|
Abstract:
|
Most corporations record each customer's spending for a product and are interested in estimating each customer's related total spending (customer wallet). One approach is to identify key predictors (business size, industry, etc) that drive the spending and develop a quantile regression to estimate the wallet. The quantile is chosen to normalize the corporation's overall market share. This approach is unable to provide measure of uncertainty and wallet may be less than spending. Here, we develop a framework using local regression with random censored data. Using local regression to model the total spending and the predictors, assume the corporation's share of wallet follows a Beta distribution. Under certain conditions, we can estimate the parameters of the Beta distribution, the local regression, and the conditional distribution of the total spending of each customer.
|
