Identification of Outliers in High Density Areas with the Use of a Quantile Regression Model

Research background: The study exemplifies the issues presented in the international trade data, which can affect the efficiency of customs controls in clearance systems.

Purpose: The purpose of the research is to find a method of identifying suspicious transactions in datasets, where the risk factor is related to the overvaluation of goods covered under export procedure and data that are affected by areas of extra high density.

Research methodology: The proposed methodology is two-step. Firstly, to eliminate areas of extra high density, with the use of a sampling scheme set reciprocally to the intensity point pattern, defined by means of a two-dimensional kernel estimator. Next, based on sampled data, a quantile regression model is built. Moreover, the inference about the type of model is provided by using the Khmaladze test.

Results: The example highlights the benefits of the use of the two-step approach in model building. The proposed methodology provides the foundation for the inference by means of the Khmaladze test. The reliable threshold for selecting the suspicious transactions can be built.

Novelty: The paper addresses some of the previously identified issues in a two-dimensional intensity assessment. Moreover, the proposed methodology based on quantile regression and the Khmaladze test provides the foundation for the customs gap measure in export data.