crime and savings in brazil: an empirical investigation joão manoel p. de mello eduardo zilberman...
Post on 16-Dec-2015
213 Views
Preview:
TRANSCRIPT
Crime and Savings in Brazil: an Empirical Investigation
João Manoel P. de Mello
Eduardo Zilberman
LACEA, 2005
Motivation
• Main focus of the crime literature is on the determinants of crime
• Another branch of this literature tries to estimate the cost associated with crimes– Material cost– Welfare cost
Motivation
• The effects of crime on economic variables have not been studied extensively
• If crime distorts agents’ decisions, it represents another source of cost to the society, generally unaccounted for by the literature.
Research Question
• How do crime and savings relate empirically?
• Results:– Property crime seems to increase savings;– Violent crime as a whole is not significant;– Savings appear to affect negatively property crimes,
but it does not affect violent crimes.
Theoretical Reasons
• Through probability of death:– Crime raises probability of death increases
consumption increases savings reduce– This channel should be stronger for violent crimes
• Through the precautionary motive:– Crime raises future flows of income are more
volatile savings raise– How important is the precautionary motive?
• Lusardi (1998) – It exists but is not very large• Gourinchas e Parker (2001) – It is important at low wealth
levels
Theoretical Reasons
• Through the marginal utility:– If marginal utility of consumption is decreasing on
crime, an increase in crime will reduce consumption Savings will be higher
– A caveat:• Instead of substitute consumption intertemporally, people
could substitute one type of good, that is “taxed” by crime, for another, that is not “taxed” by crime. No effect on savings
• However, if crime rate is expected to fall, people will postpone consumption of “taxed” goods
Theoretical Reasons
Figure 2 - Crimes per 100.000 Habitants for São Paulo
600
700
800
900
1000
1100
1200
1300
1400
1500
1600
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
Pro
pert
y C
rime
Rat
e
0
100
200
300
400
500
600
Vio
lent
Crim
e R
ate
Property Crime Rate Violent Crime Rate
Descriptive Statistics (São Paulo 2000)(reported crimes)
Fraud 4.66% Manslaughter 0.98%Extortion via kidnapping 0.03% Felony murder 2.44%Other extortions 0.10% Involuntary assault 22.37%Achieved theft 30.40% Felony assault 33.23%Attempted theft 1.17% Attempted murder 1.94%Achieved theft of vehicles 11.65% Other violent crimes 39.03%Attempted theft of vehicles 0.17%Achieved qualified theft 9.10%Attempted qualified theft 0.47%Achieved robbery 20.95%Attempted robbery 0.83%Achieved robbery of vehicles 12.16%Attempted robbery of vehicles 0.13%Robbery followed by murder 0.07%Other property crimes 8.12%
Reports: 966,788 Reports: 522,831
Table 2
Property crimes Violent crimes
Dados
• Unit of observation: city in São Paulo state, 2000.
• Savings measure: total amount of deposits in savings accounts and long-term CDs (Certificates of Deposit).
• Crime measure: total number of crimes reported.
OLS Regressions
Violent PropertyIncomePC 8.747 8.937
(2.172)*** (2.111)***
IncomePC2-0.668 -0.689
(0.184)*** (0.179)***Crime100 0.031 0.122
(0.054) (0.043)***Gini 1.825 1.672
(0.584)*** (0.564)***Rural -0.160 -0.203
(0.226) (0.223)Adults -1.649 -1.130
(1.847) (1.851)Divorce 1.125 0.792
(2.239) (2.214)WorkHours 1.353 1.267
(0.560)** (0.570)**Education -0.212 -0.320
(0.326) (0.330)DummyPoup 0.137 0.142
(0.086) (0.085)*DeficitPC 0.000 0.000
(0.000) (0.000)Banks100 0.479 0.509
(0.078)*** (0.080)***Density 0.242 0.233
(0.031)*** (0.030)***WealthPC 11.831 9.417
(8.864) (8.654)
WealthPC2-2.480 -1.821(2.348) (2.292)
Constant -45.340 -43.817(9.858)*** (9.451)***
Observations 566 566R-squared 0.60 0.60
Robust standard errors in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%
OLS Regressions: Interpretations
1. For violent crime, effects could be offsetting themselves;
2. Channels are more relevant for property crime;
3. Expectations is relevant only for property crime;
4. In Brazil, people can protect themselves better from violent crimes (example: avoiding pass in isolated areas at night).
OLS regressions: property crimes
Robbery and Theft Vehicles Fraud Extortion
Crime100 0.149 0.091 0.035 0.042(0.041)*** (0.018)*** (0.020)* (0.023)*
Observations 566 566 566 566R-squared 0.61 0.62 0.60 0.60
Robust standard errors in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%
Dealing with the reverse causality
• Focus: property crime
• Assumption: savings decrease crime
Coefficient is unambiguously positive
• Strategies for identification:– Look for exogenous variation in crime– Estimate the reverse effect
Exogenous variation in property crimes
• Instruments used:– Number of pay phones per 100.000 habitants– Drug trafficking apprehensions per 100.000
habitants– Drug consumption apprehension per 100.000
habitants– Others instruments used (updated version):
• Maternal mortality in 1984• Victims of car accidents per 100.000 habitants
Instrument: drug trafficking and drug consumption
• Inclusion condition: does the drug market affect crime?
– Robberies and thefts are means used by gangs to finance themselves;
– Gangs need young poor people and guns to protect their “market”;
– Addicted people could commit crimes to sustain their addiction.
Instrument: drug trafficking and drug consumption
• Exclusion condition: does not the drug market affect savings above and beyond demographics and income?
– Most of the drug dealers are poor and young people that probably would not save if they were not linked to this activity
– Most of the drug consumers are young people that probably would not save if they were not addicted
First Stage
IncomePC -1.743 -1.955 -2.054 -2.041 -2.242(2.629) (2.437) (2.558) (2.402) (2.249)
IncomePC20.182 0.192 0.209 0.205 0.214
(0.231) (0.213) (0.224) (0.210) (0.196)Gini 1.631 1.379 1.261 1.260 0.978
(0.568)*** (0.563)** (0.557)** (0.537)** (0.539)*Rural 0.246 0.337 0.358 0.340 0.425
(0.204) (0.203)* (0.209)* (0.198)* (0.200)**Adultos -4.438 -4.800 -4.415 -3.308 -3.742
(1.579)*** (1.554)*** (1.544)*** (1.595)** (1.581)**Divorce 4.473 3.756 3.108 1.942 0.997
(2.461)* (2.485) (2.460) (2.574) (2.567)WorkHours 0.498 0.686 0.576 0.432 0.566
(0.700) (0.703) (0.705) (0.693) (0.703)Education 1.021 0.801 0.950 0.783 0.591
(0.354)*** (0.354)** (0.343)*** (0.343)** (0.343)*DummyPoup -0.051 -0.054 -0.034 0.006 0.003
(0.088) (0.087) (0.086) (0.085) (0.084)DeficitPC 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)Banks100 -0.243 -0.283 -0.218 -0.187 -0.215
(0.070)*** (0.070)*** (0.069)*** (0.067)*** (0.067)*Density 0.066 0.040 0.065 0.066 0.044
(0.039)* (0.037) (0.038)* (0.036)* (0.035)WealthPC 21.502 20.970 21.163 18.840 18.933
(8.802)** (8.233)** (8.757)** (8.512)** (8.013)**
WealthPC2-5.822 -5.758 -5.747 -5.156 -5.257
(2.349)** (2.208)*** (2.337)** (2.275)** (2.153)**Phones100 0.294 0.2709
(0.066)*** (0.063)***DrugTraffic100 0.052 0.0079
(0.016)*** (0.020)DrugCons100 0.083 0.080
(0.016)*** (0.019)***Constant -11.190 -10.050 -10.050 -7.474 -6.764
(12.236) (11.152) (11.991) (11.090) (10.090)
Observations 566 542 566 566 542R-squared 0.32 0.34 0.33 0.36 0.38
Instrument: pay phones
• Inclusion condition: does the pay phone affect the perception of crime?
– Pay phones: easier to report crime; – Disque Denúncia program, which encourages people to
report potential and actual crimes; – One caveat:
• This instrument seems to alter the number of reported crimes, but not crimes de facto
• However, crime perception is also important to determine savings decision
– Pay phones may capture the confidence on public services, that makes people more prone to report crimes to the police.
Instrument: pay phones (first stage)
Phone100 Sewage Garabage Water HospBeds100 Bus100
0.294 -0.002 -0.007 -0.003 -0.043 0.031
(0.066)*** (0.001) (0.010) (0.003) (0.044) (0.033)
Table 7
*Controls used in these regressions are not reported here.
Sewage - % of households with sewage
Garbage - % of household with garbage collection
Water - % of household with water provision
HospBeds100 - logarithm of hospital beds per 100000 habitants
Bus100 - logarithm of buses per 100000 habitants
Effects of these variables on property crime (OLS regressions)*
2SLS regressions: second stage
Instrument: Phones100 DrugTraf100 DrugCons100 All
Crime100 0.603 1.588 0.640 0.659
(0.262)** (0.615)** (0.222)*** (0.188)***
2SLS regressions: second stage
Instrument: Phones100 DrugTraf100 DrugCons100 All
Crime100 0.603 1.588 0.640 0.659
(0.262)** (0.615)** (0.222)*** (0.188)***
Why do we believe that DrugCons100 is a better instrument than DrugTraf100?
• Some cities may have a disproportionate amount of documented trafficking inasmuch as they could be distribution centers
• Drug trafficking is a more infrequent occurrence, and produces more outliers
Other stuff on the paper
• Savings, when instrumented for by the number of banks per capita, decrease property crime, while has no effect on violent crime
• We check robustness by– Controlling for regional fixed effects– Changing the saving measure for residential
capital per capita
top related