Comparison of Sales Response Models

Abstract

이 논문의 목표는 코익(Koyck) 모형에 기초하여 매출반응모형을 비교하는 것이다 광고는 매출에 이월효과를 갖는 것으로 가정된다. 모의(Simulated) 자료를 이용하여 자기상관 오차항을 갖지않는 Brand Loyal 모형이 축소된 코익 모형과 모수의 추정값에 의해 비교되었다. 또한 소비자 특성변수들이 실제 자료를 이용한 모형추정에서 추가적으로 이용되었다. 코익모형에 기초하여 모의추출된 자료를 이용하여 각 모형들을 추정한 결과 코익모형이 모수추정에 있어서 우수하게 나타났다. 따라서 자료와 일치하지 않는 모형이 이용되어질 경우 잘못된 결과가 산출될 수 있다. IRI사(Information Resources Incorporated)의 실제 자료를 이용하여 모형추정을 한 결과 코익모형이 Brand Loyal model과 R² 값이나 RMSE(root mean square error)값에 있어서 보다 우수하게 나타났으나 크게 차이가 없었으며 여러 변수의 부호가 기대와 다르게 나타났다.

The objective of this paper is to compare sales response models based on the Koyck model. Advertising is assumed to have distributed carryover effects on sales. Brand Loyal model without serial correlation of error terms was compared with the reduced Koyck model in terms of recovery of parameters by using the simulated data. Also consumer-specific variables were added to the model in the estimation with the real data. In the simulated data generated by using the Koyck model, the Koyck model was better in recovering the parameters than the Brand Loyal model. Therefore, if the wrong model is used for estimation, the wrong result may come out. In the real data of IRI (Information Resources Incorporated), the Koyck model is better than the Brand Loyal model but not much different in R² and root mean square error terms. In addition, some signs of variables are different from the expected ones.

The objective of this paper is to compare sales response models based on the Koyck model. Advertising is assumed to have distributed carryover effects on sales. Brand Loyal model without serial correlation of error terms was compared with the reduced Koyck model in terms of recovery of parameters by using the simulated data. Also consumer-specific variables were added to the model in the estimation with the real data. In the simulated data generated by using the Koyck model, the Koyck model was better in recovering the parameters than the Brand Loyal model. Therefore, if the wrong model is used for estimation, the wrong result may come out. In the real data of IRI (Information Resources Incorporated), the Koyck model is better than the Brand Loyal model but not much different in R² and root mean square error terms. In addition, some signs of variables are different from the expected ones.