In particular, we are trying to disclose whether forecasting improvements exist or not, and their potential relationship with the size of the observed series vector, the prediction horizon and the presence of common factors.
An interesting advantage (and off the beaten track in terms of research) is the fact that using many predictors might also provide a bigger strength against the so-common structural instability of the small-sized forecasting models.
We must point out that Stock and Watson (2000) found empirical evidence of forecasting improvement in favor of great-sized models compared to those small-scaled models. This evidence, nevertheless, may not become widespread when speaking of all kinds of econometric models, and, both the model size and the prediction horizon determine the forecasting results.
Besides that, the firm financial management requires risk quantification and forecasting. In the specialized literature there are a number of proposals for performing these tasks. Nevertheless, more and more frequently financial analysts face a difficulty, which is being able to analyze, in an accurate way, data about share prices, exchange rates, etc. These data are obtained almost continuously, that is to say each time a transaction is made. In this case, the amount of data to deal with may become disproportionate compared to the traditional procedures. Thus, we propose to use a totally new technique for the analysis of financial data: the analysis of functional data.