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.
=

# Female-Targeted Hiring Subsidies, Firm Learning, and Women’s Employment

Authors: Mimosa Distefano, Lorenzo Incoronato, Anna Raute  
Discussant: None  
Video: https://www.youtube.com/watch?v=0mE9GNhwdY8&t=24149s


## Talk (06:42:29 – 07:27:18)

[06:42:29] >> Yes. Yes. Do you take it? >> [cough]  
[06:42:47] >> Okay. Uh hi everyone. Uh and many thanks for uh having us in the program. I'm very excited to present this paper which is joint work with Mimosa Dano and Anar.  
[06:43:01] Uh so uh despite improvements in recent uh years uh we still see quite stubborn gender gaps in labor markets around the world and as many papers including many  
[06:43:13] that we've seen uh in in this in this uh uh workshop uh show that a lot of these gap uh gaps arise when children arrive.  
[06:43:22] And uh in response to this uh policy interventions have largely focused on the supply side uh meaning policies such as parental leave or or childare uh interventions and uh evidence as to the  
[06:43:36] effectiveness of these policies in bringing women back to the labor market permanently is at best mixed. Uh now a possible reason for this is that is coming from the demand side. So even in the presence of these policies that  
[06:43:50] support women in the return to the labor market after having children uh firms may simply be reluctant to hire workers that have been out of the labor market for for some time. And this can be from for for various reasons. One could be  
[06:44:03] that uh long non-employment spells might signal loss of human capital and firms might have may be uncertain about the productivity of these workers. A lot of it may also arise because of stereotypes and stigma. Actually, there is some  
[06:44:18] research in sociology that shows that firms believe mothers that have taken parental leave to be potentially less committed and and and therefore may be uncertain about their productivity. Then regardless of the reasons for this um to  
[06:44:32] the extent that firms do not know are unsure about the productivity of uh women that come back to the labor market uh then this might generate what we in the paper refer to as learning trap which is a quite simple concept to the  
[06:44:45] extent that a firm can learn about a certain group of workers by uh having exper by hiring and and interacting with these workers. If firms are very uncertain, they will never manage to interact with these workers and hence they will never learn. Okay. So the  
[06:44:59] basic idea of the paper is then uh understanding whether policies that can stimulate firm demand for these workers and nudge them to actually experiment with these workers can break this learning trap. Okay. So uh what we ask  
[06:45:13] in this paper is so we focus on a specific policy which is hiring subsidy.  
[06:45:17] All right. And we ask whether iring subsidies can persistently shift firms hiring uh towards the target group of the aring subsidy which would be women that have been that have taken career breaks. So we study a policy that was  
[06:45:32] implemented in Italy quite a peculiar policy was named bonus don which consisted of a 50% payroll tax cut for firms hiring women uh with at least 24 months of unemployment and the policy was temporary. So it lasted 12 to 18  
[06:45:47] months. I will give you more detail. So in order to sort of guide our empirical analysis, we we develop a very simple uh uh uh conceptual framework where firms are uncertain about the productivity of  
[06:45:59] a group and then a hiring subsidy can uh compensate for the cost of experimentation and then allow firms to learn. Um and then we then turn to new data and we use quite unique social security employer employee data for for  
[06:46:14] for Italy. uh which has a unique feature which which is that it allows us to direct directly observe take up. So we know the workers and the firms that so the workers that are hired through the subsidy and the firms that make use of it and empirical design is quite simple  
[06:46:28] is a much different study designed around uh take up. So let me tell you what we find in a nutshell. uh we find that firms that make use of the policy uh end up persistently hiring more women from the target group even after they do  
[06:46:42] so compared to you know control group. Uh so and the coefficient so we find that treated firms hire roughly 3% more women from this group compared to control firms in the long run and this comes at no expenses in terms of labor  
[06:46:56] productivity. So this despite hiring women that have been out of the labor market for some time these firms do not appear to to lose in terms of of productivity. Uh what we also find and which is also consistent with our uh uh  
[06:47:11] learning framework is that these persistent effects depend on the initial experience that firms have with these workers. namely firms that uh uh hire a good match and I will tell you later what we mean by this are actually those  
[06:47:25] that show this persistent effects. All right. Uh I probably won't have time to talk about the individual level effects today. Uh what we find so we also focus on these workers and we find that women that are hired to the subsidy are more  
[06:47:38] likely to stay employed both at the iron firm and any firm. So we show that the subsidy actually seems to have fostered longran labor market attachment. Uh however we don't find wage effects uh which is actually consistent with a lot of research on the effect of iteing  
[06:47:52] subsidies which is that firms simply pocket the tax cut and do not pass it through to workers in the form of of higher wages. So uh let me briefly tell you how we believe we contribute to the literature. So there is a long-standing  
[06:48:06] literature on subsidy uh which has actually gained more traction in recent years with the advent of administrative data. uh to our knowledge this is the first paper that thinks of iron subsidy as a way for firms to learn about a group and then generate persistent  
[06:48:20] effects. There is one exception which is this uh barometal paper which also thinks of subsidy as a way for firms to learn but they focus on learning about individuals. So individual worker productivity whereas the key novelty that we believe we're bringing here is  
[06:48:34] that firms can use this subsidy to learn about the group which is what will then deliver persistent effects. That is our question. Yes.  
[06:48:44] Just wondering in this context is there also cost of hiring the workers or laying up the workers because maybe if there is a cost they just stick with it because they don't want >> you can hire. So the the policy allows  
[06:48:57] firms to hire workers also on temporary contracts. So if they want they can just hire workers on a uh like a fixedterm contract.  
[06:49:04] >> So no compensation if they later >> uh unless they hire them on a permanent contract. Yes.  
[06:49:10] >> But they're free to do. they shift into harmony.  
[06:49:12] >> Yes, we do >> very soon or >> uh within a year.  
[06:49:16] >> Yes. Um so um we also contribute we believe to a kind of a parallel literature that studies policies that aim to increase representation of minorities. For example, affirmative action policies.  
[06:49:30] There is for example work by by by Koran Miller that shows that affirmative action policies can have persistent effects uh even if they're temporary.  
[06:49:38] And uh we have we believe so we we we we kind of present that in subsidies as a policy that can also have persistent effects. However, with kind of a different mechanism and I will tell you a little more about it at the end. And  
[06:49:52] we also contribute to in a kind of a minor way to a more theoretical literature about experiential learning that shows that firms long behavior can be driven by their experiences with with workers. and we uh contribute to it by  
[06:50:06] showing that this experimentation can actually uh be uh favored by policy interventions.  
[06:50:13] So >> can I ask you a question?  
[06:50:14] >> Sure. Um so I was a little bit surprised when you uh said said something about well uh firms do not pass through >> Mhm.  
[06:50:25] >> uh the subsidy right because I thought you know so you are as essentially you want firm to hire these women that are returning to the labor market.  
[06:50:36] >> They probably have depreciated [snorts] right and so the firms are really uncertain. Sorry, the word is so firsth.  
[06:50:49] [laughter] >> So you were you're surprised about the lack of pass through.  
[06:50:54] >> Yeah. Because the because the subsidy is exactly to be the mechanism, right? So what you want these women because of being out they're already, you know, they're be may be less productive.  
[06:51:06] >> So you would expect firms to pay them even less. No, that that that the subsidy would be enough to get them with a foot in the door, get all this group in the door and the firm >> uh by would learn that turns out that  
[06:51:21] their average quality is pretty good >> and then this then kind of propagates.  
[06:51:26] So it's it's kind of it's the subsidy that is trying to convince the firm right to >> so >> try to try.  
[06:51:34] >> This may partly answer your question. Uh so we find was there also a followup?  
[06:51:39] Yeah. >> No, I think I think that it gets to the previous question which is you hire someone whose human capital is depreciated and you may have had very little  
[06:51:54] experience with someone let's say who was a physical therapist for a while and then was out for a while. you're not quite certain exactly what skills have depreciated.  
[06:52:07] >> So, um you might hire them at some lower wage. Okay. But now you're getting this subsidy. The question I think goes back to the question before which is I can  
[06:52:21] hire the person >> if I have the ability to just get rid of the person. Okay. that there may be uh either actual or firm level reasons for  
[06:52:36] not wanting to hire people and then let them go.  
[06:52:40] >> Mhm. >> I mean, it creates sort of a bad environment. Depends how many people are in the group. Okay.  
[06:52:47] But but you one wouldn't think that something like this, you know, where you do have a sense that this person had been a physical therapist and I was out for a while. You can see that why the  
[06:53:01] person was out for a while. You know, you know whether the person had a child, for example, why should there be such reluctance that requires you to have substance?  
[06:53:12] >> Yeah. So, I mean, I think this uh we could talk about this for very long. So I try to keep keep it concise. So uh uh so we find um that if anything workers  
[06:53:25] that are hired through the subsidy I mean there is mixed evidence but they appear to be slightly positively selected. Right? So uh that's a a first answer and I also agree with you that the the way that these workers are hired  
[06:53:38] uh um you know the fact that they can then lay them off afterwards is actually also part of the answer. Also uh I believe there is research that shows that uh human capital does not depreciate much right during the  
[06:53:52] >> but you know that I that's one it depends upon the field >> it depends upon >> the yeah on the on the actual job. Yeah.  
[06:53:59] >> But and also you're trying to you know you're asking this firm to try to learn that >> if there is a planner that knows these women are high quality >> but the firms are not  
[06:54:12] >> um willing to take the risk of hiring them. So that's what the subsidy >> was there question. Um so I don't know  
[06:54:25] how large this program is but are you going and I know you focus kind of on um productivity at the firm level these guys who are being affected by the policy but are you going to be able to say anything about general equilibrium effects and the impacts on men so is this just like reshuffleling the jobs  
[06:54:40] between them actually like growing the firm >> good point so I will tell you a little bit more about the size let me anticipate that the size is tiny >> it's tiny which is quite surprising to us but I would also be able to tell you something about men very little happens to  
[06:54:54] uh this will come with the results. >> So I guess you had plenty of time to look at the road map. So let me [laughter] uh let me tell you a little bit more about the policy. Uh so first about the setup, Italy is suddenly uh a very good  
[06:55:09] setup for us because it's a country where uh there are still quite big gender gaps especially in employment rather than than in earnings and a lot of these gaps are actually explained by the child penalty. Uh Italy also has a dual labor market where we have insiders  
[06:55:23] which have uh permanent and very protected jobs and outsiders which have more intermittent labor market participations and women are actually over represented in in the in the outsider category. So in order to uh address this a policy this this bombs  
[06:55:37] don policy was implemented in 2013 as I mentioned it consistent of a 50% payroll tax cut uh to firm social security contributions. it amounts roughly 10% of labor cost and it's similar in size to  
[06:55:50] other subsidy hiring subsidies cited in the literature. So what was the target population uh was women that had been out of the labor market so none employed for at least 24 months. Okay. And the subsidy lasted for 12 months on  
[06:56:05] fixedterm hires. So temporary contracts which could then which then increased to 18 months if these uh were converted uh to uh permanent jobs or if they were permanent jobs to begin with. Uh there  
[06:56:18] were also important requirements on the firm side. Uh the main one was that for a firm to be eligible to use the subsidy, it should have not shed workers on net in the previous 12 months. So this is actually a quite restrictive  
[06:56:33] requirement especially at the time. So Italy was going through a a recession at the time, right? So and we believe this actually and this goes back to to to to points which is that this policy was very uh was not used a lot. So we we  
[06:56:46] compute on average between 2013 and 2019 which is the end of our sample that only 4% of eligible workers were hired through the subsidy which is tiny. All right. So to be uh uh fair uh the take  
[06:57:00] up has been increasing over time if we computed this today. I believe this will be much larger. The policy is still around. It has also become more generous actually. Now it's about it covers 100% of P taxes. So uh it's basically it's  
[06:57:13] huge but you know at the time it was not so many just 180,000 workers had been hired through the policy. So very little.  
[06:57:23] >> Yes. So can a firm repeatedly hire workers with this subsidy and are you going to speak to whether they the effects that you find are from these like repeat subsidized hires?  
[06:57:34] >> So thank you for this question because this allows me to tell something that I would have told at some point. So this is precisely uh uh so this kind of persistent effect is precisely what we care about in the paper. uh we would love to distinguish  
[06:57:48] this is like top one in our to the list to distinguish uh between firms continuing to hire to the group through the subsidy which will also be might be doing simply by costsaving measure they can potentially or through a learning story. uh I will go back to this when I  
[06:58:02] you know towards the end of the talk. Uh the unfortunate part is that at some point the Italian social security issues which provides fantastic data just cuts the access and we would like to we would really love to do that [laughter]  
[06:58:14] but uh and that's but again let me tell you that towards the end I will show you evidence that reassures that the the persistent effects that we're seeing do not seem to come from this kind of cost related reasons but really from this we  
[06:58:29] believe at least from this learning part. So I will go back to this point later. So, I was just going to ask too, u I mean, the extent to which a woman who's been out of labor force for two years or more is signaling dedication seems like it depends a lot on the availability of child care for kids  
[06:58:43] under two. So, defer it if it's coming, but how how easy would it have been to get child care for for those very youngest kids?  
[06:58:50] >> So, um so child care is relative I mean Italy is as you know depends also where in Italy you leave.  
[06:59:00] >> Yeah, exactly. So um in certain areas of the country there's you know broadly based on on the network on family network and and and so on whereas in other in like more urban areas and in the north you have larger availability  
[06:59:13] of childare I would say it's fairly uh available I should have a better answer >> child could have aged out of child care >> potentially yes >> it's at least 24 months  
[06:59:26] >> it's at least 24 months potentially yes >> sure but just for for the you know for at least for at least 24 of those months. If it's an environment where child care is very scarce, that that seems like >> Yeah. And that's why she's out of work.  
[06:59:39] >> Okay. Now she's now she's going back, >> right? What what are we learning about >> heavily depreciated?  
[06:59:47] >> But thanks for asking me because I I need I should never say depreciation.  
[06:59:50] [laughter] Sorry. >> This is on to me anyway. So but thanks for the question. I think I I want to have a more precise answer to to your question. uh so okay u let me tell you briefly about the data it's much  
[07:00:04] employer employee data from uh for for Italy it's on the universe of private sector workers and firms and again that the crucial piece of information for us is that we can observe uh take so we know when a worker is hired through this  
[07:00:18] policy and so we can characterize these hires and we actually use this unique feature uh we also we have plenty of information on workers and firms and we also tell you that and This also partly speaks to the previous question. Uh we  
[07:00:31] observe maternity leave uh spells. So we identify mothers as workers that have had uh maternity leave in the previous four years. So when we when I talk about mothers later I'm talking about employed  
[07:00:45] uh mothers of young children. Uh there is new data which as you might have understood we cannot access right now.  
[07:00:51] Uh that actually le allows us to observe the full population of mothers in Italy. uh which we can then link to mystery records and and this for sure uh we will include this whenever we have access to the data  
[07:01:04] again um so let me characterize this work this firm so here in this table I'm comparing on the left column firms that have used the policy at least once between 2013 and 2019 and firms that  
[07:01:19] never used the policy and in time let me just highlight a couple things firms that use the policy appear to be more female friendly right so they have a larger share of women uh both amongst employed workers and amongst new hires.  
[07:01:32] And they also seem to be on a uh on a growth path. So they've been hiring more. These are characteristics before uh so in 2012 uh and they also seem to be paying a bit less their workers. Um  
[07:01:45] in terms of sectors here we are talking about low think of this as lowkilled services jobs like hospitality and retail. So this is the most predominant uh sectors that are uh that that that feature in our analysis. I don't I'm not  
[07:01:59] sure about time. So um maybe because it's it's okay just to have an idea of where I am.  
[07:02:05] >> 26 26 minutes. >> Okay. Left.  
[07:02:07] >> Okay. There was also a question. Yes. Just wanted to test the reason why you don't have the share of women who >> I can give the the precisely the same answer that I gave before which is we would love to include we will definitely  
[07:02:21] include the share of women say from the target group here because we want to precisely see that and this also speaks a lot to the learning part and uh we will definitely have this number in this table uh when we can it's super important.  
[07:02:34] So let me tell a little more about the conceptual framework. Uh uh so the key so here uh the key of the framework is to kind of uh showing us how we think about how a high subsidy can foster firm learning about a group. This is the  
[07:02:47] basic idea and this again builds model on models of experience based learning and the starting point is that firms are uncertain about individual productivity which we call theta i. They know about workers group membership and they take  
[07:03:01] group membership into account when making hiring decisions as in standard discrimination models. The key novelty here is that firms are also uncertain about how worker from group G matches with firm J. All right. And so firms can  
[07:03:13] also learn about that. Okay. Now uh the hiring process occurs in two stage.  
[07:03:19] There is an interview stage. I will show you this more in detail but this slide just for a broad overview. So uh here the interviews so we don't observe interviews uh let me make it clear they're just a way for us to generate  
[07:03:33] the learning trap in the model as I would show you. So upon interviewing a worker firms observe a noisy signal about the workers's productivity and forms of aposterior this is standard vision updating and then they will decide whether to hire the worker or not. Now the basic insight is the  
[07:03:47] following. If a firm is very much uncertain about this theta GJ so about how a worker from group G is productive then a firm will not or more precisely if a firm believes that this theta is very low then they will not even  
[07:04:02] interview the workers. All right and so they will never manage to learn about that worker and therefore never learn about the group and this may have firm stuck into this learning job. Now what the subsidy does is simply raising the value of interviewing workers from the  
[07:04:16] group and thereby therefore kind of incentivizing experimentation and therefore learning and to the extent that this allows a firm to learn about the group then this can generate persistent effects. Okay, this is the basic idea of the model. I will give a little bit more detail now but you know  
[07:04:30] this is the the kind of the flavor. >> So yes, this goes goes back to my GE question a little bit. So if there's enough workers that are out there for me to hire then maybe I don't care  
[07:04:44] about the learning trap. Can you >> So, so um maybe I don't care about the learning trap if there are enough workers of >> I mean on from other groups.  
[07:04:55] >> Yeah, but like so so what what in your model is telling me that I you know that that's not going on?  
[07:05:02] >> Nothing in my model >> and just a short answer. Uh just a short answer is yeah but I I I see your point.  
[07:05:10] Uh >> is there a bonus Germany?  
[07:05:13] >> There is a bonus Germany in Italy. Yes, there are so many bonuses in Italy like uh it's very hard to keep track of all of them.  
[07:05:20] >> Yes. [laughter] You know if you don't even >> there is a bonus south so there is many.  
[07:05:28] >> Sorry. >> Yes.  
[07:05:28] >> Going back to bonus Germany actually. So your group is just people who were away from workforce for the last 24 hours.  
[07:05:37] Actually >> 24 months [laughter] >> it's late. No worries.  
[07:05:45] at least 24 months. So, this can be just for men as well.  
[07:05:50] >> No, no, no, no. It's target to women. >> I understand.  
[07:05:52] >> Ah, okay. Okay. Yes. But like you can easily have another policy for men and then this >> worse >> because the question is why were they away?  
[07:06:04] >> And it's often because [clears throat] they were in jail or on the land.  
[07:06:10] [laughter] >> The base case is jail. No, but yeah.  
[07:06:19] >> Can you can you increase the bridge because I didn't hear.  
[07:06:24] >> Now it's working. >> Yes. So because why don't you show us the the summary stats for the firms who are employing the people they are actually type of firms which have more women to start.  
[07:06:35] >> Yes. So they know actually how female workers are potentially how mothers work their you know limitations constraint etc. So the the group maybe Claudia's point that men who were out of workforce  
[07:06:50] for two years are different characteristics of maybe a set of woman employees would be more helpful like what type of people we are talking about which are >> I think this goes back to the previous comment we want to characterize this firm better along this dimension. And I  
[07:07:04] agree with you. I agree. It's you know it's it's true like we I think so far we have not found a smoking gun for you know how precisely characterize these firms and I think we should produce a bit more statistics on that. I fully agree.  
[07:07:17] Uh so um you've had enough time to look at how workers productivity looks like.  
[07:07:24] uh let me tell you that there is a baseline component like a prior on the group and then there are the two tas which are individual productivity and group productivity and the firm must learn about these objects as I mentioned there is an interview a firm space a  
[07:07:37] cost kappa to interview a worker and upon interviewing they observe a noisy signal y tilda this is standard discrimination you know with bian updating the firm will form a posterior expectation about that worker's productivity which is that y hat and  
[07:07:51] then decides whether to interview a worker from group G if the expected value of doing so uh if the value of doing so is which I will define in a in a minute is bigger than the cost of interviewing then the firm will decide whether to hire the worker and they will  
[07:08:05] do so whether the expected payoff of the worker is bigger than zero and the expected payoff is simply the worker's productivity minus a fraction alpha which is the what the firm's pay to the worker right the wage is alpha times the what a firm believes the worker to be  
[07:08:19] productive and this allows us to define the value of interviewing workers from group G which is just the expectation of the surplus the different will get from doing so. All right. Now for the target group this value we assume is smaller than the cost of interviewing. So this  
[07:08:34] is the learning trap as as I as I explained. Now what a subsidy does and sorry if I'm rushing a bit here. I just want to show you the empirics is uh a subsidy. So the high subsidy the bonus don subsidizes a portion beta of the  
[07:08:47] wage. So this pushes up the value of interviewing workers. And if this is high enough then now the firm will uh will it will go above the cost of interviewing the workers and the firm will then begin interviewing workers from this.  
[07:09:00] >> This is a very strange group though. >> It's one thing if I say I don't want to hire high school dropouts or if I don't want to hire people of a certain race or people of a certain sex but I would have  
[07:09:12] to actually know a lot about this person before I put them in this group. I would have to have their CV. And many of us have looked at résumés and CVs and realized that we can't really figure out  
[07:09:26] whether someone had a gap of a certain number of years.  
[07:09:31] >> Well, at least the way we uh >> I'm just saying that it's a it's an interesting group. And what I'm suspicious about is I have a subsidy for this and now I'm going to go out and  
[07:09:44] find these people. M well I mean I mean this was the goal of the policy right and and at least and this is also based as I mentioned at the beginning on some research in sociology firms believe mothers that come back to work to be for some reason less  
[07:09:59] committed potentially because they have to take care of of their children and so this might generate these low priors from firms but you know probably we can do a better job at characterizing this group which is >> no in fact what you're saying is a  
[07:10:13] little more you're saying that these These firms believe that mothers with children of certain ages are going to be less productive. Not that women who are out of the labor force who are applying.  
[07:10:27] They should have fired all of their >> workers with young children. And of course we know there was a time >> even in this great country when that was the case.  
[07:10:39] >> Okay. And that in fact is the first n that is the first title seven case.  
[07:10:46] Okay. >> Which is Martin Maretta.  
[07:10:49] >> Okay. So, >> okay.  
[07:10:53] >> What I mean it's uh I I see your point. >> Thanks. Bruce, >> can I >> Yes, sure.  
[07:11:00] >> One possible mechanism that maybe addresses this is unrelated to motherhood. If you're out of the labor market for a certain period of time, the interview might be less informative. So, like in your model, do you have anything  
[07:11:15] that distinguishes between the signal being less informative for certain groups relative to the prior level and like is there a way to distinguish that >> so there is a noise in the signal so but uh we don't so in the model here we  
[07:11:30] don't distinguish between certain we would probably think of extending of extending it >> well I'm I'm suggesting this in part because this kind of provides a mechanism that doesn't have the same I think issues that you're bringing here which is that  
[07:11:43] >> you don't you have a learning trap in part because the interview isn't as informative for this type.  
[07:11:48] >> Mhm. >> And >> okay, I see >> already employed mothers you have information. You have a signal that's less >> okay thank you. I yeah I noted it down.  
[07:12:03] So let me tell you one last thing about the model which is that if you write down the posterior expectation. So upon observing uh the signal the firm not only forms a posterior about individual productivity but also about group  
[07:12:16] productivity. And let me tell you just very briefly that a firm updates its belief upwards about the group if they're positively surprised by the worker. Okay. And this generates another  
[07:12:30] empirical prediction here. So uh so that the firm updates the belief upwards if they uh if the signal is uh uh higher than the prior and let me tell you one  
[07:12:41] last thing which is that upon forming a posterior upon updating their belief about the group then in the next period the firm will have a a new distribution.  
[07:12:51] So the distribution of a firm's expectation about the group's productivity will be centered around a new mean. So in period two even if the subsidy is not there anymore a firm that has updated their belief will access that that group of worker anyways.  
[07:13:06] Right? So again this is what I've told you already. Uh and so now in the empirical part we want to test these two things. We want to test whether firms that use the subsidy thereafter increase uh their hiring of workers from the  
[07:13:19] group whatever that is and and uh and second that these effects should be more pronounced for firms that are that have better initial matches again consistent with this experiential learning mechanism. So let me move to the  
[07:13:31] empirical part. Now uh the empirical design is quite simple. It's a I will show you uh the math. Can you just Okay. Very briefly unless you have a question about the previous part.  
[07:13:43] >> I just want to test you have variation in how easily observable productivity is across firms.  
[07:13:50] >> You mean productivity of the workers? >> So we do not have variation.  
[07:13:55] >> Okay. So the the the empirical part is quite simple. So this is a firm level design.  
[07:14:02] We compare a firm that uses a subsidy to a firm that does not. This has all sorts of selection issues. I will tell you more about it later. Uh and and then we want to make sure that these firms uh you know look similar under observable characteristics. That's the key idea of  
[07:14:16] the of the of the design. So treated firms are those that use the bonus don at least once between 2013 and 2019. And here what I call event time or time zero is when they you or T star is when they use the subsidy. All right. Control  
[07:14:31] firms are firms that look similar. So we implement a simple propensity score matching on previous wages, employment and female share and also like growth pot. The second requirement that impose  
[07:14:44] on control firms is that at time zero they also hire a woman but not through the subsidy. Uh this is not like a innocent choice. You can make different choices here. Uh here the basic idea is that treated firms. So we want control  
[07:14:56] firms to uh you know again look similar to treated firms and also some of the outcomes that I will show you will not be defined for control firms if we don't impose that they also are women at time zero. Okay. Let me tell you that you  
[07:15:10] know results that I will show you you know they they don't vary much when we uh kind of play around with this choice of the control group. uh we end up with a final sample of roughly 40,000 firms.  
[07:15:22] Uh you don't have to worry about differential firm survival. They're all surviving in the period that we look at.  
[07:15:27] And this is the equation. It's a much staggered event design. Uh we also experiment with several fixed effects.  
[07:15:33] We include the industry year fixed effect, province fixed effect, payear fixed effects and results do not change.  
[07:15:40] We also have an alternative design where we just do get rid of the control group.  
[07:15:44] We only look at treated firms and we exploit staggered timing in in in the adoption of the reform. This rule rules out selection into the reform. These are not rule out dynamic selection into the reform. We are fully aware of it which  
[07:15:57] is why and I will briefly comment on that later. We both here in the paper we are you know uh careful with you know implying too much in terms of causal evidence here. we are a bit more u uh confident about co causality in the  
[07:16:12] individual level analysis. So what do we find the outcome here is the average uh non-employment spell l spell length of female new hires at the firm. So this is basically telling us  
[07:16:25] that in the years prior to the use of the reform from the treated firm the treated firm was hiring women that came from very similar non-employment spell length compared to control firms. Then at time zero a treated firm hire women  
[07:16:39] that have on average 1.5 more years of non-employment spell of pre previous non-employment spell than control firms.  
[07:16:47] So this is a big effect is a 50% effect. But of course this is sort of mechanical. The firm is adding you know by law firms that have been out of the labor market for for longer. But even afterwards as you can see the treated  
[07:16:59] firms keep on adding more women uh so oh sorry women that have been out of the labor market for a little longer.  
[07:17:06] >> Oh their friends >> referral. We get this point about referral uh too. And we for now we cannot rule that out.  
[07:17:15] Uh this is a similar outcome. This is just the number of new hires that have been out of the labor market for very long. So this is at least three years.  
[07:17:22] And again we see similar patterns. And this is mothers again similar pattern.  
[07:17:27] So firms keep on hiring more mothers afterwards. Okay. This is this is log.  
[07:17:31] So this is a 2% effect. Okay. Uh I had a question earlier on about uh men. Uh so this is main new hire. So we're not uh so we I cannot tell you  
[07:17:45] much about incumbents right now but as you can see firms uh you know if anything they seem to hire even more men afterwards and as a result they increase in size uh after the use of the policy.  
[07:17:56] Again we are very careful about not implying too much returns look great but of course we do matching so we don't want to claim too much. Okay. So, uh, ah, this is firms value added which also goes up. So, value added the worker  
[07:18:10] doesn't move. So, firms do not appear to lose in terms of labor productivity. How long do I have because it's blank? So, it's >> 10 minutes.  
[07:18:19] >> Okay, good. Um, and again, this is something that I told you already. These results are robust across the various robustness checks, including just using the treatment group. uh again and here I also want to flag the our limitation that we cannot account fully for dynamic  
[07:18:34] selection to treatment. In a way here our goal is not really to say the causal effect of the bombs donate is this but here we really care about understanding how a change in the group of workers hired at the firm uh which is brought  
[07:18:48] about by the subsidy then subsequently changes hiring behavior of the firm which then brings me to uh uh to um tell you more a little bit about about learning and also addressing some of the comments that I got earlier on which I  
[07:19:01] promised that response. So recall what the model told us. Yes.  
[07:19:05] >> Um on the outcomes, did you also look at u you know like total employment by the fans? Are they substituting information?  
[07:19:17] >> It's >> this one on the right.  
[07:19:22] >> Yes. Okay. >> So >> what's your theory that they're hiring more men?  
[07:19:27] >> So uh I don't have a precise a definite answer. It could be that the firm is, you know, saving some money >> and using that to expand.  
[07:19:34] >> Yeah. >> Could it also be like treated firms were already going to hire >> and they see, okay, it's a good deal now. Let's hire more women.  
[07:19:43] >> Sure. I mean, again, this is also a reason why we are careful about implying too much causality here. Even though prisons look good, but again, I you know, I I'm sympathetic with your with your point.  
[07:19:56] So uh let me go back to what I believe will will be the last part of the presentation which is learning as I'm let me just uh refresh uh you about what the model would say which is that the extent to which firms persistently hire  
[07:20:10] more from this group after using the subsidy depend on how they were surprised by the group to begin with.  
[07:20:17] Okay. Now we want to kind of implement that empirically which is of course not very easy. Uh so first of all we focus on early adopters. So firms that because we want to look at the long run we focus on firms that use the subsidy earlier on. Uh and then we do the following. We  
[07:20:32] want to proxy for match quality. So and what we do is run miner wage regression.  
[07:20:37] So what we do is we pick all subsidized hires at treated firms between 2013 and 2016. So this cohort. So we pick all these workers. We run simple uh we run min wage regressions when we regress their wage on observable workers and  
[07:20:52] firm characteristics and we pick the residual which is the deviation of the w the hiring wage from the predicted wage which is at least you know uh which goes  
[07:21:05] close to what at least we believe to a measure of firm surprise or better a matter of of of the the call in the model the firm sets the hiding wage based on the signal that they receive.  
[07:21:16] at the interview. Right? So this is the basic idea. Then we classify firms whether they hired workers with a high or low residual. This is based on the on the median of the residual distribution.  
[07:21:28] But we also experiment with different definitions using tiles.  
[07:21:32] Then we run a very simple definitive. We pick only treated firms. So here forget about the control group. We pick only treated firms. And when we compare treated firms that hired a good match to treated firms that hired a bad match.  
[07:21:47] All right. Importantly, this definitive coefficient gamma 2 excludes the time zero effect. Remember that peak that you we see when treat frames use the policy. We this is not going to be included because again this is mechanical. We really care about what  
[07:22:02] happens after time one. All right. In a robustness check, we bring back the control group and we run a triple differences. So we run a a specification where we compare a good match treated  
[07:22:15] firm to its match control, a bad match firm to its match control and then we net out these two differences and we find results that are uh similar to those that I will show you uh now. Yes.  
[07:22:29] Is this is this kind of residual measure of match quality correlated with firm wage premiums for other workers? So, is is this picking up match quality or is this picking up just that firms that that pay relatively higher wages and  
[07:22:44] maybe are extracting more productivity from their average worker?  
[07:22:48] >> So, I will I will I do not have an answer to you.  
[07:22:54] uh we could compute like AKM regressions and and then and then see if there is a correlation there. Yes, I will uh tell you something about I mean in the next slide, but it's probably not a good answer to what you're saying. So, I'm  
[07:23:08] not going to frame it as an answer to your question, but let me write it down because I need to that's another point on our to-do list.  
[07:23:15] >> Uh which is already quite long. >> You're allowed not to take questions until the end.  
[07:23:20] >> Okay, good. But yeah, five minutes. >> Okay, but this is the this is the last time.  
[07:23:24] >> Okay, this is the last uh slide that I want to show you. I anticipated I would not have time to for the individual analysis. So, let me just tell you what we find here. So, what we find here is that good match treated firms hire more  
[07:23:36] workers through the subsidy 2% more workers uh than bad matched firms uh after using the policy. And this go go goes back to I believe probably was man's question which is um at least the  
[07:23:49] way I would uh one one possible worry is that these persistent effects are just coming from the fact that these firms use the subsidy more because they want to save money right but again this should be the same amongst the group of remember here we're comparing treated  
[07:24:04] firms to each other so this incentive to save money might actually be the same between good match and bad matched firms so you know one interpretation is therefore then that these effects are not driven by these costsaving explanations but really about our  
[07:24:18] learning story. Okay. Uh and so they keep on using the subsidy more. They hire more women from unemployment from long-term unemployment. They are more mothers. They do not grow more. So these effects do not really uh reflect these firms growing more but simply adding  
[07:24:33] more workers from from from this group. Uh so a possible interpretation for these persistent effects which is for example what uh uh has been brought about as an explanation for the persistent effects  
[07:24:46] of affirmative action policies is that firms learn how to screen workers. We rule this out by uh looking at uh basically so we run miner wage regressions on all workers and we find  
[07:25:00] that these firms uh do not hire workers that have higher miner wage residual in their previous job but only in their existing jobs which is again I you know complimentary evidence to our learning story at least we believe  
[07:25:14] u let me just show you the event study for workers uh the the empirical analysis is quite simple again it's a much different study design because uh using the policy is a firm decision. We believe from a worker's perspective our results are a bit more causal. Okay,  
[07:25:29] again this can be debate debated but uh we don't have time for debate. Uh let me just tell you let me just the design is quite simple so I won't tell you much.  
[07:25:38] We do some matching here. Uh this is effect on wages. Again there is some small evidence of a decline but you know it's it's imprecise. So we believe there is no pass through. uh but in terms of employment at uh any firm and also the  
[07:25:52] time zero firm we find positive and persistent effects okay so I have one minute so I'm just going to conclude unless you have flesh questions about the workers which I don't see so let me just wrap up  
[07:26:05] by uh you know saying that at least to our knowledge this is the first paper that thinks of this uh active labor market policies starting subsidy as a way for firms to learn and and uh our framework because we uh model learning  
[07:26:19] about the group then also generates this persistent effects of these of these interventions and we believe you know this is a not only relevant for uh equity reasons but also and this goes back to the GE points you know there are  
[07:26:32] papers that uh show that uh increasing labor market participation for women is also important for efficiency reasons so potentially there might also be efficiency benefits from this thank you for your attention >> [applause] >> So this paper concludes our program for  
[07:26:53] this year and my voice is low so even with the mic. Uh I just wanted to thank all the paper presenter for for really great presentation and the audience for very  
[07:27:07] uh hidden participation and uh we hope to see you all next year.  
[07:27:16] [applause]