*Result*: On the significance test of negative binomial regression model.

*When constructing a regression model, improper independent variable selection often leads to the damage of the prediction performance of the model outside the sample. In order to improve the prediction accuracy, we explore the variable selection strategy based on significance tests, aiming at enhancing the out-of-sample prediction ability of the negative binomial regression model. Through analysis of the statistical characteristics of log-likelihood function values of the negative binomial regression model under two scenarios, we construct a significance test statistic dedicated to negative binomial regression. The designed algorithm is especially suitable for forecasting problems involving counting data as dependent variables, such as emergency response efficiency evaluation and small sample rapid prediction. In order to verify the effectiveness of this method, the empirical analysis using the sales data of WalMart stores shows that the proposed method can accurately screen the optimal subset of independent variables in the negative binomial regression model, thus significantly improving the prediction efficiency of the model. [ABSTRACT FROM AUTHOR]*