Forecasting models for updating monthly time series: The case of Russian tourism in Italy

This paper addresses how to update monthly time series of Russian tourism in Italy using both univariate and multivariate forecasting models. Updating these time series is necessary because the final data on monthly tourist flows in accommodation establishments (arrivals and overnight stays) are published by ISTAT with a significant delay, which prevents timely analyses of these flows. One solution to this problem is to forecast future values for arrivals and overnight stays. To obtain these values, appropriate forecasting models have been used. From the simplest to the most complex, these include the following: 1) univariate SARIMA models; 2) dynamic multiple regression models; and 3) transfer function models. Each of these models is proposed in two versions: one does not perform the adjustment of outliers, and one does perform such an adjustment. This has made it possible to assess whether and to what extent outliers have a significant impact on forecasts. In addition, the applications include different time intervals and tourism segments. In particular, we have tried to determine the effect of the different lengths of the observation interval on the forecasts by comparing forecasts based on longer interval (eleven years, 2001-2011) with those based on shorter and more recent interval (five years, 2007-2011). With regard to the comparison of forecasts between the different tourism segments under consideration, we have examined how such forecasts vary depending on the type of accommodation establishment (hotels and other accommodations) and tourist nationality (Russian, German, foreign and Italian). Finally, the accuracy of forecasts from one to twelve months, evaluated with the appropriate accuracy indices, has been compared with the goodness of the estimates, and consequently, differences between the forecasting capacity of the models proposed and their goodness of fit to past values have been highlighted