In this paper we discuss the use of elastic maps as support tool in the decision process underlying the selection, optimization, and management of financial portfolios. In particular, we suggest an allocation scheme which is interely driven by neural networks, in contrast to the traditional model where investors distribute their money among assets chosen according to the mean and variance of their returns. Our optimization procedure is based on the selection of assets from clusters originated by the nets, according to their proximity to the nodes of the map; this, in turn, is the criterion thanks to which we assign the proper weight to each asset into the portfolio. In order to check the profitability of the approach, we have empirically tested the method with stocks from the Italian Stock Exchange; market reference index has been then used to build proper performance benchmarks. Our main results may be summarised as follows: (
) our approach has revealed to be generally more informative than classical mean−variance method, since it allows to take into account additional variables in the selection procedure; (
) our procedure can work both in a static framework (i.e. for one time choice), and into a dynamic context (i.e. to the purpose of re−calibration of original decisions). The overall performances appear to be superior to the benchmark in both the static and dynamic case.