Hedonic Price Index for Mobile Phone Based on E-Commerce Data

Abstract

Big data provides an opportunity to build datasets that fit specific requirements for research. One of the global concerns about big data is how to utilize it for measuring price index. Due to the proliferation of e-commerce sites, a vast amount of prices information followed by the product specifications are available on the internet. This information would allow researchers to calculate hedonic indices for the products undergo rapid quality change like a mobile phone. A method to gather big data is by implementing a large-scale data collection on the web with an automated scraping process. In this study, a web scraper was built using python language to extract information about product, price and characteristics from various e-commerce displayed on the IPrice.co.id website. The data extraction was conducted weekly from January to June 2020 and gathered more than 34,000 records in total. A database containing comprehensive specifications of 458 types of mobile phones was developed based on information from gsmarena.com to validate the scraped data. The method used for calculating the hedonic price index is the double imputation method. One of the steps in this method is to construct a hedonic regression model for datasets consists of two consecutive periods. Due to the presence of outliers and leverage points, robust regression with MM (Multi-Stage Method) estimation was used to get a reliable and efficient estimate of the parameters. Due to the lack of information on the product quantity for weighting, several alternative weighting scenarios were taken. The first is weighting based on the release year mobile phone, the second is by using market share as a proxy for weighting, and the third is the combination of both. The result shows that price movement based on online transactions captured in this study do not reflects the one from offline transactions captured in the Consumer Price Index (CPI) produced by Badan Pusat Statistik (BPS). The weekly indices have a wider fluctuation than the CPI. Weighting based on both release year of mobile phone and market share produces the most stabilized index.