가격 내생성을 반영한 자동차 선택모형 (Empirical Application of the Automobile Choice Model by Accounting for Price Endogeneity)

서지정보

· 발행기관 : 한국마케팅학회
· 수록지 정보 : 마케팅연구 / 25권 / 3호 / 51 ~ 69페이지
· 저자명 : 장광필

초록

기존의 시장구조 분석모형은 대체적으로 포장소비재를 분석대상으로 하고 있으며, 제품시장을 협소하게 정의함으로써 분석에 포함된 경쟁상표의 숫자가 제한적이었다. 또한, 제품의 물리적 속성과 소비자 선택과의 관계에 대한 분석은 요인분석적 시장구조 분석모형을 통해 시도하였지만 그 내용은 실제 물리적 속성자료를 사용한 것이 아니라 요인분석을 통해 속성차원의 의미를 주관적으로 해석한 것이었다. 이 연구는 선행연구에서 다루지 않은 부분을 보완하는 차원에서 내구재인 자동차를 분석대상으로 하고 다양한 데이터 소스로부터 제공받은 자료를 이용하여 고려상표군 정보를 선택확률에 포함시켜 340개 대안의 선택모형을 추정하고 자동차의 물리적 속성과 소비자 속성의 상호작용이 자동차 선택에 미치는 영향을 분석하였다.
선행연구와의 차별점은 자동차의 물리적 속성과 소비자의 인구통계학적 변수이외에 자동차나 운전에 대한 태도변수와 자동차 속성에 대한 평가를 추가적으로 모형에 포함하여 보다 다양한 가설의 검정이 가능해졌으며 자동차 선택에 있어서 물리적 속성뿐만 아니라 속성에 대한 평가가 미치는 영향도 분석할 수 있게 된 것이다.
방법론적으로는 다항로짓 모형을 중심으로 자동차 속성에 대한 소비자 민감도의 이질성을 관측된 소비자 속성과 관측되지 않은 소비자 속성을 통해 반영하였다. 모형추정 방법으로 SML(Simulated Maximum Likelihood)을 통해 자동차 속성과 소비자 속성의 교차항 계수를 추정하고 나머지 브랜드가치에 해당하는 절편, 는 CM(Contraction Mapping)방식으로 추정하였다. 다음 단계로 브랜드가치를 종속변수로 하고 자동차 속성을 독립변수로 하는 회귀분석 모형을 3SLS로 추정하였다. 선행연구에서는 가격 내생성의 문제를 해결할 목적으로 도구변수를 사용하여 수요측면의 가격효과만을 분석하였지만 (single equation), 본 논문에서는 3SLS를 적용하여 수요측면의 가격효과뿐 만아니라 공급측면에서 브랜드가치가 가격에 미치는 영향도 동시에 분석하였다 (simultaneous equation). 단순 OLS의 추정결과와는 달리 중요한 구조적 모수 (structural parameter)의 부호, 즉, 가격 변수의 계수가 유의하며 음의 부호를 갖는 것으로 추정되었으며 동시에 브랜드가치가 높을수록 높은 가격을 책정하는 것으로 나타났다.

영어초록

Most of previous studies on the market structure were based on frequently purchased packaged goods market. These studies defined product markets so narrowly that the resulting markets included a very limited number of alternatives and made no attempt to relate physical product attributes to consumer choice (Even though previous studies based on the factor-analytic market structure models investigated the relationship between physical attributes and consumer choice, those approaches attempted to subjectively interpret underlying dimensions of the inferred vehicle attributes rather than to use real vehicle attributes and to directly relate those attributes to consumer choice). To fill in this under-researched area, the current study investigates the automobile industry which is characterized by its large number of competing vehicle models (which are durable goods) and attempts to relate vehicle attributes and consumer attributes to automobile choices by incorporating consideration set information in building the likelihood function.
Empirical application of mixed logit model to the 2008 Vehicle Shopping Survey (VSS) data was carried out by accounting for response heterogeneity via random coefficient specification and by interacting the vehicle attributes with consumer characteristics. In addition, consideration set heterogeneity across consumers was accounted for by including the stated considered vehicle models along with the purchased vehicle in the logit choice probability. Estimation of the model parameters is based on the SML (Simulated Maximum Likelihood) and the CM (Contraction Mapping) procedure. The CM procedure is absolutely necessary to estimate the brand-specific intercepts specifically when using the likelihood function that incorporates the considered vehicles.
What distinguishes the current study from previous studies in terms of the empirical data used in the analysis is as follows. First, in addition to the vehicle attributes (e.g., horse power) and consumer demographics (e.g., age and sex) that were used in the previous studies, the current study includes attitudinal variables (e.g., degrees of agreement with the following statement; I prefer a car that has better fuel economy. My first consideration in choosing a vehicle is that it must be fun to drive.) and attribute evaluations (e.g., rating on 10 point scale as to rapid acceleration/passing power) obtained from the Vehicle Quality Survey (VQS). Because of multicollinearity amongst attribute evaluations, factor analysis was implemented and 9 factors were extracted from the original attribute evaluations. The resulting factor scores, instead of the original attribute evaluations, were used in the empirical application and these factors complemented so-called hard attributes like horse power. By using additional data sets, the current study enabled us to test practically more relevant hypotheses and to draw managerially useful implications from empirical results.
Methodologically, multinomial logit model was employed and heterogeneity corrections were implemented in such a way that heterogeneity reflects both observed and unobserved consumer characteristics. At the first stage, the parameters for the interactions of consumer characteristics (observed and unobserved) with the vehicle attributes were estimated via the SML while the vehicle model-specific intercepts were estimated via the CM procedure. The CM procedure was implemented in such a way that predicted market shares based on estimated intercepts exactly match the actual shares. At the second stage, these intercepts (which were interpreted as brand values) were regressed on the vehicle attributes to establish the relationship between brand values and the vehicle attributes (brand value regression).
In the brand value regression, the price variable (one of the vehicle attributes) is most likely to be correlated with the regression error term because price is not exogenous but endogenous, resulting in a biased estimate for the price variable, i.e., positive and significant estimate. To solve this typical problem caused by price endogeneity, 3SLS (Three Stage Least Squares) method was applied. In the previous studies, the instrumental variable (IV) regression was used to analyze the demand-side relationship, i.e., negative impact of price on brand value (single equation). The current study attempts to analyze both the demand and supply-side relationship simultaneously (simultaneous equation). Specifically, one more regression involving price as dependent variable was estimated simultaneously with the brand value regression.
As reported in Chintagunta et al (2005), the IV regression is more robust than 3SLS method because the former is free from misspecification error by not assuming any price-generating process. On the other hand, 3SLS may suffer from misspecification error typically due to choice set misspecifcation or omitted inventory variable as pointed out in Chintagunta et al (2005). However, the use of 3SLS method in the current study is justifiable for the following reasons. First, the proposed model explicitly incorporates choice set information. Second, unlike the case of frequently purchased consumer goods, household inventory level is irrelevant in automobile choice. Third, if we are interested in multiple relationships, system of equations approach is required.
Unlike the typical OLS estimation that ignores the price endogeneity problem, the 3SLS method resulted in correct and significant parameter estimates, more importantly, the price parameter being negative and significant. In addition, the brand value parameter turned out to be positive and significant, implying that in the supply-side, higher brand value enables the manufacturer to charge higher price.

참고자료

· 없음