Empirical analysis of online sales data with implications to pricing

Abstract

Sales dynamics are investigated on a multi-seller online platform through analyzing the relationship between price and quantity. 376 days long sales transaction data of smart phones by Apple, Samsung and Huawei are observed. Since there is simultaneous causality between price and quantity, instrumental variables (IV) are identified and used in regression models to fix the bias in results. Two-stage Least Squares Regression (2SLS) models are formed including price as endogenous variable and quantity. Instrumental variables which are correlated with endogenous variable (price) and independent of the error term are put in these models with controls. A parametric analysis is conducted through several products with daily data. Price and ratio of stock-out sellers of the product are found significantly effective on sales quantity. Moreover, simultaneous bias which is fixed by 2SLS regression models is shown over several models through comparison with Ordinary Least Squares (OLS) regression results. Additionally, since the data observed includes zero sales price data, the selection bias (which would exist if this information is excluded) is fixed and shown. Besides describing the dynamics, these models are also used for predicting sales quantity. On the other hand, consumer choice models are formed as Multinomial Logit to describe consumer choice behaviour on the online platform during the time observed. Model results confirm that customers are more determined in their product selection and first decide on the product then compare among its sellers. Some features like brand reputation, seller service are found to be significantly effective on consumer choices.