Integer-valued GARCH processes for Apple technology analysis

Purpose – The keywords from patent documents contain a lot of information of technology. If we analyze the time series of keywords, we will be able to understand even more about technological evolution. The previous researches of time series processes in patent analysis were based on time series regression or the Box-Jenkins methodology. The methods dealt with continuous time series data. But the keyword time series data in patent analysis are not continuous, they are frequency integer values. So we need a new methodology for integer-valued time series model. The purpose of this paper is to propose modeling of integer-valued time series for patent analysis. Design/methodology/approach – For modeling frequency data of keywords, the authors used integer-valued generalized autoregressive conditional heteroskedasticity model with Poisson and negative binomial distributions. Using the proposed models, the authors forecast the future trends of target keywords of Apple in order to know the future technology of Apple. Findings – The authors carry out a case study to illustrate how the methodology can be applied to real problem. In this paper, the authors collect the patent documents issued by Apple, and analyze them to find the technological trend of Apple company. From the results of Apple case study, the authors can find which technological keywords are more important or critical in the entire structure of Apple's technologies. Practical implications – This paper contributes to the research and development planning for producing new products. The authors can develop and launch the innovative products to improve the technological competition of a company through complete understanding of the technological keyword trends. Originality/value – The retrieved patent documents from the patent databases are not suitable for statistical analysis. So, the authors have to transform the documents into structured data suitable for statistics. In general, the structured data are a matrix consisting of patent (row) and keyword (column), and its element is an occurred frequency of a keyword in each patent. The data type is not continuous but discrete. However, in most researches, they were analyzed by statistical methods for continuous data. In this paper, the authors build a statistical model based on discrete data.