4. Results and discussion
The empirical results are presented and discussed in the light of the conceptual framework above defined. Table 2 shows the estimated coefficients of Models 1e3, as well as some goodness-of-fit and predictive power measures. The overall R-squared in Model 1 and 2 is close to the between R-squared in Model 3, indicating that the cross-sectional fit does not change. The additional explanatory power of Model 3 is expressed by the within R-squared (0.245). Besides the R-squared, the Median Absolute Percentage Error (MdAPE) complements the analysis comparing the predictive power of the models (Armstrong, 2012). The ability to predict actual pricing behavior increases, as expected, moving from Model 1 and 2 to Model 3. Unsurprisingly, predictions of Model 1 are far from real prices because catalogue prices only represent a hypothetical maximum reference. The MdAPE lowers to 16% in Model 2 and shows an additional decrease to a level below 10% in Model 3, thanks to the explicit consideration of the within-hotel variability.