| Abstract [eng] |
Value-added tax brings the biggest revenue of all the taxes. Nevertheless, value-added tax evasion is one of the biggest in Lithuania of all the European Union. And it is not substantially decreasing. Thus, the research of this work is oriented towards decreasing tax evasion, by increasing the probability to effectively find evaders. The research analyses value-added tax payers’ typical behavior in the most vulnerable to evasion economy fields. Value-added tax payment, counting, evasion methods related to this tax and separate economic fields’ behavior theoretical aspects is reviewed in the work. Also, data analysis methods, which are developed for the similarity analysis, are reviewed. Construction, agriculture and electronics sales fields’ data of value-added tax declarations is analysed by identifying payers’ similarities and companies that are suspected to evade tax. Companies’ data is from tax declarations between 2011 January and 2016 December, together with added employees number as external variable. Derivative indexes are also used in the analysis. The structure of the analysis is: gathering the data, data preparation, review of the data, outliers detection, clustering, typical and atypical value-added tax payers‘ behavior identification, validation, introduction of categorical variables and introduction of the results. The analysis uses self-organizing map and hierarchical clustering for the separation of groups and decision tree for the validation of results and creation of separation rules. The research identified that the majority of companies exist, which has the common behavior according to the variables used. All variables‘ medians of all three economical fields‘ value-added tax payers is around the middle in all the quarters compared to all clusters. Agriculture products sales field typical tax payers stand out as the most quarterly seasonal. Self-organizing map and hierarchical clustering result is strongly similar to the decision tree based rules separation. The companies, which are suspected to evade the tax, is identified according to the atypical value-added tax payers‘ clusters‘ indicators. The information of the location of business does not give any additional useful insights in this research. Although, the information of the size of the companies shows that the big and average sized companies is not included in the typical payers‘ clusters in all three economical fields. |
