Exploring quality adjustment in PPI cloud computing

Abstract

Cloud computing services (hosting) is an important component of the Producer Price Index (PPI) for data processing, hosting, and related services. The current approach to changes in cloud services is to estimate a price change between the old and new service that is equal to the average of price changes for similar products in the PPI. This method has limitations because it does not specifically account for changes to characteristics such as microprocessor, memory, and data storage, which are key price-determining characteristics that see rapid technological change. In order to estimate price changes more accurately in this ever-evolving industry, we develop time dummy hedonic models. Our models use publicly available data while future BLS models could instead use confidential respondent data for their estimates. One of the challenges of developing models is accounting for the performance of the microprocessors used in the cloud services. One of the cloud service providers in our dataset has a performance benchmark for their cloud services. We calculate models that use this benchmark and compare them to models that use the characteristics of the microprocessors in order to evaluate the appropriateness of using microprocessor characteristics as explanatory variables in future models. The comparison between these two types of models is critical, because none of the other service providers currently has a microprocessor benchmark. We also use a statistical learning technique to select variables for our models. Finally, we discuss the feasibility of applying these models to the PPI in the future.