R&Dsimulab: a micro-policy simulator for an ex-ante assessment of the effect of public R&D policies

R&Dsimulab is a micro-policy simulator for an ex-ante assessment of public Research & Development (R&D) policy effect on companies’ R&D activity. It is an agent-based computational model based on the interaction between a public agency, entitled to manage a direct (or grant-based) R&D policy, and a given set of companies eligible for receiving a monetary support to increase their actual level of R&D activity. On the part of policymakers, such model can be used to build and compare ex-ante evaluation scenarios related to alternative policies aimed at fostering the R&D activity of companies undergoing a given public R&D support. R&Dsimulab can be run either using a pre-defined set of parameters, thus exploring outcomes’ sensitivity to parameters’ changes, or by a calibration based on empirical evidence. R&Dsimulab assumes that agents (the public agency and the companies) maximize an objective function under reasonable constraints, and assumes that companies doing R&D are placed within a network of firms where possible positive or negative externality effects can arise. To our knowledge, no previous models of this type have been proposed so far in the literature. Therefore, R&Dsimulab constitutes a first attempt to build a policy simulator for an ex-ante assessment of R&D policy effects, whose scientific and policy-oriented scope can be worth exploring. R&Dsimulab is an agent-based simulative model. The agents constituting the model are: one public agency, which provides public funds to support private R&D companies, and a set of eligible-for-fund private companies. Both types of agents take decisions by maximizing an objective function under reasonable constraints. In particular, the model run under these assumptions: a.Agency behaviour It is assumed that the direct objective of the public agency is that of maximizing the total level of R&D (i.e., the sum of all companies’ R&D spending, that we indicate by R) using a given amount of monetary support S which has to be optimally allocated within firms. The agency knows the company ability to do R&D and its centrality within the network, but it has only an imperfect knowledge of all firms’ R&D network relationships. As objective, the agency wants to determine two things: (i) which companies are worth to support and which are not (i.e., selection-process); (ii) which share of S has the agency to provide to each firm selected for support. Thus, the agency comes up with two optimal solutions: (i) the N1 (out of N) selected companies; the optimal allocation of the subsidy S within the N1 selected companies. b. Companies’ behaviour Companies choose the level of R that maximizes the profit. Thus, the optimal R is the one equalizing the marginal rate of return and the marginal capital cost of doing R&D. The optimal level of R&D is in turn a function of the R&D support that a firm might potentially receive. We assume that each company owns an optimal level of the subsidy, thus making the R&D optimal equation as a concave function of the public support (a parabola, for instance). Finally, we also assume that R&D spillovers among firms may take place, due to companies’ relationships within an R&D Network, where the R&D flows from one company to another according to the strength of the relationship between firms. Therefore, each company R&D includes both an idiosyncratic component and an “additional” component due to the presence of R&D externalities. c. Externality or network effect As companies are located within an R&D network, different network topologies can produce different policy effects. The network impacts on R in two ways: (i) on the one hand, the more a company is central in the network, the more a lower barrier to do R&D is assumed (thus reducing the fixed costs of doing R&D); (ii) on the other hand, different network topologies could provide different R&D performance. Therefore, running a series of simulations under different policy scenarios can provide some guidance to detect the emerging properties in the R&D effect’s pattern, especially when one considers specific model’s parameterizations. R&Dsimulab uses Monte Carlo methods to provide sound conclusions about simulation results. Are specific configurations of the network more likely to produce larger R&D effect than other types of settings? In order to answer questions like this, we run a number of R&Dsimulab simulation exercises. For example, one could be interested in identifying whether, ceteris paribus, a quasi-random network is or is not more conducive to higher levels of R&D than, for instance, networks characterized by the emergence of specific nodes playing as hubs. It may thus be interesting to assess whether the policy effect on R will show an increasing or decreasing pattern as a function of the network’s “hubness”. Other experiments could also include the assessment of policy effect when other significant network parameters are changed or when one considers different network topologies, such as “scale-free” or “small-world” networks. Moreover, once a measure of the actual companies’ network is available and an empirical calibration of the model’s parameters achieved, one may also provide an assessment of the impact of the R&D support policy on a real study context, thus using R&Dsimulab as a tool for an effective ex-ante evaluation of the R&D policy considered.