The previous hypotheses will be tested by considering the 107 Italian provinces over the period 2010-2018 using a spatial autoregressive model (SAR), both without and with instrumental variables. The dependent variable is the money laundering rate, ML Rate, which is the number of money laundering crimes reported to the judicial authorities per 100,000 resident population (ISTAT, 2019). Such a measure may be imperfect, as money laundering, being illegal, tries to be invisible; however, the number of reports is nonetheless useful as a proxy for the magnitude of the phenomenon (Masciandaro, 1998 and 2000 and Ardizzi et al., 2014). The predictors considered are: i) Waste Convicted, the number of people convicted for illegal treatment, disposal and trafficking of waste per 100,000 resident population (ISTAT, 2019); ii) the level of the crime rate, Crime Rate (ISTAT, 2019) Indeed, criminals may be assumed as rational investors (Gilmour, 2016 and Dalla Pellegrina et al., 2020); as such, they will invest in illegal activities only if it is more convenient than choosing legal alternatives. Deterrence and, in particular, its effectiveness plays a crucial role in determining such a convenience. Moreover, to address the deterrent effect on criminal behaviour, the analysis uses the variable Clearance, that is the ratio of money laundering crimes committed by persons known to all recorded money laundering crimes and ML Conviction, the number of convicted defendants for money laundering crime by a final judgment to all recorded money laundering crimes. Furthermore, the regressions control also for the unemployment rate (ISTAT, 2019) at provincial level (Unemployment). Table A.1 (Appendix A) summarises all the variables above mentioned.
Estimates are obtained using spatial autoregressive (SAR) techniques to account for the characteristics of the data. On the one hand, organised criminality is persistent over time, as it tends to reproduce itself, especially in areas where it has been historically rooted. The case of Italian mafias is particularly exemplificative of such a phenomenon (Sciarrone, 2009; Sciarrone and Storti, 2014 and Allum et al., 2019), also as a consequence of the relationships existing between mafias and civil society (Sciarrone, 2010). On the other hand, while organised criminality affects some Italian provinces more than others, inter-provincial spillovers are possible, especially across provincial borders: criminality is likely to expand across provincial borders, although criminal organisations may form cartels to divide up the territory (Fiorentini and Peltzman, 1995). Indeed, Willis (1983) had already shown that, in England and Wales, crime rates feature spatial correlation, due to similarities of some variables (such as unemployment rates) across contiguous regions. For these reasons SAR estimation seem the best methodology. With particular reference to Italian provinces, Ardizzi et al., (2018) show that contiguous provinces present similar rates of cash anomalies, suggesting that inter-provincial spillovers are possible.
Waste Convicted may however be endogenous, as provinces where laundering money is easier, may induce more criminals to operate there. Consequently, IV-SAR is used to account for this possible problem. The instrument chosen is the urbanisation rate at provincial level, provided by ISTAT. In facts, Gillis (1996), Bisi and Buscemi (2004) and Malik (2016) show that criminality tends to concentrate in urbanised areas, as there the opportunities of illegal operations and gains are more than in other areas. Finally, Urbanisation is uncorrelated with the money laundering rate (Pearson correlation coefficient equal to 0.201[4]) in the sample used in the analyses.
It is noteworthy that in 2017 Italy ratified the European Directive 2015/849/EC, which contains specific rules on fighting money laundering. Such a discontinuity with respect to the previous legislation may have had consequences both in terms of detections and number of crimes committed. For this reason, the analyses consider dummies variables, considering the value 0 for years before 2017 and 1 for years after.
For the sake of robustness, the estimations presented in the next section regress the money laundering rate on either the number of waste-related crimes (per 100,000 inhabitants)[5] or that of people convicted for committing such crimes. The reason behind this choice is that the number of convicted might underestimate the real dimension of the phenomenon, as often criminals remain unpunished. For these estimations, in order to check for robustness of deterrence measure, has been used the variable (Uif), instead of Clearance,that is the ratio of suspicious transactions reported to the Italian FIU over the total of money laundering crimes (UIF, 2019). While we expect a positive correlation between the number of the convicted and that of crimes, Tables A2, A3 and A4 show that it is not very high (around 0.4), suggesting that the two variables may lead to different results. Such a correlation is, however, not surprising. Indeed, illegal money may come either from the activities of criminal organisations or those of individual criminals (such as thieves). In the first case, large amounts of illegal proceeds are generated by relatively few people; in the second, instead, there is higher correlation between the number of (small) crimes and that of money launderings. Table 1 shows the descriptive statistics for the variables considered in both the estimations. Moreover, tables 2 and 3 present correlation matrices for the same variables linked to the two analyses.
TABLE 1 DESCRIPTIVE STATISTICS
Variable
|
Obs
|
Mean
|
Std. Dev.
|
Min
|
Max
|
|
|
|
|
|
|
ML Rate
|
954
|
4.754
|
11.253
|
0
|
122.824
|
Crime Rate
|
954
|
3850.751
|
1140.257
|
0
|
8482.3
|
Waste Convicted
|
954
|
8.294
|
12.107
|
0
|
131.877
|
Waste Crimes
|
954
|
39.188
|
67.278
|
0.092
|
756.095
|
Uif
|
954
|
104.301
|
235.398
|
0
|
4273
|
|
|
|
|
|
|
ML Conviction
|
954
|
1.131
|
1.791
|
0
|
18
|
Clearance
|
954
|
0.597
|
0.736
|
0
|
14
|
Unemployment
|
954
|
0.111
|
0.055
|
0.026
|
0.314
|
TABLE 2 CORRELATION MATRIX (1)
|
ML Rate
|
Crime Rate
|
Waste Crimes
|
Uif
|
ML Conviction
|
Unemployment
|
ML Rate
|
1.000
|
|
|
|
|
|
Crime Rate
|
0.237
|
1.000
|
|
|
|
|
Waste Crimes
|
0.411
|
0.051
|
1.000
|
|
|
|
Uif
|
-0.117
|
-0.042
|
-0.058
|
1.000
|
|
|
ML Conviction
|
-0.098
|
-0.076
|
0.018
|
0.154
|
1.000
|
|
Unemployment
|
0.111
|
-0.201
|
0.296
|
-0.065
|
-0.045
|
1.000
|
TABLE 3 CORRELATION MATRIX (2)
|
ML Rate
|
Crime Rate
|
Waste Convicted
|
Clearance
|
ML Conviction
|
Unemployment
|
ML Rate
|
1.000
|
|
|
|
|
|
Crime Rate
|
0.237
|
1.000
|
|
|
|
|
Waste Convicted
|
0.447
|
0.065
|
1.000
|
|
|
|
Clearance
|
-0.042
|
-0.008
|
0.109
|
1.000
|
|
|
ML Conviction
|
-0.098
|
-0.076
|
0.078
|
0.444
|
1.000
|
|
Unemployment
|
0.111
|
-0.201
|
0.102
|
0.120
|
-0.045
|
1.000
|
In this section the study implements a crime model that posits a relationship between the annual reported crime in each province and a set of explanatory variables. The basic econometric specifications, for both the models considered, is given by the following equations:
𝑀𝐿 Rate𝑖𝑡 = β0 + β1Crime Rate𝑖𝑡 + β2 Clearance𝑖𝑡 + β3 ML Conviction𝑖𝑡 + β4Waste Convicted𝑖𝑡 +
+ β5Unemployment𝑖𝑡 + 𝑌𝑒𝑎𝑟𝑡 + ε𝑖 (1)
𝑀𝐿 Rate𝑖𝑡 = β0 + β1Crime Rate𝑖𝑡 + β2 Uif𝑖𝑡 + β3 ML Conviction𝑖𝑡 + β4Waste Crimes𝑖𝑡 +
+ β5Unemployment𝑖𝑡 + 𝑌𝑒𝑎𝑟𝑡 + ε𝑖 (2)
where the subscripts i and t represent province and year respectively. Provi are a set of provincial fixed effects, while time fixed effects are introduced with dummies Yeart to capture possible shocks which may influence all provinces in a given year. In particular, the analysis uses the year dummies to determine if there are some shocks considering the period before and after the main European Directive anti-money laundering.
The spatial regression model has been chosen to describe the relationship between the independent variables and the dependent variable by involving location effect of the data, like spatial diffusion, spillover, interaction, and dispersal processes. In this regard, Table 4 shows the results of the Moran’s index associated to each variable considered in the principal model. In the case of ML Rate and Uif, the p-value shows that the test is not significant, so it is possible to affirm that the phenomenon described by this variable is restricted to each single province that it represents. On the other hand, for all the other variables considered, the table of Moran’s indexes (Moran’s I) allows to reject the null hypothesis of zero spatial autocorrelation. Therefore, it is possible to affirm that in these cases the phenomena represented by these variables have effects not only in a single province, but also in the surrounding areas. In particular, the table shows that the values of I are greater than those of E (expected index) and that the values of z are positive. This means that the existent autocorrelation is a positive spatial autocorrelation.
TABLE 4 VALUES OF MORAN’S TEST
Variables
|
I
|
E(I)
|
sd(I)
|
z
|
p-value
|
|
|
|
|
|
|
ML Rate
|
-0.010
|
-0.001
|
0.007
|
-1.209
|
0.113
|
Crime Rate
|
0.175
|
-0.001
|
0.007
|
23.556
|
0.000
|
Waste Crimes
|
0.198
|
-0.001
|
0.007
|
27.310
|
0.000
|
Waste Convicted
|
0.084
|
-0.001
|
0.007
|
11.567
|
0.000
|
ML Conviction
|
0.020
|
-0.001
|
0.007
|
2.863
|
0.002
|
Uif
|
0.003
|
-0.001
|
0.007
|
0.527
|
0.299
|
Clearance
|
0.054
|
-0.001
|
0.007
|
7.851
|
0.000
|
Unemployment
|
0.601
|
-0.001
|
0.007
|
80.412
|
0.000
|
[4] The more the value is close to zero and away from -1 and 1, the weaker the correlation will be.