Assessing the Impact of Financial Aids to Firms: Causal Inference in the Presence of Interference

Regional and national development policies are an important feature for setting up and supporting local enterprise. Policy makers at national or sub-national level plan interventions aiming at removing barriers (e.g., credit rationing) and promoting investments and firms' performances, such as occupation and turnover. We consider the problem of assessing the effectiveness of financial aids provided to firms. An assumption usually made in causal inference is the Stable Unit Treatment Value Assumption (SUTVA). SUTVA combines the “no-interference” assumption that one unit's treatment assignment does not affect another unit's outcomes with the assumption that there are “no hidden versions” of the treatment. For no-interference to hold, whether or not one firm receives a financial aid should not affect other firms' outcomes, such as their performances or their business strategies. Firms operating in the same geographical area and sector of activity are likely to interact each other. This implies that assigning an incentive to a firm might affect also its competitors, violating the no-interference assumption. The assumption that there are no hidden versions of the treatment implies that there exists only one version of financial aid provided to firms, so that we can consider a binary treatment variable (to be exposed or not to the policy). Here, we assume that this assumption is met, but we focus on violations of SUTVA due to interference. In policy evaluation studies, violation of SUTVA is usually overlooked. Here, we consider the estimation of causal effects with interference, proposing models for interaction among units. This approach involves specifying which firms interact with each other, and the relative magnitudes of these interactions. In some cases it may be plausible to assume that interactions are limited to units within well-defined, possibly overlapping groups, with the intensity of the interactions being equal within the same group. Our approach assumes that intensity of interactions depends on a distance metric, based on geographical distance and firms' characteristics. Specifically, we assume that the potential outcome for a firm depends on the treatment assignment of that firm and some function of treatment assignments of the other firms. This function will be used to weight matched firms in the estimation of causal effects of interest. Our approach also allows us to consider alternative causal estimands that can be of interest for policy makers. As a motivating example we re-analyze data on a policy intervention implemented in Italy on Tuscan artisans firms. Given the characteristics of the Tuscan labour market, where most artisan firms are small-sized, and generally operate in a limited geographical area, it is crucial to consider possible spill-over effects of the policy.