00:00 Hello everybody,
00:01 um,
00:02 to,
00:02 uh,
00:03 Gurav Khana's seminar.
00:04 He's from UCSD and he's in a time zone where it's 10:30 in the morning.
00:09 Um,
00:10 we're very happy to have him.
00:12 Um,
00:12 thanks to all of you also,
00:15 yes,
00:15 for yesterday.
00:16 It was a very nice seminar,
00:19 um,
00:19 and it went very well.
00:20 So just the ground rules quickly for those of you who weren't there yesterday.
00:24 No questions the 1st 10 minutes.
00:26 After that,
00:27 we'll regularly take little breaks to have a few questions.
00:31 Um,
00:32 it worked very well yesterday.
00:34 If you just raise your hand,
00:36 it moves you to the top in the participant list so that
00:39 I can see your hand and I can just call on you
00:41 when we have the breaks.
00:43 Feel free to also ask questions in the middle.
00:46 Still raise your hand and I'll ask you if your question is burning or it can wait.
00:50 If you say burning,
00:51 I can actually interrupt Gav and you can just jump in,
00:53 even if it's not a natural break.
00:55 With that,
00:56 um,
00:57 we have Gav Khanna today.
00:59 He's going to be talking to us about his work in
01:01 Uh,
01:02 Medellin is actually part of a broader,
01:05 bigger picture,
01:07 uh,
01:07 kind of study,
01:09 but he's going to talk about the latest work from that.
01:11 If at the end there's some time left,
01:13 he's actually going to also touch on
01:16 two other papers that are already published that are smaller,
01:18 that are related,
01:20 so that you can get a sense of his like broader work on this topic.
01:23 Uh,
01:23 with that,
01:24 I hand it over to Gora.
01:26 Thank you,
01:26 Bert.
01:26 Yeah,
01:27 thank you.
01:27 Thanks so much for coming and for having me.
01:29 Um,
01:30 so yeah,
01:31 so like Burke said,
01:32 this is kind of
01:33 part of a bigger research agenda that we started about,
01:37 uh,
01:37 a couple of us started,
01:38 you know,
01:38 about 4 years ago.
01:39 And since then,
01:40 the team has grown as we've written more and more papers.
01:43 I particularly wanted to highlight Danielle and Audrey,
01:46 who are still grad students,
01:47 and they'll be on the junior market in about another year or so.
01:50 So
01:50 keep an eye out for them.
01:52 Um.
01:53 But,
01:53 you know,
01:54 essentially,
01:54 we came,
01:55 you know,
01:55 while I was still in grad school,
01:57 we started having these discussions when I went to first,
02:00 when I first went to Medellin,
02:01 as to
02:02 trying to understand why do these,
02:05 uh,
02:05 you know,
02:06 young men essentially
02:08 choose this life of crime in a setting like Medellin.
02:11 So,
02:12 uh,
02:13 when we kind of
02:14 approach this question,
02:17 the
02:18 The most salient thing that came up was that look,
02:20 there's already this old theory that goes up,
02:23 that goes back at least half a century,
02:26 talking about how
02:28 choosing crime versus other legitimate alternatives was
02:32 essentially an economic decision.
02:34 And these economic incentives are quite salient in a decision like that,
02:37 right?
02:38 But that theory
02:40 was
02:41 a model of individual behavior.
02:43 Um,
02:44 but then if you looked at all the,
02:46 all the empirical research on it,
02:48 all the causal stuff was
02:50 using this kind of
02:51 area-based aggregate relationships,
02:54 right?
02:54 So there would be papers about
02:56 local recessions and plant closings and trade shocks.
03:00 Um,
03:01 but these area-based relationships are essentially
03:04 estimating a different Elasticity altogether.
03:06 It's no longer this individual level elasticity of labor supply choice.
03:13 Uh,
03:13 because when you use an area-based relationship,
03:15 there are these GE effects,
03:17 there are
03:17 spillover to the control group,
03:20 there are,
03:22 there's a local recession,
03:23 there's no resources for policing,
03:24 lower resources to expropriate.
03:26 It might be right side.
03:27 So it's a different elasticity,
03:29 right?
03:29 Um,
03:31 but the reason we use these area-based relationships is because,
03:33 you know,
03:33 the causal,
03:34 trying,
03:35 trying to find good individual level variation
03:38 is really difficult,
03:39 you know,
03:39 you need exogenous variation
03:41 in incentives at the individual level,
03:44 and you need really good
03:46 individual level administrative data.
03:48 So what we started doing,
03:49 and I'm giving you a bigger overview of the research agenda,
03:52 what we started doing,
03:53 uh,
03:54 you know,
03:54 in 2016,
03:56 is
03:57 trying to convince the government to
03:59 allow us
04:01 access to this administrative data that gives
04:02 us everyone's Social Security numbers and birthdays,
04:06 so that we can link many different types of administrative data sets,
04:10 including all arrest records,
04:12 um,
04:12 uh,
04:13 histories of where people work,
04:15 uh,
04:15 credit histories,
04:16 uh,
04:17 these large censuses that they do that tell us about their wealth,
04:21 access to social programs,
04:22 and insurance,
04:24 where they went to school,
04:26 how they did in school,
04:27 you know,
04:28 we've been,
04:28 over the last 4 years,
04:29 we've been trying to essentially put together
04:31 a lot of different data sets that then allow us to answer a lot of these questions,
04:36 right?
04:37 Um.
04:38 And we're doing this in the context of Medellin,
04:40 which,
04:40 you know,
04:40 is a,
04:41 I don't know how many of you have been there.
04:42 It's a really beautiful city.
04:44 It's situated in a valley,
04:45 uh,
04:46 in Colombia.
04:46 It's the 2nd largest city in Colombia.
04:48 Uh,
04:49 a lot of the public transportation uses these kind of cable cars,
04:52 as you can see,
04:53 I'm sorry about the picture quality.
04:54 My phone is not
04:56 the most expensive phone.
04:57 So this is,
04:58 uh,
04:58 not the best picture,
05:00 but,
05:00 but it gives you a sense of,
05:02 you know,
05:02 the kind of the,
05:03 the topography.
05:04 A very interesting topography of the city.
05:06 You know,
05:07 it's a beautiful city,
05:08 but it has
05:09 a pretty
05:11 tragic history of crime and violence.
05:15 There's a history of cartels and
05:17 militias and paramilitaries,
05:20 um,
05:20 you know,
05:21 um,
05:21 even after
05:23 they killed Escobar,
05:25 crime fell,
05:26 but
05:27 it was still pretty high.
05:29 Um.
05:30 Even as late as 2010,
05:32 um,
05:32 when crime levels had come down a lot,
05:34 it was still one of the 10 most dangerous cities in the world.
05:37 So this is in 2010.
05:39 Um,
05:40 and
05:41 a lot of the governance essentially is done
05:44 by Local street gangs.
05:46 So there's about more than 300 combos of street gangs that essentially also
05:51 uh uh
05:52 control a lot of the public service delivery in these areas,
05:55 right?
05:55 So the gangs are very salient
05:58 in,
05:58 in a place like Medellin.
06:00 Um,
06:01 and a lot of the people involved in the gangs are young men,
06:04 you know,
06:05 uh,
06:06 a large fraction of young men in the setting have been arrested at some point,
06:10 and this is from our data set,
06:11 about 21% of them have been
06:13 arrested at some point in time.
06:15 Um.
06:17 And
06:17 economic incentives are really salient here,
06:19 right?
06:20 So there's this new work by Chris Blackman,
06:22 which is
06:24 essentially showing the industrial organization of the gang network and shows how
06:29 these gangs are really like profit seeking organizations,
06:32 right,
06:33 where
06:33 the gang leaders are almost
06:36 in the,
06:37 you know,
06:37 the top 10% of owners in the city.
06:40 But even if you read a lot of the kind of the anthropological work,
06:43 a lot of the interviews,
06:45 uh,
06:45 economic incentives are quite salient,
06:47 right?
06:47 So people are talking about what the reason why
06:49 they join gangs is for these profitable careers,
06:53 uh,
06:53 that these gangs recruit these,
06:55 uh,
06:56 idle youth who don't have a great job.
06:59 Uh,
06:59 and they,
06:59 you know,
07:00 they approach them
07:01 almost like with the Bearian menu of choices.
07:04 Saying,
07:05 hey,
07:05 are you going to,
07:06 you know,
07:06 work for us,
07:07 or are you going to do a normal job,
07:09 right?
07:10 Um,
07:10 so these economic incentives are really salient when you kind of
07:13 try to think of why these young men join gangs,
07:17 you see it in all these interviews,
07:19 as well.
07:20 So looking at all of that,
07:21 we said,
07:22 OK,
07:22 look,
07:23 we're going to write a few papers about this.
07:25 So,
07:25 so the first kind of the first paper that we worked on
07:28 is
07:29 Looking at,
07:30 um,
07:30 you know,
07:31 changes to relative wages and relative benefits.
07:35 Um,
07:35 how does that affect the choice to join gangs,
07:38 right?
07:39 So there we're leveraging,
07:41 essentially changes to formal employment benefits,
07:43 and seeing how those changes in these relative wages,
07:46 these relative returns to different types of jobs,
07:49 affect this Baarian choice between uh gangs and non-gangs,
07:53 right?
07:53 The next paper is about,
07:56 we wrote that was about,
07:57 you know,
07:58 When you lose your job,
07:59 how does that affect this quote unquote choice to join again,
08:03 right?
08:03 So when you lose your job,
08:05 we're they're leveraging mass layoffs.
08:07 So we kind of create this employer-employee match data set,
08:10 and we see when kind of your firm is having mass layoffs or shutting down,
08:15 uh,
08:15 you know,
08:15 you lose your job.
08:17 What we follow these workers over time,
08:18 and we see,
08:19 you know,
08:20 uh,
08:20 when,
08:21 when you kind of get involved in crime,
08:23 then we also look at your family members,
08:25 and then we look at things that kind of mitigate that choice.
08:28 So like access to credit,
08:30 other programs that might mitigate this kind of uh
08:33 uh getting involved in crime.
08:35 Uh,
08:35 but the paper I want to talk to you guys more about today is
08:39 really about how access to jobs in different parts of the city affect this choice,
08:44 right?
08:45 So,
08:45 do I have easy access to getting to where the good jobs are?
08:50 Or am I mostly relegated to being in a gang neighborhood,
08:53 in which case,
08:54 my employment options are related,
08:57 are constricted to being
08:59 involved,
09:00 uh,
09:00 to being involved in a gang,
09:02 right?
09:02 So everywhere,
09:04 essentially,
09:04 what we're doing is we're using
09:06 this kind of individual level data and individual level
09:09 variation
09:11 to try to look at,
09:12 you know,
09:12 to try to think of,
09:14 uh,
09:14 to show that,
09:15 you know,
09:15 in Medellin,
09:16 you should think of crime really as an occupational choice.
09:19 It's not just that someone is walking down the street,
09:21 and they get involved in crime.
09:23 It's almost like you're choosing an occupation in a context like this,
09:26 where this economic incentives are really salient.
09:29 Um.
09:30 So I think um I'm gonna,
09:32 you know,
09:33 I'm gonna,
09:33 like Brooke said,
09:34 I,
09:34 I've moved the discussion of paper 1 and 2 to the end of the presentation,
09:38 because
09:39 I want to spend most of my time talking about paper 3.
09:41 If we have time at the end,
09:43 we'll talk a little bit about paper 1 and 2.
09:46 But um I wanted to go into talking about paper.
09:49 Now,
09:50 I don't know,
09:50 Burke,
09:51 if you think that it might be worth pausing here for a second
09:54 to see if anyone has any questions about the broader research agenda,
09:58 or should we,
09:59 should we
10:00 go right into the main paper?
10:03 I think maybe you go another 5 minutes or so,
10:05 and
10:06 maybe I'm not seeing any hands yet.
10:08 That that makes sense.
10:09 I think that makes sense,
10:10 because,
10:10 yeah,
10:10 so,
10:11 so,
10:11 so this paper,
10:12 which is,
10:13 you know,
10:13 about spatial mobility,
10:14 so access to economic opportunity in different parts of the city,
10:18 and,
10:18 uh,
10:19 and whether or not folks get involved in crime,
10:21 right?
10:21 So,
10:22 It comes a lot,
10:24 you know,
10:24 we,
10:24 we,
10:24 we started off with starting off with this kind of
10:27 literature and economics segregation.
10:29 So segregation of economic opportunity and employment.
10:33 And there's this long literature,
10:35 mostly in the US context,
10:36 really,
10:37 that talks about how
10:39 economic opportunity
10:41 and
10:41 legitimate employment opportunities are unequally
10:45 and spatially distributed within cities,
10:47 right?
10:48 But the one big takeaway from this literature is that
10:51 good causal empirics is very hard to do,
10:54 because
10:55 there are these vicious cycles.
10:57 So the segregation is both a cause
11:00 and the consequence
11:01 of between opportunities,
11:03 right?
11:04 Because when people leave,
11:05 uh jobs might also
11:07 leave when certain other types of jobs come in,
11:09 then other people might move in.
11:11 So there's because of this interplay between the types of jobs,
11:15 and the types of workers,
11:17 there's going to be these kind of cycles
11:19 where you get different kinds of equilibrium.
11:21 So
11:21 causal impacts is very difficult to do in a context like this.
11:25 At the same time,
11:26 though,
11:26 because we know all these canonical crime models talk about how actually,
11:32 when you,
11:32 when you have transportation infrastructure,
11:34 you might think
11:35 in these canonical models,
11:37 what,
11:37 what that translates to is
11:40 uh
11:40 better infrastructure means I have more access to opportunities,
11:45 or alternative opportunities,
11:47 legitimate opportunities
11:48 that might actually reduce my involvement in crime,
11:51 right?
11:51 So you might think that
11:52 from a public policy point of view,
11:54 expanding the transportation network might actually be
11:57 a good way of reducing crime.
11:59 But when you actually look at what cities are doing,
12:02 you actually see that a lot of cities have blocked projects,
12:07 uh,
12:07 that expand the public transportation infrastructure,
12:09 right?
12:10 They're concerned about the fact that
12:13 these public expansions in public transportation.
12:15 will actually bring crime to affluent neighborhoods,
12:18 you know,
12:18 you've seen it in Baltimore,
12:19 and Atlanta.
12:21 I don't know if the Georgetown Metro story is apocryphal or not,
12:24 but,
12:25 but you've seen it in many different uh cities,
12:27 at least in the US,
12:28 but
12:29 also in other parts of the world.
12:31 So,
12:31 so there is this tension,
12:33 right?
12:33 So
12:33 where we come in is we're trying to kind of try to resolve this tension.
12:37 We're trying to ask this research question.
12:40 Does improving the connectedness between these segregated
12:43 and unequal parts of the cities,
12:46 does that import opportunity
12:48 to these gang areas?
12:49 Or is it exporting crime
12:51 to these other areas,
12:52 right?
12:53 Um,
12:54 and so that's kind of the question that we're,
12:56 we're going to try and answer in,
12:58 in this context,
12:59 right?
13:00 Um,
13:02 but in trying to answer this question,
13:04 there's going to be
13:05 a few different challenges,
13:06 right?
13:07 So firstly,
13:08 there's just a research design challenge,
13:10 which is that
13:12 you're working in a world,
13:13 when you expand transportation,
13:14 you're working in a world where there is no control group,
13:17 right?
13:18 Every region of the city
13:20 is either directly or indirectly treated in some way or the other,
13:24 right?
13:25 Um,
13:26 And,
13:26 and this is especially true,
13:28 because
13:29 all the,
13:29 you know,
13:30 all the different types of variables you're looking at,
13:32 so returns to formal sector jobs versus returns to crime,
13:36 the quantities of formal sector workers and crime,
13:39 um,
13:39 those are all jointly determined in the spatial generated equilibrium framework,
13:44 right?
13:44 Um,
13:45 and that's exacerbated by the fact that there might be these cross-section.
13:49 externalities.
13:51 So
13:52 more crime might actually mean that formal work becomes less productive,
13:56 you know,
13:56 firms are so busy
13:58 trying to engage in protection services or
14:01 board up their shops that they can't really
14:04 be that productive.
14:05 So because it's cross-sectoral externalities,
14:08 these kind of spatial linkages get exacerbated in many ways,
14:12 right?
14:13 So you need a little bit of structure to kind of discipline what's going on here.
14:17 So we're going to build on these quantitative spatial GE
14:20 models to try and answer some of these questions.
14:23 But you still need good identification.
14:25 So for identification,
14:27 we're going to use the fact that
14:29 the city of Medellin rolled out their public transportation system over,
14:33 you know,
14:34 the last two decades,
14:35 essentially.
14:35 And so we're going to use that rollout to identify
14:39 the model's parameters to try and,
14:41 you know,
14:41 say something intelligent about what's going on.
14:45 But to do this properly,
14:47 you really need good data.
14:49 You need to know,
14:50 like I'll show you,
14:51 you really need to know where people live.
14:54 We need to know
14:55 where they work,
14:56 and we need to know where they commit crimes.
14:59 So
15:00 that's kind of data that you haven't really seen anywhere else,
15:03 right?
15:03 So you need to know not only where you live,
15:05 but when you commit crimes and where you work.
15:07 So
15:08 it's,
15:08 um,
15:09 it's,
15:09 um,
15:10 and at the same time,
15:11 What I show you is,
15:13 uh,
15:13 to really get at the nuances,
15:14 you need to have variation
15:16 in the,
15:17 uh,
15:18 the different types of opportunities,
15:19 both criminal and formal across different neighborhoods in the city,
15:23 and good variation in the transportation costs.
15:25 So that's where the identification is going to come
15:27 in from the rollout of the cable car system.
15:30 So,
15:30 like I said,
15:31 we've been putting together
15:32 this administrative data that tells us where people live,
15:35 where they work,
15:35 where they commit crimes.
15:37 And we're doing this in a setting where there is
15:40 a lot of good variation in um where crime takes place,
15:44 where criminal opportunities are,
15:47 and good variation in the transportation network,
15:50 right?
15:50 So over,
15:50 over the last two decades,
15:52 they've been building this cable car system
15:54 that really then allows us to identify a lot of uh these parameters.
15:59 Um,
16:00 so,
16:00 so
16:02 where we stand really,
16:03 sorry,
16:03 go ahead.
16:04 There are a couple of hands,
16:05 and this seems like a good place where before you go to the literature,
16:09 um,
16:09 Roman,
16:10 Roman,
16:10 you're first,
16:11 go ahead.
16:13 Hi,
16:13 Gura.
16:14 Uh,
16:14 this is Roman.
16:14 Uh,
16:15 I have a question about human capital accumulation.
16:17 Like,
16:18 usually in these months,
16:19 you only have like jobs and,
16:21 and resident choices,
16:23 but many of these crime activities are committed by people that are
16:26 like in high school or that are going to educate themselves.
16:30 So I,
16:30 I was wondering if you are incorporating.
16:32 Any of these into a generally clear model
16:35 because usually we only care about the access of jobs,
16:37 but we don't care about the access to human capital,
16:40 so I,
16:40 I don't know if you have anything to say about that.
16:43 Yeah,
16:43 that's a great point.
16:44 No,
16:44 so in this model,
16:45 we don't,
16:46 uh,
16:46 we're not,
16:47 um,
16:48 changing whether or not you go to school or anything.
16:51 Anything,
16:52 um,
16:53 we could do that.
16:54 We,
16:54 I mean,
16:55 we're writing another paper,
16:57 which is about,
16:57 you know,
16:58 who you go to school with,
16:59 a paper on peer effects,
17:00 and,
17:01 um,
17:02 and who you go to school with,
17:03 and whether you get involved in crime,
17:05 if your friends get involved in crime.
17:07 Uh,
17:07 so we have that information.
17:09 Um,
17:10 But we actually haven't modeled the change in the,
17:14 like,
17:14 we haven't endogenized
17:16 um
17:17 what you do in terms of human capital.
17:19 Uh,
17:19 it's,
17:20 we're essentially going to be using the occupation
17:24 modeling occupation choice,
17:25 but not
17:27 the human capital accumulation itself.
17:29 Yeah.
17:32 Thanks,
17:32 Gav.
17:32 Chala,
17:33 go ahead.
17:34 Yeah,
17:34 Gav,
17:35 a quick question,
17:36 I assume the,
17:37 the,
17:37 the cable car system is very dependent on the topography,
17:41 right?
17:42 Why not use topography?
17:43 Maybe you're going to talk about it,
17:44 but I just want to put a
17:46 pointer there.
17:47 Why not use the topography directly?
17:49 Why do we need to roll out?
17:50 Because there's endogeneity there,
17:51 right,
17:51 where they set it up.
17:52 Yeah,
17:53 so,
17:53 no,
17:53 I think it's a great question.
17:54 So,
17:55 so,
17:56 so essentially,
17:57 if,
17:57 if you look at the topography in Medellin,
17:59 it's just,
18:00 it's one big valley,
18:01 right?
18:01 So there's,
18:02 um,
18:02 at the,
18:03 at the base of the valley,
18:05 there's a kind of downtown where a lot of the jobs are located.
18:08 But
18:09 other than that,
18:10 they're essentially,
18:12 uh um
18:13 there,
18:14 it's It's just essentially two sides of that valley.
18:17 Now,
18:18 of course,
18:19 you're right that there are certain slopes where it's just too steep to build,
18:24 and we could use that.
18:25 But the problem with that is,
18:27 that I think is a stronger determinant of the
18:30 types of people who live on those slopes,
18:32 because you can't build certain types of buildings on those slopes,
18:35 right?
18:36 So,
18:36 I can't build a mansion
18:38 on a certain type of slope,
18:40 whereas I can build
18:41 a shack,
18:42 or I can build slums on certain types of slopes,
18:44 right?
18:45 So when you look at the variation between the
18:48 gradient and the types of people who live there,
18:51 we don't think it's that exogenous.
18:53 Um,
18:53 and so we didn't,
18:55 uh,
18:55 so we didn't want to use that variation,
18:57 uh,
18:58 necessarily.
19:00 OK
19:02 OK,
19:03 thanks again.
19:04 Uh,
19:04 there are a couple more questions and I'm gonna stop it there.
19:07 So key first and then Harrilo.
19:09 OK.
19:21 Sorry,
19:21 should I,
19:21 is there a question,
19:22 or?
19:25 Oh,
19:25 OK,
19:25 sure,
19:25 sure,
19:26 sorry,
19:26 I didn't hear you call me.
19:27 Yes,
19:28 um,
19:28 thank you very much.
19:29 A couple of points.
19:30 One is that you bring up the point that,
19:32 uh,
19:33 crime and jobs are.
19:35 Uh,
19:36 both opportunities for the
19:38 young men to choose.
19:40 So I suppose what they're trying to say is that
19:42 crime
19:43 is countercyclical.
19:45 So if you,
19:46 as a motivation,
19:48 I'm sure you can find this,
19:49 show us a graph,
19:50 a serious graph.
19:51 The crime rate and GDP growth is going against each other,
19:56 countercyclical.
19:57 So that when employment is high,
19:59 crime rate is low.
20:00 Second thing is that
20:02 from the title of the paper,
20:03 it looks as if they're talking about organized crime,
20:06 but from the description,
20:07 it sounded more like a
20:09 more individual petty theft kind of crime.
20:13 Do you distinguish both because organized crime,
20:15 I suppose it's almost like a
20:17 big firm organization type of thing that
20:20 is requires much more capital involvement.
20:23 Recruitment efforts and all that.
20:25 Small time crime,
20:26 individual young men wake up in the morning,
20:28 take the cable to
20:30 down to downtown and mock someone.
20:32 Those are
20:33 spontaneous on the spot kind of decisions.
20:36 So,
20:36 are you distinguishing
20:38 between these two types of crime?
20:40 Thank you.
20:41 So we,
20:42 we can do that.
20:43 Um,
20:44 so,
20:44 so I will say that both we find,
20:46 so.
20:47 In,
20:48 in our earlier paper,
20:49 the,
20:49 you know,
20:49 the one I mentioned about job loss,
20:51 um,
20:52 we find that both types of crime are countercyclical,
20:55 uh,
20:55 um,
20:56 yeah,
20:56 and you,
20:57 you see that elsewhere as well,
20:58 you know,
20:58 the reason why
21:00 crime in Rio is also going up now is,
21:02 uh,
21:03 uh,
21:03 you know,
21:03 there's,
21:04 uh,
21:04 Rio is doing badly economically.
21:06 Um,
21:07 both types of crime are countercyclical.
21:09 Um,
21:10 but there is in Medellin,
21:12 it's very,
21:14 When,
21:14 if you're doing petty theft,
21:16 you essentially need the sanction of the gang,
21:19 right?
21:19 Uh,
21:20 because essentially,
21:22 the,
21:22 you think of it as they control the territory,
21:25 right?
21:25 So,
21:26 so you,
21:27 you can't get it,
21:28 you can't get away with it.
21:29 Unless you are quote unquote,
21:31 essentially
21:32 uh okayed by the gang.
21:34 Now,
21:35 there's some interesting distinctions that come up when I show you the,
21:39 you know,
21:39 the kind of the types of crime that we're looking at.
21:42 Um,
21:43 uh,
21:44 and
21:45 Uh,
21:45 and I think there is a lot of good nuance you can get by using that.
21:50 But for now,
21:51 we're just using it as,
21:53 do you have access to
21:55 these kinds of jobs?
21:56 Or
21:57 is it more that you have access to criminal opportunities?
22:02 Perfect,
22:02 thanks.
22:03 Um,
22:03 last question goes to Harry,
22:05 and if I can remind,
22:06 remind people to just lower your hand after your question,
22:09 thanks.
22:10 Hi.
22:11 So here's my question.
22:12 Uh,
22:13 so I imagine that
22:14 for this model,
22:15 we want to understand where people locate within the city
22:18 and how they move around within the city.
22:20 So moving around somehow is constrained by the existence of,
22:23 you know,
22:23 transportation opportunities,
22:24 transportation networks,
22:25 etc.
22:26 But what are your assumptions regarding,
22:28 you know,
22:29 location choices?
22:30 It's important probably in the in the US literature actually
22:33 there's a lot of discussion whether there's a free market
22:36 or if people's housing location choices are constrained
22:39 either by things like,
22:40 you know,
22:40 territorial discrimination,
22:42 redlining,
22:43 or.
22:44 In this particular case,
22:45 I was wondering if you account for
22:48 some type of inertia in terms of housing,
22:50 the coexistence of the formal and the informal housing sectors,
22:53 whether the poor can actually relocate within the city anywhere,
22:56 who's competing with whom in the city.
22:58 So
22:59 this is my question.
22:59 Yes,
23:00 I think it's a great question.
23:01 So the relocation decision
23:04 depends on Land prices,
23:07 uh,
23:07 access to different types of jobs.
23:10 So I might want to relocate,
23:12 because I want to be close to where my jobs are.
23:14 Now,
23:14 if,
23:15 and then access to
23:17 transportation lines,
23:19 if I,
23:19 if I
23:21 don't live near a job,
23:22 at least I should live near a metro station that gets me to a job.
23:25 Um,
23:26 and And local amenities.
23:28 So those are the things that are going into the neighborhood choice decision,
23:31 essentially.
23:32 We don't have any kind of government restrictions
23:36 that says these people cannot can live here,
23:38 and these people can live there.
23:41 Um,
23:41 the poor
23:42 are not living in Beverly Hills,
23:44 because it's too expensive for them to live in Beverly Hills,
23:46 it's just land prices.
23:51 Great.
23:52 OK,
23:52 so,
23:53 so the kind of,
23:54 like I said,
23:54 the,
23:55 you know,
23:55 the two literatures that we're trying to link together,
23:57 you know,
23:57 there's two parallel literatures.
23:59 One is this literature on segregation that has made a lot of headway,
24:03 saying um how neighborhoods are really important.
24:06 And then there's this other literature on crime that says these
24:09 relative returns are important when you think of occupation choice,
24:12 and we think we're really trying to
24:14 link them together.
24:15 But a methodologically,
24:17 where we're coming in is,
24:18 you know,
24:19 we're building on these kinds of spatial GD models,
24:21 right,
24:22 uh,
24:22 that have I
24:24 made a lot of progress in the last decade or so,
24:28 including a lot of things like externalities,
24:30 you know,
24:30 and Romance's paper that introduces this kind of sector choice.
24:33 So we're going to have the sector choice,
24:35 essentially based on the way that Roman has
24:38 kind of done it.
24:39 Um,
24:40 But then,
24:41 including,
24:42 importantly,
24:42 what we're going to have here is we're going to have these externalities
24:46 that,
24:46 um,
24:47 that are cross-sectoral externalities,
24:49 and that
24:51 generates some interesting dynamics that,
24:53 that,
24:53 you know,
24:53 so let me give you an example of what I mean by cross sectoral externalities.
24:58 So suppose you have
25:00 a lot of
25:01 individuals involved in crime in a neighborhood,
25:04 right?
25:04 So
25:05 they're involved in crime,
25:06 they live there,
25:07 and say they commit crimes in that neighborhood.
25:10 Um,
25:11 what that might do is that might lower
25:13 productivity of formal sector firms in that neighborhood,
25:16 right?
25:16 There's too much crime,
25:17 formal sector firms can't really
25:19 do their things.
25:20 They're so involved in protection services.
25:23 So it loses formal sector productivity,
25:25 and firms are less likely to locate there,
25:28 there might be fewer formal sector jobs in that neighborhood,
25:30 right?
25:31 So
25:31 it's just,
25:32 you can't really do formal sector work.
25:34 If there's fewer formal sector jobs in that neighborhood,
25:37 then
25:38 young men are not likely to join the formal sector in that neighborhood.
25:41 They're more likely to get involved in crime.
25:44 So what you get is this kind of what we call this crime sinkhole equilibrium,
25:48 uh,
25:49 you know,
25:49 it goes back to this kind of shellings or paper on this
25:52 tipping point type of segregation models,
25:55 where you get this kind of low-level equilibrium
25:57 where more crime means fewer formal sector jobs,
26:00 which means more crime,
26:00 which means fewer formal sector jobs,
26:02 right?
26:02 So you get this vicious cycle.
26:05 What transportation does is it kind of breaks that link
26:10 between where the jobs are located
26:12 and where
26:13 the people are living.
26:15 So,
26:15 even if I don't have formal sector jobs in my neighborhood,
26:19 if there is a good cable line that takes me downtown where I can access the jobs,
26:24 then it's fine.
26:25 I don't have to join the gang.
26:27 I can still,
26:27 you know,
26:28 I can take that cable every day,
26:29 and I can go
26:30 and do a job somewhere else,
26:32 right?
26:32 So the transportation sector helps break you out
26:35 of that low-level crime scene ho equilibrium,
26:37 right?
26:38 So that's where this kind of externality
26:41 kind of becomes quite important.
26:43 Especially because you can also kind of get this other kind of benign equilibrium,
26:48 where,
26:48 for instance,
26:49 if there's fewer formal sector jobs in your local neighborhood,
26:53 then maybe the returns to crime in that neighborhood go down.
26:56 There's less things to expropriate,
26:58 there's no one to extort,
27:00 there's no one to offer protection services to,
27:02 when there are no formal sector firms in that neighborhood.
27:05 So it lowers the returns to crime.
27:07 And then you get into this kind of,
27:09 you might get into a less crime equilibrium,
27:11 and it starts a different type of cycle,
27:12 right?
27:13 So you get at these different types of equilibrium,
27:15 and the transportation sector is quite important in breaking these
27:18 links between where jobs are located and where people live,
27:21 right?
27:22 So that's going to be kind of an important part of what happens in the model.
27:26 And so we're doing this,
27:27 like I said,
27:28 so the cable system in Medellin
27:30 is important because kind of,
27:32 of the topography of the city.
27:33 Again,
27:33 this is my terrible phone camera.
27:36 This is from last November when I was there.
27:38 Um,
27:38 but,
27:39 you know,
27:39 I've taken this picture from a cable,
27:41 uh,
27:42 you can see,
27:42 you know,
27:43 kind of the cable going over the different slums that are uh on the hillside.
27:48 Um.
27:49 Um,
27:49 and,
27:50 uh,
27:51 both of them have cables in it.
27:52 So they're going,
27:53 a lot of them
27:54 on the left picture,
27:55 you can actually see that's going towards downtown,
27:57 that's a flat area of the valley,
27:58 that's the bottom of the valley,
28:00 where a lot of the high rise buildings are,
28:02 that's where
28:03 the downtown sector is,
28:04 uh,
28:05 you know,
28:05 a lot of the jobs are there.
28:06 There are other sectors with a lot of jobs as well,
28:08 but
28:09 most of the jobs are there.
28:11 Um.
28:12 And uh this cable system was rolled out over two decades,
28:16 right?
28:16 So,
28:17 so
28:18 between 2003 and 2019,
28:20 they built a lot of cable lines.
28:22 Um,
28:23 so they first built
28:24 a cable line going up this hill.
28:27 This is an old Donna neighborhood.
28:31 So it was
28:32 a pretty high gang neighborhood.
28:35 They built another cable line here.
28:38 This is also a different kind of gang neighborhood.
28:42 Then they build a bus rapid transit system over here,
28:46 over,
28:47 you know,
28:47 in around 2012.
28:49 And then,
28:50 uh,
28:51 I'm not showing you the,
28:52 the,
28:53 then in 2016,
28:54 they built a tram line here.
28:56 And right now they're constructing a cable line over here.
28:58 And those are the things we're going to be
29:00 using when we do some of the counterfactual,
29:01 but for the estimation,
29:02 essentially,
29:03 we're going to be using
29:04 uh the rollout of,
29:06 uh,
29:06 you know,
29:07 these kinds of these cable lines,
29:09 essentially.
29:10 Um,
29:11 but to do this,
29:12 like I said,
29:13 we need to match
29:15 this administrative data using
29:17 Social Security numbers and birthdays,
29:19 right?
29:19 So,
29:20 so some of the main of some of the main data sets that we use,
29:24 or
29:25 one of them is called a CISban survey,
29:27 which is
29:28 a census of 70% of the poorest households in the city,
29:32 where we
29:33 have,
29:34 importantly,
29:34 we have,
29:34 you know,
29:35 we have a lot of information on their wealth demographics.
29:38 You know,
29:38 family structure,
29:39 etc.
29:40 And one of the important things is the geolocation of where people live,
29:45 right?
29:45 So we know where people live.
29:47 Um,
29:48 and then we have the universe of arrests,
29:50 um,
29:51 um,
29:51 between 2012 and 202015,
29:53 which tells us the coordinates of where exactly the crime takes place.
29:57 So
29:58 together,
29:58 we know where people live,
30:00 where they commit a crime.
30:02 Um,
30:02 when we get the employer employee,
30:04 we create an employer-employee mass data set,
30:06 where we know
30:07 where people work,
30:08 what their wages are,
30:10 and uh for all registered firms,
30:12 we know the location of these firms as well.
30:15 And,
30:15 so the different data sets span slightly different time periods,
30:18 but
30:18 they cover,
30:19 you know,
30:20 these vertical lines of the different cables that get built or
30:23 other sources of variation that you're going to be using.
30:25 Uh,
30:26 so they span a lot of the different sources of variation that we have.
30:30 And of course,
30:30 we have the uh
30:32 land prices,
30:32 we have cadatro data
30:34 on land prices.
30:35 Uh,
30:36 so that's kind of the data we're putting together,
30:38 using the Social Security numbers and birthdays.
30:41 Uh,
30:42 so let me show you kind of a couple of facts.
30:45 And at the end of the facts,
30:46 I can pause again for a few questions.
30:47 So,
30:48 so the first fact really is that there is a bunch of segregation.
30:51 Um,
30:52 on the left graph,
30:53 this is the fraction of the neighborhood
30:56 that's ever been arrested.
30:57 Arrested for a crime,
30:58 right?
30:59 So this is where you live.
31:00 This is based on where you live.
31:01 Um,
31:02 you know,
31:02 you see a lot of folks
31:04 in the northern part have been involved in some crime or the other,
31:08 um,
31:08 in the east and the west as well,
31:10 on the more so on the outskirts.
31:12 Whereas when you look at the destinations of formal sector jobs,
31:16 you see a lot of it is downtown,
31:17 right?
31:17 That's the central business district.
31:19 There's a few other
31:21 business districts around,
31:22 but a lot of the jobs happen
31:25 in,
31:25 in kind of in the central business district.
31:27 So it's,
31:28 there's
31:28 differences in where the jobs are,
31:30 and where
31:31 a lot of people who are involved in crime are located.
31:35 The other thing is,
31:36 when you look at the travel time,
31:38 um,
31:39 you know,
31:39 for the different types of jobs that the different
31:41 types of activities that people are located in,
31:44 um,
31:44 so since we know the origin and destination of all the activities,
31:47 we can measure these travel times.
31:49 Uh,
31:49 for formal sector jobs,
31:50 people are traveling a lot further.
31:53 Um,
31:54 for crime,
31:54 you're more likely to commit crime in your own neighborhood.
31:57 Now,
31:57 this is going to be important when we talk about
31:59 the elasticity with respect to travel time,
32:03 right?
32:03 So for different activities,
32:04 you might be,
32:05 uh,
32:05 for certain activities,
32:07 you're more,
32:07 you're likely to just committed.
32:09 In your local street gang,
32:10 and for certain other activities,
32:12 you have to go downtown to to do the job.
32:15 So in the model,
32:16 that's going to be quite an important thing.
32:18 Uh,
32:18 for certain types of crime,
32:19 though,
32:19 you still do see some people traveling.
32:21 So this is property crime that where people do travel a little bit more.
32:26 Um,
32:27 but the important thing is,
32:29 there's a nuance as to how we think
32:32 the cable should affect
32:34 uh whether people get involved in crime or not,
32:36 right?
32:37 So let's compare these two graphs.
32:39 So here in 2004,
32:41 a cable was built.
32:43 Um,
32:44 and the green line is the neighborhoods that were near this station.
32:48 So,
32:48 right,
32:49 near the newly built cable stations,
32:50 right?
32:52 Now,
32:52 what,
32:53 where was that cable built?
32:54 It was built in the Don Berner neighborhood,
32:55 where you are now essentially connecting a gang neighborhood
32:59 to
32:59 uh
33:00 where the jobs are located,
33:02 right?
33:03 And
33:04 what you see is in that gang neighborhood,
33:06 crime actually fell after the cable was built.
33:09 So when you connect a gang neighborhood to where jobs are located,
33:12 crime goes down in those gang neighborhoods,
33:15 right?
33:16 But then look at a different cable that was built in 2008.
33:20 Here,
33:20 what you were doing was you actually ended up connecting a poor neighborhood
33:24 to a gang neighborhood,
33:26 so to where the gangs are located.
33:28 And if you look at the poor neighborhood,
33:30 what you actually see is young kids in
33:32 that poor neighborhood got more involved in gangs,
33:35 they got more involved in crime.
33:37 So,
33:38 so when you build a cable,
33:40 depending on where you connect,
33:41 what types of neighborhoods,
33:43 is going to give you different answers,
33:45 right?
33:46 So if I just looked at this graph,
33:47 I would have said,
33:47 oh,
33:48 look,
33:48 crime goes down when I build a cable.
33:49 If I looked at this graph,
33:50 I'd say,
33:50 oh,
33:51 crime goes up when I build a cable.
33:52 So there's a lot of nuances going on over here,
33:55 right.
33:56 You
33:56 know,
33:56 so for instance,
33:57 if I did a simple different and dip,
33:59 right?
33:59 So suppose I just looked at
34:01 what happens to crime in a neighborhood,
34:04 based on how far
34:05 you are from a newly built uh station,
34:08 a cable station,
34:09 and,
34:09 you know,
34:09 after the station gets built,
34:12 I could do this differently,
34:13 right?
34:13 I could do this different different.
34:14 It will tell me,
34:15 um,
34:16 crime goes down
34:18 in places near the newly built stations,
34:21 right?
34:21 I could look at any crime or violent crime,
34:24 look at the,
34:25 whether,
34:25 whether you're looking at the destination of the crime,
34:27 or whether you live near,
34:29 you know,
34:29 where the residents of the people who are committing the crime,
34:33 you know,
34:33 it would tell me that
34:35 crime goes down on average.
34:38 But I just showed you that average hides a lot of different nuances,
34:42 right?
34:42 On average,
34:43 sure,
34:43 crime might be going down,
34:44 but it can't be the complete picture,
34:46 because in certain places,
34:48 crime is actually going up as a result of this,
34:50 right?
34:51 Um.
34:52 The other thing is that when you do a different diff like this,
34:56 you're making a very strong assumption on what
34:58 your treatment and kind of control group is,
35:00 right?
35:01 Suppose I think of the treatment group.
35:03 So for instance,
35:03 suppose I think of it here,
35:04 I picked the treatment group to be,
35:06 say,
35:06 1.5 kilometers away from the station,
35:08 right?
35:09 But
35:10 what if,
35:11 what if crime spills,
35:13 suppose,
35:13 say crime goes down 1.5 kilometers near the station.
35:17 But it actually,
35:18 what that means is just being displaced
35:20 to places further away from the station.
35:22 So that means crime has gone up in the control group,
35:26 right?
35:26 So there's a subtle violation here,
35:28 because what's happening is that crime is going down in the treated group,
35:31 but by construction crime is going up in the control group because of the spillover.
35:35 So the effect of crime,
35:37 you're saying is you're saying that the
35:39 um
35:40 the cable reduced crime by double the amount that it actually did.
35:44 Right.
35:45 Uh,
35:45 so there's no real control group here,
35:47 right?
35:48 So you could try to play around with what
35:50 that that
35:51 bandwidth around the station is.
35:54 Is it 1 kilometer away from the station or 2 kilometers away from the station?
35:58 But that bandwidth is also going to differ by the type of neighborhood.
36:02 In a low crime neighborhood,
36:05 there might not be any effect.
36:06 In a high crime neighborhood,
36:08 you have different bandwidths,
36:09 right?
36:09 So it might be that sure,
36:11 crime goes down right near the station.
36:13 But further away from the station is not really a good,
36:17 uh,
36:18 uh,
36:18 you don't want that to be a part of your treatment group,
36:21 right?
36:22 Um,
36:22 and this already tells you that
36:25 when you compare low baseline crime areas to high baseline crime areas,
36:29 um,
36:30 You would,
36:30 you could do this kind of event studies.
36:32 But essentially,
36:33 what it's telling you is,
36:33 sure,
36:34 in high baseline areas,
36:35 there is
36:36 more room for adjustment.
36:38 There is more room for crime to come down.
36:40 When I build a cable,
36:41 crime goes down.
36:43 Uh,
36:43 but you don't see
36:45 any such effect like that in,
36:46 in the low baseline crimes,
36:48 right?
36:49 So you could do all of this,
36:51 and I could stop here and write a paper in just these,
36:54 with these reduced form results.
36:56 But essentially,
36:57 it's missing a lot of nuances,
37:00 right?
37:00 Like I showed you,
37:02 it's ambiguous as to what would happen with crime.
37:04 In certain cases,
37:05 crime goes down,
37:06 and certain other cases,
37:07 crime goes up.
37:09 And that's partly because there's no real control group here.
37:12 Every
37:13 neighborhood in the city is indirectly treated in
37:16 some ways when you build these new cables,
37:18 right?
37:19 Because they're linked spatially to these other neighborhoods,
37:22 right?
37:23 At the same time,
37:24 there are going to be these
37:25 cross-sectoral externalities as well,
37:28 um,
37:30 that,
37:30 whereby crime can affect other types of jobs,
37:32 and that can be affecting these other types
37:34 of jobs in surrounding neighborhoods as well.
37:36 So you want to be really careful about all of these.
37:39 And then the last thing,
37:40 which I didn't even raise,
37:41 I didn't show you any evidence,
37:42 well,
37:43 you saw some spikes in the crime rates.
37:45 The spikes in the crime rates are because there were gang wars that break out.
37:49 Now,
37:49 what if that gang war just happened to break out the moment I built that station,
37:53 you know,
37:53 coincidentally.
37:54 So you could have these neighborhood by time shocks
37:57 that make this kind of analysis very difficult
37:59 in a reduced form framework.
38:01 So you need a little bit more of a structure to kind of discipline all these issues,
38:07 if you want to do this properly.
38:09 And so,
38:09 I'm going to go into what that structure means,
38:12 but I want to stop here,
38:13 I think,
38:14 because I,
38:15 I,
38:15 I've kind of done a lot of the reduced form stuff here.
38:17 I want to stop here before I move forward.
38:21 Thanks,
38:22 Gorav.
38:22 That's actually perfect timing because we can spend about 5 minutes
38:25 on questions and then that leaves you half an hour more.
38:29 So,
38:29 uh,
38:29 Devaki,
38:30 you had your hand up first,
38:31 so go ahead.
38:33 Um,
38:34 hi,
38:34 thanks.
38:34 This is Evaki.
38:36 Um,
38:36 I wanted to ask you how you treat,
38:38 uh,
38:39 informal sector versus crime in the model.
38:42 So,
38:42 a
38:43 a person could be going into a gang and joining crime,
38:45 but there's a lot of,
38:47 uh,
38:47 jobs in developing countries which are
38:50 basically a small shop set up to avoid tax.
38:52 So what would you count it as?
38:54 Uh,
38:55 that's kind of a clarification question.
38:57 Yeah,
38:58 so in our data,
38:59 we won't pick up something like
39:02 a street cart.
39:04 But we'll pick up a small shop.
39:06 Um,
39:07 so we'll know the location,
39:09 kind of,
39:10 the form of the small shop.
39:12 So,
39:13 essentially,
39:13 we'll miss a bunch of people who,
39:17 if you're just hawking on the street,
39:20 or if,
39:20 you know,
39:20 you're selling backpacks by walking down the street,
39:23 uh,
39:24 so we,
39:24 we'll miss that kind of informal work,
39:26 which
39:27 It does happen in Medellin,
39:28 but we don't have them.
39:30 What we have in our,
39:33 not in the form data,
39:33 what we have in our suspend surveys is how much income they earn from that.
39:38 So we can get a sense,
39:40 we've been trying to figure out
39:42 how much are we missing here,
39:44 uh by looking at the household survey side of things,
39:47 because there they report
39:48 the sources of income.
39:50 Um,
39:51 So we can try to get a sense of that.
39:53 But right now,
39:53 from the forum side,
39:54 we don't have that information.
40:02 Yes,
40:04 yeah,
40:04 hurry,
40:04 go ahead.
40:05 OK.
40:06 I have a clarification question regarding the data.
40:09 Uh,
40:09 you mentioned that you use the Cister data
40:12 in order to recover information and,
40:14 you know,
40:14 housing prices.
40:15 But I mean,
40:16 I was wondering,
40:17 I imagined that you would only capture the formal housing there
40:19 which may just be part of the city,
40:21 so you may lose,
40:22 you know,
40:23 you may not have in your,
40:24 uh,
40:24 you know,
40:25 in your data set.
40:25 Information on housing prices in slums.
40:28 And also,
40:28 if you find areas where you have both formal and informal,
40:31 there are plenty of reasons to think
40:33 that the informal prices are very different from the formal prices.
40:35 I mean,
40:35 there should be no discount for tenure and security.
40:37 Tradability
40:38 of informal housing is very different.
40:40 Bargaining power
40:41 between owners and renters differ a lot,
40:43 you know,
40:43 in the informal housing sector.
40:45 So I was just wondering how you go about those challenges.
40:48 Yeah,
40:49 that's a great question.
40:50 So,
40:50 so,
40:50 OK,
40:50 so,
40:50 so our,
40:51 the land price data comes
40:53 from Cadaster uh data,
40:56 and um
40:57 they do these household surveys.
40:59 And now in the household surveys,
41:01 again,
41:01 they ask,
41:02 you know,
41:02 how much
41:04 rent you've paid and everything.
41:06 And then a lot of details about the house.
41:08 Um,
41:09 so what we,
41:10 so
41:11 for now,
41:11 what we've been using is that rental information and the capacity information.
41:16 Uh,
41:16 and it lines up pretty well.
41:17 But you're right that
41:18 essentially what we want to do is,
41:20 um,
41:22 uh,
41:22 we,
41:22 we might be interested in seeing,
41:24 based on the characteristics of the house,
41:27 can we also predict how valuable that house is,
41:30 right?
41:31 So,
41:31 but it,
41:31 but it,
41:32 yeah,
41:32 it's totally,
41:32 it's,
41:33 it's a great question.
41:34 Um,
41:35 but the,
41:35 but,
41:36 but it's a combination of the Cadaster data and this household survey data as well.
41:42 Uh,
41:43 Samuel,
41:43 go ahead.
41:48 Uh,
41:49 hi,
41:49 Kav.
41:49 Um,
41:50 thanks for presenting.
41:51 Uh,
41:52 a few questions.
41:53 First is,
41:54 um,
41:55 the map of the city went by fairly quickly in the lines,
41:59 but,
41:59 um,
42:00 one question that I wanted to ask is
42:02 how did you guys think about whether you've,
42:04 you guys have gotten the boundaries of the,
42:07 of the cities or the commuting zone,
42:08 right?
42:08 Um,
42:09 so we have a project.
42:10 Um,
42:11 at,
42:12 uh,
42:12 at the bank looking at sort of the effect of the Tanzania BRT,
42:15 the bus rapid transit project in Burla,
42:17 and we noticed that
42:19 sort of there's a lot of residential
42:21 people that
42:23 are taking other buses or other sort of transit
42:26 into the endpoints of the BRT and then getting to work,
42:30 and it seemed to me that some of the lines
42:32 and especially the endpoints of the cable lines were sort of
42:36 situated on the boundary of the,
42:38 of the Midland there.
42:39 Um,
42:41 that's my first question.
42:41 I have,
42:42 I have a couple other ones,
42:43 but maybe I'll wait to ask them or,
42:45 or,
42:45 uh,
42:46 fire them away now,
42:47 but maybe you can get started off on the first one.
42:50 Yeah,
42:50 so,
42:50 so,
42:51 so,
42:52 so for some of our data sets,
42:54 uh,
42:54 so the reason we
42:56 Uh,
42:57 choose
42:58 the city boundaries,
42:59 essentially,
42:59 because for some of our data sets,
43:00 we only have information
43:02 as long as the activity takes place within this boundary.
43:06 But we still do know,
43:08 in some of our data sets,
43:09 we still do know.
43:10 So if you don't show up,
43:12 um,
43:13 uh,
43:14 uh,
43:15 in,
43:16 no,
43:16 we,
43:16 we still have a tag of if you show up from outside the city.
43:19 So we can check how big of a deal.
43:22 That is,
43:22 like,
43:22 what fraction of people are we missing,
43:25 um,
43:26 that come from the suburbs.
43:28 Uh,
43:29 off the top of my head,
43:29 I don't remember what it is,
43:31 but the constraint has been,
43:32 uh,
43:33 uh,
43:33 we use the city because
43:35 we want,
43:36 we want to,
43:37 you know,
43:37 we don't want to do a survey,
43:38 we're just using as much,
43:40 we want to use the administrative data here.
43:42 So we want to use
43:44 where the essentially the administrative data ends in terms of the city boundary.
43:50 OK,
43:50 thanks.
43:52 Oh,
43:53 maybe I'll wait to ask my other questions later.
43:57 Yes,
43:57 um,
43:57 that would be great.
43:58 Thank you so much.
43:59 OK.
44:01 OK,
44:01 my turn.
44:02 So my question is,
44:03 in your model as well as in the empirics,
44:06 do you think of this as a static one-time decision of the individual,
44:10 or there's a repeated game going on?
44:13 Is committing crime like a point of no returns?
44:15 Once you commit crime,
44:17 you go down the criminal path where you cannot come back to former employment.
44:21 Or there's actually
44:23 repeated gas that
44:25 now when the economy is not doing well,
44:28 I commit crime when it is convenient.
44:30 When the economy is doing well,
44:32 I come back to former employment.
44:33 Is there any punishment,
44:35 discount factor
44:36 for this kind of decision
44:38 for both the supplier side,
44:40 the,
44:40 uh,
44:41 the,
44:41 the young men's side as well as the demand side,
44:43 the firm's perspective.
44:44 Do they hire people with criminal records?
44:48 Any discount factor
44:49 in both theoretical and empirical
44:51 treatment in your model?
44:53 Yeah,
44:53 that's a great question,
44:54 really.
44:54 Yeah,
44:54 so,
44:55 so in the model,
44:56 you,
44:57 you do change,
44:58 you can change your decisions as to kind of what sectors you work in.
45:03 Um,
45:04 it's not a dynamic model in the sense that you don't.
45:07 You know,
45:07 you don't have future expectations on what's going to happen to wages,
45:11 as such when you're making these choices,
45:13 but you do change.
45:14 So you think of it as a bunch of static models every year.
45:18 Um,
45:19 but the other empirical question is,
45:22 do people hire folks if they've ever been arrested?
45:25 That's a great question.
45:27 Um,
45:27 in our other paper,
45:30 You do see some of that happening,
45:32 but in our interviews,
45:33 we've actually heard that it,
45:34 it does become pretty difficult to do that.
45:36 So,
45:38 I don't have a good answer to that in the sense
45:39 that you do see it happening in the sense that,
45:41 you know,
45:42 if 20% of young men have been arrested at some point in time,
45:45 um,
45:46 I think for phones,
45:47 it's like,
45:48 no,
45:48 we're going to,
45:49 we're going to have to hire people who may have been arrested at some point in time,
45:53 you know,
45:54 um,
45:55 our,
45:55 our
45:56 one of our RAs
45:57 was at one point a part of the gang,
45:59 you know,
45:59 so it's like,
46:00 they've been,
46:01 I mean,
46:01 well,
46:02 we hired him also because he was,
46:04 but,
46:05 but,
46:05 uh,
46:06 But,
46:06 um,
46:07 but,
46:07 uh,
46:09 you,
46:09 you do have people involved.
46:12 Uh,
46:12 but
46:13 when you talk to
46:14 them,
46:14 also,
46:15 they say,
46:15 no,
46:15 it is difficult for them to get hired if they have an arrest record.
46:18 So there's both of those things going on.
46:20 So it's not that
46:21 there's not a clean answer to this,
46:24 unfortunately.
46:27 OK,
46:28 so
46:29 let me quickly describe
46:30 the kind of framework,
46:32 uh,
46:33 in with the help of a few flow charts.
46:35 Now that I've said,
46:36 I've shown you the kind of the reduced farm results,
46:38 but I've said that
46:39 it's missing a lot of nuances.
46:41 Let me tell you a little bit about the framework,
46:42 right?
46:42 So,
46:43 so
46:44 think of it as
46:46 Suppose we start off with the fact that,
46:47 look,
46:48 in certain neighborhoods,
46:49 there are those who are,
46:50 who are involved in crime and those who are working in the formal sector.
46:53 And so,
46:54 so this is based on where they live.
46:55 So the origin means where you live.
46:58 So they are criminals at,
46:59 at an origin,
47:00 you know,
47:00 in a certain neighborhood,
47:01 there are people who live
47:03 there that are either committing crimes or working in the formal sector.
47:08 Based on how
47:09 easy it is to get to different neighborhoods,
47:12 uh,
47:12 it's going to determine where you commit crimes
47:15 or where you do your formal sexual work.
47:17 Now,
47:17 I already showed you that
47:19 most crime destinations are usually pretty close to where you live,
47:22 whereas people do travel pretty far for formal sexual work.
47:26 So these elasticity.
47:27 Cities might be different.
47:28 So you,
47:29 you might be
47:30 more likely to,
47:32 uh,
47:32 if you're involved in the crime sector,
47:33 you might not want to be traveling too far.
47:36 Um,
47:37 but where formal sector work takes place,
47:39 then determines um what happens to production,
47:43 what happens to foreign production,
47:45 the output in those neighborhoods,
47:46 right?
47:47 Um.
47:49 So to identify some of these parameters,
47:51 we're going to be using the rollout of these
47:53 cable car system that I kind of showed you,
47:55 uh,
47:56 showed you the rollout of the cable system.
47:58 We're going to be helping,
47:59 using those that show you exactly how,
48:02 from identification point of view,
48:03 how that helps identify these parameters.
48:06 Uh,
48:06 but then what happens is,
48:08 I said there was this externality,
48:09 right?
48:10 So
48:11 where crime takes place
48:14 is going to affect
48:15 formal sector productivity in those destination neighborhoods,
48:19 right?
48:19 So if crime takes place in a certain neighborhood,
48:22 forums might be less productive in that neighborhood,
48:25 and that's going to affect destination level output.
48:28 Now,
48:28 to identify this crime externality,
48:30 we're going to rely on a different paper that we're working on,
48:33 paper number 5 or 6,
48:35 I forget,
48:36 using this data,
48:37 but different paper that we're working on,
48:39 which essentially is going to leverage
48:41 the capture and extradition of crime logs,
48:44 where when they capture these
48:46 Crime laws,
48:46 what it did was it led to turf wars and led to spikes in crime.
48:51 Um,
48:52 and we want,
48:52 we want to look at them for formal sector production as a consequence
48:56 of what happens to formal sector production when there's a spike in crime.
49:00 So
49:00 that's going to help identify this kind of crime externality parameters.
49:04 So all of that is going to affect destination level output.
49:08 Um,
49:09 and,
49:09 and of course,
49:10 it's going to also affect uh property prices.
49:13 Destination level output is going to affect
49:17 what wages are in these different different destinations,
49:20 right?
49:20 So,
49:21 if firms can produce and be productive in certain neighborhoods,
49:24 wages might be higher in those neighborhoods,
49:27 as it's going to affect the destination level of wages.
49:30 But based on the transport time,
49:33 For each neighborhood that I'm living in,
49:35 I know
49:37 the what the destination level wages
49:39 are available in the different neighborhoods,
49:42 and how long it takes me to get to those neighborhoods,
49:45 right?
49:46 So,
49:47 by each original destination,
49:48 I know the option value that I have for
49:51 the different neighborhoods that I can work in,
49:53 right?
49:54 So
49:54 I see that option value
49:56 um
49:57 of,
49:58 of being involved in the formal sector,
50:00 I see that option value of being involved
50:02 in,
50:03 in
50:03 the criminal sector.
50:05 And I choose
50:07 Whether or not to be in the formal sector,
50:10 or,
50:11 uh,
50:12 or in the criminal sector,
50:13 right?
50:13 Uh,
50:14 that parameter,
50:15 you're going to be using this kind of uh
50:18 industry level kind of shocks to kind of help identify
50:21 that parameter as to whether,
50:23 what kind of sectoral choice you use,
50:26 uh,
50:26 you know,
50:26 the sectoral choice parameter.
50:28 But essentially it's going to help all of this.
50:31 I mean,
50:31 this arrow is not fair,
50:33 because it actually a lot of things determine what your residential choices,
50:37 but you should think of it as,
50:38 I see
50:40 where the different jobs are located.
50:42 I see how far,
50:43 how long it takes me to get to different types of jobs,
50:45 so I get a sense of which neighborhoods
50:48 I should be in.
50:50 I see the land prices and everything,
50:52 and I'm going to choose my
50:54 uh
50:56 uh
50:56 my
50:57 residential neighborhood.
50:58 Now,
50:58 if I remember correctly,
50:59 in the Dar es Salaam,
51:00 I think it was quite important with the BRT if I remember the the paper correctly,
51:05 but here,
51:05 it doesn't seem to be making much uh uh too big of a difference.
51:09 Uh,
51:09 the residential choice doesn't seem to be responding as much,
51:12 at least
51:13 empirically.
51:14 Um.
51:15 OK,
51:16 so,
51:16 so,
51:17 so,
51:18 so that's kind of the overview of what the model is kind of trying to do.
51:21 But let me go into a few technical details
51:25 here and there to try and explain something,
51:27 right?
51:28 So,
51:28 so like romance paper,
51:29 we have this occupational choice.
51:32 In this kind of uh spatial GE setup,
51:34 right?
51:35 And,
51:35 so these individuals are choosing whether or not they should
51:38 be in the formal sector or the criminal sector.
51:41 They're also going to choose which destinations they should be working,
51:44 uh,
51:45 where to work and where they want to live,
51:47 right?
51:48 Um,
51:48 so Roman has already solved a lot of problems for us by kind of solving this model.
51:53 But where we kind of,
51:55 where we,
51:56 what we're trying to do here is then,
51:58 This helps us understand,
52:00 you know,
52:00 this,
52:00 the indirect utility
52:02 that
52:02 people have,
52:04 um.
52:05 Of being in a certain neighborhood,
52:06 working in a certain destination in a certain sector,
52:09 uh,
52:10 and how that depends on wages,
52:12 the commute times,
52:14 this kind of idiosyn idiosyncratic draw,
52:17 right?
52:18 So I might have a draw for working in a certain sector in a certain neighborhood,
52:21 for instance,
52:23 um,
52:23 and kind of the land prices,
52:25 right?
52:25 So how expensive it is,
52:26 uh,
52:27 where I live,
52:28 um,
52:29 right?
52:29 So that's kind of the indirect utility.
52:32 A lot of,
52:32 of course,
52:33 the trick to these models,
52:35 as all the deck TI people will know,
52:37 is that
52:38 this,
52:38 this
52:39 error term is drawn from this kind of nested freshet distribution.
52:44 But
52:45 that structure
52:47 is just useful in terms of
52:49 producing very intuitive um
52:52 uh probabilities to work with,
52:54 right?
52:54 So even if you don't like
52:56 using these kind of fresher models,
52:58 what you,
52:58 what I'm trying to going to show you is that
53:01 The,
53:02 what it produces is what
53:04 um
53:04 uh a reduced form person would have also defined in terms of what,
53:08 what we want to study.
53:09 So let's think about
53:10 what we're looking at here,
53:12 right?
53:13 So,
53:13 this is the share of people
53:15 who live in Origin O,
53:18 work in destination D in Sector S,
53:21 so the formal sector,
53:22 the crime sector,
53:23 for instance,
53:24 right?
53:25 So that's going to be
53:27 a product of three different kinds of probabilities,
53:29 essentially.
53:30 One is
53:31 the probability that you work in destination D,
53:35 given
53:36 where you live
53:37 and what sector you've chosen.
53:40 Right?
53:40 So
53:41 why would I choose
53:43 destination D?
53:44 Well,
53:44 if wages in destination D are higher,
53:46 I'll go to destination D and work there.
53:49 Or if commute times,
53:50 which is Tao,
53:51 if commute times to destination D are lower,
53:54 then I'd rather choose destination D
53:58 over
53:59 All these other destinations,
54:00 right?
54:00 So the share of people who choose destination D
54:03 is
54:04 this divided by
54:06 the different options that you have,
54:07 all the different types of destinations,
54:09 right?
54:09 That the is just that kind of labor supply elasticity.
54:12 How sensitive am I to changes in wages,
54:15 or changes in commute times,
54:17 right?
54:18 So this tells me
54:19 the destination choice,
54:20 right,
54:21 the choice of destination.
54:23 But then
54:24 you come to kind of
54:26 the sector choice,
54:27 right?
54:27 So
54:28 should I work in the crime sector,
54:30 or should I work in the formal sector?
54:33 And
54:33 that's going to
54:34 depend on
54:36 Um,
54:36 you know,
54:37 the option value of being a criminal,
54:39 or of being
54:41 a former sex worker,
54:42 given
54:43 which neighborhood I work,
54:44 I live in,
54:45 right?
54:45 So given that I live in Origin O,
54:48 what is the option value of becoming a criminal,
54:50 and what is the option value of,
54:52 um,
54:53 of becoming a formal sexual worker?
54:55 That option value is just a function
54:57 of all the different wages I can earn in the different neighborhoods,
55:02 and the transportation time to those different neighborhoods.
55:05 Right?
55:06 So if that option value is higher
55:08 for the criminal sector,
55:09 I might choose
55:11 to be a part of the criminal sector,
55:12 right?
55:12 Of course,
55:13 I have an amenity value also of being a criminal or not.
55:16 Uh,
55:17 but
55:18 the main thing is that,
55:18 you know,
55:18 the option value is quite important.
55:21 And so that's the sector choice.
55:23 And once I've chosen the sector of choice,
55:26 it helps me,
55:27 it helps determine where I should live,
55:29 right?
55:29 So
55:30 I might choose to live
55:32 in a certain neighborhood,
55:33 because it's a very well connected neighborhood,
55:35 it gives me,
55:36 it's well connected to good jobs.
55:39 It's really close to downtown where a lot of formal sector jobs are.
55:42 So I want to live near downtown.
55:44 Of course,
55:45 if land prices are high,
55:47 then you don't want to live near downtown.
55:48 So,
55:49 so,
55:50 This helps me determine kind of what the residential choices,
55:54 the amenity value of living downtown,
55:56 the land prices downtown,
55:58 but then the option value
56:00 of living downtown,
56:01 and what option value is that?
56:03 Well,
56:03 it's just the
56:05 option value of the option values of the different.
56:07 Jobs that I can do,
56:09 right?
56:09 So,
56:10 so that's essentially,
56:11 you can split up kind of
56:13 this decision into these,
56:15 into these shares,
56:17 right?
56:17 So,
56:18 so it's a very intuitive thing.
56:19 The fresher is nice only because it makes things
56:22 intuitive,
56:23 empirically in this way,
56:24 right?
56:25 Um,
56:27 So,
56:28 so,
56:28 on the production side,
56:29 though,
56:29 you kind of have
56:31 this,
56:31 this formal sector
56:33 output,
56:34 that's we assume is produced called Douglas,
56:37 uh based on the number of workers,
56:40 and like kind of the
56:42 commercial floor space.
56:44 But the productivity on the formal sector output side,
56:47 the productivity
56:48 is a function of
56:50 the crime externality.
56:51 So this is the number of,
56:52 say,
56:52 folks involved in crime in that neighborhood.
56:55 And it kind of,
56:57 it decays.
56:58 So for instance,
56:59 it's not just that.
57:01 Maybe crime affects only
57:04 has a negative externality on your
57:06 formal sector productivity in your neighborhood.
57:08 There's a kind of a decay function,
57:09 it might also spill over to neighboring neighborhoods as well,
57:13 right?
57:13 So productivity
57:14 can go down when there's more crime,
57:17 for instance.
57:18 But the crime returns themselves are also endogenous,
57:22 right?
57:22 So they depend on formal sector work,
57:25 and they depend on
57:27 whether there are other criminals in your neighborhood or not.
57:31 So
57:31 why might it depend on other formal sector workers?
57:34 Well,
57:34 if there are more formal sector workers,
57:36 there's more people to expropriate,
57:38 there's more
57:39 protection services you need to offer,
57:41 there's more people to extort.
57:43 Uh,
57:44 so maybe more formal sector work means more returns to crime.
57:48 Uh,
57:48 on the other hand,
57:49 if there are more criminals,
57:51 maybe either maybe there's some kind of
57:53 agglomeration economies for the crime sector,
57:55 that the more criminals you are,
57:56 you can set up your own cartel,
57:58 or it is that
57:59 there's,
58:00 you know,
58:00 it's a downward sloping labor demand curve,
58:02 where
58:03 the more criminals you are,
58:04 the less valuable you are as a criminal yourself,
58:06 right?
58:06 There's just too many of you,
58:08 you should be working in a different neighborhood
58:09 if you want to be involved in crime,
58:11 right?
58:12 So,
58:13 so,
58:14 The crime returns themselves are endogenous as well,
58:17 right?
58:17 So that,
58:17 you kind of have these
58:19 cross-sectoral spillovers,
58:22 uh,
58:23 because the crime sector affects the formal sector and
58:25 the formal sector affects the crime sector as well,
58:28 right?
58:29 Um,
58:30 so,
58:30 given the setup,
58:31 there's going to be 4
58:33 important elasticities,
58:35 right?
58:36 So let me try to explain
58:38 what these elasticities are.
58:39 I think,
58:40 to me,
58:40 at least,
58:41 one of the most important ones is this,
58:44 um,
58:45 this sensitivity to commute,
58:47 to commute costs,
58:49 right?
58:49 So do the commuting time,
58:50 right?
58:51 So this,
58:52 so,
58:53 how do we get this equation?
58:54 Well,
58:55 it's simply this equation,
58:56 right?
58:56 So it's just
58:58 this share,
58:59 to the share of people
59:01 who choose to be in destination D,
59:04 given where they live,
59:05 and the sector that they work,
59:07 I take logs of this.
59:09 I get this,
59:10 that denominator is what we call a market access measure
59:17 consumer market access measure.
59:19 But
59:20 That the
59:21 is
59:22 that elasticity,
59:23 it's that sensitivity to changes in commute times,
59:26 or changes in wages,
59:28 right?
59:28 So the share of people,
59:29 so if,
59:30 for instance,
59:31 commute times go up a lot
59:32 to go from origin O to destination D,
59:35 then fewer people
59:36 will work in destination D,
59:38 right?
59:38 If you're sensitive to that commute time,
59:40 if the data is high,
59:41 then you're less likely to go work in that destination,
59:43 right?
59:45 The good thing about this kind of data structure
59:47 is that it's
59:49 origin,
59:50 but destination,
59:51 by time,
59:52 by sector level.
59:54 Right?
59:55 So,
59:56 you can put in
59:57 destination by time level,
59:59 fixed effects,
1:00:00 origin by time level,
1:00:01 fixed effects.
1:00:02 You can control for origin by destination level,
1:00:05 things like distances between origin and destination level.
1:00:08 Um,
1:00:09 why is that important?
1:00:11 Remember the issues I raised with identification.
1:00:13 So,
1:00:13 for instance,
1:00:15 if there's a gang war that breaks out
1:00:18 in this destination,
1:00:19 in this time period that you suddenly open the cable,
1:00:22 I can put in a destination by time level fixed effect,
1:00:26 that controls for anything that's happening
1:00:29 to the destination by time level,
1:00:31 at the destination by time level.
1:00:32 So I control for destination by time level shocks,
1:00:35 because my variation is still coming
1:00:38 from the link between origin and destination
1:00:40 over time,
1:00:41 right?
1:00:42 How that,
1:00:42 how that origin destination pair commute times between
1:00:46 that origin and destination pair changes over time,
1:00:48 right?
1:00:49 So,
1:00:50 you're on
1:00:51 formal ground in terms of identification,
1:00:54 when you put in things like the destination by time level fixed effects,
1:00:58 or original by time level fixed effects.
1:01:00 These are done by sectors separately.
1:01:02 So let's just say,
1:01:03 if you're,
1:01:04 if you're involved in crime,
1:01:06 or if you're involved in formal work,
1:01:08 right,
1:01:09 how sensitive you are to these commute times,
1:01:12 right?
1:01:12 Um,
1:01:13 and that helps you back out,
1:01:15 that helps you estimate what these elasticities are,
1:01:17 right?
1:01:18 So it helps you,
1:01:18 uh,
1:01:19 uh,
1:01:20 uh,
1:01:20 it,
1:01:20 it tells you what these elasticities are.
1:01:23 Um,
1:01:24 and I think
1:01:25 in Roman's paper,
1:01:26 it might be,
1:01:27 Roman can remind me.
1:01:28 About 4.5 for the informal sector,
1:01:30 where we have a much higher
1:01:32 elasticity when it comes to the crime sector,
1:01:33 about 7.9.
1:01:35 for the formal sector,
1:01:36 I think it's very similar,
1:01:37 about 1.3% for formal sector,
1:01:39 people
1:01:40 are,
1:01:41 um,
1:01:42 uh,
1:01:42 you know,
1:01:43 people
1:01:43 are willing to travel further
1:01:46 for the formal sector,
1:01:47 whereas for the crime sector,
1:01:48 if it takes too long to go somewhere to commit a crime,
1:01:51 you're not really going to go commit a crime.
1:01:53 And that's going to be important when we do these model kinderfacturers.
1:01:57 OK.
1:01:59 The second,
1:01:59 the 3rd elasticity that's important,
1:02:01 is the first two are these uh elasticity.
1:02:03 The 3rd elasticity is
1:02:05 the sectoral choice elasticity,
1:02:07 right?
1:02:08 So remember,
1:02:08 these are the option values.
1:02:10 So if I
1:02:11 When the option value of becoming a formal sector worker changes,
1:02:15 versus when the option value of becoming a criminal changes,
1:02:18 right?
1:02:19 So,
1:02:20 but the option value,
1:02:20 if I become a formal sector worker,
1:02:22 what are the wages I can earn in different parts of the city?
1:02:26 And how long does it take me to get to those different parts of the city,
1:02:28 right?
1:02:28 That's kind of,
1:02:29 you think of that as your
1:02:30 option value of becoming a formal sexual worker.
1:02:33 As that changes,
1:02:35 as those change relatively,
1:02:36 you can help.
1:02:38 Uh,
1:02:38 you can estimate
1:02:40 what's,
1:02:41 uh,
1:02:41 you can estimate,
1:02:42 uh,
1:02:42 the change in
1:02:44 whether people are choosing the formal sector or the criminal sector,
1:02:47 and that CAPA is regulating that sectoral choice,
1:02:50 right?
1:02:51 So for that,
1:02:52 we,
1:02:52 we use this kind of shift share instrument,
1:02:55 because
1:02:55 we know where people work,
1:02:58 right?
1:02:58 So we know all the formal sector options.
1:03:00 So We know there are different industries in different neighborhoods.
1:03:04 So we can look at
1:03:05 changes in industrial growth
1:03:08 in other parts of the country,
1:03:09 even,
1:03:10 right,
1:03:10 as those changes in industrial growth in other parts of the country.
1:03:14 So suppose the textile sector is growing,
1:03:16 and in Bogota,
1:03:18 does that mean that in certain neighborhoods,
1:03:20 there are going to be more jobs?
1:03:22 In the certain neighborhoods,
1:03:23 uh,
1:03:24 as the textile industry grows,
1:03:26 right?
1:03:26 So it's just a shifts share instrument,
1:03:28 essentially,
1:03:29 um,
1:03:29 to help
1:03:30 estimate the sectoral choice,
1:03:33 right?
1:03:33 So,
1:03:34 so given that,
1:03:35 you know,
1:03:36 we can estimate what happens
1:03:38 to sectoral choice,
1:03:41 when you use this kind of shifts share instrument.
1:03:45 Um,
1:03:46 and then the kind of the last parameter is how crime affects formal sector wages,
1:03:51 right?
1:03:52 So this is the externality parameter.
1:03:54 Um,
1:03:54 so for that,
1:03:55 you need
1:03:56 a shock
1:03:57 to crime in a neighborhood,
1:03:59 right?
1:03:59 So you need a shock to crime,
1:04:01 because
1:04:01 then you need to look at what happens to formal sexual output when crime goes up.
1:04:06 So we have another paper
1:04:08 um about this,
1:04:09 but
1:04:10 essentially,
1:04:11 um,
1:04:12 uh,
1:04:12 Don Berner was one of the big crime lords,
1:04:14 and he had a lot of combos under him.
1:04:17 Um,
1:04:18 and when he gets arrested and extradited to the US,
1:04:21 uh,
1:04:22 They,
1:04:23 you pull out a crime log,
1:04:24 and
1:04:25 a turf war breaks out on your turf,
1:04:28 right?
1:04:28 So,
1:04:29 so following the extradition,
1:04:31 in areas,
1:04:33 so
1:04:33 I'll explain again,
1:04:35 why these are
1:04:36 in Don Bona areas,
1:04:37 there was a big spike
1:04:38 in homicides.
1:04:40 So,
1:04:41 in the pre-period,
1:04:42 you see that
1:04:43 the different types of areas are very similar homicide rates.
1:04:46 But after Don Bunner gets extradited,
1:04:48 there's a huge spike in homicides
1:04:50 in
1:04:51 the Don Bona areas,
1:04:52 because,
1:04:53 you know,
1:04:53 people are essentially vying for that turf.
1:04:56 And so we use that variation.
1:04:58 So we compare those Don Bona areas to these other
1:05:01 areas to look at what happens to formal sectoral productivity
1:05:05 in those areas relatively,
1:05:07 right?
1:05:07 So we have that form level data.
1:05:09 So,
1:05:10 So that's how we kind of look at this,
1:05:12 get at this externality parameter,
1:05:14 right?
1:05:14 Um,
1:05:15 I can go into detail as to how these areas are determined,
1:05:18 but essentially,
1:05:19 it
1:05:20 In the late 90s,
1:05:22 the government demobilized the paramilitary groups
1:05:25 and housed them in different areas.
1:05:28 And a certain set of paramilitary groups joined Don Burner later,
1:05:32 and so they got involved in the cartel.
1:05:34 And so you have
1:05:35 good variation in what the Don Burner areas are and what the other cartels are.
1:05:40 And when Don Bunner gets extradited,
1:05:42 you see them this kind of tough war breaking.
1:05:44 Anyway,
1:05:44 so that's kind of how we identify this externality parameter,
1:05:47 right?
1:05:48 OK,
1:05:49 so
1:05:50 that's how you do that kind of,
1:05:51 yeah,
1:05:52 yeah.
1:05:53 Can I just,
1:05:54 uh,
1:05:54 there's been a couple of hands for a while.
1:05:57 Like,
1:05:57 can I have them quickly call on them and then I'll give you 5 minutes to kind of wrap up.
1:06:01 Is that fair?
1:06:02 Absolutely.
1:06:02 OK.
1:06:03 Thanks.
1:06:04 Uh,
1:06:04 I apologize to everybody who had their hands up,
1:06:07 but I'm going to prioritize people who haven't spoken yet.
1:06:09 So Lupe first and then David McKenzie.
1:06:13 Thank you.
1:06:14 Uh,
1:06:14 I have a question on,
1:06:15 uh,
1:06:15 in this context,
1:06:17 uh,
1:06:17 the majority of the employment is informal,
1:06:20 about 60% of employment is informal.
1:06:23 So,
1:06:23 I wonder,
1:06:24 given that gangs,
1:06:25 uh,
1:06:25 gang members are likely to have low human capital and maybe
1:06:29 have less opportunity to,
1:06:31 to enter the,
1:06:32 the formal market,
1:06:33 how,
1:06:34 how
1:06:35 considering that may change your model or the parameters of your model.
1:06:40 Yeah,
1:06:41 so,
1:06:41 OK,
1:06:41 so,
1:06:42 yeah,
1:06:42 I think this is somewhat related to David Key's question.
1:06:44 So we,
1:06:45 so,
1:06:46 by formality,
1:06:47 of course,
1:06:48 um,
1:06:50 There's no good way to define formality,
1:06:52 but
1:06:52 we do pick up
1:06:54 a big fraction of the workforce in our data.
1:06:57 Um,
1:06:58 uh,
1:06:58 like I said,
1:06:59 what we don't pick up people who are say hawking on the streets,
1:07:02 or if you have a shopping cart or something like that.
1:07:05 And,
1:07:06 but the interesting thing is,
1:07:07 in our earlier papers,
1:07:08 uh,
1:07:09 that,
1:07:09 that I kind of give you a hint of,
1:07:10 you know,
1:07:10 the one about formal sector job loss or changes in formal sector benefits,
1:07:15 You see that they are,
1:07:17 you know,
1:07:17 quite strongly related,
1:07:19 like,
1:07:19 you know,
1:07:19 a large fraction of the people in our
1:07:21 data set are just earning minimum wage,
1:07:23 you know,
1:07:23 these are minimum wage workers.
1:07:25 Um,
1:07:26 and so they are strongly linked to the,
1:07:29 you know,
1:07:29 if they lose their job,
1:07:31 one of the best outside options for them is you join the local street,
1:07:35 right?
1:07:35 Uh,
1:07:36 uh,
1:07:37 they're,
1:07:37 you know,
1:07:38 by formal,
1:07:39 essentially it means,
1:07:40 is your shop that you're working at registered or not,
1:07:43 right?
1:07:44 Um,
1:07:45 and,
1:07:45 uh,
1:07:46 so
1:07:47 anything that has a structure has to be registered,
1:07:50 right?
1:07:51 Um,
1:07:51 and so,
1:07:52 so all of those people are in our data,
1:07:54 um.
1:07:57 But it's a great,
1:07:57 it's a great question.
1:08:00 And then my question was just on these,
1:08:02 all these elasticities,
1:08:03 are you,
1:08:04 was how much they're varying by individual characteristics,
1:08:07 we think,
1:08:07 you know,
1:08:07 gender,
1:08:08 age,
1:08:09 uh,
1:08:09 wealth levels,
1:08:11 you,
1:08:11 you know,
1:08:11 your,
1:08:12 your elasticity to all of these things is probably
1:08:14 gonna vary according to some of these characteristics.
1:08:17 Yeah,
1:08:17 that's a great question.
1:08:18 So,
1:08:18 so we have been,
1:08:19 so we have been doing,
1:08:21 yes,
1:08:21 a gender,
1:08:22 age,
1:08:22 education level,
1:08:22 those are the exact three things that we're,
1:08:24 so,
1:08:24 so the,
1:08:25 the nice thing about these models is,
1:08:27 you can nest
1:08:29 characteristics of people,
1:08:31 right?
1:08:31 So you can essentially have different elasticities for different types of people.
1:08:36 Um,
1:08:37 and,
1:08:37 um,
1:08:38 So you can do it by gender,
1:08:40 you can do it by age,
1:08:42 you can do it by education levels.
1:08:44 And those are the things we are,
1:08:45 we're uh doing,
1:08:47 uh,
1:08:48 you know,
1:08:48 in our earlier work,
1:08:50 the elasticity,
1:08:51 the,
1:08:51 the
1:08:52 The labor supply lices are just a lot stronger for young men,
1:08:56 right?
1:08:57 Women are getting involved in crime,
1:08:59 but a lot of that is prostitution.
1:09:00 But
1:09:01 for most of the gang crime that we're looking at is mostly
1:09:04 uh young men uh getting involved in that.
1:09:08 And so,
1:09:09 we agree that it is going to be an important factor
1:09:12 of um when we do that,
1:09:15 when we move forward and do this.
1:09:16 Education,
1:09:17 there's,
1:09:17 it's kind of tricky.
1:09:18 There's
1:09:19 an interesting thing where we actually find that
1:09:22 Uh,
1:09:22 some
1:09:23 very educated people get involved in gangs,
1:09:25 you know,
1:09:25 they're like accountants who become bookkeepers,
1:09:28 and,
1:09:28 anyway,
1:09:28 that's a
1:09:29 side story that we don't want to go into right now.
1:09:31 But with education,
1:09:32 it's not always necessarily
1:09:34 as clear cut as we,
1:09:36 we,
1:09:36 we might think,
1:09:37 but,
1:09:37 but it's a good type of thing that we should think of,
1:09:40 you know,
1:09:40 adding more nuance to this setup.
1:09:43 Um,
1:09:44 thanks,
1:09:44 Gorav.
1:09:45 Roman and Samuel,
1:09:46 I see you guys,
1:09:47 but if I can ask you to stay on maybe to the end of the talk and
1:09:50 then just kind of stay an extra 5 minutes to chat with Gorav after people have left,
1:09:55 I'm gonna let him just finish in like 56 minutes.
1:09:57 Is that OK,
1:09:58 Gav?
1:09:59 Yeah,
1:09:59 I think that,
1:09:59 that,
1:10:00 that would be great.
1:10:00 Thanks.
1:10:01 Yeah,
1:10:02 uh,
1:10:02 so,
1:10:02 so,
1:10:03 so.
1:10:04 These option values
1:10:06 by different types of activity differ in different
1:10:09 neighborhoods,
1:10:09 right?
1:10:10 So,
1:10:10 so if you're in crime,
1:10:13 you you might be,
1:10:14 if you're living in a neighborhood in the east,
1:10:16 it's more lucrative maybe to
1:10:19 be in crime,
1:10:21 rather than in the formal sector.
1:10:22 Um,
1:10:23 so what we're going to do is we're going to do a few counterfactuals,
1:10:25 right?
1:10:26 Uh,
1:10:26 now,
1:10:26 this is where,
1:10:28 uh,
1:10:28 so,
1:10:28 so the first two counterfactuals that we do,
1:10:31 so the,
1:10:32 the first counterfactual that we do is essentially in 2016.
1:10:35 So we estimate the data using data to 2015.
1:10:39 In 2016,
1:10:39 they built this tram line,
1:10:42 uh,
1:10:42 that connects,
1:10:43 uh,
1:10:43 connects the downtown to the eastern part of the city,
1:10:47 or the eastern part where there was a lot of crime,
1:10:49 essentially,
1:10:50 right?
1:10:50 Um,
1:10:51 and so we're going to look at the,
1:10:53 the welfare and sectoral choice as a result of this connection,
1:10:57 right,
1:10:58 uh,
1:10:58 of this tramline,
1:10:59 right?
1:10:59 So
1:11:00 what you essentially see is that,
1:11:03 um,
1:11:04 connected neighborhoods,
1:11:05 there's a big decline
1:11:07 in the probability of
1:11:09 being becoming a criminal.
1:11:10 So if If I lived there,
1:11:12 so we're talking about,
1:11:13 this is a statement about the origin,
1:11:14 right?
1:11:15 So if I lived there,
1:11:16 there's a much lower probability that it'd be involved,
1:11:19 that I'd become
1:11:20 a criminal,
1:11:21 after this tramline gets built,
1:11:23 right?
1:11:23 So this is a counterfactual that we're doing,
1:11:25 it's just a model exercise,
1:11:26 right?
1:11:27 Uh,
1:11:27 you're much,
1:11:28 much less likely to be involved in the criminal sector if you live there.
1:11:33 But
1:11:33 what you do see,
1:11:34 unfortunately,
1:11:35 is that
1:11:36 there is
1:11:37 some export of crime when you look at the destinations,
1:11:40 right?
1:11:41 So,
1:11:42 so,
1:11:43 There are far fewer crimes being committed as a destination level.
1:11:48 There's far fewer crimes being committed in these destinations.
1:11:50 But when you look at the destinations of crimes,
1:11:52 there's actually an increase
1:11:55 in crimes being committed downtown and an increase in crimes
1:11:58 being committed in a few other parts of the city.
1:12:01 So
1:12:02 there is a bit of
1:12:03 That kind of that
1:12:04 export of crime to other neighborhoods that is unfortunately going on,
1:12:09 right?
1:12:09 So,
1:12:10 so sure,
1:12:11 on aggregate,
1:12:13 much fewer people are involved in crime,
1:12:16 right?
1:12:16 Crime levels go down in cities,
1:12:18 fewer youth are involved in crime,
1:12:20 but the crime does get spread out.
1:12:23 Where the destinations of crime are taking place,
1:12:25 that does get spread out to different parts of the cities,
1:12:27 right?
1:12:27 And that
1:12:28 depends on these commute times,
1:12:29 right?
1:12:29 So as you change the commute times,
1:12:31 the fraction
1:12:33 of people involved in crime,
1:12:35 you know,
1:12:35 kind of
1:12:36 uh changes almost,
1:12:38 there's a bit of a nonlinearity here,
1:12:40 but I think that's
1:12:41 Because
1:12:41 of the kinds of externalities and all that we have in the model,
1:12:44 but,
1:12:44 uh,
1:12:45 but it's somewhat linear.
1:12:46 Uh,
1:12:47 but the destinations of crimes,
1:12:48 there are,
1:12:49 there is this kind of export,
1:12:50 right?
1:12:51 So with those,
1:12:52 uh,
1:12:52 the conditional on the sector choice,
1:12:54 criminals do subsidy away from local crime to somewhat more distant crime.
1:12:59 But think of a slightly different counterfactual,
1:13:01 where we build,
1:13:02 where currently it's stalled because of COVID,
1:13:05 but they're building a cable line here.
1:13:07 Um,
1:13:08 then,
1:13:09 the sector chart shows that there's a much smaller reduction
1:13:13 in,
1:13:14 uh,
1:13:14 in the probability of being involved in crime,
1:13:16 right?
1:13:16 So there's only a 7% reduction,
1:13:19 rather than 11% reduction.
1:13:21 And that's because of the fact that
1:13:24 these neighborhoods are already relatively more connected,
1:13:29 right?
1:13:29 So,
1:13:30 so when you're thinking of the optimal cable to be
1:13:32 built where you want to build the next cable,
1:13:35 you want to really start thinking about targeting places
1:13:38 that
1:13:39 have low market access,
1:13:40 that they're,
1:13:41 they're less connected,
1:13:43 essentially,
1:13:43 to the uh to the network,
1:13:45 right?
1:13:46 Um,
1:13:47 so there's a smaller reduction over here.
1:13:50 Um,
1:13:51 so today,
1:13:51 I'm only going to show you these two kinds of factuals.
1:13:53 We're working on a few other kinds of factuals that
1:13:55 really exploit the multiple equilibrium nature of what's going on.
1:13:59 So there's,
1:14:00 because of the nonlinearity,
1:14:01 as a,
1:14:02 as a function of the cross-sectoral externalities,
1:14:05 um,
1:14:05 you,
1:14:06 you get
1:14:07 based on
1:14:08 the baseline structure of your neighborhoods.
1:14:11 So if it's a very high gang neighborhood,
1:14:14 you get a different effect
1:14:15 when you,
1:14:16 when you reduce transportation times,
1:14:18 than if you're already a very rich neighbor,
1:14:20 right?
1:14:21 So that nuance,
1:14:22 um,
1:14:23 I haven't been able to show you with this kind of factuals yet,
1:14:26 but that's something we're still working on.
1:14:28 But that's,
1:14:29 uh,
1:14:29 that's kind of the next thing that we're,
1:14:31 we're trying to get at.
1:14:32 So
1:14:32 let me just wrap up really quickly,
1:14:34 because I think I'm over time now.
1:14:35 But,
1:14:36 but
1:14:37 You know,
1:14:38 this links back to our bigger research agenda,
1:14:41 that's saying that,
1:14:42 look,
1:14:43 when you're thinking of crime,
1:14:45 at least in certain settings like Medellin,
1:14:47 not in many settings,
1:14:48 but in a few settings like Medellin,
1:14:50 that,
1:14:51 you,
1:14:51 you should start thinking of it as an occupation choice,
1:14:54 right?
1:14:54 And that their economic opportunity quite becomes quite important,
1:14:58 right?
1:14:59 And then there's this other literature,
1:15:00 like I said,
1:15:01 on segregation that says that look,
1:15:03 neighborhoods are really important.
1:15:05 Where you live
1:15:06 has a strong,
1:15:07 strong impact on
1:15:09 what you're going to grow up to end up doing,
1:15:11 what your wages are in the future.
1:15:13 And,
1:15:14 and part of that is because of segregation,
1:15:16 because you don't have really good access to other types of neighborhood,
1:15:19 to other types of opportunities in other parts of the city.
1:15:22 Right?
1:15:22 So if you can break that link by expanding the transit system,
1:15:26 then you can do a lot more,
1:15:28 right?
1:15:29 And so what we're trying to essentially
1:15:30 show here is that the transportation infrastructure
1:15:34 can have these meaningful impacts on this kind
1:15:36 of sectoral choice when it comes to crime,
1:15:38 but also overall welfare,
1:15:40 uh,
1:15:40 as a result of the kind of crime reduction,
1:15:43 um,
1:15:43 as a consequence of this.
1:15:45 OK,
1:15:45 I'm going to stop here because I think I'm a minute over.
1:15:47 But
1:15:48 thank you.
1:15:50 Thanks so much,
1:15:50 Gurav.
1:15:51 Um,
1:15:51 I give you a virtual round of applause on behalf of everybody.
1:15:55 Um,
1:15:56 thanks,
1:15:57 everyone.
1:15:57 We'll bring it to a close here.
- add-style
- lp-body-content