Using Markov-Switching Models to Identify the Link between Unemployment and Criminality1

Using Markov Switching Autoregressive models the behaviour of four crime variables and unemployment rate during the period of study is investigated and different regimes for each variable determined. Using some nonparametric measures such as the Concordance Index (Harding and Pagan, 2002) and Independence of Chronologies (Bodman and Crosby, 2005), among others, the independency of cycles of unemployment rate and crime variables is tested. The results of this stage show that there is no relationship between unemployment rate and burglary and motor. However, for larceny and robbery the results are mixed and inconclusive. At the second stage, Markov Switching Vector Autoregressive models are also used to determine the states for both unemployment rate and each one of crime variable simultaneously. The results of this stage show that the effect of unemployment rate on larceny and motor depends on the state of the variables. For larceny this effect is either positive or null, and for motor it fluctuates among negative, null, and positive. Also the result shows that regardless of the state of the variables, the effect of unemployment on burglary and robbery is negative and null, respectively.