In Model 1, the dependent variable is represented by the standard catalogue price (Pcat) of the basic room. Since for each hotel we have a unique price observation, tactical variables are not relevant. Similarly, in Model 2, while considering all the real prices charged during the observed period, the dependent variable is computed as the median value (Pmed) applied by each specific hotel. Thus, again, we have a unique price observation for each hotel, i.e., dynamic variability is removed. Model 3 exploits the full potential of the panel dataset and uses a random-effect regression analysis with clustered errors to explain both cross-section and time variability of real prices (Preal).
These prices are observed for any hotel i and at any time t. Thus, in this case, also contextual variables come into play. The term ui represents the individual specific error of the random regression model. A SarganeHansen test of over identifying restrictions does not reject the appropriateness of the random effect versus the fixed effect model (p 1⁄4 0.16).