Econometric modeling study demonstrates improved purchase prediction across e-commerce shoppers, highlighting the utility of integrating deep learning with sequential search theory.
Key Points
To develop and validate the Sequential Search Transformer, an end-to-end framework integrating deep learning with sequential search theory to model consumer decision-making and evaluate platform policies.
Formulated an end-to-end deep structural econometric architecture that tracks consumer behavior across sessions and sequentially resolves utility uncertainty for searched items.
Established theoretical identification proofs showing all model parameters can be identified under the proposed framework.
Trained and evaluated the model on detailed clickstream data from a major U.S. e-commerce platform and ran counterfactual policy experiments.
The proposed Sequential Search Transformer outperformed existing deep learning architectures and structural search models in predicting consumer search patterns and purchase decisions.
Policy experiments showed the model effectively optimized product recommendation algorithms and new product promotional strategies, increasing retail revenue.