How the Internet Lowers Prices - Essay Example

The article "How the Internet Lowers Prices: Evidence from Matched Survey and Auto Transaction Data" by Zettelmeyer, Scott Morton and Silva-Risso (2005) is intended to the issues why, how and in what extent Internet helps consumers to reduce retail prices in such established industry as auto retailing. The article continues the idea stated in the previous article of the authors that new vehicle buyers who use the Internet pay 2.2% less for their car than those who do not. In this article authors calculated the average total Internet effect as 1.5% of the purchase price, or 22% of dealers' average gross profit margin. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.2) The question posed by the authors is how using the Internet lowers retail car prices. A number of activities are involved in process of car purchase: information search, dealer selection, price negotiation and others. The authors investigate in what extent use of Internet in these activities helps consumers to reduce car retail price. Research is made based on transaction data on 1,500 car purchases in California and responses to a survey which asked buyers detailed questions about their Internet usage, their attitudes towards information search and bargaining, and their demographics. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.3) Authors use direct measures of search behavior and consumer characteristics; car price, or Price, is dependent variable. Standard hedonic regression of Price's logarithm is used for modeling: ln(Pricei) = Xi+ Di + Si + i, where the X matrix is composed of car and transaction variables: car model, month, and region fixed effects, car costs, and controls for time of purchase and whether the car was traded in. The D matrix contains demographic and census characteristics of buyers: gender, age, education, income, race, house ownership, median house value and type of occupation. The S matrix contains survey responses that indicate the search behavior and Internet use of a buyer. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.12) First authors investigate the role of Internet by including a corresponding indicator InternetUse. They find that buyers who reported having used Internet to help shop for a new vehicle pay average 1.16% less than other buyers with p-value < 0.001. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.12) Authors conclude that Internet effect is unlikely to result from differences between online and offline buyers in their demographic and other differences. Also the Internet effect is unlikely to result from actual or dealer-perceived differences between online and offline consumers. The aim of further exploration is to identify what consumers do by means of Internet to lower prices. One of the main ideas of the article is that obtaining a referral to an online buying service causes approximately 0.7% price reduce. Another important result is finding that use of Internet only benefits consumers who dislike bargaining. While consumers with a high disutility of bargaining pay 1.5% less when they use Internet for collecting information, consumers who like bargaining do not lower the car price due to using Internet for information. If those consumers also obtain a request from an online buying service, they save additional 0.74% of the car price. Therefore, consumers who dislike bargaining may save 2.24% of the car price and consumers who like bargaining may save 0.74% from the total Internet effect. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.22) Identifying the Variables Despite some factors (consumer occupation, online experience, etc.) are not included in the model, authors regarded almost all significant variables. Authors cited 30 variables that have price effects of search and purchase behavior. Authors made a good work looking for the most significant factors of the model. They estimate how much of the effect of InternetUse on prices is attributable to a consumer being better informed (Informed variable). They find that these two variables have high and significant correlation coefficient of 0.5, so use of Internet significantly helps consumers to become better informed. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.19) An issue that consumers benefit from use of Internet because it provides them with information that helps them better negotiate with a dealer discovered by authors is also significant. But authors find that Internet does not help consumers to find low price dealers, because almost all car dealerships engage in price negotiations it's not possible to make direct price comparison. Here authors do not take into consideration ChatRoom variable that has coefficient of -0.73 that shows its significance. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.33) Secondary information sources, such as chat rooms or bulletin boards may give consumers information about dealers with low prices. Also correlation between search in Internet and search among dealers for lower prices was not observed. Authors correctly identify online buying services as the only type of online site, which is associated with lower price: the OBSSite coefficient is -0.90 and significant with p-value of 0.001. They find that submitting a referral to an independent online buying service (OBSReferral) is associated with 0.72% lower prices in addition to the savings of 0.76% associated with using the Internet; but submitting a referral to manufacturer's website does not impact on car retail price. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.33) Heteroskedasticity Authors regard various car models in their research, e.g. midsize sedan, luxury sedan, pickup, SUV, etc. These cars have different price variations, for example $500 price difference for midsize sedan and luxury sedan are not the same. When OLS to is applied heteroskedastic models the estimated variance is a biased estimator of the true variance. That is, it either overestimates or underestimates the true variance, and, in general it is not possible to determine the nature of the bias. The variances, and so the standard errors may therefore be either understated or overstated. To avoid heteroskedasticity authors use percent deviation of the dealer's cost of purchasing the vehicle from the average vehicle cost of that car in the dataset - VehicleCost. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.11) Multicollinearity In the regarded model nominally different measures actually quantify the same phenomenon to a significant degree. For example, authors included variables GatherMuchInfo and ReadCarMagazines in the model, which have very similar coefficients in Table 2. Also variables InternetUse and InternetForInfo are used simultaneously, but it's hardly to imagine that consumer uses Internet only to make an online booking, not for gathering information. A principal danger of such data redundancy is that of overfitting in regression analysis models. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.33) Omitted Variables Authors did not attach the survey, so I cannot estimate if the independent variable selection is optimal. Not including an independent variable that actually does belong in the model causes biased estimates of the variables that are included, so the model becomes noisier. Although almost all regression models have omitted variables, no model check for significance of omitted variables is made in the article. Endogeneity Endogeneity in regression models closely concerned with omitted variables, so if the regarded hedonic regression model may have omitted variables it is most likely to have some endogeneity. It means that the error terms are not independent and are correlated with the dependent variable. Additionally, the correlation between the dependent variables can create significant multicollinearity, which violates the assumptions of standard regression models and results in inefficient estimators. Differences in significance level in hedonic regression model's coefficients (1%, 5% and 10%) shows that different variables are more or less dependent from unobserved factors of the model. For example, OBSSite, AfraidTakingAdv. and Education are significant at 1%, so they are independent from the model. Informed, ReadCarMagazine and Income2 are significant at 10%, so they are dependent from the model in some extent. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.33) The subject of the research is how Internet helps to lower prices in auto retailing industry, so it must be regarded jointly with other factors that also help to lower prices. The situation may occur that substantial group of consumers who used (or did not used) Internet got low price due to another reason (seasonal discount, referral program, etc.), which is correlated with use of Internet. The impact of unobserved factors is not investigated enough in the presented hedonic regression model in the article. Taking the logarithm of independent variable Price makes their distributions more centered and symmetrical and has the effect of diminishing the otherwise inordinate influence of high prices for luxury cars on the coefficient estimates. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.12) Due to evening-out of the Price function almost linear dependence between two main factors (obtaining a referral and disutility of bargaining) was discovered. Authors use two datasets: transaction data and results of survey and combine these two datasets together to generate the conclusion. This lets authors to assess how respondents and non-respondents differ along census and demographic parameters. If these parameters are identically distributed in 5250 consumers in transaction data and 1435 consumers in survey-based data, the sample is essential. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.7, 33) Since these two samples are dependent, only one of them must be used for hedonic regression model, as it is done in the article. But there's no confidence that the sample of 1435 consumers is significant. In general, hedonic regression model in the article is good, because relevance and independence of the two main factors is shown properly and other regarded factors are less significant. Really, OBSReferral variable has coefficient of -0.72 and is independent from consumer's ability to bargain (BargainingDisutility and DislikeBargaining). Using of Internet benefits buyers who dislike bargaining by 1.5%. Sum of these two factors is 2.2% for buyers who dislike bargaining and 0.7% for buyers who like bargaining. The average value of both factors combined is (0.72% + 2.2%) / 2 = 1.46% that corresponds with average 1.5% of total Internet factor benefit. (Zettelmeyer, Scott Morton and Silva-Risso, 2005, p.26) Bibliography Zettelmeyer, Florian, Scott Morton, Fiona and Silva-Risso, Jorge (2005). How the Internet Lowers Prices: Evidence from Matched Survey and Auto Transaction Data. Cambridge: National Bureau of Economic Research Read More