B1767
Title: A comparison of biased estimation methods for predicting tourism income of Turkey
Authors: Esra Polat - Hacettepe University (Turkey) [presenting]
Abstract: One of the problems encountered in regression models is multicollinearity. If there is a multicollinearity problem, the variance of the least squares estimator may be very large and subsequent predictions rather inaccurate. In this case, biased estimation methods could be used to overcome the problem of inaccurate predictions. Biased estimation methods such as Ridge Regression (RR), Principal Component Regression (PCR) and Partial Least Squares Regression (PLSR) are used with the consequent trade-off between increased bias and decreased variance. RR is based on adding a biasing constant $k$ to diagonal elements of X'X. PCR and PLSR methods discard the irrelevant and unstable information and use only the most relevant part of the $x$-variation for regression. Tourism sector in Turkey has shown great progress since 1980s. Contribution of foreign currency, while the country was having economic problems, helped to decrease foreign debt and unemployment. The annual data set of Turkey (1985-2014) including the factors (number of foreign tourists, total bed amount of tourism facilities having tourism operation license, number of tourism agencies, exchange rate of euro and exchange rate of US dollars) affecting the tourism income, is examined. RR,PCR and PLSR methods are compared in terms of fitting to data and predictive ability. Therefore, best model, giving the best prediction of tourism income, is selected.
