Crime and the Economy in Mexican States
Heterogeneous Panel Estimates (1993-2012)1
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund
  • | 2 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Corresponding Author’s E-Mail Address: CVerdugo@imf.org

This paper studies the transmission of crime shocks to the economy in a sample of 32 Mexican states over the period from 1993 to 2012. The paper uses a panel structural VAR approach which accounts for the heterogeneity of the dynamic state level responses in GDP, FDI and international migration flows, and measures the transmission via the impulse response of homicide rates. The approach also allows the study of the pattern of economic responses among states. In particular, the percentage of GDP devoted to new construction and the perception of public security are characteristics that are shown to be associated with the sign and magnitude of the responses of economic variables to crime shocks.

Abstract

This paper studies the transmission of crime shocks to the economy in a sample of 32 Mexican states over the period from 1993 to 2012. The paper uses a panel structural VAR approach which accounts for the heterogeneity of the dynamic state level responses in GDP, FDI and international migration flows, and measures the transmission via the impulse response of homicide rates. The approach also allows the study of the pattern of economic responses among states. In particular, the percentage of GDP devoted to new construction and the perception of public security are characteristics that are shown to be associated with the sign and magnitude of the responses of economic variables to crime shocks.

I. Introduction

Recently, the problem of crime has been a source of concern for international organizations, policy makers, and the populations in Mexico. Mexico has recently put forward an ambitious structural reform agenda, including initiatives that aim at improving the rule of law. In light of this, understanding the relationships between crime and overall economic activity is as important as ever.

Despite the increase in Mexico’s crime rates over the sample period of this study (from 1993 to 2012), over similar periods crime has been much higher elsewhere in the Americas, as measured by homicide rates—one of the most commonly used indicators for comparing levels of crime (see UNODC 2014).2 As illustrated in Figure 1, Honduras has a homicide rate nearly four times that of Mexico, El Salvador’s rate is three times as high, and Venezuela’s is more than twice as high. Even Colombia has a homicide rate that is nearly 50 percent greater than Mexico’s.

Figure 1.
Figure 1.

Selected Countries: Intentional Homocide Rates

(per 100,000 inhabitants)

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

Source: UNODC Global Study on Homicide (2014), Heinle et al.(2014) and authors’ calculations

The incidence of crime varies widely across Mexican states. According to the Mexican National Institute of Statistics (INEGI), during most of the last decade, from 1997 to 2007, Mexico’s homicide rate plunged from 14 homicides per 100,000 to a much lower level of 8 per 100,000 in 2007. However, over the next few years, Mexico’s relatively successful story in decreasing homicides reversed itself with more than 21.5 per 100,000 recorded in 2012. As discussed in numerous studies (see for example Robles et al., 2013, Mejia et al., 2012b; Guerrero, 2011a; Dell, 2012; Calderon et al., 2013), this remarkable surge in homicide rates is considered to be a direct result of the dramatic increase in violence associated with drug trafficking organizations and other organized crime groups. To a large extent, the growth of the drug trafficking organizations was likely in response to shocks taking place not only internally, but also in other latitudes: the drug demand from the U.S. market; the success of other governments, such as Colombia’s in regaining control of the country from crime; and the closing of the drug-trafficking routes in the Caribbean achieved by the U.S.

It has long been recognized that crime has effects on the economy (for references see the literature review in section II below). In the short run, crime has the potential to inhibit the accumulation of part of the physical and human capital stock, the allocation of which may be further distorted by the infiltration of criminal organizations into the formal economic system. From a dynamic perspective, crime potentially increases the risk and uncertainty of the business environment, which in turn may hinder the accumulation process and lower the long-run growth rate of the economy.3 For example, according to Hallward-Driemeier and Stewart (2004) and Daniele and Marani (2011), the primary effect of organized crime is to increase the costs of doing business. Other adverse effects of crime include extortion (Brock, 1998 and Daniele and Marani, 2011), kidnapping of workers (Clegg and Gray, 2002), and disruptions in supply chains (Barnes and Oloruntoba, 2005; Czinkota, 2005; Globerman and Storer, 2009; and Branzei and Abdelnour, 2010). Local demand may decline due to emigration, business relocation, and firm closures (Greenbaum et al., 2007). Also, Ashby and Ramos (2013) find a negative effect of organized crime on foreign investment in Mexico.

Figure 2 reflects the evolution of real GDP (RGDP) growth and homicides rates for different periods between 1993–2012. In particular, the last period (2007–2012) illustrates an important increase in the median homicide rate at the national level. It also shows that a number of states have homicide rates well above the national median. It is interesting to observe, however, that for some states with increases in homicide rates the economic growth rates are above the national median, while others fall below the median RGDP growth rate for the period.

Figure 2.
Figure 2.

Mexico: Real GDP Growth vs. Homicide Rates1/

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

Source: INEGI, CONAPO, and authors’ estimates1/ For the methodology used to estimate the Real GDP, see the technical Appendix A.

Against this backdrop, the objective of this paper is to investigate the relationship between crime and other economic activities in Mexico at the state level from 1993–2012, with an eye toward the broader potential role of crime for domestic stability as a whole.4 For example, does crime lower overall economic activity in Mexico in the long run, at both the individual state and aggregate levels, and if so, by what mechanisms? How important are shocks at the state level versus the national and international level for the relationship between crime and overall economic activity? What role, if any, do international factor flows in the form of foreign direct investment (FDI) and international migration play in the channel by which crime impacts overall economic activity at both the state and national levels? How do the dynamic relationships between crime, FDI, and overall economic activity differ across the states of Mexico and what socio-economic factors if any may account for these differences? In this regard, it is important to note that, in its primary analysis, this paper does not study a particular category of crime shocks, but rather studies the response of both homicide data and economic data in response to general crime shocks.

In addressing these questions, we recognize that the relationships are potentially complex in terms of the dynamic interdependency between crime and economic activity, and that these dynamic interdependencies may differ substantially among different states. In light of this, as described in section III of our paper, we employ the panel SVAR methodology of Pedroni (2013), which is designed to accommodate these issues for panels comprised of relatively short and possibly noisy data. Furthermore, by using a structurally identified VAR approach, subject to our econometric identification scheme, we are able to study the role of shocks to crime in general, rather than on the basis of more narrowly defined proxies. Thus, while we use recorded homicide data as one of our endogenous variables, in the VAR analysis the crime shocks are defined as any crime which induces a movement in the homicide rate subject to the econometric identifying restrictions that we place on the other economic variables in the system, as we discuss in further detail in section III of the paper.

On the basis of such an approach, we find that crime has played an important role in driving the variation in several key economic variables at the state level. For example, we find that on average state-specific and common national and international crime shocks together are responsible for up to 5.5 percent of the variation state-level GDP in the years following the shocks. We also find considerable heterogeneity among the states, with up to one-fourth of all states experiencing as much as 11 percent of the variation in their GDP to crime shocks. Furthermore, we show that the specific impulse responses are heterogeneous, but with up to a quarter of all states seeing on average a 1.6 percent per annum decrease in GDP for the first five years following a crime shock. We also find that crime shocks have a considerable, but also very heterogeneous relationship to FDI flows at the state level. When we study the pattern among states of the magnitude of the responses of GDP and FDI to crime shocks, two important characteristics stand out, namely the relative importance of the construction sector as a percentage of the state economy and the degree of a perception of public security as reflected in survey responses. On average, the prevalence of construction is associated with a diminished negative impact of crime on overall economic activity, while a sense of insecurity, as reflected in public survey data, is on average associated with an increased negative impact of crime on economic activity and FDI in particular. These findings are potentially important for the ongoing process of judicial reform in Mexico as well as for the continuing efforts toward a further strengthening of the Anti-Money Laundering/Countering the Financing of Terrorism (AML/CFT) regime in Mexico. We should note of course that our study is focused on the period from 1993 to 2012 and therefore does not account for the reforms that have taken place in Mexico since 2012.

The remainder of the paper is organized as follows. In the next section, we present the empirical framework for examining the effect of crime on the economy. In section III, we describe our empirical methodology and strategy for identifying crime shocks, other key structural shocks in a panel VAR context, and limitations of the analysis. Section IV describes the data. The paper’s empirical results are presented and discussed in section V, while section VI summarizes and concludes. A technical appendix includes a brief description of the methodology used for constructing a sufficiently long series for real GDP using a single base year, which we also consider to be an important contribution in support of the primary objective of the paper.

II. Related Literature on the Relationship Between Crime and Other Economic Activity

The importance of crime in determining a country’s economic progress has long been recognized both in the academic literature and in policy-making circles. Some contributions have tried to establish a relationship theoretically between crime, growth, and development (e.g., Bourguignon, 2001; Fajnzylber et al., 2002; Mauro and Carmeci, 2007) and some studies quantify economic and social cost of organized crime for different countries (Fajnzylber et al., 2002c; Buvinic and Morrison, 1999; Glaeser, 1999; and International Centre for the Preventing of Crime, 1998; Londono and Guerrero, 2000; and Rios, 2011). Other scholars, such as Prasad (2012), try to examine the link between economic controls and black markets, by exploring the effects of India’s liberalization experiment on its murder rate. A number of studies have analyzed the transmission channels through which crime, either directly or indirectly, impacts economic growth (see e.g., Goulas and Zervoyianni, 2013; Detotto and Otranto, 2010; Detotto and Vannini, 2009; Czabanski, 2008; Brand and Price, 2000; and Anderson, 1999).

Nevertheless, despite the growing literature, empirical studies have not yet produced a definitive conclusion regarding the impact of crime on economic growth. A way to measure the crowding-out effect of crime is to estimate its impact on the economic performance of a country, a region, or a municipality. We can distinguish two approaches. The first approach is to compare the overall macroeconomic performance of countries or regions with high levels of crime to that of countries with low levels of crime, controlling for other explanatory variables (this approach comes from Barro, 1996). For example, Peri (2004), using a large data set (from 1951 to 1991), shows that the annual per capita income growth is negatively affected by murders after controlling for other explanatory variables. Gaibulloev and Sandler (2008) measure the impact of domestic and transactional terrorism on income per capita growth for 1971–2004 in a panel of 18 Western Europe countries.

The second approach consists of univariate and multivariate time series methodologies. Recently, there have been many contributions to this approach, wherein crime is considered along with various macro variables. For example, Detrotto and Otranto (2010) show that crime negatively impacts economic performance in Italy. The findings suggest that the economic costs of crime include a significant fixed component, and that the dynamics of the economic cost is time-varying, but always significant. Chen (2009) implements a VAR model to examine the long-run and causal relationships among unemployment, income, and crime in Taiwan. Narayan and Smyth (2004) implement a Granger causality tests to examine the relationship among seven different crime typologies, unemployment, and real wage in Australia within an AutoRegressive Distributed Lag (ARDL) model. Kumar (2013) empirically examines the causality between crime rates and economic growth using a reduced form equation and an instrumental variable based on state level data in India. Pinotti (2011) examines post-war economic growth of two regions in southern Italy which were exposed to the presence of mafia organizations after the 1970s, and applies synthetic control methods to estimate the counterfactual growth performances in the absence of organize crime.

Other studies have quantified different economic effects of organized crime. For example, Hallward-Driemeier and Stewart (2004) and Daniele and Marani (2011) show that the primary effect of organized crime is to increase the costs of doing business. The potential adverse effects are through assessment of regional security risks (Kotabe, 2005), security budgets (Spich and Grosse, 2005 and Czinkota et al., 2010), extortion (Brock, 1998 and Daniele and Marani, 2011), kidnapping of workers (Clegg and Gray, 2002), disruptions in supply chains (Barnes and Oloruntoba, 2005; Czinkota, 2005; Globerman and Storer, 2009; and Branzei and Abdelnour, 2010), and decreases in local demand (Greenbaum et al., 2007). Local demand may decline due to emigration, business relocation, and firm closures (Greenbaum et al., 2007). Brock (1998) finds relatively higher FDI in regions of Russia where the level of crime is lower. Madrazo Rojas (2009) finds empirical evidence of a negative association between violent organized crime and foreign direct investment (FDI) in Mexican states. Daniele and Marani (2011) find support for a negative relationship between total regional crime and FDI in Italy. Many of these scholars, including Fajnzylber et al. (2000), Detotto and Otranto (2011); Forni and Paba (2000), Cardenas (2007), Ashby and Ramos (2013), and Robles et al. (2013), use the number of recorded intentional homicides as a crime indicator.

Finally, for Mexico, Dell (2012) examines the direct and spillover effects of Mexican policy towards the drug trade, and finds that crime creates a contagion effect between those municipalities closer to drug trafficking routes. Ashby and Ramos (2013) find that organized crime deters foreign investment in the financial services, commerce, and agricultural sectors, but not the oil and mining sector, for which they find increased crime associated with increased investment. Robles et al. (2013) find that there is a threshold for crime, below which individuals and companies internalize the cost of security and protection in accordance with their economic capacity. Once the threshold is surpassed, companies and individuals will modify their investment decisions, production, labor participation, and employment, all of which have a negative impact on economic activity.

In our paper, we use a relatively novel approach to investigate the effect of crime on economic performance in Mexican states in a manner that more systematically accounts for dynamic endogeneity. We seek to do so subject to the double challenge of employing credible identifying restrictions while deriving results for a large group of possibly quite heterogeneous states. Conventional dynamic panel methods are not appropriate in light of the fact that they require the dynamics of individual state responses to be identical among all the states. Instead, we employ a panel methodology that allows individual states’ responses to structural shocks to be heterogeneous. In particular, to address these issues in the context of structural identification, we use the panel SVAR methodology developed in Pedroni (2013). Recent examples of the use of this methodology in policy relevant empirical applications include among others, Pedroni and Verdugo (2011), which analyzes the effect of drug production on the formal economy in Peru, and Mishra et al. (2014), which analyzes the effectiveness of monetary policy in low income countries.

III. Estimation and Identification Strategy

A. Overview of the Methodology

As noted, the relationship between crime and economic activity is quite complex. The two can be deeply intertwined, such that crime has an impact on economic activity at the same time that economic activity has an impact on crime. Furthermore, the nature of this endogeneity is likely also to be dynamic, in the sense that the feedback between crime and economic activity occurs gradually over time, and with different intensities over different time horizons.

Adding to this complexity, when we study the relationship between crime and economic activities at the aggregate state level in a country such as Mexico, we need to recognize that the nature of these dynamic relationships need not be the same in different states and may differ substantially among states. The differences may arise for a number of reasons. For example, the structure of the economies differs at the state level. Factors affecting crime differ geographically at the state level. The nature and the intensity of both crime and economic activity can differ across states. Finally, the mechanisms by which crime and economic activity interact with one another over time can differ regionally among states.

Another dimension to this complexity arises from the fact that both crime and economic activity may respond differently to unobserved innovations in various types of economic activity and crime depending on whether the innovations originate locally or somewhere else. For example, changes in economic activity nationally or internationally, as well as changes in other factors that drive crime nationally or internationally can be expected to impact criminal and economic activity at the state level. This creates a potential further complexity in the form of a cross sectional, or spatial dependence of crime and economic activity among the states of Mexico.

For these reasons, our empirical approach is one that accommodates potentially complex dynamic endogeneities that differ among states, and which are responding to potentially unobserved shocks that occur either at the local state level or at the national and international level. A standard econometric tool that accounts for dynamic endogeneities in general is the structural VAR approach. The structural VAR approach begins by estimating the dynamic relationship among the variables by use of a system of equations that represent a sufficient number of lags of each of the variables in each of the equations, so that what remain as residuals in the equations are disturbances that are uncorrelated over time. Next, in order to relate these disturbances to structural shocks that have economic meaning, one exploits economic identifying restrictions. However, the approach requires time series data of substantial length, beyond what is available for the current analysis. A natural solution is a panel approach, which treats each state as a member of the panel, and compensates for the lack of a substantial time series dimension by exploiting the fact that the dynamic relationships are observed repeatedly among members of the panel.

However, in taking a panel approach, we must take care in how we treat the individual members of the panel. Most fundamentally, we must account for the fact that the states of Mexico differ from one another as reflected in potentially heterogeneous dynamic relationships between crime and other economic activity. Failing to account for heterogeneity in the estimation of dynamic relationships results in a well known econometric problem that can lead to inconsistent estimation and inference of the relationships.5 Consequently, we do not want to simply pool the Mexican state-level data as one would for conventional dynamic panel methods, which rely on the assumption of homogeneous dynamics among the members of the panel. Instead, the methodology we use must account for this heterogeneity directly.

Rather than considering the heterogeneity as an obstacle, the method we use exploits this heterogeneity as an asset that can help us to uncover some of the differing mechanisms by which crime and economic activity interact in the states of Mexico from 1993–2012. In particular, the methodology that we use is based on the panel structural VAR approach developed in Pedroni (2013). Specifically, the approach models heterogeneous state-specific dynamic responses to unobserved shocks that occur either at the state level or at the national and international level. In this manner, the technique accommodates both the heterogeneity and the cross sectional dependence that arises from the responses to shocks that are common across states.6 The shocks are treated as structural and unobserved. They are identified and estimated via a method of structural identification analogous to the conventional structural VAR approach. The panel methodology then exploits the statistical relationship of the structural shocks to decompose them efficiently into shocks that are common to the members of the panel versus shocks that are idiosyncratic to individual members of the panel. The relative importance of the idiosyncratic versus common shocks is permitted to differ for each member of the panel, and each member is permitted to respond in a heterogeneous member specific manner to both the common and idiosyncratic shocks.

As is typical in structural VAR approaches, the responses to the structural shocks are represented as impulse responses, and the importance of the shocks are represented as dynamic variance decompositions. In the context of our panel approach, our identification provides us with sample estimates of a set of state-specific responses and variance decompositions to both the idiosyncratic and common structural shocks for each of the 32 states of Mexico. This sample distribution of state-specific responses allows us to study the economic conditions and characteristics of the states that are associated with particular patterns among the responses. For example, using the distribution of individual state responses we can investigate which state characteristics are associated with larger or smaller responses of economic activity to unexpected changes in crime. Of course, in doing so, we must take into account the fact that the responses and decompositions are estimated and are subject to uncertainty from the sampling variation associated with the estimation. Accordingly, we use a bootstrap estimator which produces confidence bands not only for the distribution of impulse responses and variance decompositions, but also for the subsequent analysis of the state characteristics associated with patterns in these distributions. For a more detailed discussion of the methodology, we refer readers to the discussions in Pedroni (2013) as well as the empirical application of the technique in Mishra et al. (2014).

B. Overview of the Identification Strategy

Next, we discuss the identification strategy associated with our panel methodology. While we employ the identification scheme and corresponding methodology to the case of the states of Mexico, in principle, the same can be applied to any of a number of countries facing crime. In this regard, as a structural VAR-based method, a key feature involves the identification of the unobserved structural shocks to which the observed state level data is responding. Proper identification ensures that the impulse responses and variance decompositions can be given economic causal interpretations that account for the interdependent endogenous dynamics. It is also an essential element of our panel methodology in that with relatively short lengths of data it will allow us to efficiently decompose the structural shocks into identified common and idiosyncratic components.

A key strategy for successful identification is to consider the variables of interest in relationship to the shocks that drive them. In our case, we are primarily interested in the responses of crime and overall economic activity, as represented by GDP. However, we are also interested in studying the role that key international factor movements, such as capital flows in the form of FDI and labor flows in the form of migration, play in the relationship between crime and economic activity. For each of these variables, we will investigate the responses to the key structural shocks that potentially impact the economy. Among these, we have classified shocks into four broad categories. The first two we view as shocks to economic activity. In the regional and macroeconomic literatures, these are often loosely referred to as aggregate supply (AS) and aggregate demand shocks (AD). They are distinguished from one another on the basis of whether they correspond to shocks that permanently increase or decrease total economic activity (AS) versus shocks that have only a transitory effect on total economic activity (AD).

The next two shocks we view as perturbations in crime and migration that originate independently of other economic activity. The crime shocks can be thought of as shocks to either the supply or demand for criminal activity for a given level of economic activity,7 and similarly the migration shocks can be conceptualized as shocks to either the supply or demand for net migration for a given level of economic activity. As we discuss in the next section on data sources, our GDP variable is measured as log per capita state GDP, our FDI variable is measured as log per capita cumulative FDI state inflows, our crime variable is measured as log per capita state homicides and our migration variable is measured as log per capita international net migration into the state. Each of the four categories of shocks is permitted to affect each of the four variables over time. Thus, in our setup, GDP responds to crime and migration shocks as well as economic activity shocks, and homicides respond to crime and migration shocks as well as economic activity shocks, and so forth.

However, the shocks are neither directly observed nor proxied. Rather, as in the structural VAR literature in general, we infer the shocks based on the pattern of responses among the observed variables. Doing so requires us to place a few minimal restrictions on the timing of the permissible responses of the variables to the shocks. These are known in the structural VAR literature as the identifying restrictions. The most commonly used identifying restrictions come in the form of exclusion restrictions on either the immediate impact effect of some of the shocks on some of the variables, or on the very long-run steady-state effect of some of the shocks on some of the variables. The former are loosely referred to as “short-run” restrictions while the latter are loosely referred to as “long-run” restrictions. As best as possible, these restrictions should be motivated on the basis of sound a priori economic reasoning. All of the transition dynamics between the initial impact and the eventual long-run steady state time horizon are then typically left completely unrestricted, and are permitted to be fully endogenous in terms of the feedback among the variables.

For the purpose of this paper, we use a mix of both short-run and long-run identifying restrictions and build on some of the typical restrictions that have been used periodically in the structural VAR literature. For example, to distinguish between AS shocks and AD shocks, we use the restriction that AD shocks do not cause permanent movements in GDP, while AS shocks do. This is in keeping with the literature and is consistent with their conceptual definitions. Similarly, to distinguish these two economic activity shocks from the crime shocks, we employ a restriction that reflects the idea that while both the crime and economic activity shocks can have an immediate impact effect on economic activity, only the crime shocks have an immediate impact on homicides. This is consistent with the idea that crime shocks potentially can induce rapid changes to the economy while homicides respond more slowly to changes in economic conditions with a lag, following the initial period of the shock. In similar spirit, we distinguish migration shocks via a similar short-run restriction, namely that migration can respond quickly to crime shocks, while homicides change more slowly in response to migration shocks. Notice that in this manner, the response of the economy in the form of GDP, net migration, and FDI responses to crime shocks, is left unrestricted.

It is also worth noting that, as with any structural VAR identification scheme, short-run and long-run exclusion restrictions as we have described only identify the shocks up to a sign, meaning that our restrictions are sufficient to distinguish the four shocks from one another, but are not sufficient to determine whether the particular shocks were positive or negative shocks, for which we need additional restrictions. Consequently, we identify the signs of the shock as follows. A crime shock is identified as a positive shock if it increases homicides in the impact period, and a migration shock is identified as a positive shock if it increases net international in-migration in the impact period. Similarly an aggregate supply shock is identified as a positive shock if it permanently increases GDP in the long-run steady state time horizon. Finally, an aggregate demand shock is identified as a positive shock if it increases GDP in the short-run impact period. Each of these identifies the sign in a manner that is consistent with the economic conceptualization of the shock.

As with all structural VAR analysis, since the shocks are conceptual and are not directly observed, they are unit free shocks. To associate economic units with the shocks, one must therefore look to the response variables. Furthermore, one can use any of the response variables for this purpose. Thus, for example, one can study the effect on GDP of a crime shock that leads to an “x” percent increase in homicides. Or alternatively, one can study the effect on GDP of a crime shock that has a “y” percent increase in migration rates, and so forth. As is conventional in structural VAR analysis, we report impulse responses to the structural shocks symmetrically without associating units and allow the reader to choose any of the response variables as a means for scaling the shocks to an economic unit.

Since each of the variables is related to each of the four shocks, identification schemes are often depicted in matrix form. Thus, schematically, our set-up can be characterized as follows: where we have used a 0 to denote a zero restriction for the particular entry, we have used a + to indicate a sign restriction, such that while the value of the response is unrestricted, the sign of the response is dictated by our definitions of what constitute positive versus negative shocks. Finally, a * is used to denote a completely unrestricted value for the particular entry. (See Figure 3)

Figure 3.
Figure 3.

SVAR Identification Scheme

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

Thus, as we see from the schematic representation, our VAR system includes a total of six exclusion restrictions on either the short-run impact or long-run steady state responses, and a total of four sign restrictions. Along with the standard assumption that the structural shocks are orthogonal, this is sufficient to exactly identify the VAR system of shocks and impulse responses, with all of the dynamic transition responses completely unrestricted. For the panel framework, we allow a similar set of restrictions to apply to both the response of the variables to the idiosyncratic state-specific shocks as well as the common national or international shocks, so that for each structural shock we have both a common and idiosyncratic component for a total of eight structural shocks. For each structural shock, we then obtain impulse responses for each of the four variables for a total of 32 impulse responses for each of the 32 states of Mexico. As discussed in the previous subsection, once we have the collection of impulse responses for each state, in the next stage we use a bootstrapped estimator to investigate state-specific characteristics associated with the heterogeneous patterns of responses.

In the conventional structural VAR literature, it is well known that results can be sensitive to the choice of identifying restrictions, and it is no different for panels. Accordingly, in our analysis, we confirm that our key findings are robust to viable variations in the identification scheme. In this regard, an additional attractive feature of this identification scheme is that various blocks of the system can be also investigated separately to confirm robustness of some of the key underlying results. For example, if we are only interested in the relationship between per capita crimes and economic activity in general, we can examine a bivariate structural VAR which includes only log per capita homicides and log per capita GDP with similar identifying restrictions to confirm that the key patterns hold in the subsystem. We elaborate on this further in the results section of our paper.

Finally, it is worth noting a few important limitations. For example, the approach is naturally limited due to the measurement error inherent in the data, which we further discuss in the next section. Furthermore, as with any structural VAR analysis, interpretations are dependent upon the particular econometric identification scheme, which employs homicide data in part to capture the response of crime shocks, and is subject to assumptions that must be made on whether the class of relevant shocks has been captured by the set of econometric restrictions that are imposed on the data. In this regard, the research aims to contribute to the analysis of the relationships, but is not intended to be conclusive in its findings.

Before proceeding to our results, next we discuss the details of the various data sources for Mexico.

IV. Data Sources

A. Real GDP for the Period During 1993 to 2012

A key challenge in implementing our panel VAR is to estimate a sufficiently long series of RGDP for each of the Mexican states. INEGI has only recently published the RGDP aggregated data for 2003–2012 using base year 2008, but to date has not yet released the GDP aggregated data from 1993 to 2003 using the base year 2008.

There are a number of approaches to obtain real GDP values for multiple base years, among them, the annual chain-linked approaches.8 In this paper, we present two different approaches. In the first approach (RGDP-1), we link two aggregated RGDP series for each state: (i) the first series contains aggregated RGDP data from 1993 to 2003 with base year 1993, and (ii) the second series contains aggregated RGDP data from 2003 to 2012 with base year 2008. For year 2003, for each state, we simply compute the ratio between the aggregate RGDP with base 2008 and the aggregate RGDP with base 1993, and then multiply the aggregated RGDP series with base 1993 by this ratio to obtain RGDP series for 1993–2002 with base 2008. However, this method does not account for the heterogeneous ratios among all RGDP sectors. In order to fix this, in the second approach (RGDP-2), we apply the above method to calculate the ratio for each GDP sector to estimate the sector values with base 2008 for years 1993–2002. Then we sum up these sector values to obtain the aggregated RGDP estimates with base 2008 for years 1993–2002.

The methodological description of these two different approaches is included in the Technical Appendix 1. We also discuss each of the various robustness checks undertaken. In Figure 9, we compare each of the Mexican RGDP estimates using different approaches. For cases in which we use sectoral GDP estimates for the purposes of our second stage analysis (see section V), we restrict our attention to 2003–2012.

Figure 4.
Figure 4.

Mexico: Comparative Homicide Rates from Different Sources

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

Source: INEGI, SESNSP, CNDH, CONAPO, Heinle et al. (2014), and authors’ calculations.
Figure 5.
Figure 5.

Mexico: Homicide Rates vs. Incidence of Crime

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

Source: INEGI, SESNSP, CONAPO, and authors’ calculations
Figure 6.
Figure 6.

State Level Response of GDP to Idiosyncratic Crime Shocks

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

Figure 7.
Figure 7.

State Level Quantiles Responses to Idiosyncratic Shocks

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

Figure 8.
Figure 8.

Mexico: Response Patterns

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

Source: INEGI, Mexico’sSecretary of Economy, ENSI, ENVIPE, and authors’ calculations
Figure 9.
Figure 9.

Approaches for Linking Real GDP Series for Mexican States

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

B. Measures of Crime

Intentional Homicides

The second important task in implementing our panel VAR is the choice of a crime variable which is measured relatively well and which is likely to move in response to general crime shocks as we define them in our econometric identification scheme described in section II. Toward this end, and also following numerous other scholars (Fajnzylber et al., 2000; Detotto and Otranto, 2011; Forni and Paba, 2000; Cardenas, 2007; and Robles et al., 2003), the number of recorded intentional homicides are used here as the crime variable. The homicide rates are chosen for their highest reliability among all crime variables.9 Homicide data are of special interest because these crimes are usually thought to be the least affected by the problems of under-reporting and under-recording that afflict official crime statistics (Fajnzylber, 2000). Even for the United States, experts have frequently focused on homicides as a proxy for crime, not only because “it is a fairly reliable barometer of all violent crime,” but also because “at a national level, no other crime is measured as accurately and precisely.” (Fajnzylber, 2000; and Fox and Zawitz, 2000). However, as Heinle et al. (2014) point out, what is of particular concern regarding Mexico’s sudden increases in homicides in recent years (2007–2012) is that much or most of this could be attributable to organized crime groups. Although drug-related homicides are widely used to describe Mexico’s security challenges from 2007 to 2012, there is no formal definition of this concept in Mexican criminal law. 10 Finally, it is important to note that it is not possible to distinguish between the homicides that are a result of criminal activity, much less any particular category of criminal activity such as organized crime, and those deaths that are the product of social conflict, demographic phenomena and so forth.

Official data on homicides in Mexico are available from two sources. Also, a number of non-governmental sources have collected estimates of the number of homicides that are specifically related to the drug trafficking in Mexico. On the official side, public health records filed by coroner’s offices can be used to identify cases where the cause of death was unnatural, such as cases of gunshots wounds, stabbings, etc. While all the datasets have limitations, the most consistent, complete, and reliable source of information in Mexico is the autonomous government statistics agency, INEGI, which provides data on death by homicide (Heinle et al., 2014). We compile a data set of these INEGI data for 32 Mexican states during the period 1993 through 2012.

A second source of data on homicide comes from criminal investigations by law enforcement to establish a formal determination of criminal wrongdoing, and the subsequent conviction and sentencing of suspect charged with these crimes. The Executive Secretary of the National System for Public Security (SESNSP) compiles and reports data on cases involving homicide that are identified by law enforcement. In recent years, SESNSP has its homicide data on a monthly basis.11

As we can observe in Figure 4,12 there is a noticeable difference between public health and law enforcement homicide statistics, which appears to be attributable to the different timing and methodologies by which cases are classified. Still, data from the two sources (INEGI and SESNSP) are closely correlated and offer fairly consistent measures of the trends in overall homicide. Neither of the two official sources on homicide statistics identifies whether there is a connection to organized crime in a particular case. However, both government and independent sources have attempted to do so by examining other variables associated with a given crime.13

Some statistics on organized crime related homicides are available from SESNSP for a few years from 2006–2013. These are only comparable to some nonofficial organizations estimates from Mexico’s National Human Rights Commission (CNDH) for 2000–2008.14

Other Measures

As discussed in the previous section, once we have the collection of impulse responses for each state, in the next stage we use a bootstrapped estimator to investigate state-specific characteristics associated with the heterogeneous patterns of responses. Among those state-specific characteristics, we are particularly interested in exploiting all the incidence of crime data from the SESNSP database, in particular, the incidence of crime related to crimes against health and the federal law against organized crime (LFCDO), kidnapping, extortion, aggravated assault, and robbery.15 In Mexico, one of the possible classifications of crimes corresponds to their jurisdictional nature. Following this logic, crimes can be catalogued as local crimes (fuero comun) or federal crimes (fuero federal).16 This classification will indicate the authority responsible for their investigation and prosecution. Fuero comun crimes are those affecting individuals directly. Examples of such crimes are assault, robbery, threats, sexual crimes, frauds, homicides, extortion, kidnapping, among others.17 On the contrary, fuero federal crimes are characterized as the ones affecting health, the economy and the national security or interests of the federation. These include drug trafficking, organized crime, environmental crimes, firearm crimes, crimes committed by public servants, money laundering, people and child smuggling, and electoral crimes, among others.18 19 The SESNSP reports crime incidence information from initiated investigations by authorities, using data from the Institutional System of Statistical Information (Sistema Institucional de Información Estadística, SIIE). Since the crime incidence data refers to crimes of the federal and local jurisdiction, the information allows drawing conclusions on the macro and micro criminal situation in each state of the Mexico and at the national level. The conducts contemplated in the analysis by SESNSP refer to federal jurisdiction crimes which are typified in the Federal Criminal Code and Federal Special Laws. It has been mentioned that the reported data apparently refer to initiated investigations by MPF. Therefore, this likely would not give an accurate picture of the criminal scenario in Mexico, since data relating to the result of subsequent first and second judicial instances, and amparo procedures, are not reported.20

With regard to fuero federal, of particular interest is the incidence of crime related to crimes against health (i.e., drugs) and organized crime due to the economic activity associated with these types of crime. The top part of the Figure 5 shows the scatter plot of annual totals intentional homicides and incidence of federal crimes related to drug-related issues (production, traffic, etc) and organized crime from 2007 through 2012. However, there are some states in which the homicide rate is increasing, but the incidence of crime is decreasing. The latter could be related to two types of underreporting: (i) crimes are unobserved by the victim or authorities, (ii) crimes are known, but not reported. (See Shirk and Rios Cazares, 2007).

From the fuero comun perspective, we are interested in two crimes that generate violence: robbery and assault.21 The bottom part of the Figure 5 shows the scatter plot of annual totals intentional homicides vs. incidence of local fora crimes related to assault and robbery.22

C. Migration

We are also interested in the role that migration plays in the relationship between crime and economic activity. The Consejo Nacional de Poblacion (CONAPO) releases the demographic data at the state level from 1990 to 2010. It also forecasts the demographic data for 2011–2030. These data include the population, immigration, and emigration between Mexican states; immigration from and emigration to foreign countries; natural and social birth and death rates; life expectation; etc.

D. Foreign Direct Investment

As discussed in Section III, we are factoring in the role that FDI plays in the relationship with crime and economic activity. Mexico’s Secretary of Economy releases nominal FDI at the state level by sectors, origins, and investment types. The currency is in U.S. dollars. There are three investment types: new investments, reinvestment of profits, and intercompany accounts. Over 140 countries and economies are listed as the FDI origins. The U.S. consumer price index is used to deflate the nominal FDI to generate the real FDI.

E. State-Specific Characteristics

For the analysis of the stated characteristics associated with patterns in the distributions of crime, GDP, FDI, and migration, we use the following list of variables:

Average schooling years: A large body of evidence suggests that education and labor market opportunities influence criminal activity. Someone with a poor education and bleak labor market opportunities is more likely to commit a crime. According to Lochner and Moretti (2004), schooling significantly reduces criminal activity. In order to understand whether average schooling years matters for the Mexican states’ GDP, FDI, and migration responses to crime, we use the INEGI’s Statistical Yearbook By State, which compiles average schooling for adults from 2003-2012.

Unemployment rates: Although time series have failed to uncover a robust, positive, and significant relation between unemployment and crime (Fleisher, 1966; and Erlich, 1973), most studies based on cross-sectional and individual data point in that direction (Freeman, 1986). In order to understand whether the unemployment explains the Mexican states’ responses of GDP, FDI, and migration to crime, we use unemployment rates data compiled by INEGI based on the National Survey on occupation and employment (ENOE) from 2003 to 2012.

Bank Deposits: As discussed in International Monetary Fund (2001), criminal organizations and individuals sometimes rely on the legitimate banking system to hide illegally obtained assets. Financial institutions can be used as an instrumentality to keep or transfer the proceeds of a crime (IMF 2001). Although the circumstances vary from state to state in Mexico, sometimes where the criminal profits are laundered may be a different location from where the predicate crime was committed. To understand whether the size of banking assets explain the Mexican states’ GDP response to crime, we use the deposit data from the National Banking and Securities Commission of Mexico (CNBS) during 2000 through 2010. The data include the amount of deposits for immediate repayment purpose or investment purpose.

Marginalization indices: There is strong evidence that social exclusion renders individuals vulnerable to criminal behavior (Hale, 2005). We use CONAPO data on marginalization indices for each five-year period since 1990. The indices are based on: (i) literacy rates; (ii) percentage in dwellings without drainage or toilet; (iii) percentage of households without power or without piped water or some level of overcrowding in houses; and (iv) percentage of population with income up to two minimum wages.

Perception of Lack of Public Security:23 Periodic victimization surveys are the best tool that policy makers have for both detecting early trends and identifying the groups that are most at risk (Fajnzylber et al., 2000). INEGI published the household survey data on the perception of the lack of public security in ENSI (2005 and 2009), and Encuesta Nacional de Victimización y Percepción sobre Seguridad Pública (ENVIPE) (2010, 2011, and 2012). One common question in these surveys is “Do you think that living in your state is secure, insecure, or no response?” We use the percentage of the “secure” responses to measure the perception of public security in the state.

Government spending on security: Conventional wisdom focuses on the direct effect of greater policing in reducing the probability of a successful crime. Becker (1968) assumes that private and public preventive measures are substitutes and that if the state spends substantially on crime prevention, individuals need to spend less to achieve a given rate of arrest. For example, Lin (2009) suggests that a police force reduces crime. We take this into consideration and employ a measure of the government sending on security in the analysis of characteristics associated with patterns in the distribution of our main variables. Mora (2009) lists the public spending on security at the state level for years 2007, 2008, and 2009. The expenditure includes situational prevention, law enforcement, social reintegration, judiciary, ombudsman, and contributions to the Fund for Public Safety (FASP).

Informality: Yishay and Pearlman (2011) find a potential ambiguous response of formalization to robbery risks in Mexico, based on the dual potential effects of formality on raising the targeting of firms by criminals and improving protection by and recourse to official authorities. Despite this, we consider a measure of informality in the analysis of characteristics associated with patterns in the distributions of crime, GDP, FDI, and migration. The informality is reported by the ENOE.24

Disaggregated GDP: the disaggregated GDP includes the following sectors: agriculture, livestock and forestry fishing and hunting; electricity, gas and water supply for final consumer products; construction; manufacturing industries; commerce, transport, postal and storage; mass media information; financial services and insurance; real estate services; professional services, scientific and technical; corporate management companies; support services business and waste management and remediation services; education services; health and welfare services; leisure services, cultural and sports, and other recreational; temporary housing services and preparation of good and beverages, other services; and government activities and international organizations. For a detailed description of disaggregated GDP sector see Table 2 in the appendix and INEGI (2013a).

Table 1.

GDP Sectors for years 1993 and 2008

article image
Source: INEGI 2013b and 2013c
Table 2.

Names of Mexican States

article image

The disaggregated FDI:25 the disaggregated GDP includes the following sectors; agriculture, livestock and forestry fishing and hunting; electricity, gas and water supply for final consumer products; construction; manufacturing industries; wholesale; retail trade; transport, postal and storage; mass media information; financial services and insurance; real estate services; professional services, scientific and technical; corporate management companies; support services business and waste management and remediation services; education services; health and welfare services; leisure services, cultural and sports, and other recreational; temporary housing services and preparation of good and beverages, other government activities. For a detailed description of disaggregated FDI sectors see the reference Secretaría de Economía (2014)

In Table 3 of the appendix, we summarize the list of variables used in this paper as well as the data sources.

Table 3.

Data Sources

article image
article image
Notes: Most data is used in log per capita form. For the first stage VAR estimation, Homicides, GDP and Net In-migration were used in log per capita growth rates, and the subsequent impulse responses were accumulated. These transformations are discussed in the text of the paper.

V. Results of the Empirical Analysis

In this section, we summarize some of the key findings of the empirical analysis. Since we are interested in both the short-run and long-run dynamic responses of our variables and use a combination of short-run and long-run identifying restrictions, one of the first empirical features of the data that we examine is the stationary properties. If variables are stationary, it implies that in response to any shock, over time they will revert back to their mean values. Consequently, for a variable that is stationary, no shock can cause a permanent change. Only transitory changes are possible. By contrast, for variables that are nonstationary due to the presence of a unit root, at least some shock must be present that can cause a long-run permanent change in the variable. This carries important practical implications for our analysis. Firstly, there can be no lasting long-run change in a stationary variable; and secondly, a stationary variable does not contain information about shocks that may be responsible for causing a lasting long-run change in a nonstationary variable that follows a unit root process. Consequently, for the first stage of our analysis, we investigate the stationary properties of the data.

In particular, we use standard panel unit root tests such as the one developed by Im, Pesaran, and Shin (2003) to test our data.26 Not surprisingly, we are unable to reject the null hypothesis that log per capita GDP follows a unit root for each of the states of Mexico, and that this holds regardless of which method we use to link the two GDP periods before and after 2003. Given that the tests have high power, this implies that GDP likely follows a unit root process at the state level, and therefore it is possible for some shocks in our VAR analysis to permanently lower or raise state level GDP.

By contrast, for the crime data, only per capita homicides, as measured by INEGI were similarly found to follow a unit root process at the state level. By contrast, other crime measures, such as the SESNSP data on homicides, violent crimes, organized crime, or crimes against health appear to be stationary in the sense that the null hypothesis of a unit root in all states is rejected in favor of stationarity. This implies that these other series revert to their respective means in response to any changes, and therefore any shocks in a VAR system can have only transitory effects on these. Similarly, shocks that drive these data cannot have long -term effects on variables that follow unit root process, such as GDP. If there are long-term causal effects present between crime and economic activity measured by GDP, they are likely to be reflected in the INEGI homicide data, and not these other measures. For these reasons, as well as the reasons discussed in the previous section, we use the INEGI homicide data as our primary crime measure in our identified structural VAR estimation. However, this is not to say that the information contained in these other measures might not interact to play a role in influencing the nature of the dynamic responses of GDP and INEGI recorded homicides at the state level, and therefore we employ these other variables in our second stage analysis to investigate the pattern of heterogeneous responses among the states.

Next, for our migration data, we find that most of the measures for internal migration are stationary. By contrast, per capita international net in-migration appears to follow a unit root at the state level, so that it is possible that it contains information regarding shocks that can cause permanent changes in the INEGI homicides data. Since our other variables are measured in logs, we similarly convert our migration data to log form. To address the fact that our migration data is net in-migration, and therefore can take on positive or negative values, we use a fairly standard approach to convert such series to logs. Specifically, we take the largest negative value and this value plus 1.0 to all values of per capita net in-migration, so that upon taking natural logs all values are positive. Adding a constant to the series in this fashion affects the value of the fixed effect intercepts in a VAR estimation, but does not affect the dynamic impulse response or variance decomposition estimates, which are the objects of interest. Similar to the crime data, we use some of the other stationary migration data for our second stage analysis to investigate whether they contain information which helps to explain the pattern of heterogeneous responses among states.

As the fourth variable in our VAR system, FDI is relatively unrestricted econometrically. Since we find that the flow variable for FDI is stationary, and we are interested in the relationship to the stock, we use accumulated net FDI for the purposes of our analysis. Since this variable is also measured as a net inflow, we use a similar method as we did for international per capita net in-migration to compute the log of per capita accumulated FDI.

Consequently, as alluded to in section III, the four state-level variables that we use in our most general VAR specification are log per capita INEGI recorded homicides, log per capita international net in-migration, log per capita GDP, and log per capita accumulated net FDI. The four categories of structural shocks are crime shocks, migration shocks, AS shocks, and AD shocks. For a more detailed motivation for these shocks and for a discussion of the structural identifying restrictions used to distinguish these shocks, please see section III.

One of the first things that we consider is the relative importance of the various shocks in driving the variations in the observed variables. This is reflected in the variance decompositions, which are depicted in Figure 11. Not surprisingly, the primary shock driving variations in homicides are the crime shocks, and the primary shocks driving variations in GDP are economic AS shocks. For example, the median percentage of GDP variation explained by state level aggregate supply shocks is roughly 70 percent. What is of note is that the median percentage of variation in state GDP driven by state-specific crime shocks is roughly five percent with another half a percent driven by common national or international crime shocks. If we consider the seventy-fifth percentile, this rises to as much as 10 percent due to state-specific crime shocks and another one percent due to national or international crime shocks. In other words, this tells us that for at least a quarter of the Mexican states, crime shocks are responsible for driving up to 11 percent of the variation in state GDP. To put this in perspective, according to Figure 11 these values are roughly comparable to the percentage variation in state GDP that is attributable to economic aggregate demand shocks, which would include private sector changes in demand for resources, as well as changes in government spending that stimulate the economy. Similarly, we see that crime shocks play a substantial role in driving the variation in both migration and FDI. In the case of migration, for at least a fourth of states, state-specific crime shocks initially account for 10 percent of the variation in the initial period of the shock, rising to close to 20 percent five years after the shock. For FDI, the median state variation explained by crime shocks is not much, at roughly one percent. However, for the states most affected, which lie in the top quartile, up to eight percent of the variation is driven by crime shocks in the initial period of the shock, rising to as much as 15 percent by five years after the shock.

Figure 10.
Figure 10.

State Level Quantiles for Impulse Responses to Idiosyncratic and Common Shocks

(Four variable system, 1993-2012, GDP approach 2)

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

Figure 11.
Figure 11.

State Level Quantiles for Variance Decompositions for Idiosyncratic and Common Shocks

(Four variable system, 1993-2012, GDP approach 2)

Citation: IMF Working Papers 2015, 121; 10.5089/9781513541969.001.A001

Similarly, one can examine which shocks beyond the crime shocks are most important for driving homicide rates. Here again, we see a large spread among the states. Whereas the variation in crime in some state, in other words those in the lowest quantile, attributable to other shocks is relatively low, in a few states, the contribution of is substantial. For example, in the top quartile states, over 10 percent in the variation in crime can be attributed to state-specific migration shocks, with a little over three percent each to state-specific AS shocks and AD shocks at the longer time horizons following the shocks.

Of course, while these variance decompositions point to the relative importance of the various shocks in explaining variations in the observed variables, they tell us nothing about the signs, magnitudes, or duration of the responses to these shocks. For this, we turn to the impulse response analysis. The quantile impulse response estimates for our general four variable VAR are reported in Figure10 for both the idiosyncratic and common shocks. However, before exploring these, it is worth delving a bit deeper into some basic robustness checks to confirm that our results are not specialized to some of the choices that we have made regarding the variables and the identification scheme. Toward this end, we investigated multiple variations in the identification and multiple variations in the data choices to confirm that results were similar. For example, as discussed in the previous section, we examined five different methods for chaining together the state-level GDP estimates into a single base year from 1993 to 2012. We explored our panel VAR estimation for all of these different methods and found that the key results were qualitatively invariant to the choice of method for chaining the GDP segments. As described in section III, a nice feature of our identification scheme is that it also permits us to examine subsystems of the general VAR, based on two and three variable blocks, to confirm that our primary results also hold for these specifications.

To see an example of this analysis, in Figure 6, we report one of the key sets of impulse responses, namely the quantile responses of log per capita GDP to crime shocks. Since the variance decompositions point to a more prominent role for the idiosyncratic state-specific crime shocks, we focus here on the responses to these idiosyncratic crime shocks. Thus, for example, Figures 6a and 6b show the quantile responses of state GDP to state-specific crime shocks based on two different methods for chaining together the GDP segments. While there are small quantitative differences, the shape of the response, and even the overall magnitudes are remarkably similar. For example, we see that the point estimate for the median response among states is initially slightly negative and eventually moves to approximately -0.005. The twenty-fifth percent quantile starts out at -0.004 and continues to drop to around -0.16 by five years after the shock. To translate this into economic terms, this implies that for at least a quarter of the Mexican states, a ceteris paribus shock to crime reduces real per capita GDP by 1.6 percent per year after a period of five years following the shock.

As discussed in section III B, to understand the magnitude of the shocks themselves, one can scale them by the units associated with any of the response variables, as depicted in the appendix figure 10. Thus for example, since the median state homicide rate increases by 0.2 percent in the initial period in response to the crime shock, one can say that a crime shock that induces an initial 0.2 percent increase in homicides causes a 0.5 percent decrease in GDP for the median state. Notice that this does not imply that a change in homicides causes GDP to move by this amount. Indeed, if the cause of the movement in homicides is due to one of our other shocks, such as an aggregate supply shock or an aggregate demand shock, then the multiplier between homicides and GDP will be very different. Finally, notice that one need not scale the crime shock by homicides. For example, since the median state accumulated FDI decreases by 0.1 percent by the fifth year following the crime shock, one could just as well describe these results by saying that a crime shock which decreases accumulated FDI by 0.1 percent over a five-year period for the median state causes a 0.5 percent decrease in GDP for the median state. It should be clear from this discussion that crime shocks are not being measured nor proxied by homicide rates. Rather, homicide rates, in conjunction with the other variables, assist in identifying crime shocks based on the structural VAR identification scheme. We elaborate more on the interpretation of these results later, as well as the confidence bands associated with these estimates.

First, we continue to examine the qualitative robustness of the results. For example, in Figures 6c and 6d, we see a similar comparison of results for the two GDP methods when we use a three variable specification for the VAR that excludes log per capita international net in-migration. Since we use similar identifying restrictions for the remaining variables, the interpretation of the shocks may be slightly altered for these relative to the four variable specification. In particular, some migration shocks are now potentially lumped into what we call the crime shock, while only AS and AD shocks are controlled for in the other shocks. Accordingly, the impulse responses change quantitatively somewhat relative to the four variable specification. However, they are still remarkably similar in pattern to the four variable specification, and the results for the two GDP measures are again remarkably similar. Similarly, we also examined a streamlined two variable system, with just log per capita homicides and log per capita GDP. Under this specification, we no longer control for AS and AD shocks separately, but rather control for a single economic activity shock that encompasses both AS and AD. Again, there are small quantitative differences, but the basic pattern remains with a clearly heterogeneous response of GDP to the crime shocks, with again remarkable similarity among the two GDP measures.

The two variable system also affords us another useful opportunity relative to the three and four variable systems. Specifically, since the number of degrees of freedom required to estimate the lag structure for a two variable VAR system is considerably less than for a three or four variable VAR system, for the two variable system we are able to examine shorter spans of time. Thus, we take the opportunity to examine the impulse responses when we restrict the sample to 2003 to 2012. We choose this period for three reasons, namely (i) since the break in the base period for the original data occurs in 2003, it allows us a robustness check to confirm that chaining the GDP measures together across the two different periods does not alter our results; (ii) using 2003 as the sample break splits our sample into exactly two equal segments of 10 years; and (iii) we are particularly interested in the later period and take the opportunity to confirm that our general results are likely representative not only of the entire period, 1993–2012, but also of the more recent period, 2003-2012. Finally, to check our chaining methods, we also ran each of the full sample VAR specifications with a dummy break for the pre- and post-2003 segments and confirmed that the results are not affected by the use of a dummy for the two periods.27

Next, we examine in greater detail the various impulse responses that are of primary interest, for which we use the more general four variable VAR specification with our RGDP-2 data. Specifically, in Figure 7, we look in greater detail at the impulse responses of each of the variables to the state-specific crime shocks, as well as the response of crime to each of our other three shocks. In each case, we also report the bootstrapped 90 percent confidence bands based on 5,000 bootstrap draws to confirm that the heterogeneous spread in the individual state responses is statistically significant. Thus, for example in Figure 7b, we see the responses of state-level log per capita GDP to state-specific crime shocks. This time, however, we have also included the 90 percent confidence bands around both the upper and lower quantile responses. As we can see, these are both significantly different than zero. Thus, we can say with 90 percent confidence, equivalent to statistical significance at the 10 percent level or better that for at least a quarter of the states log per capita GDP decreases in response to crime shocks, while for at least a quarter of the states log per capita GDP increases. Furthermore, for the 25 percent or more states where log per capita GDP falls in response to a crime shock, we can say with 90 percent confidence that annual GDP falls by at least half a percentage point, and that this decrease appears to be permanent. On the other hand, for at least 25 percent of the states, we can say with 90 percent confidence that GDP actually rises by a small amount, though we cannot say that it is necessarily greater than around a fifth of a percent of GDP.28 Our subsequent second stage analysis investigates for which types of states we can expect the response to be positive versus negative.

First, we discuss the other key impulse responses. For example it is interesting to see that there are other statistically significant spreads in the individual state responses to the various shocks. For example, when we look at Figure 7c, we see that the point estimates for the state median of the cumulative response of FDI to crime shocks is slightly negative as indicated by the red line. The blue band illustrates that there is a 90 percent confidence band that is significantly below zero for at least 25 percent of the states. For these states, by the time of the fifth year following the shock, cumulative FDI has decrease by somewhere between one percent and six percent. At the same time, the green band indicates that for at least 25 percent of the states’ FDI actually increases following a crime shock. In the second stage analysis, we investigate which state characteristics are associated with this pattern of responses of FDI to crime shocks.

The prevailing pattern in all of the impulse responses depicted in Figure 7 is that regardless of whether the state median is above or below zero, there is in each case a group of states whose responses are either positive or negative with statistical significance at the 10 percent level or better (i.e., the 90 percent confidence band is entirely above or below the zero line). This is true for the response of migration to crime shocks as well as the response of homicides to each of the structural shocks. This is an important feature of the complexity of the relationship between crime and economic activity in Mexico. The relationship is by no means uniform, and there are differences in the patterns of responses that are statistically significant with a relatively high level of confidence.

Next we discuss the second stage analysis in which we look for characteristics of the state that are associated with the heterogeneous responses among the different states. For example, what characteristics of the state are associated with a positive or negative response of log per capita GDP to a crime shock? Similarly, what characteristics of the state are associated with larger or smaller changes in log per capita accumulated FDI in response to a similar crime shock? To investigate this, one might consider computing correlations, or even simple linear regressions between the response estimates for each of the 32 states and measured characteristics of each of the states. However, the problem with this approach is that it treats the impulse responses as raw data and does not adequately account for the uncertainty associated with the fact that state-level impulse responses are themselves estimated. To address this, we instead use a bootstrapped estimator for the regression correlation between the state-specific responses and the measured state characteristics. Specifically, the bootstrapped estimator uses the same bootstrapped sampling distribution that was used to compute the confidence bands for the impulse responses and uses this to resample the regression correlation. Thus, for each of the 5,000 bootstrap draws, we compute 5,000 impulse responses, which we regress 5,000 times on the state characteristic measures. The 5,000 slope estimates give us a sample distribution for the bootstrapped regression correlation estimator. The tails of this distribution give us the 90 percent confidence bands for the regression correlation estimator and allow us to test the significance of the relationship in a way that accounts for the underlying uncertainty associated with the estimated impulse responses.

The state-specific characteristics that we consider are described in section IV and include average schooling years, unemployment rates, bank deposits, marginalization indices, perceptions of public security, government spending on security, and a measure of the importance of economic informality. In addition to these, we also investigate state characteristics associated with the concentration of different economic sectors as a percentage of state GDP, the concentration of the sectors to which FDI flows, as well as the various other crime measurements and migration measurements previously discussed. Whenever we have multiple observations over time for any of these variables for a given state, we use the sample average for that state. Whenever the measures are economic quantities, we use them in log per capita form.

A number of different characteristics appear to be statistically significant under certain scenarios. However, with so many characteristics and so many different impulse response configurations, we must take care not to inadvertently data mine the results. For example, at a five percent significance level, with so many characteristics and responses, we expect some significant responses to appear randomly and spuriously for certain specifications. Instead, we focus here only on the results that appear consistently significant for all of the various specifications. Among these, two stand out as particularly robust. The first is the relationship between the response of GDP to crime shocks and the percentage of GDP devoted to construction.29 Figure 8a shows a scatterplot for this relationship. Specifically, the vertical axis depicts the percentage of state GDP devoted to construction. On the horizontal axis, we represent the response of log per capita GDP in the second year following the crime shock. Since the crime shock can have differing effects on homicides in different states as evidenced by the impulse responses in Figure 10, we also scaled this response by the magnitude of the first period homicide response to the crime shock. Thus, the horizontal axis can be interpreted as the percentage change in per capita GDP in the second year following a crime shock scaled by the percentage change in homicides caused by the same crime shock in the period in which the shock occurred. The positive association in the scatterplot implies that for states in which construction is more prominent, a crime shock is more likely to have a positive (or less negative) effect on overall state-level GDP.3031

The bootstrap estimator for this relationship indicates that the bootstrapped median estimator for the slope is 1.091 for the GDP-1 based VAR and 0.924 for the GDP-2 based VAR. The bootstrapped ninety-fifth percentile value for the GDP-1 based VAR is 2.065, and the fifth percentile value for the same is 0.112. Thus, very conservatively, taking into account all of the uncertainty associated with first stage VAR analysis, we can say that we have greater than 90 percent confidence that the relationship is positive and statistically significant. The values for the GDP-2 based VAR are similar, though with not quite as tight a confidence band, with the ninety-fifth percentile value at 1.944 and only the tenth percentile value in the positive range, also at 0.112.

The other state characteristic that stands out as robustly significant in many different scenarios is the sense of public security measure, and in particular in regard to the responsiveness of FDI to crime shocks. Accordingly, in Figure 8b, we present a scatterplot of the relationship between the second period response of FDI to a crime shock, again scaled by the effect that the crime shock has on homicides, and the corresponding measure of the sense of public security in the state. Here the positive association implies that the greater the sense of security, the more likely the per capita accumulated FDI response is to be less negative to a crime shock. Conversely, the higher the sense of insecurity, the more likely it is that a crime shock will induce a larger cumulative decrease in FDI.

The bootstrapped median estimator for the slope is 0.006 for the GDP-1 based VAR and 0.005 for the GDP-2 based VAR. The bootstrapped ninety-fifth percentile values for the two are both 0.01, and the fifth bootstrapped fifth percentile values are 0.002 and 0.001 respectively. Thus, in both cases we can say conservatively that, after taking into account all of the uncertainty associated with first stage VAR analysis, we are 90 percent confident in both cases that the relationship is statistically significant and positive.

VI. Summary and Conclusions

In this paper we have used a panel structural VAR approach to study the dynamic relationship between crime and economic activity, including FDI flows and international migration flows at the state level in Mexico from 1993–2012. We find that these relationships are both highly inter-related and also potentially very diverse among the different states of Mexico. On the whole, crime is shown to be a relevant factor of the economies of many of the states and to have an impact on the overall economic activity at the state level. However, we find that the relationships between crime and economic activity at the state level are not as simple as one might have thought, and are instead fairly nuanced.

In general, the overall picture that emerges from our analysis is that crime has been intertwined with other forms of economic activity in the states of Mexico from 1993-2012, including international factor flows. However, the relationships are complex and by no means uniform across the country. As with any important feature of an economy, a better understanding of the relationships at the aggregate economic level would be useful for the purposes of strengthening the policy response to crime and increasing potential economic growth.

In particular, the economic responses to crime are quite different among the states, as is the relative importance of crime for the economies of the various states of Mexico, with different quantiles of states experiencing different variations in GDP, FDI, and migration flows in response to crime shocks. To reiterate some of our quantitative results, we find that for the median state, a typical state-specific crime shock induces a roughly one-half percent temporary decrease in per capita GDP which persists up to two years after the shock before gradually dissipating after the third year. However, the median state responses obscure the fact that for some states the magnitude of the response is much greater. For example, for the most negatively affected quantile of states, we can say with 90 percent confidence that a typical crime shock induces at least one-half percent decrease in GDP that is permanent and does not dissipate. Furthermore, we find that the factor flow dynamics associated with these crime shocks, for both international migration and FDI, are by no means uniform across states, with a substantial number of states experiences factor flows in opposing directions.

Consistent with other findings in the literature, our study also shows that the percentage of GDP devoted to construction in a particular state is associated with the size and magnitude of the economic response to crime, as is also the overall sense of public security at the state level as measured by survey data. Finally, it is worth noting that our sample covers the period from 1993 through 2012, and therein does not account for the reforms that have occurred in the period since then. Furthermore, our analysis is primarily intended to focus on the state level.

A few of the policy relevant implications of our analysis are fairly straightforward. For example, one of the more robust findings in our second stage analysis of the patterns of responses among the states to crime shocks is that the overall sense of security on the part of the public as reflected in survey responses appears to be important. Systematically, when the sense of security is higher, the impact of an equal-sized crime shock has a less negative impact on overall economic activity at the state level and in particular on FDI flows. This points to the potential importance of the perception and public confidence in the quality of institutions which provide for public security. Mexico has already put forward an ambitious structural reform agenda, including initiatives improving the rule of law. Judicial reform at the federal level in Mexico was approved by Congress in 2008, and it is expected that this reform will be in place in all states by 2016. It will be interesting to see whether these judicial reforms contribute to an improvement in the sense of overall security and therein help to mitigate the impact of crime on overall economic activity at the state level.

Finally, in light of the economic aspect of crime and the degree to which we have shown how crime and economic activities have been interwined at the state level, both in response to local idiosyncratic as well as national and international common shocks, it will be interesting to see the extent to which current efforts toward a further strengthening of the AML/CFT regime, particularly in areas of high ML/TF related risks, might also help to further improve future potential economic growth prospects for Mexico.

VII. References

  • Abadie, Alberto and Javier Gardeazabal, 2003, “The economic costs of conflict: A case study of the Basque Country,American Economic Review, Vol. 93, No. 1, pp. 113132.

    • Search Google Scholar
    • Export Citation
  • Anderson, David A., 199, “The aggregate burden of crime,Journal of Law and Economics, Vol. 42, pp. 611642.

  • Andreas, Peter, 1998, “The Political Economy of Narco-Corruption in Mexico,Current History, April, pp. 160165.

  • Andreas, Peter and Joel Wallman, 2009, “Illicit markets and violence: what is the relationship?Crime, Law and Social Change, Vol. 52, No. 3, pp. 225229.

    • Search Google Scholar
    • Export Citation
  • Angrist, Joshua and Adriana Kugler, 2008, “Rural Windfall or the New Resource Curse? Coca, Income and Civil Conflict in Colombia,Review of Economics and Statistics, Vol. 90, No. 2, pp. 191215.

    • Search Google Scholar
    • Export Citation
  • Anselin, Luc, 1995, “Local Indicators of Spatial Association—LISA,Geographic Analysis, Vol. 27, No. 2, pp. 93115.

  • Ashby, Nathan J. and Miguel A. Ramos, 2013, :Foreign direct investment and industry response to organized crime: The Mexican case,European Journal of Political Economy, Vol. 30, Issue C, pp. 8091.

    • Search Google Scholar
    • Export Citation
  • Astorga, L., 2005, El siglo de las drogas (Ciudad de México: Plaza y Janés).

  • Astorga, L., “A Pax Mafiosa in Mexico?,Project Syndicate, August 11, 2010. Available via the Internet: http://www.project-syndicate.org/commentary/a-pax-mafiosa-in-mexico-.

    • Search Google Scholar
    • Export Citation
  • Australia, Australian Bureau of Statistics (ABS), 1998, Introduction of Chain Volume Measures in the Australian National Accounts, Information Paper No. 5248.0 (Canberra).

    • Search Google Scholar
    • Export Citation
  • Badinger, Harald and Peter Egger, 2011, “Estimation of higher-order spatial autoregressive cross-section models with heteroscedastic disturbances,Regional Science, Vol. 90, No. 1, pp. 213235.

    • Search Google Scholar
    • Export Citation
  • Banco de Informacion Economica, Instituto Nacional de Estadistica y Geografia. Available via the Internet: http://www.inegi.org.mx/sistemas/bie/default.aspx.

  • Barnes, P. and R. Oloruntoba, 2005. “Assurance of security in maritime supply chains: conceptual issues of vulnerability and crisis management,Journal of International Management, Vol. 11, pp. 519540.

    • Search Google Scholar
    • Export Citation
  • Barone, Raffaella and Donato Masciandaro, 2008, “Worldwide anti-money laundering regulation: estimating the costs and benefits,Global Business and Economics Review, Vol. 10, No. 3, pp. 243264

    • Search Google Scholar
    • Export Citation
  • Barro, Robert J., 1996, “Democracy and Growth,Journal of Economic Growth, Vol. 1, pp. 127.

  • Becker, G S., 1968, “Crime and Punishment: An Economic Approach,Journal of Political Economy, Vol. 76, pp. 169217.

  • Ben Yishay, Ariel and Sarah Pearlman, 2011, “Crime, Informality and Microenterprise Growth: Evidence from Mexico.

  • Brock, G. J., 1998, “Foreign direct investment in Russia’s regions 1993–95. Why so little and where has it gone?The Economics of Transition, Vol. 23, pp. 397415.

    • Search Google Scholar
    • Export Citation
  • Buhaug, Halvard and Jan Ketil Rod, 2006, “Local determinants of African civil wars, 1970-2001, Political Geography, Vol. 25, No. 3, pp. 315335.

    • Search Google Scholar
    • Export Citation
  • Buhaug, Halvard and Kristian S. Gleditsch, 2008, “Contagion or Confusion? Why Conflicts Cluster in Space,International Studies Quarterly, Vol. 52, No. 2, pp. 215233.

    • Search Google Scholar
    • Export Citation
  • Bourguignon, F., 2001, “Crime as a social cost of poverty and inequality: a review focusing on developing countries,” in S. Yusuf, S. Evenett, and W. Wui (eds), Facets of Globalization International and Local Dimensions of Development, World Bank Discussion Paper Series (WDP) 415 (Washington: World Bank).

    • Search Google Scholar
    • Export Citation
  • Bourguignon, F., 2000, “Crime, violence and inequitable development,” in B. Pleskovic and J. Stiglitz (eds), Annual Bank Conference in Development Economics: 1999 (Washington: World Bank).

    • Search Google Scholar
    • Export Citation
  • Branzei, O. and S. Abdelnour, 2010, “Another day, another dollar: enterprise resilience under terrorism in developing countries,Journal of International Business Studies, Vol. 41, pp. 804825.

    • Search Google Scholar
    • Export Citation
  • Brock, G. J., 1998, “Foreign direct investment in Russia’s regions 1993–95. Why so little and where has it gone?The Economics of Transition, Vol. 23, pp. 397415.

    • Search Google Scholar
    • Export Citation
  • Buvinic, Mayra and Andrew Morrison. 1999, “Technical Notes: Violence Prevention.” (Washington: Inter-American Development Bank (IDB)).

    • Search Google Scholar
    • Export Citation
  • Brand, S. and R. Price, 2000, “The economic and social cost of crime,Home Office Research Studies #217 (London: Home Office).

  • Calderón, G., A. Díaz-Cayeros, B. Magaloni, and G. Robles, 2013, “The temporal and spatial dynamics of violence in Mexico,Working Paper (Stanford, California: Stanford University).

    • Search Google Scholar
    • Export Citation
  • Camara de Diputados, “Encadenamiento de series históricas del producto interno bruto de Mexico 1970–2001,Centro de Estudios de las Finanzas Públicas (CEFP), April 2003. Available via the Internet: http://www.cefp.gob.mx/intr/edocumentos/pdf/cefp/cefp0122003.pdf

    • Search Google Scholar
    • Export Citation
  • Càrdenas, M., 2007, “Economic growth in Columbia: a reversal of fortune?Ensayos Sobre Poltíica Económica, Vol. 25, pp. 220259.

    • Search Google Scholar
    • Export Citation
  • Chatterjee, I. and Ray, R., 2009, “Crime, corruption and institutions,Discussion Paper No. 20/09 (Victoria, Australia: Monash University).

    • Search Google Scholar
    • Export Citation
  • Castillo, J. C. and D. Mejía, 2012a, “Narcotráfico y violencia en México: Las razones más allá de Calderón,Blog Foro Económico, October 23, 2012. Available via the Internet: http://focoeconomico.org/2012/10/23/narcotrafico-y-violencia-en-mexico-las-razonesmas-alla-de-calderon/

    • Search Google Scholar
    • Export Citation
  • Castillo, J. C., D. Mejía, and P. Restrepo, 2012b, “Illegal drug markets and violence in Mexico: The causes beyond Calderón,” (Bogotá, Colombia: Universidad de los Andes).

    • Search Google Scholar
    • Export Citation
  • Carter, David, 2010, “The Strategy of Territorial Conflict,American Journal of Political Science, Vol. 54, No. 4, pp. 969987.

  • Chabat, J., 2006, “Mexico: The security challenge,Centro de Investigación y Docencia Económicas (CIDE) Working Paper 140 (Mexico City, Mexico: Centro de Investigación y Docencia Económicas).

    • Search Google Scholar
    • Export Citation
  • Chabat, J.,Combatting drugs in Mexico under Calderón: The inevitable war,CIDE Working Paper 205, December 2010. Available via the Internet: http://www.libreriacide.com/librospdf/DTEI-205.pdf

    • Search Google Scholar
    • Export Citation
  • Chen, Shyh W., 2009, “Investigating causality among unemployment, income and crime in Taiwan: evidence from the bounds test approach,Journal of Chinese Economics and Business Studies, Vol. 7, pp. 115125.

    • Search Google Scholar
    • Export Citation
  • Cifuentes Cruz, Sonia Milena, Marie Monique Pinot de Libreros, Andrés David García Avila, Humberto Mora Zambrano, Jorge Armando Herrera Barrera, 2014, “Compilacion y sintesis de las metodologias internacionales aplicadas a procedimientos de retropolacion,” Available via the Internet: http://www.dane.gov.co/candane_2014/images/DT_DANE/wp_cuentas_retropolacion.pdf

    • Search Google Scholar
    • Export Citation
  • Clegg, B. and S. J. Gray, 2002, “Australian expatriates in Thailand: some insights for expatriate management policies,International Journal of Human Resource Management, Vol. 13, pp. 598623.

    • Search Google Scholar
    • Export Citation
  • Correa, V., A. Escandon, R. Luengo, and J. Venegas, 2002, “Empalme PIB: Series Anuales y Trimestrales 1986–1995, Base 1996,Documento Metodológico, Working Paper No. 179 (Santiago, Chile: Banco de Chile).

    • Search Google Scholar
    • Export Citation
  • Coscia, M. and V. Ríos, 2012, “Knowing where and how criminal organizations operate using webcontent,study presented at the 21st ACM International Conference on Information and Knowledge Management (CIKM 2012), Maui, United States, October 29–November 2, 2012.

    • Search Google Scholar
    • Export Citation
  • CNBV (La Comisión Nacional Bancaria y de Valores), Boletín Estadístico de Banca Múltiple, 2000–2012. Available via the Internet: http://www.cnbv.gob.mx

    • Search Google Scholar
    • Export Citation
  • CONAPO (Consejo Nacional de Poblacion), Nuevos patrones de la migración Mexico—Estados Unidos. Available via the Internet: http://www.conapo.gob.mx/work/models/CONAPO/intensidad_migratoria/pdf/Nuevos_patrones.pdf

    • Search Google Scholar
    • Export Citation
  • CONAPO (Consejo Nacional de Población), Datos de Proyecciones 2010–2015, June 24, 2014. Available via the Internet: http://www.conapo.gob.mx/es/CONAPO/Proyecciones_Datos

    • Search Google Scholar
    • Export Citation
  • CONAPO (Consejo Nacional de Población), Índices de Marginación, May 7, 2012. Available via the Internet: http://www.conapo.gob.mx/es/CONAPO/Indices_de_Marginacion

    • Search Google Scholar
    • Export Citation
  • CONEVAL (Consejo Nacional de Evaluación de la Política de Desarrollo Social), 2012, Medición de la Pobreza por Municipio. Available via the Internet: http://web.coneval.gob.mx

    • Search Google Scholar
    • Export Citation
  • Cornelius, Wayne A. and David A. Shirk, 2007, Reforming the administration of justice in Mexico (University of Notre Dame Press)

  • Czabanski, Jacek, 2008, Estimates of cost of crime: history, methodologies, and implications (Berlin: Springer).

  • Czinkota, M.R., 2005, “Positioning terrorism in management and marketing: research propositions,Journal of International Management, Vol. 11, pp. 581604.

    • Search Google Scholar
    • Export Citation
  • Czinkota, M.R., G. Knight, P.W. Liesch, and J. Steen, 2010, “Terrorism and international business: a research agenda,Journal of International Business Studies, Vol. 41, pp. 826843.

    • Search Google Scholar
    • Export Citation
  • Daniele, Vittorio, and U. Mariani, 2011, “Organized crime, the quality of local institutions and FDI in Italy: a panel data analysis,European Journal of Political Economy, Vol. 27, pp. 132142.

    • Search Google Scholar
    • Export Citation
  • Dell, M., 2012, “Trafficking networks and the Mexican drug war,mimeo (Boston, Massachusetts: Harvard University).

  • Díaz-Cayeros, A., B. Magaloni, A. Matanock, and V. Romero, 2012, “Living in fear: Mapping the social embeddedness of drug gangs and violence in Mexico,Working Paper (Stanford, California: Stanford University).

    • Search Google Scholar
    • Export Citation
  • Detotto, C. and E. Otranto, 2009, “Does more crime mean fewer jobs? An ARDL model,CRENos Working Paper 2009/05.

  • Detotto, C. and E. Otranto, 2010, “Does crime affect economic growth?Kyklos, Vol. 63, pp. 330345.

  • Detotto, C. and M. Pulina, 2012, “Does more crime mean fewer jobs and less economic growth?European Journal of Law and Economics, Vol. 36, pp. 183207.

    • Search Google Scholar
    • Export Citation
  • Detotto, C. and Marco Vannini, 2009, “Counting the cost of crime in Italy,Workshop on Applied Analyses of crime: Implications for cost-effective criminal justice policies, Porto Conte Research, October 26, 2009.

    • Search Google Scholar
    • Export Citation
  • Donnelly, Robert A. and David A. Shirk, 2010, Police and Public Security in Mexico (San Diego, California: Trans-border Institute).

  • Dube, Arindrajit, Oeindrila Dube, and Omar Garcia-Ponce, 2013, “Cross-Border Spillover: U.S. Gun Laws and Violence in Mexico,American Political Science Review, Vol. 107, No. 3, pp. 397417.

    • Search Google Scholar
    • Export Citation
  • “Ebbing Mexican wave, The: The world’s biggest migration has gone into reverse,” The Economist, November 24, 2012. Available via the Internet: http://www.economist.com/news/special-report/21566781-worlds-biggest-migration-has-gone-reverse-ebbing-mexican-wave

    • Search Google Scholar
    • Export Citation
  • Enamorado, Ted, Luis F. Lopez-Calva, Carlos Rodriguez-Castelan, 2013, “Crime and growth convergence: evidence from Mexico,Policy Research Working Paper 6730 (Washington: World Bank).

    • Search Google Scholar
    • Export Citation
  • Enders, Walter, Todd Sandler, and Gerald F. Parise, 1992, “An econometric analysis of the impact of terrorism on tourism,Kyklos, Vol. 45, pp. 531554.

    • Search Google Scholar
    • Export Citation
  • Ehrlich, Isaac, 1973, “Participation in Illegitimate Activities: A Theoretical and Empirical Investigation,Journal of Political Economy, Vol. 81, pp. 52165.

    • Search Google Scholar
    • Export Citation
  • Escalante, F.,Homicidios 2008–2009: La muerte tiene permiso,Nexos, January 1, 2011. Available via the Internet: http://www.nexos.com.mx/?p=14089.

    • Search Google Scholar
    • Export Citation
  • Fajnzylber, P., D. Lederman, and N. Loayza, 2002a, “Inequality and violent crime,Journal of Law and Economics, Vol. 45, pp. 139.

    • Search Google Scholar
    • Export Citation
  • Fajnzylber, P., D. Lederman, and N. Loayza, 2002b, “What causes violent crime?European Economic Review, Vol. 46, pp. 13231357.

  • Fajnzylber, P., D. Lederman, and N. Loayza, 2002c, “Crime and Victimization: An economic perspective,Journal of Law and Economics, Vol. 45, pp. 139.

    • Search Google Scholar
    • Export Citation
  • Fajnzylber, P., D. Lederman, and N. Loayza, 2000, “Crime and Victimization: An economic perspective,Economia: Journal of the Latin American Economic Association, Vol. 1, No. 1.

    • Search Google Scholar
    • Export Citation
  • FATF-GAFI, 2007, “Money Laundering & Terrorist Financing Through the Real Estate Sector” Available via the Internet: http://www.gafisud.info/documentos/eng/doc_interes/tipologias/003-Money%20Laundering%20and%20Terrorist%20Financing%20through%20the%20Real%20Estate%20Sector%202007.pdf

    • Search Google Scholar
    • Export Citation
  • FATF-GAFI, 2008, Mutual Evaluation Report: Anti-Money Laundering and Combating the Financing of Terrorism, Mexico (France).

  • FATF-GAFI, 2014, Mutual Evaluation of Mexico 7th Follow-Up Report: Update and Full Analysis, February 12–14, 2014 (France).

  • Felbab-Brown, V., 2009, “The violent drug market in Mexico and lessons from Colombia,Documento de Políticas 12 (Washington, D.C., United States: Brookings Institution).

    • Search Google Scholar
    • Export Citation
  • Fleisher, Belton M, 1966, “The Effect of Income on Delinquency,American Economic Review, Vol. 56, pp. 11837.

  • Forni, Mario and Sergio Paba, 2010, “The sources of local growth: Evidence from Italy,Giornale degli Economisti, Annali di Economia, Vol. 59, pp. 149.

    • Search Google Scholar
    • Export Citation
  • Fox, James Alan, and Marianne W. Zawitz, 2000, Homicide Trends in the United States (Washington: Bureau of Justice Statistics).

  • Franco Vivanco, E., 2013, “The economic impact of Mexico’s war on drugs: A cost-benefit analysis,mimeo (Stanford, California: Stanford University)

    • Search Google Scholar
    • Export Citation
  • Freeman, Richard, 1986, “Who Escapes? The Relation of Churchgoing and Other Background Factors to the Socioeconomic Performances of Black Male Youths from Inner-City Tracts,” In The Black Youth Employment Crisis, R. Freeman and H. Holzer (Chicago: University of Chicago Press).

    • Search Google Scholar
    • Export Citation
  • GAFISUD, 2012, GAFISUD-EGMONT Typologies Joint Meeting, May 8–11, 2012 (Quito, Ecuador). Available via Internet: http://www.gafisud.info/documentos/eng/doc_interes/gafi-tipologias/009-Joint%20typologies%20GAFISUD%20EG%20meeting%202012%20concept%20paper.pdf

    • Search Google Scholar
    • Export Citation
  • GAFISUD, 2010, Tipologías regionales GAFISUD—2010, December. Available via Internet: http://www.gafisud.info/documentos/eng/doc_interes/gafi-tipologias/006-Informe%20de%20Tipolog%EDas%20Regionales%20de%20GAFISUD%202010.pdf

    • Search Google Scholar
    • Export Citation
  • GAFISUD, 2010, Actividades y Profesiones No-Financieras Designadas (APNFD) 2010. Available via Internet: http://www.gafisud.info/documentos/eng/doc_interes/gafi-tipologias/005-Informe%20de%20Tipolog%EDas%20Regionales%20de%20GAISUD%20-%20APNFD

    • Search Google Scholar
    • Export Citation
  • Gaibulloev, Khusrav and Todd M. Sandler, 2008, “Growth consequences of terrorism in Western Europe,Kyklos, Vol. 61, pp. 411424.

    • Search Google Scholar
    • Export Citation
  • Glaeser, Edward, 1999, “An Overview of Crime and Punishment” (Washington: World Bank).

  • Globerman, S. and P. Storer, 2009, The Effects of 9/11 on Canadian-U.S. Trade: An Update through 2008 (Washington, D.C.: Brookings Institution). Available via the Internet: http://www.brookings.edu/research/papers/2009/07/13-canada-globerman

    • Search Google Scholar
    • Export Citation
  • Goulas, E. and A. Zervoyianni, 2013, “Economic growth and crime: does uncertainty matter?,Applied Economics Letters, Taylor & Francis Journals, vol. 20(5), pp. 420427.

    • Search Google Scholar
    • Export Citation
  • Greenbaum, R.T., L. Dugan, and G. LaFree, 2007, “The impact of terrorism on Italian employment and business activity,Urban Studies, Vol. 44, pp. 10931108.

    • Search Google Scholar
    • Export Citation
  • Grillo, Ioan, 2011, El narco: Inside Mexico’s criminal insurgency (London: Bloomsbury Press).

  • Guerrero-Gutiérrez, Eduardo, 2010, “Cómo reducir la violencia en México,Nexos, November 3, 2010. Available via the Internet: http://www.nexos.com.mx/?p=13997.

    • Search Google Scholar
    • Export Citation
  • Guerrero-Gutiérrez, Eduardo, 2011a, “La raíz de la violencia,Nexos, June 1, 2011. Available via the Internet: http://www.nexos.com.mx/?p=14318.

    • Search Google Scholar
    • Export Citation
  • Guerrero-Gutiérrez, Eduardo, 2011b, “Security, Drugs, and Violence in Mexico: A Survey,seventh North American Forum, Washington, D.C., United States. Available via the Internet: http://iis-db.stanford.edu/evnts/6716/NAF_2011_EG_(Final).pdf

    • Search Google Scholar
    • Export Citation
  • Guerrero-Gutiérrez, Eduardo, 2012a, “2011: La dispersión de la violencia,Nexos, February 1, 2012. Available via the Internet: http://www.nexos.com.mx/?p=14705.

    • Search Google Scholar
    • Export Citation
  • Guerrero-Gutiérrez, Eduardo, 2012b, “Epidemias de violencia,Nexos, July 1, 2012. Available via the Internet: http://www.nexos.com.mx/?p=14884.

    • Search Google Scholar
    • Export Citation
  • Hale, C., 2005, “Economic Marginalization and Social Exclusion,” C. Hale et al. (eds), Criminology, p. 328 (Oxford: Oxford University Press).

    • Search Google Scholar
    • Export Citation
  • Hamilton, James D., 1994, Time Series Analysis (New Jersey: Princeton University Press).

  • Hallward-Driemeier, Mary and David Stewart, 2004, How Do Investment Climate Conditions Vary Across Countries, Regions and Types of Firms?” Available via the Internet: http://siteresources.worldbank.org/INTWDR2005/Resources/477407-1096581040435/hallward-driemeier_stewart_variations_in_investment_climate.pdf

    • Search Google Scholar
    • Export Citation
  • Harvey Andrew C., 1990, Forecasting, Structural Time Series Models and the Kalman Filter (Cambridge: Cambridge University Press). Available via the Internet: https://ideas.repec.org/b/cup/cbooks/9780521321969.html

    • Search Google Scholar
    • Export Citation
  • Heath, J., 2012, “Lo que indican los indicadores: Como utilizar la información estadística para entender la realidad económica de Mexico.” Available via the Internet: http://www.inegi.org.mx/prod_serv/contenidos/espanol/bvinegi/productos/estudios/indican_indi/indica_v25iv12.pdf

    • Search Google Scholar
    • Export Citation
  • Heine, Kimberly, O. Rodriguez, and D. Shirk, 2014, Drug Violence in Mexico: Data and Analysis through 2013. Justice in Mexico Project: Department of Political Science and International Relations (San Diego: University of San Diego)

    • Search Google Scholar
    • Export Citation
  • Im K, Pesaran H, Shin Y., 2003, “Testing for unit roots in heterogeneous panels,Journal of Econometrics 115: 5374.

  • International Monetary Fund, 2001, “Financial System Abuse, Financial Crime and Money Laundering—Background Paper” (Washington, International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2011a, “Anti-Money Laundering and Combating the Financing of Terrorism (AML/CFT)—Report on the Review of the Effectiveness of the Program,http://www.imf.org/external/np/pp/eng/2011/051111.pdf (Washington, International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2011b, “Acting Chair’s Summing Up, Anti-Money Laundering and Combating the Financing of Terrorism (AML/CFT)—Report on the Review of the Effectiveness of the Program,https://www.imf.org/external/pubs/ft/sd/index.asp?decision=EBM/11/55 (Washington, International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2013, Anti-Money Laundering and Combating the Financing of Terrorism—Inclusion in Surveillance and Financial Stability Assessments—Guidance Note, http://www.imf.org/external/np/pp/eng/2012/121412a.pdf (Washington, International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Instituto Nacional de Estadística y Geografía, 1993–2000, 2003, 2006, 2007–2011, 2008, 2012, 2013, Anuario de Estadisticas por entidad federativa Datos estadísticos (Mexico City).

    • Search Google Scholar
    • Export Citation
  • Instituto Nacional de Estadística y Geografía, 2006, Sistema de Cuentas Nacionales de Mexico. Producto Interno Bruto por Entidad Federativa. Metodologia.

    • Search Google Scholar
    • Export Citation
  • Instituto Nacional de Estadística y Geografía, 2010, Series del PIB Trimestral a Precios Constantes, Retropolacion para el periodo 1993–2002, Ano base 2003.

    • Search Google Scholar
    • Export Citation
  • Instituto Nacional de Estadística y Geografía, 2011, Marco Concepctual del Censo de Poblacion y Vivienda 2010.

  • Instituto Nacional de Estadística y Geografía, 2011, 2012, ENVIPE-Encuesta Nacional de Victimizacion y Percepcion sobre Seguridad Publica.

    • Search Google Scholar
    • Export Citation
  • Instituto Nacional de Estadística y Geografía, 2013a, Sistema de Cuentas Nacionales de Mexico, Cuentas de Corto Plazo y Regionales: Fuentes y Metodologias.

    • Search Google Scholar
    • Export Citation
  • Instituto Nacional de Estadística y Geografía, 2013b, Modernizacion de las Cuentas Nacionales Cambio de Base de 2003 a 2008.

  • Instituto Nacional de Estadística y Geografía, 2013c, Sintesis metodológica del Sistema de Clasificacion Industrial de America del Norte, Mexico, SCIAN.

    • Search Google Scholar
    • Export Citation
  • INEGI and STPS (Ministry of Labour and Social Welfare), 2005, 2007, 2009–2014, Encuesta Nacional de Ocupación y Empleo. Available via the Internet: http://www.inegi.org.mx/est/contenidos/Proyectos/encuestas/hogares/regulares/enoe/

    • Search Google Scholar
    • Export Citation
  • Institute for Economics and Peace, Mexico Peace Index. Available via the Internet: http://www.visionofhumanity.org/#/page/indexes/mexico-peace-index

  • Koopman, Siem J., Neil Shephard, and Jurgen A. Doornik, 1999, “Statistical Algorithms for Models in State Space Using SsfPack 2.2,Econometrics Journal, Vol. 2, No. 1, pp. 107160.

    • Search Google Scholar
    • Export Citation
  • Kotabe, M., 2005, “Global security risks and international competitiveness,Journal of International Management, Vol. 11, pp. 453455.

    • Search Google Scholar
    • Export Citation
  • Kumar, S., 2013, Crime and Economic Growth: Evidence from India. Available via the Internet: http://mpra.ub.uni-muenchen.de/id/eprint/48794

    • Search Google Scholar
    • Export Citation
  • La Jornada, 2008, “En Sinaloa, guerra interna entre narcos desgaja a La Federación.” Available via the Internet: http://www.jornada.unam.mx/2008/05/13/index.php?section=politica&article=006n1pol.

    • Search Google Scholar
    • Export Citation
  • La Jornada, 2012, “Disputa entre zetas y el cártel del Golfo en Tamaulipas ha dejado decenas de víctimas.” Available via the Internet: http://www.jornada.unam.mx/2012/05/05/politica/003n1pol.

    • Search Google Scholar
    • Export Citation
  • Lin, M.J, 2009, “More Police, Less Crime: Evidence from U.S. State Data,International Review of Law and Economics, Vol. 29, No. 2, pp. 7380.

    • Search Google Scholar
    • Export Citation
  • Lochner, Lance and E. Moretti, 2004, “The Effect of Education on Crime: Evidence from Prison Inmates, Arrests, and Self-Reports,” American Economic Review, Vol. 94, March, No. 1, pp. 155189.

    • Search Google Scholar
    • Export Citation
  • Londoño, Juan Luis and Rodrigo Guerrero, 1999, “Violencia en America Latina: epidemiología y costos,Working Paper R-375 (Washington, D.C., United States: Banco Interamericano del Desarrollo).

    • Search Google Scholar
    • Export Citation
  • Lopez, R. and S. Velarde,Una aproximación a los patrones de migración interregional en Mexico, 1990–2010,La situación demográfica de México 2013. Available via the Internet: http://www.conapo.gob.mx/es/CONAPO/Una_aproximacion_a_los_patrones_de_migracion_interregional_en_Mexico_1990-2010

    • Search Google Scholar
    • Export Citation
  • Madrazo Rojas, F.M., 2009, “The effect of violent crime on FDI: the case of Mexico 1998–2006,” unpublished Master’s thesis, Georgetown Public Policy Institute, Georgetown University, Washington, D.C.

    • Search Google Scholar
    • Export Citation
  • Mauro, L. and G. Carmeci, 2007, “A poverty trap of crime and unemployment,Review of Development Economics, Vol. 11, August, No. 3, pp. 450462.

    • Search Google Scholar
    • Export Citation
  • McCollister, K., M. French, and H. Fang, 2010, “The cost of crime to society: New crime-specific estimates for policy and program evaluation,Drug and Alcohol Dependence, Vol. 108, Nos. 1–2, pp. 98109.

    • Search Google Scholar
    • Export Citation
  • McDonald, J., 2005, “The narcoeconomy and small-town, rural Mexico,Human Organization, Vol. 64, No. 2, pp. 115125.

  • McLennan, W., 1998, “Introduction of Chain Volume Measures in the Australian National Accounts, Information Paper 5248.0 (Canberra, Australia: Australian Bureau of Statistics).