“…However, the validity of its identification assumption is a general concern in this strand of the literature (Mayda et al, 2016), and it is particularly problematic here, since demographic data are not available before 1999 and therefore the base year (1999) is relatively close to the beginning of the estimation period (2001). Although having few years lags is common in this literature (Lonsky, 2018), I address this data limitation by proposing a new instrumental variable for the immigrant share in NIL a at time t , using: where Immigrants t denotes the total number of foreign presences in Milan in year t ; is share of total immigrants in NIL a in the base year (1999); and is its predicted value using the physical characteristics of local buildings, measured 10 years before the first election (1991). Z 1 a , t combines two different sources of exogeneity of immigrants' location.…”