generally would not leverage the second prediction.  
[01:26:06] And then most directly, if learning about demand is important, then we should expect experimentation to lead to learning. And so a one-time return offer would be expected to have persistent effects on new product stocking.  
[01:26:18] And so I'll design the learning experiment in order to test this class of predictions.  
[01:26:24] And then lastly, there's a question of whether returns solve risk aversion or biased beliefs. So suppose that offering returns in period t led to a subsequent persistent increase in stocking. Then this could be explained by the targeted mechanism that firms are riskaverse and  
[01:26:38] experience resolved uncertainty about demand. But it may also just be that they had misperceptions about demand which were corrected once they had experience in the market. And so if it's a story of misperceptions among riskneutral firms, we'd expect a  
[01:26:52] positive relationship between uncertainty about demand and experimentation reflecting learning value. Whereas if it's a story of risk aversion preventing experimentation, we would expect a relationship that's negative.  
[01:27:04] If it's this confounding mechanisms, we'd expect returns to matter principally for firms that had pessimistic beliefs about expected profits, whereas we'd expect it to matter for firms with very uncertain beliefs if risk aversion is what matters.  
[01:27:18] And then third, and most directly, we should expect experience to lead to positive updating of expected profits if it's a story of misperception versus a reduction in posterior variance about profitability. If risk aversion, demand uncertainty prevent the adoption of new  
[01:27:32] products. >> If you're attracted to any model has potentially an insurance policy. So if the firm would respond by choosing effort in terms of selling effort something like that, then this might not  
[01:27:46] make even for the guy that operating the insurance policy, right? So are you going to are you going to worry about that? Yeah, let me let me get to how I address this in experimental design, which I agree is is something that I will will rule out by my experimental  
[01:28:00] design, but is something that is a bit stylized since that may also contribute to the absence of such policies by by real world insurers. Um, and so I'll collect detailed belief data within the learning experiment so that I can disentangle whether any persistent effect can be attributed to risk  
[01:28:14] aversion as opposed to misperceptions. So let me now turn to the empirics beginning with whether risk aversion affects firm stocking decisions. So to test this I'll leverage the insurance experiment where the goal is to generate a mean preserving contraction of  
[01:28:27] profits. To do so I'll design an insurance contract where a firm stock helmets and fail to sell them. Then they receive a transfer to offset part of their losses. Our offer control firms helmet stock without access and insurance shops the ability to stock  
[01:28:41] with the contract where they'll then test if insurance increases helmet uptake. There are several challenges with making this offer mean preserving contraction in practice. The first is with capital constraints since charging uh upfront premiums would affect  
[01:28:55] liquidity demands. And so to shut this down, I'll offer firms a choice over future options. I promise control firms an unconditional payment and a follow-up. And I offer insurance shops the choice between the same unconditional payment or a free insurance contract which is state  
[01:29:10] contingent but pays out at the same period in time.  
[01:29:14] Second, there are concerns that offering insurance may send a private signal about demand, changing beliefs. And so to prevent this from confounding our inform both insurance and control shops about the existence of the contract and its random assignment procedure such  
[01:29:28] that the same information is presented to all firms and the only thing that differs is whether they can actually access the contract.  
[01:29:35] And then lastly, there may be concerns about biased beliefs or manipulation, especially this type of moral hazard.  
[01:29:41] And so let me address those on the next slide after I detail some of the design.  
[01:29:46] So to detail the insurance contract, I begin by promising an unconditional payment, a piece of eye to all shops um at a follow-up period, which is unconditional on both whether they decide to stock helmets or not as well as they whether or not they sell out or  
[01:29:59] not. Conditional on choosing to stock. Then I offer insurance shops the additional option to forgo PI for a hedging contract where if they stock and sell out they receive nothing but they that means that demand was high so they  
[01:30:13] get high profits whereas if demand is low and they fail to sell out they receive a thousand Kenyon shillings to offset part of their losses and then the trick is to calibrate PI such that insurance is a mean preserving contraction. What I do is leverage  
[01:30:27] beliefs about the suggestive beliefs about the probability of selling out conditional on choosing to stock where if I set PI to a thousand times the probability of failing to sell out then the expected value of the two offers is the same and so insurance would be a  
[01:30:40] mean preserving contraction. What I do in practice is actually increase the guaranteed payment by about 10% so that insurance would be strictly dominated if firms are risk neutral which makes the test stronger and creates robustness against measurement error in the  
[01:30:54] illicited beliefs. There are at least two remaining concerns. One is that firms may strategically manipulate the offer by um accepting helmets and then intentionally not selling them in order to get a payout or reducing effort. And so to to  
[01:31:08] prevent this, I prohibit restocking of firms collect the insurance payout and audit a fifth of the shops in order to make the design manipulation proof. And this is quite binding empirically as shops choose in practice to decline the insurance payout when they had one  
[01:31:22] helmet remaining in order to restock in several instances. And this makes the strict dominance relationship stronger.  
[01:31:29] Second, you may be concerned that beliefs are mismeasured. And so expost will find that insurance adopters would have been much better off on average having declined the insurance offer than opting into it. Which implies that insurance is strictly dominated um under  
[01:31:43] both rational expectations and the suggestive beliefs. And the beliefs are quite predictive of actual outcomes with an elasticity of selling out um with respect to the beliefs of point4 among those that adopt the contract and 02  
[01:31:56] overall >> the different expected profit. So the thousand is like the sample.  
[01:32:05] >> So then yeah no so the thousand is the what you get if you fail to sell out. So that's set for everyone >> based on the sample expectation.  
[01:32:16] >> The PI is calibrated based off of individual firms. That's right. Yeah.  
[01:32:20] And so it should if those are are measured reasonably well, it should be strictly dominated for every firm and then I add a buffer essentially to allow for measured question. And so I implement this design with a sample of  
[01:32:32] 350 firms in West Kenya. Um I recruited permanent retailers not selling helmets.  
[01:32:37] And so about a quarter of them held permanent employees. Um and they're they're a bit larger than the typical firm in the context because of their permanence. Um I offered all shops helmet stock after baseline and in a single follow-up survey. And I offered insurance to half of the shops uh  
[01:32:52] selected at random with stratification by market.  
[01:32:56] And so in the first major empirical result of the study, I find that offering insurance leads to a large and statistically significant increase in helmet adoption, which implies a rejection of firm risk neutrality. So I measured adoption over two time horizons. First firms immediate  
[01:33:10] decisions within 24 hours where I find that offering insurance leads to a two and a halffold increase in stocking. And then I gave firms two weeks to come to a final decision so they could do things like attempt to pre-ell helmets before  
[01:33:23] stocking to consumers to give them other options to to alleviate risk. And I find that there's about a 50% expansion in stocking that's statistically significant even after this longer time horizon.  
[01:33:36] Yes. >> Yeah, that's a great question. And so  
[01:33:58] the one thing I'd say is that I to a large extent follow the microeconomic literature. We're in development. The tests that that we employ across many areas of of firm behavior assume risk neutrality in a way that the interpretation of results depends on  
[01:34:12] them. And so for instance, one in like instance of this is in a misallocation literature where the variance in returns to capital across firms is interpreted as supply side access issues in in accessing credit markets. But this paper  
[01:34:26] suggests that may also just be a function of heterogeneous risk preferences. And then another is for instance in industrial organization um in in like competitive conduct testing those tests tend to assume profit maximization in order to identify if  
[01:34:40] firms are colluding versus competing in a way where collusion could also simply be a function of risk preferences of firms. Um the the other thing that I'd say is that even if there there may be an assumption that risk aversion may matter to some extent there's very little empirical evidence about how  
[01:34:55] important it is. And what I find is that it seems to be a very important distortion in retail markets that's likely to consequentially affect incentives for manufacturers to introduce goods in the first place. And so kind of I think that the quantitatively that matters outside of  
[01:35:09] just the what buyers might be. >> Uh so the kind of older literature on agriculture sort of the sharecropping literature also spends a lot of time on risk conversion. That's one potential justification of sharecropping. But it kind of differs from your point because  
[01:35:23] in that situation we assume tenant farmers are risk averse but they're risk averse in a sense where they're not learning anything they're just risk averse is the weather could be variable.  
[01:35:32] In your case that could be a mechanism in addition to the learning about demand. Is there any possibility of sort of separating out sort of general risk aversion about the risky returns with full knowledge versus this fact that you don't really know about this particular  
[01:35:47] product of the health? >> That's precisely where I'm getting to in the next experiment. And so let me me speed up a little bit and I might defer some questions so that I can get to that. But good but great question.  
[01:35:55] >> Um and so then the next thing that I do is look at heterogeneity in responses to insurance by firm characteristics. And I find that even larger and older firms uh exhibit significant responses to insurance. And in fact, if anything, the  
[01:36:09] responses seem to be a bit larger, which is explained by the fact that they have fewer other binding constraints. And this suggests that even when firms have resources available to invest, many of them don't uh invest in riskier opportunities unless they're they're protected against the downside. This is  
[01:36:24] consequential to policy design since it suggests that policies like cash transfers could be made much more effective by bundling them um with something that would would lower the risk of higher return investments um instead of giving cash on its own.  
[01:36:38] So we now turn to this question of whether risk aversion prevents retailers from adopting new products. And so exactly as the point was just made, this result that I've shown so far could be explained by the fact that demand for all goods is just really volatile. And so it's true of all goods across the  
[01:36:51] market or it may be unique to the fact that helmets are new. And so it's a a feature of not having complete information about the market. And so I test so the learning experiment examines whether risk aversion prevents experimentation needed to learn about  
[01:37:05] new products and in doing so traces out the effects of risk aversion on stocking decisions over a longer time. Room.  
[01:37:24] in that sense like it's actually like we don't think about science.  
[01:37:39] >> Yeah. No, so I think it's both. I think it's that conceptually we've made the argument that competition may drive out risk averse firms because there there are many of them and so risk neutrality may be a reasonable first order approximation and so I think the empirical results speak to the fact that  
[01:37:53] it is very important um in a way that may may affect the results that we get and then then second so I think it both has that kind of academically and then also I think it's just very important policy and it suggests that retail markets just don't function very well  
[01:38:07] because of risk aversion in a way that I don't think has been bit studied for >> um so what I do here is I think that there  
[01:38:22] are some settings where we can just design tests in a way which is robust to firms being riskaverse and so I think that's kind of the gold standard. So like one example is in in contact testing and this follow-up work what we do is just do the typical conduct testing mechanisms on top of offering a  
[01:38:36] consignment policy to just eliminate risk in the market in the first place.  
[01:38:40] And so I think some like creativity in terms of thinking about tests in a way that um does not require assuming anything about risk aversion is robust to it is is probably the best way to do this. In other settings, just estimating risk aversion using parameters could be  
[01:38:53] valuable to to get the sensitivity >> deferred.  
[01:38:59] But have you done some sort of calibration of what level given the minimum stocking charge or other cost like what level of risk converging would justify this?  
[01:39:06] >> I have the short answer is that if you assume firms are very capital constrained, it's pretty modest at like a CRA coefficient of around one which aligns with kind of lab choice schemes very well. If you assume that they can smooth consumption over their lifetime, then you would you would need loss  
[01:39:20] aversion. So in the learning experiment, I study a sample of shops located near the helmet factory where the market features are ideal to test if risk aversion undermines learning. Uh mainly shops had access to helmets for more than two  
[01:39:34] years from the manufacturer. So other barriers to adoption were limited. Um and yet there was very limited retail adoption of the product at under 3% of screen shops. So it's a setting where the natural diffusion and to some extent fails. I offer helmets at prevailing  
[01:39:48] market prices and then implement two treatments to test if risk aversion prevents learning about helmet profitability.  
[01:39:54] So the first treatment is a return policy where in phase one control firms are offered helmet stock with no ability to return unsold stock. Treated firms are given the ability to return unsold stock for a refund at a follow-up survey.  
[01:40:07] Then the key feature of the design is then there's a second phase where I again offer all firms access to stock at prevailing prices with no returnability for anyone in order to set up a a test for persistence and stocking conditions are symmetric.  
[01:40:20] The second arm is what I call a supplier commitment policy where it's the same as the control in phase one. So stocking risk is unaffected and it differs in that I commit at the beginning of the experiment that once it ends about seven months later I can help shops restock  
[01:40:34] from the manufacturer which raises the the value of learning since it makes information more useful. Um and then finally I give a fourth of shops access to both of the interventions in order to test the model's predictions that the supplier commitment um shouldn't have  
[01:40:48] much of an effect if firms can already access returns. or what does help mean that subsidy >> when you say help restock?  
[01:40:56] >> Good good question. So firms were just concerned that because they were small the manufacturer like would not return to them and continue offering them stock even though in principle they should. Um and so what we said is we'll just make sure that they're actually coming and if they're not accessing you, we'll call them and have them come back. And so it  
[01:41:11] speaks to just like a lot of um kind of skepticism about supplier relationships.  
[01:41:16] This actually seemed to be a real thing in practice. I was skeptical Xanti, but that actually did happen.  
[01:41:23] So, I'll first examine if learning about demand is important to product adoption.  
[01:41:28] To do so, I'll estimate a regression of stocking in phase one on assignment to returns, the supplier commitment, and an interaction between the two. Um, where the model predictions are that uh return should increase experimentation, the  
[01:41:42] supplier commitment should increase experimentation by increasing the continuation value of learning. And recall this third prediction that the supplier commitment shouldn't matter if firms can access returns because it's no longer risky to stock. And then the  
[01:41:55] final prediction is that experimentation should lead to learning. And so I'll test this by estimating regression of stocking in phase two um and then examining if there's a persistent effects of returns.  
[01:42:07] So first I find that the phase one effects match the experimentation predictions. The control stocking rate is about 6.8%.  
[01:42:14] This jumps, this more than doubles when firms can access returns. There's about an 80% increase in stocking when firms are given the supplier commitment alone.  
[01:42:24] And then consistent with this prediction that the supplier commitment shouldn't have an effect firms can access returns, the stocking rate conditional and having both of the interventions um looks approximately the same to to the returns policy alone.  
[01:42:37] And so, so far I've shown you evidence that risk aversion affects firm stocking decisions and that firms perceive uncertainty about demand um from helmets they may be able to overcome from learning. But I haven't shown that this leads to a distortion where firms end up  
[01:42:50] stocking a different set of products than they would if they were engaging in more experimentation.  
[01:42:55] And so another major result of the study, I find that the effects of the returns policy are large and persistent after the intervention ends where there's about a 70% increase in phase 2 stocking or seven percentage points associated with having been assigned  
[01:43:09] returns in phase one. And both revealed preference evidence and firm reports suggests that this unlocked a growth opportunity for many firms. So what I'm showing you in green is that the intervention doubled the share of shops that decided to permanently enter the  
[01:43:22] helmet market by the end of the study. twothirds of phase one adopters restocked with 80% intending to. And to give you a sense of the magnitudes, uh firms reported that helmets accounted for about a 10% of entrance profitability on average with plans to  
[01:43:37] expand at 20%. And so these are self-reports, not treatment effects to be clear, but this just suggests a modest but permanent increase in profitability.  
[01:43:51] absent your insurance policy is moving downwards there is some substantial risk.  
[01:43:59] >> Yeah. So I do have evidence in the paper that a small number of firms suffered pretty big losses because they didn't sell anything. Um and then there's so I'm underpowered to detect increases in overall profits because the first stage is like a 10 percentage point expansion  
[01:44:13] in stocking which is really sizable in terms of adoption but really difficult for kind of looking at noisy profit measures.  
[01:44:25] Uh I haven't looked at quantile treatment effects. I just do is like the self-reports about the number of losses.  
[01:44:31] And they're firms that both stocked and didn't sell anything. And then if you ask them how much did this hurt your overall profitability, it's a pretty big negative profit effect. And so it is more suggestive that the experiment wasn't kind of designed to look at treatment effects on profits because they're noisy and because I'm looking at  
[01:44:45] a kind of a first stage in the margin of like 10 percentage points. But that is a limitation of the paper.  
[01:44:50] >> Are you making a pricing decision or is that >> Yeah.  
[01:44:54] >> So there's a learning by doing component to everything that that insurance will also help.  
[01:44:58] >> Yeah. So it could in principle I find very little evidence of learning by doing. Let me defer that until afterwards in the interest of time so I can get to a few other results but that was was a possibility and so I have have several checks I'd be be happy to talk about after.  
[01:45:13] And so I've shown that inducing shops to try selling helmets with return policy leads most firms to discover they're profitable. And this could be rationalized by two stories. First, it could be that risk aversion demand uncertainty prevent experimentation as I've been been arguing. But it could  
[01:45:26] just be the story of misperceptions that firms underestimated how profitable helmets were. And so you use belief measures to test what types of firms the returns policy caused to try stocking and then how the beliefs of adopters evolved. And very briefly, the beliefs  
[01:45:41] are constructed using a frequentist approach with visual aids. And so I collect not only a measure of firms expected sales or expected profits, but also measures of individual firms uncertainty about profitability, um, which I'll encode in a standard  
[01:45:54] deviation of an individual firm's beliefs about sales.  
[01:45:58] And so first, the model predicts that if it's a story of misperceptions, the control relationship between uncertainty and stocking should be positive because of learning value. Whereas if risk aversion is undermining experimentation, we'd expect it to be negative. So I estimate a regression of phase one  
[01:46:13] stocking on expected sales standard deviation of a in individual firms beliefs about sales and then these interacted with returns. So the first result is that I find that a one unitit increase in the standard deviation of  
[01:46:25] sales is associated with about a 90% reduction in the probability of stocking helmets in phase one which is exactly in line with the prediction we would expect from risk aversion but not if it was a story of misperceptions.  
[01:46:40] Second, we'd expect the shops affected by returns to be those with relatively pessimistic beliefs if it's a story of misperceptions versus those that view stocking as risky if risk aversion is the principal force. Looking at the interaction between returns and expected  
[01:46:54] sales, there's no evidence that the policy crowds in firms with systematically more pessimistic beliefs.  
[01:47:01] But when we look at um the the same measure with variance particularly with the most granular controls in column three which includes fixed effects for very granular buckets of expectations there's evidence that  
[01:47:14] the policy causes firms that are much more uncertain about demand to try stocking and kind of in particular note that if you take the point estimates seriously this suggests that it flips the relationship from one that's negative suggesting that risk aversion dominates to one that it's positive that  
[01:47:29] is firms that are more uncertain are more likely to talk, which would be consistent with it restoring firms internalizing the the learning value of experimentation.  
[01:47:39] And then most directly, we would expect the change in beliefs with experience to be one of positive updating of expected profits. Um, if it's a story of misperceptions, whereas we should see evidence that there's a contraction of posterior uncertainty about demand if it's the fact that firms have overcome  
[01:47:54] uncertainty in a risk averse that explains the persistence.  
[01:47:58] And so to test this, I estimate a regression of the change in beliefs from baseline to the follow-up surveys on phase one stalking where I instrument for phase one stalking using treatment assignment. So that I'm estimating a causally identified local average treatment effect of experience on  
[01:48:12] beliefs among the compliers with the intervention where note that the compliers are grouped with relatively uncertain but also relatively optimistic beliefs from the priority results. And so first what I'm plotting in blue um is the the point estimates for  
[01:48:26] expectations. And so the takeaway is that there's no evidence that expectations become systematically more positive among adopters. In fact, if anything, the point estimate is very slightly negative at the endline survey.  
[01:48:47] um and statistically significant where the poolled point estimates combining the two surveys implies that there is about a 73% reduction um in the variance of beliefs associated with gaining experience in the market. And so you may be concerned that this is just a feature  
[01:49:01] of like experimental demand effects. And so what I do um in the paper is show that firms also become much better at forecasting their actual realizations once they have experience in the market.  
[01:49:12] And if you back out an implied reduction in uncertainty about demand from that, you get almost exactly the same estimate coming into about an 80% reduction in uncertainty.  
[01:49:21] And so to recap, the beliefs match the predictions if risk aversion explains persistent stocking effects. Whereas they violate all three of the conditions which would be needed to rationalize the results based off of mean beliefs changing.  
[01:49:33] >> Well, reduce their variance.  
[01:49:40] Um, >> you have two.  
[01:49:46] >> Yeah, that's a good question. I don't think they're super correlated. I'm struggling to remember exactly what the result is. I did look at that at some point and thought that it wasn't um particularly consequential. It looks like changes in mean are both relatively  
[01:50:00] modest. Um, and also descent kind of seem to be be fairly orthogonal to uncertainty. The bigger predictor of contracting uncertainty is just like whether they were good at completing the the mechanism and kind of spreading it broadly across time. But then conditional on like articulating  
[01:50:14] uncertainty, well, it seems like pretty much everyone contracts substantially once they have experience.  
[01:50:20] And so these results suggest that risk aversion caused a breakdown in firm learning preventing the discovery of a profitable product.  
[01:50:27] And so to conclude in my last minute, I've um shown two experiments indicating that Kenyon firms are risk averse, preventing new product experimentation from occurring efficiently. The insurance experiment provides evidence that risk aversion affects firm stocking  
[01:50:40] decisions. And the learning experiment indicates that this can undermine the discovery of profitable products preventing them from diffusing efficiently.  
[01:50:49] So I think there are two important policy implications from this as well as academically. The first is that the slow retail adoption of products may harm incentives for upstream firms to innovate in the first place. And so the ex to the extent to which risk aversion slows or prevents the retail adoption of  
[01:51:03] new products that reduces the residual demand available to manufacturers which may prevent them from either taking products which were introduced in rich countries and introducing them to consumers even that may demand them um or inventing products that are specific to low-income countries in the first  
[01:51:18] place. And then second, risk aversion may matters for many areas of small firm behavior and it's I think something that has been kind of underststudied to date.  
[01:51:27] For instance, unknown returns to machines may lead to less manufacturing adoption of new technologies and the fact that hiring an impermanent employee involves paying them a known wage, but uncertain returns may contract for confirmed consolidation and expansion in  
[01:51:40] terms of dimension of employees. Since these are of course kind of speculative but I think speak to the directions that the taking risk aversion seriously could help um expand our understanding of firm behavior. Thank you so much and welcome any questions during break 11.