A Comparison of Forecasting Performance of PPML and OLS estimators: The Gravity Model in the Air Cargo Market

Using international air cargo data from Turkey, this study compares the forecast performance of three different approaches in the air transport literature for the basic gravity model parameter estimation. The first approach uses ordinary least squares to estimate the gravity model, which is frequently utilized in air transport literature. The second approach, like the first, employs the log-linear estimate technique, but unlike the first, it adds a small amount to the observations with a zero-valued dependent variable and includes them in the analysis. The third method is to estimate the gravity model using the Poisson pseudo maximum-likelihood estimator, which is an alternative to the ordinary least square estimator. The forecast performance of the models developed after estimating the equation with three different approaches was compared with error metrics and the Diebold-Mariano test. As a result of the study, the Poisson pseudo-maximum-likelihood estimator was observed to be the estimator with by far the best forecast performance for the total amount of cargo carried. However, the forecast performance of models differs for some cities.