STA 290 Seminar Series
Thursday, May 14, 2015, 4:10pm, MSB 1147 (Colloquium Room)
Refreshments at 3:30pm in MSB 4110 (Statistics Lounge)
Speaker: Prasad Naik (Graduate School of Management, UC Davis)
Title: Robust estimation for conservative marketing resource allocations
Abstract: How should brand managers make conservative marketing resource allocations? Should they increase or decrease advertising budgets? Should the budgets be allocated differently to multiple media? If so, how? To answer these questions, left unexplored in the extant marketing literature, we present a framework to incorporate conservatism in marketing decision-making via robust estimation of dynamic models. The proposed robust estimation approach eschews the assumption of a probability distribution function for the error terms when estimating dynamic sales response models. Specifically, applying minimax control theory, we derive distribution-free disturbances and the conservative filter, whose limiting special case yields the standard Kalman filter. The derived filter is conservative in the sense that it minimizes the largest possible forecast errors rather than the usual squared errors. As a result, conservatism optimally widens the range of sales forecasts to consider when making decisions. We empirically validate the proposed approach using market data from Canadian Blood Services. Using several tests for randomness, we test for the randomness of empirical disturbances and cannot reject the alternative conceptualization of distribution-free disturbances. Empirically, we find that advertising effectiveness increases and the carryover effect decreases as conservatism increases, suggesting a compensatory effect that generalizes across geographical regions. Finally, we find that optimal advertising budgets should be larger in all but one region in the pursuit of conservatism.
Keywords: Robust Estimation, Times Series, Kalman Filter, Optimal Control, Nonprofit