| 讲座简介： | We propose two new adaptively aggregated forecasting strategies through exponential reweighting and quadratic reweighting in exponential family panel data (binary choice, count, etc.) models. The oracle inequalities for the two proposed aggregated forecasting strategies are derived. We show that the exponential reweighting based strategy enjoys promising Kullback--Leibler risk bound adaptation in the sense that it automatically achieves the best possible performance among all the candidate forecasting procedures up to an additive term that will vanish as the within-subject sample size increases. Whereas, under the quadratic risk function, we find that the exponential reweighting based strategy may not be able to achieve the similar adaptation property but our quadratic reweighting based strategy can overcome this deficiency and yield promising risk bound adaptation. Under mild conditions, we also establish the risk bound properties of the two proposed procedures in the presence of pre-screening. Simulation studies and a real-world example in analyzing television viewers' binary decision sequence of watching drama episodes verify the superiority of our methods over existing model selection methods. |
