The Best Price You ll Ever Get The Employee

“The Best Price You’ll Ever Get”: The 2005 Employee Discount Pricing Promotions in the U.S. Automobile Industry∗ Meghan R. Busse+ University of California, Berkeley Duncan Simester++ Massachusetts Institute of Technology Florian Zettelmeyer+++ University of California, Berkeley and NBER September 2007 Abstract During the summer of 2005, the Big Three U.S. automobile manufacturers offered a customer promotion that allowed customers to buy new cars at the discounted price formerly offered only to employees. The initial months of the promotion were record sales months for each of the Big Three firms, suggesting that customers thought that the prices offered during the promotions were particularly attractive. In fact, many customers paid higher prices following the introduction of the promotions than they would have in the weeks just before. We fail to find evidence that the simultaneous increase in prices and sales is due to advertising, decreased financing costs, industry trends, or other explanations. We conclude that the most likely explanation is that the promotion changed customers beliefs about current versus future prices, convincing them that the employee discount promotions were an attractive time to buy. We present several scenarios that could lead to such beliefs. We are grateful for helpful comments from seminar participants at Washington University, UC Berkeley, Harvard Business School, and Dartmouth, and from participants in the NBER IO Program Meeting, the Global Marketing Thought Leaders conference in New Zealand, and the Summer Institute in Competitive Strategy at Berkeley. We are particularly grateful to Steve Berry and Jennifer Aaker for their detailed reviews of the paper, to Jorge Silva-Risso, and to an anonymous automobile manufacturer for their numerous contributions to the project. + Contact: meghan@haas.berkeley.edu. Busse gratefully acknowledges the support of NSF grants SES0550508 and SES-0550911. ++ Contact: simester@mit.edu. +++ Contact: florian@haas.berkeley.edu. Zettelmeyer gratefully acknowledges the support of NSF grants SES-0550508 and SES-0550911. ∗ 1 Introduction U.S. auto manufacturers offer promotions on a regular basis. This is in part because auto manufacturers’ production is fairly inflexible. Production lines in many cases are specifically tailored to a single model, they are designed to run at high efficiency, and there are limited savings to be obtained by scaling back production—labor contracts, for example, guarantee workers almost their full salary, whether production lines run or not. If planned production levels are much too high given demand, car companies will scale back. But for less dramatic oversupply, companies will simply produce the cars, and try to design promotions that are attractive enough to move the cars out of inventory. During the summer of 2005, the Big Three U.S. automobile manufacturers had substantially higher levels of inventory than usual. That summer, all three manufacturers offered an “employee discount” promotion that allowed the general public to buy cars at below-MSRP prices that were formerly offered only to employees. These promotions led to record sales. Sales in the first month of Chrysler’s promotion were the highest monthly sales ever; for Ford, they were the highest sales of any July ever; for GM, they were the highest monthly sales in nearly 20 years. Even though promotions are a very regular phenomenon in the industry, something about this promotion in particular seemed to strike a chord in customers, and get them into the showroom and back out again with a new car. Perhaps the simplest explanation for the record-high sales would be record-low prices. Indeed, one of the promotional taglines for Ford’s version of the employee discount promotion—“You get the best prices of the year. Period.”—seems designed to convey exactly this message about prices. However, in this paper we will show that the actual prices that customers paid before and after the start of the promotion indicates that prices were not lower on average under the employee discount pricing promotion than they had been in the period just prior. Very generous customer rebates that were available just prior to the start of the employee discount promotions meant that the prices offered under the employee discount promotions often were not much lower than the final prices customers had been paying with the rebates. While prices did go down for some models, for many models there was no statistically significant difference in price, and for about 40% of models, prices actually increased. Yet even on these models, where prices increased, sales also increased about 24%. Although this may seem surprising if one believes that the demand curve for automobiles slopes downwards, there is nothing about observing a price increase accompanied by a sales increase that is necessarily inconsistent with downward sloping demand. For example, an industry-wide increase in demand for cars (due to seasonal effects or economic expansion) could lead to prices and sales increasing together. An increase in persuasive advertising (advertising that raises valuations, as opposed to informing customers of the existence of a product or of a discounted price) could also raise prices and sales at the same time. A reduction in financing costs is another avenue that could lead to contemporaneously increased prices and increased sales. The fact that the employee discount promotion offered customers relatively low prices without having to haggle is a fourth reason that sales could have increased even though prices increased. As we will show in this paper, the empirical evidence supports neither decreased prices nor Page 1 any of these alternative explanations for why sales increased during the employee discount promotions. Instead, we propose an entirely different kind of explanation, which is suggested by the fact that cars are durable goods. When buying a durable good, customers can optimize not just which product to buy, but also—if prices fluctuate over time—when to buy. It is fairly well-known by customers that for most car models, transaction prices decline over the model year (Copeland, Dunn, and Hall, 2005). The decline is neither completely predictable, nor monotonic, but generally speaking, the later in the model year, the higher the likelihood of being able to obtain a larger discount off the list price (known in the car industry as the manufacturer’s suggested retail price, or MSRP). Thus, a car customer will typically trade off the benefit of having the vehicle in hand sooner versus the gain of waiting and possibly obtaining a lower price. If customers do indeed consider this tradeoff, then price expectations will play an important role in car purchases. How would expectations explain the sales increases that occurred in conjunction with the employee discount promotions? Suppose that the employee discount promotion begins at the beginning of June. Seeing a big increase in sales between May and June means that many people thought that June was a good time to buy a car, but not so many thought that May was a good time to buy a car. Phrasing this in terms of expectations, the price that customers believed they could obtain in May must not have been particularly low relative to the prices that they expected in May to be able to obtain in the following months. However, the price that customers believed they could obtain in June must have been particularly low relative to the prices that they expected in June to be able to obtain in the following months. As we describe later in the paper, there are several ways in which could occur the change in customers’ beliefs about current versus future prices that would be necessary to induce a sudden increase in sales. In this paper, we will argue that the evidence suggests that the increase in sales that happened in conjunction with the employee discount promotions is most consistent with the cause being one of these changes in expectations, rather than a change in some non-expectations related factor (such as advertising or financing). The implication of this paper is that it is possible for firms to have big effects on their sales by manipulating customers’ beliefs about current versus future prices, even without changing prices themselves. While this is not the first paper to show that firms pursue such strategies, this paper’s contribution is to show that firms can do so successfully even in product categories in which customers conduct a lot of information search and where there is significant money at stake. Cars are an example of such a product category; customers typically engage in information search for 15-20 hours before buying and the difference between a “good” and a “bad” price for a given car at a given dealership can be several thousand of dollars. For durable goods purchases, the key price comparison is the relative price of buying now versus in the future. While information search may not yield much information about future prices, it should yield information at least about current prices. We will argue that the employee discount promotions primarily affected customers’ beliefs about current prices, and that the resulting change in beliefs about current versus future prices is what led to the dramatic sales increases. As we will discuss, part of the reason this is possible is that the car industry engages in price obfuscation. There is an “arms race” of sorts between customers trying to find out Page 2 reliable price information, and firms (particularly dealers) trying to make that information difficult to obtain. Our paper suggests that the observed outcome of this escalation leaves customers still sufficiently uninformed that their beliefs can be influenced. The paper proceeds in the next section, where we briefly describe automobile pricing and the employee discount pricing promotions. In Section 3, we review the relevant literature, and then in Section 4 we present the data used in the analysis. Section 5 establishes the main empirical result and we perform robustness checks. In Section 6, we review alternative explanations, and in Section 7, discuss the price expectations explanation. The paper concludes in Section 8. 2 Automobile Pricing and the Employee Discount Pricing Promotions In this section, we describe some characteristics of retail automobile pricing in the U.S. that are relevant to understanding the employee discount pricing (EDP) promotions. We then describe the EDP promotions themselves. 2.1 Automobile Pricing In the U.S. market, dealer franchise laws mandate that auto manufacturers and dealers must be independent parties. Dealers buy cars from manufacturers at invoice prices that, due to these same laws, generally vary little over units or across dealers. Dealers are then free to set the retail prices at which they sell to customers.1 Almost all dealers negotiate prices individually with customers.2 Knowing that they will be expected to negotiate with dealers, customers spend significant amounts of time gathering information before buying a car. Although this search process is extensive, there are several industry characteristics that hinder search. First, the price that a customer obtains for a specific vehicle depends on factors that can only be learned by engaging in negotiations. Second, the total wealth outlay by the customer depends on multiple price components, including the negotiated price, the amount the customer receives for his or her trade-in, and the value of manufacturer rebates. Third, the final price of a car changes over time, particularly because of changes in manufacturer rebates. These characteristics make it difficult for customers to answer the two key questions that are important for timing the purchase of a durable good. First, what is the total amount the customer would have to pay for a car if s/he bought now? Second, is this an attractive price compared to the option of delaying and waiting for future discounts? Consumer uncertainty on these two points is what gives sellers the opportunity to influence expectations about current versus future prices. 2.2 The Employee Discount Pricing Promotions As described above, manufacturers do not directly set retail prices. Consequently, the EDP promotions were implemented as an opt-in rebate program. Dealers who chose to 1 2 For a description of how the EDP promotions operated in this environment, see the next subsection. There are some exceptions. These “no-haggle” dealerships will play a role in our later analysis. Page 3 participate could sell cars for no more than the employee discount price. A rebate from the manufacturer to the dealer compensated dealers for accepting lower prices. General Motors was the first of three manufacturers to introduce an EDP promotion. The GM event was announced on June 3, 2005, and started on this date. Most of our analyses of the GM promotion focus on changes in sales and prices in the weeks just before and after June 3. Ford introduced its “Ford Family Plan” on July 5, the same day that the GM plan was initially scheduled to end, while Chrysler introduced its “Employee Pricing Plus” plan on July 6. Both firms initially announced that their plans would only last until August 1. However, like GM they subsequently extended their programs into September. Our analysis of the effects of Ford’s and Chrysler’s promotions will focus on the weeks just before and after July 5th and 6th. 3 Literature Review This paper is related to previous work on price search, price signaling, and price obfuscation. It is also related to literatures that concern search and promotions specifically in the automobile industry. 3.1 Price Signaling, Price Search, and Price Obfuscation Our paper concerns an environment in which customers do not have perfect price knowledge, leading them to draw inferences about current and future prices from pricerelevant information, and thus opening the door to the possibility of sellers trying to influence their beliefs about current and future prices. There are existing literatures that examine different aspects of this story. For example, settings in which customers do not have good price knowledge are described by Dickson and Sawyer (1990) and Anderson and Simester (2003b). Use of signals to infer information about prices is modeled by Bagwell (1987) and Simester (1995). Finally, efforts on sellers’ part to influence buyer beliefs about prices is addressed in a literature on “price cues” such as sale signs (Inman, McAlister and Hoyer, 1990; and Anderson and Simester, 1998, 2001), and 9-digit price endings (Jain and Srivastava, 2000; Schindler and Kibarian, 1996; and Anderson and Simester, 2003a). Our paper also concerns issues of price search. Automobiles are a product for which customers engage in a lot of price search before buying. Much of the modern literature on the relationship between pricing and search costs has its origins in Diamond (1971). The more recent empirical work on price search has focused on the role of the Internet, including Brynjolfsson and Smith (2000) and Brown and Goolsbee (2002). Part of the reason that new car customers search, but are still imperfectly informed about prices is because various features of the industry—such as the fact that prices are negotiated—blunt the effectiveness of search activities. A recent literature on price obfuscation suggests that customers’ lack of full price information may be in part attributable to the actions of the firms. See, for example, Ellison and Ellison (2004) and Ellison (2005). Page 4 3.2 Search in the Market for New Automobiles A series of studies have documented consumers’ search activities when purchasing a new automobile. Ratchford and Srinivasan (1993) report that the average customer spends 12.6 hours searching for price information. When combining the search for price and model information, this average increases to over 21 hours and includes 4.6 dealer visits. Similar findings are reported by Bayus (1991), who uses a separate sample of replacement buyers of new automobiles. Ratchford, Lee and Talukdar (2003) later compare survey data from 1990 and 2000 and show that the Internet has led to a slight reduction in the time spent searching, falling from 18.6 hours in 1990 to 15.6 hours in 2000. They also report drops in the number of models and number of dealers considered. It is clear that many customers now use the Internet as their primary tool to obtain pricing information. J.D. Power reported that in 1999, 40% of new automobile buyers used the Internet during their purchasing process. This increased to 54% in 2000; and 67% in 2006 (J.D. Power, 2000 and 2006). Zettelmeyer, Scott Morton and Silva-Risso (2006) report that 72% of their survey respondents used the Internet to help shop for a new vehicle. While most consumers now use the Internet, the Internet has not eliminated the price discovery process at dealerships. Customers who use Internet referral services are only slightly less likely to report that their vehicle negotiations involved a long-series of offers and counter-offers (J. D. Power, 2000). 3.3 The Impact of Manufacturer Promotions on the Market for New Automobiles We know of only a few studies that directly analyze manufacturer promotions for new cars. Thompson and Noordewier (1992) analyze the sales effect of newly introduced endof-model year incentives offered by domestic auto manufacturers in 1985. They find that these incentives were very effective during the end of the model cycle in 1985, somewhat less effective in 1986, and largely ineffective during 1987, suggesting that the incentives became less effective over time. Bruce, Desai, and Staelin (2005) show conditions under which manufacturers would want to offer trade promotions to their dealers and Bruce, Desai, and Staelin (2006) introduce one reason for why manufacturers would want to offer cash rebates, namely that the rebate helps consumers compensate for negative equity in the vehicle they are trading in. Finally, Busse, Silva-Risso, and Zettelmeyer (2006) estimate the pass-through of auto manufacturer promotions from 1998 to 2000. They find that promotional payments to customers lower transaction prices significantly more than promotional payments to dealers. The authors attribute the difference in passthrough to asymmetries between customers’ and dealers’ information about the incentives. 4 Data We obtained detailed transaction data collected by a supplier of data to the automotive industry. The firm collects transaction data from a 25% sample of dealers, designed to be representative of national sales, in the major metropolitan areas in the US. Transactions are uploaded nightly from internal dealer accounting systems and cover all new car Page 5 transactions at the sampled dealerships.3 As we will describe in more detail later, we use a total of 8 weeks of data during the period May-July 2005, for a total of 290,910 observations. For each transaction we observe the exact vehicle purchased, including: nameplate, model, model year, trim level, body type, number of doors, and engine size. We also observe the price paid for the car, the dealer's cost of obtaining the car from the manufacturer, demographic information on the customer, detailed information on the trade-in vehicle if the customer used a trade-in, and the profitability of the car to the dealership. The data also contain census-based demographic information about buyers. To identify the employee discount promotions offered by GM, Ford, and Chrysler, we obtained from a car manufacturer copies of the actual incentive announcements sent to dealers by all three domestic manufacturers. These announcements identify which cars the incentives apply to, when they start and when they are scheduled to end. Finally, we will later investigate whether our findings can be explained by changes in advertising expenditure. We do so using data that we purchased from a third-party data source describing advertising expenditure by the three manufacturers. 4.1 Dependent Variables We begin by studying how prices and sales differ across cars and over time. For each transaction we observe the price paid (before sales tax), including factory installed accessories and options, and including any dealer-installed accessories contracted for at the time of sale that contribute to the resale value of the car.4 Conceptually, we would like our price variable to measure the customer's total wealth outlay, which we will refer to as the “final price.” In order to compute the final price we make two modifications to the observed transaction price. First, we subtract off the customer cash rebate (if any) because the manufacturer pays this amount on the customer's behalf. Second, we add (subtract) to the purchase price any loss (profit) the customer made on his or her trade-in. When a customer loses money on the trade-in transaction, part of his or her payment for the new vehicle is the value of the trade-in vehicle. Specifically, we define the Trade-In Buyer Loss as the estimated wholesale value of the trade-in vehicle (as booked by the dealer) minus the trade-in price. This Trade-In Buyer Loss is not always positive as dealers may trade off profits on the new vehicle transaction for profits on the trade-in. The final price is thus the contract price minus the customer cash rebate (if any) plus the trade-in buyer loss (if any). For the analysis of sales, we aggregate individual transactions to obtain measures of unit sales volume as our dependent variable. In particular, we will calculate sales on the level of manufacturers (e.g. GM, Ford) and at the level of a make-model-model year (e.g. 2005 Honda Accord, 2006 Chevrolet Malibu). 3 4 Dealers provide their data in exchange for information about local market conditions. Dealer-installed accessories that contribute to the resale value include items such as upgraded tires or a sound system, but would exclude options such as undercoating or waxing. Page 6 4.2 Controls We analyze the effects of the EDP promotions on prices using hedonic price regressions. In the regressions, we control for car fixed effects, which allow us to compare only identical products to each other. We define a “car” as a unique combination of make, model, model year, body type, transmission, displacement, number of doors, number of cylinders, and trim level (for example, one “car” is a 2002 Honda Accord sedan with automatic transmission, a 2.2 liter engine, 4 doors, 4 cylinders, and the EX trim). We have 4,565 thus-defined cars in our sample. The only characteristics not captured by the fixed effects are factory- and dealer-installed options which vary within trim level. The transaction price we observe covers such options, but we do not observe what options the car actually has. In order to control for price differences attributable to options, we include as an explanatory variable the dealer's cost of purchasing the vehicle from the manufacturer. Our measure of cost also takes into account any variation in holdback and transportation charges.5 The Vehicle Cost measure does not reflect the 5-6% of the MSRP that GM, Ford, and Chrysler paid to participating dealers in the period during the EDP promotions. Hence, our Vehicle Cost variable will be the same for identical cars, whether a car was sold prior to or during the EDP. To control for time variation in prices, we define a dummy variable Weekend which specifies whether the car was purchased on a Saturday or Sunday to control for price differences within a week. If there are volume targets or sales on weekends, we will pick them up with these variables. We also control for the number of months between a car's introduction and when it was sold. This acts as a proxy for how new a car design is and also for the dealer's opportunity cost of not selling the car. Based on the distribution of sales after car introductions, we distinguish between sales in the first four months, months 5-13, and month 14 and later and assign a dummy variable to each category. We control for the region in which the car was sold. Our data lists 27 such regions (e.g. Baltimore/Washington, Chicago, Northern California, Southern California). We also control for a large number of census-based demographic characteristics at the level of the zip code. We merge these data from the 2000 census by the zip code of the buyer in the transaction data. Specifically, we control for the average race, education, occupation, income, household size, house value, house ownership, number of vehicles per household, commute time to work, unemployment, poverty status, and English proficiency of residents of the zip code in which the buyer resides. Table 1 presents summary statistics for the data. These summary statistics cover the 8 weeks of data analyzed in this paper (two weeks before and two weeks after the GM EDP promotion start and the Ford/Chrysler EDP promotion start, respectively). 5 “Holdback” is the industry term for a percentage of the invoice price that is held by the manufacturer for a period and then rebated to the dealer. It serves the purpose of creating a small margin for the dealer even if the car is sold at the invoice price. Page 7 5 Empirical Results Our analysis proceeds in four major pieces. First, we analyze the effect of the EDP promotions on transaction prices. We will show that prices actually increased on average between the weeks just before and the first weeks of the EDP promotions. Although final prices do decrease for some models, for many more there is no statistically significant change in price, and for about 40% of models, price actually increase at the start of the EDP promotion. Next, we examine what happened to unit sales over this same time period. We will show that unit sales increased dramatically overall, and that sales increased even for models whose prices increased. It is this latter group that most strongly indicates to us that a large part of the success of the EDP promotions must have been due to something other than lower prices. Our third step is to examine a variety of hypotheses that would explain concomitant increases in sales and prices. These explanations include an overall increase in industry demand, increased advertising, decreased financing costs, and the “no-haggle” feature of the EDP promotion. These explanations all turn out to be refuted by the data, or unable to explain the full sales effect of the EDP promotion. Finally, we turn to our favored explanation of the success of the EDP promotions: that they influenced customer expectations of current versus future prices in a way that convinced customers that the EDP represented a particularly attractive opportunity to buy. 5.1 Price Effect We estimate the effect of the EDP promotions on transaction prices using a hedonic price regression approach. We regress the natural log of the final price6 paid by customer i for vehicle j at time t on a vector of customer demographics (Xit), a vector of time effects such as whether the car was purchased on the weekend and the age of the model (Xt), the cost to the dealer of acquiring the car from the manufacturer, regional dummy variables, and detailed car fixed effects which are the cross product of make, model, model year, body type, transmission, displacement, number of doors, number of cylinders, and trim level (φj).7 ln(pijt) = α EDPt + β 1 Xit + β 2 Xt + β3 Vehicle Costijt + β4 Regionij + φj +εijt. EDPt is a time-varying indicator that is equal to 1 during the period that an EDP promotion is in effect. The coefficient of interest is α, from which we can estimate the percentage change in the transaction price before and after the introduction of the EDP promotions. We estimate the effect of the employee discount promotion separately for each model by interacting make-model-model year dummies with the employee discount promotion indicator. We will refer to the make-model-model year dummies as “model” dummies to distinguish them from the more granular “car” dummies (φ): (1) 6 The “final price” is the transaction price minus any customer rebate plus the trade-in buyer loss, as described in section 4.1. 7 Section 4 contains a more detailed description of these variables. Page 8 ln(pijt) = αj µj ⋅ EDPt + β 1 Xit + β 2 Xt + β3 VehicleCostijt + β4 Regionij + φj +εijt. (2) where µj represents the “model” dummies. To control for possible changes in demand conditions we use a regression discontinuity approach. Regression discontinuity takes advantage of discontinuities around the introduction of a treatment. For example, if a scholarship is available to students with SAT scores of 1400 and above, then we could consider students who got an SAT score of 1400 and received a scholarship and students who got SAT scores of 1390 and did not receive a scholarship to be essentially randomly assigned to the treatment and control group. The key identifying condition is that there are no unobservable characteristics relevant to the outcome of interest that change discontinuously around the treatment discontinuity (Hahn et al., 2001; Imbens and Angrist, 1994). In the case of EDP promotions, we use transactions that occurred just before the promotion took effect as the “control” observations and transactions that occurred immediately after the introduction of the promotion as “treatment” observations. The identifying assumption in this context is that there was nothing else that occurred at the same time as the EDP promotion started that would have had a discontinuous effect on the price. Since demand conditions change gradually over time, and manufacturers observe these changes gradually as information filters up from the dealer network, we would expect that the start date of the promotion would be essentially randomly chosen within a short period of evolving demand.8 We estimate a separate specification for the GM promotion, which started on June 3, 2005, and the Ford and Chrysler promotions, which started on July 5 and 6, 2005, respectively.9 In Busse, Simester, and Zettelmeyer (2007) we report the results of estimating Equation 1 for the GM promotion time period and for the Ford/Chrysler promotion time period, estimating α separately for GM, Ford, and Chrysler. We find that prices rose for GM at the start of its EDP promotion by about 1.3%, that prices rose for Chrysler by about 2.5% and fell for Ford by about 0.2% at the start of their EDP promotions (all statistically significant at the 1% level). We conclude that on average prices rose, or fell a very small amount, for all three manufacturers in conjunction with the EDP promotion. The average effects for the three manufacturers, however, may hide considerable heterogeneity across models, so we now turn to the model-specific estimates of price change. The results for the weeks around the introduction of the GM promotion are reported in Table 3, where we summarize the µj ⋅ EDPijt coefficients from Equation 2. Recall that these coefficients estimate the increase in ln(price) for each model following the introduction of GM’s EDP promotion. Looking within GM’s 88 models, 38.6% of the coefficients are positive and significant, meaning that the start of the EDP coincided with 8 See also Busse, Silva-Risso, and Zettelmeyer (2006), who use a similar approach to estimate the price effects of automobile manufacturers’ promotions. 9 For the GM EDP promotion we use Saturday, May 14, 2005 through Friday, May 27, 2005 as the pretest (control) period, and Saturday, June 4, 2005 through Friday, June 17, 2005 for the posttest. For the Ford and Chrysler promotions, we use Saturday, June 18 through Friday, July 1 for the pretest, and Saturday, July 9 through Friday, July 22 for the posttest. Page 9 a statistically significant price increase for that model. A Gaussian test of equality between the fraction of positive and statistically significant interaction terms for GM (38.6%) and for the rest of the manufacturers excluding GM (8.2%) rejects equality at the 1% level (z = 6.90). We conclude that prices for the majority of GM models were higher in the two weeks after the introduction of the EDP promotion than in the two weeks before the promotion, and this pattern contrasted with industry trends. Table 3 reports the analogous results for the Ford and Chrysler employee discount pricing promotions. In these data, 23.4% of the “model”-promotion interaction terms are positive and statistically significant for Ford models, and 67.7% for Chrysler models. This compares to 13.8% for GM models in the same period, and 5.9% for non-domestic models. The fraction for the entire set of cars excluding Ford and Chrysler models is 8.4%. A test of the equality between the fraction of positive and significant interaction terms for Ford vs. the non-Ford/non-Chrysler cars (23.4% vs. 8.4%) rejects equality at the 5% level of significance (z = 2.65), while a test of the Chrysler vs. non-Ford/nonChrysler sample (67.7% vs. 8.4%) rejects equality at the 1% level (z = 8.74).10 Overall, this section shows that average prices increased, or fell very little, for each of the three manufacturers in conjunction with their EDP promotions. Price increased on between 23 and 68% of the models, depending on the manufacturer, and these fractions are much higher than the fractions of car models not on EDP whose price increased at the same time, suggesting that the price increases are not part of a contemporaneous industry trend. 5.2 Manufacturer-level unit sales In this section, we investigate what happened to sales in conjunction with the EDP promotion. Figure 1a shows the change in total monthly unit sales between 2004 and 2005 separately for GM, Ford, and Chrysler.11 The lighter bars show the months of the promotions, June-September for GM and July-September for Ford and Chrysler. As the top panel of Figure 1a shows, GM’s unit sales had been mostly down during January-May 2005 relative to the same months in 2004. However, during the first month of the EDP promotion, June 2005, unit sales were more than 40% higher than they had 10 All of our price specifications control for consumer demographics, which raises the following issue for interpreting our results. If consumer demographics change with the EDP, it would be possible for the price of a model conditional on demographics to rise for all demographic groups, but for more "low price" demographic consumers to buy the model, so that average price paid in the market for that model declines. To investigate whether this is the case we do two things. First, in Section 6.5 we analyze whether the demographic characteristics of consumers before and during the EDP promotion differ (we find that there are some differences but they are small). Second, we re-estimate Equation (2) without demographic variables. We find that the pattern of findings is unchanged. Even without demographic controls, cars sold by manufacturers who are starting an employee discount pricing promotion are more likely to exhibit price increases than cars sold by other manufacturers. 11 The data used in Figure 1 are data on total monthly unit sales for the Big Three U.S. auto manufacturers for 2004 and 2005 which were obtained from company press releases (available on the respective company websites). These are not figures for sales only within our 25% dealership sample. Page 10 been during June 2004, and 20% higher during July 2005 relative to the previous July. As the next two panels show, neither Ford nor Chrysler showed a sales spike in June 2005 relative to 2004; Ford’s sales were down a few percentage points and Chrysler’s up a few percentage points relative to June 2004. When Ford and Chrysler initiated their own employee discount promotions in July 2005, both experienced sales effects similar to what GM experienced in the first month of its employee discount promotion: Ford’s unit sales were about 35% higher in July 2005 than in July 2004, and Chrysler’s sales just over 30% higher. Sales figures for the five other major manufacturers (Toyota, Honda, Nissan, Hyundai, and Volkswagen) do not show evidence that the EDP promotions had a business stealing effect (See Figure 1b). For all of the manufacturers except Volkswagen, 2005 unit sales are an increase on 2004 sales for almost all of the months of 2005, and the EDP promotion months (June-September) do not stand out as either larger or smaller than during other months of the year. For Volkswagen, 2005 unit sales generally decrease relative to 2004, but June-September are actually smaller decreases relative to 2004 than the immediately preceding months. Furthermore, these sales patterns are not attributable to large competitive price cuts by these five manufacturers. None of the five lowered their prices (or increased their customer rebates) markedly during the GM EDP event. We conclude from these initial manufacturer-level comparisons that the EDP programs led to large short-run increases in sales for all three domestic firms. 5.3 Sales Response for Cars whose Prices Increased While the previous subsection showed that sales increased overall for each of the Big Three manufacturers in conjunction with their respective EDP promotions, we are particularly interested in what happened to models whose prices increased at the start of the EDP promotions. If the promotion led to a simple movement along the demand curve, these models’ sales should have decreased. 5.3.1 The GM EDP promotion Using the models that were identified in Section 5.2 as having experienced significant price increases between the pretest and posttest periods, we replicate the sales change analysis reported in Table 3, restricting attention to these models. The results are reported in Table 4. A total of 34 GM models had significant price increases, of which 26 models (76.5%) also experienced increased unit sales over the same period. There is considerable variation in the sales volume of the different models, and so we also calculated this proportion when weighting the models by sales during the pretest period. This weighting amplified the effect: we observed unit sales increases for 88.6% of the models for which there was a significant price increase. Notably, this sales increase was specific to GM models. Around the same time period (the start of the GM EDP promotion), 6 Chrysler models and 13 non-domestic models had significant price increases. None of these 19 models had an increase in unit sales over the same period. There were 5 Ford models with significant price increases, of which only one had a sales increase. With or without the weighting, a t-test rejects at the 1% Page 11 level that the proportions of models with sales increases were the same for the GM and non-GM models.12 As we would expect, GM models that did not have price increases experienced larger sales increases than those that did. Of the 85 GM models covered by the EDP promotion, the 34 models for which price increased had an average sales increase of 24%. The 51 models that did not experience a price increase had an average increase of 42%. In summary, following the introduction of GM’s employee discount pricing promotion, sales increased for most of the GM models for which prices increased. Moreover, these sales increases contrasted with the sales pattern experienced by other major manufacturers. This evidence suggests several conclusions. First, it appears that GM’s EDP promotion affected buyers’ purchasing decisions in a way that led them to be willing to increase their purchases of GM cars, despite price increases. Second, GM was not experiencing an industry-wide change in market conditions. For other manufacturers, we see the more familiar result that price increases do not lead to sales increases. 5.3.2 The Ford/Chrysler EDP promotion In the lower half of Table 4 we repeat this analysis for the Ford and Chrysler EDP promotions. There are 19 Chrysler models for which prices increase during this time period, and for 52.6% of them unit sales also increase (41.8% when weighted by prepromotion sales). For Ford, 11 models have price increases, of which 36.4% (50.0% when weighted) experience sales increases. These findings again contrast with the other manufacturers in the market. There were 12 GM models with increased prices, none of which experienced sales increases, and 12 non-domestic models, 16.7% of which had sales increases (9.6% when weighted). We again reject the null hypothesis that the proportion of models with sales increases was the same for the manufacturers that introduced the EDP promotion (Ford and Chrysler) than for other manufacturers. 5.4 Robustness 5.4.1 One-week windows We have argued that demand changes gradually enough in this industry that by restricting attention to two-week windows around the introduction of the EDP events it is unlikely that the sales changes can be explained by underlying demand changes (other than those introduced by the EDP promotions). This argument is strengthened by the evidence that the sales changes are specific to the firms that introduced the EDP programs, and do not 12 We can also use a non-parametric sign test to test whether increases in sales are significantly more likely than decreases. This test rejects the null hypothesis that unit sales did not increase for the GM models. We also reject the null hypothesis that unit sales did not decrease for the other manufacturers’ models. We obtain similar findings when using a signed rank test to weight larger changes more than smaller changes. These results are reported in Busse, Simester, and Zettelemeyer (2007). Page 12 extend to other manufacturers. However, to further test the robustness of our results we repeated the analysis using one-week (rather than two-week) data windows.13 The results are very similar to what we find with the two-week windows. The fractions of GM, Ford, and Chrysler models that have positive and significant price increases are within a few percentage points of the two-week window results reported in Table 3. The sales change on items that had significant price increases are also very similar to the results for the two-week periods (Busse, Simester, and Zettelmeyer, 2007). We conclude that the results are robust to shortening the length of the evaluation periods. 5.4.2 Other times of the year As another robustness check, we replicated our analyses for other time periods that did not coincide with the start of an EDP promotion. Specifically, we use data from various two-week “pretest” periods and compare them to data from matching two-week “posttest” periods, leaving out a week in between. We use 7 pre-test periods with start dates between March 4, 2005 and June 17, 2005 and 7 pre-test periods from the same time period in 2004. None of these comparisons yielded the same pattern of results. Instead, models that experienced statistically significant price increases were more likely to experience sales decreases than sales increases. These findings suggest that our findings are unique to the introduction of the EDP programs. 5.5 Components of Price The actual change in final prices is so different from what the EDP program seemed designed to convey, and so different from what customers believed the effect of the EDP program to be on prices that it is of interest to understand how exactly final prices changed. The final price is made up of three components: the negotiated price that the dealer and customer agree upon (the contract price), direct-to-customer manufacturer rebates, 14 and the trade-in buyer loss, which is positive if the customer negotiates a price that is lower than the true value of the trade-in, and negative if the customer negotiates a trade-in price above the true value of the trade-in. 15 Equation 4 describes this relationship: (Final price) = (Contract price) – (Rebate) + (Trade-in value) – (Trade-in price) Net new vehicle price Trade-in buyer loss (4) We investigate how the EDP promotions affected each of these components by regressing each of them on the manufacturer times EDP interaction terms, and on the other control variables: 13 For the GM event, the pre-introduction week is May 21-27, 2005, and the post-introduction week is June 4-10, 2005. For the Ford/Chrysler EDP, the pre-introduction week is June 25-July 1, 2005 and the postintroduction week is July 9-15, 2005. 14 15 When the manufacturer did not offer a rebate the Consumer Rebate equals 0. The Trade-in Buyer Loss of a consumer who did not trade in a car is equal to 0 since the customer experiences neither a gain nor a loss. Page 13 Final Priceijt or Contract Priceijt or Consumer Rebateijt or Trade-in Buyer Lossijt = αGM GMj ⋅ EDPt +αFord Fordj ⋅ EDPt + αChrysler Chryslerj ⋅ EDPt + αO Othersj ⋅ EDPt + β 1 Xit + β 2 Xt + β3 VehicleCostijt + β4 Regionij + φj +ηijt. (5) In Table 5 we report the α coefficients from these regressions. (Unlike previous specifications, we do not use logs here, so as to make magnitudes more easily interpretable.) The final price increased by $344 on average across all GM models at the start of the GM EDP promotion, and by $578 for Chrysler at the start of its EDP promotion. Prices were statistically unchanged for Ford (point estimate of - $28) for Ford at the start of its EDP promotion. (Within the sample of models with significant price increases, the averages are naturally larger: from $593 to $650 across the three manufacturers). Despite the fact that final prices do not decrease, contract prices do fall quite substantially. By $881 for GM, by $802 for Ford, and by $1,044 for Chrysler. (The changes in contract prices for models whose final prices increased by statistically significant amounts are very similar, within $100 for each of the manufacturers.) If we were to look only at contract prices—the prices negotiated between customers and dealers for the new car—we could conclude that the EDP promotions had exactly the effect that one might have expected. They lowered prices, and by large amounts. Since final price did not fall, however, there must have been either decreases in customer rebates or increases in trade-in buyer losses. In fact, both occur. Column 4 shows the change in rebate values. Rebate values fell by between $496 and $1,189 for the three manufacturers. For GM and Chrysler, rebate values fell by more than contract prices, increasing the net new vehicle prices. (Rebate values fell by more in the sample of models whose prices increased statistically significantly. The decreases range from $1,131 to $1,302.) The change in net new vehicle price, however, is not the entire story. Trade-in buyer losses also increased in conjunction with the EDP promotion, both in the full sample of models and in the subsample of models whose prices increased. The increases in trade-in buyer loss range from $277 to $433 in the full sample, and $329 to $471 in the subsample. In the subsample of models whose prices increased, about half to two-thirds of the effect on the final price is the increase in trade-in buyer loss, and the rest is the increase in the net new vehicle price. Our hypothesis that changes in price expectations explain at least part of the EDP sales effect does not have make specific predictions about trade-in values. We surmise that the EDP promotions may have focused customers on the new car price, leading them to let their guard down when negotiating the trade-in price. Said another way, customers may have been so sure that they were getting a good deal on the car they were buying, that they felt less pressure to make sure that they were also getting a good deal on the trade-in car they were selling. Although we believe that the trade-in buyer loss should be included in the final price measure in order for final price to measure customers’ actual wealth outlay for the car, trade-in transactions do introduce complications. If the EDP promotion affects how customers negotiate over their trade-in, that complicates how we interpret the comparative static of what the effect of the EDP is on price. In light of this, we repeat our Page 14 analysis of the effect of EDP on the price components by looking at the subsample of transactions that do not involve trade-ins, about 55% of transactions. These results are reported in Table 6. For the GM transactions around the GM event, the average final price increased by $693, while Ford and Chrysler’s averages increased by $366 and $398 (respectively) around their events. Notice that the Ford and Chrysler final price changes are smaller than those reported in Table 4 (which include the transactions with trade-ins). There are a range of possible explanations for these differences. Dealers are willing to shift profits between the trade-in and new car transactions, and the results may indicate that dealers were less willing to agree to a discount from the EDP price for customers not using trade-ins. The results may also be explained by differences in the types of customers who use and do not use trade-ins. While these raise potentially interesting issues, further exploring these issues is beyond the scope of the paper. Instead, we draw a more modest conclusion that the increase in final price does not appear to be fully explained by the change in the trade-in buyer loss. Even amongst transactions that do not use trade-ins, final prices for many models increased after the introduction of the EDP promotions. 6 Possible explanations for contemporaneous price and sales increases In this section, we consider several possible explanations for why prices and sales could have increased at the same time for a substantial fraction of car models in conjunction with the EDP promotions. We consider the following possibilities: 1) the EDP promotion coincided with an industry-wide increase in demand, 2) advertising increased in conjunction with the EDP promotion, possibly increasing both willingness-to-pay and sales, 3) financing costs decreased in conjunction with the EDP promotion, 4) customers were attracted by the “no-haggle” feature of the EDP promotion, and 5) differences in customers between those who bought before and during the EDP promotion. 6.1 Controlling for Industry Trends One potential explanation for the increase in sales in conjunction with the increase in prices at the time of the EDP is some industry-wide factor that changed both price and sales, or that enabled sales to increase even as prices were increasing. One way to address this is to use a difference-in-differences style of analysis by looking at the changes in sales for the three manufacturers of interest relative to the changes in sales that were happening at the same time for other manufacturers. In order to implement this, we start with the models for GM, Ford, and Chrysler whose prices are estimated to have increased in conjunction with the GM and the Ford/Chrysler EDP promotions. For each of those models, we construct the percentage change in sales for that model’s vehicle segment,16 leaving out that particular model: Sj #Q " #Q = #Q k! j k! j 0 k k! j 1 k 0 k (2) 16 Models are divided into 8 vehicle segments, such as Compact, Midsize, SUV, etc. Page 15 1 where Qk represents unit sales in the post-introduction period for model k in the same segment as model j, Qk0 refers to the pre-introduction period. Next, we calculate the differences between the percentage change in a model’s own sales, and the percentage change in vehicle segment sales: "j = Q1 ! Q 0 j j Q0 j ! Sj (3) At the time of the GM promotion, 28 of the 34 GM models whose prices increased statistically significantly had sales increases relative to sales in the segment. For the 19 Chrysler models whose prices are estimated to have increased during the Ford/Chrysler EDP period, 16 models experienced sales increases relative to the change in sales in their segment. The results are similar for Ford. For the 11 Ford models whose prices are estimated to have increased during the Ford/Chrysler EDP period, all 11 models experienced sales increases relative to the change in sales in their segment. Notice that our control for the change in vehicle sales in the rest of the segment is conservative: we are averaging the changes in sales over all models, including models whose prices have decreased—and whose sales would therefore be expected to increase—and comparing them to the subset of models on EDP promotion that have price increases. Nevertheless, the subset of EDP models still have greater sales increases than the average increase in the segment, even though some of the non-EDP models had price decreases. 6.2 Advertising Expenditure, Press Coverage & the Role of Customer Attention A second possible explanation for the concomitant increase in sales and prices is that the firms may have increased their advertising expenditure to coincide with the EDP promotions. To investigate this possibility we purchased detailed data from TNS Media Intelligence describing weekly advertising expenditure in the automobile industry. In Table 8, we report the advertising expenditure for GM during the two-week pre-test period before the start of the GM EDP promotion, and for the two-week period at the beginning of the GM EDP period. We also report Ford’s and Chrysler’s advertising expenditure in the pre-test and post-test periods surrounding the start of the Ford and Chrysler promotions. For comparison, we list the advertising expenditures for the same calendar period in 2004, the year before the EDP promotions. As Table 8 shows, advertising expenditure decreased for all three manufacturers between their respective pre-test and post-test periods. For GM, we see that advertising expenditure decreased by 21% between the pre-test and post-test periods, similar to the 17% decline for the same calendar periods in 2004. For Ford, advertising expenditure decreased by 2% between the pre-test and post-test periods, compared to a 6% increase over the same period the previous year. Finally, Chrysler’s advertising expenditure decreased by 24% between the pre-test and post-test periods, compared to a 1% decrease the previous year. As further evidence that the firms did not increase their advertising expenditure, in Figure 3 we illustrate the monthly advertising expenditures for 2005 for each of the Big Three Page 16 manufacturers, with the EDP promotion months highlighted. These figures reveal that, if anything, the EDP months were low advertising months compared to the months before and after. Before concluding that the simultaneous increase in price and sales cannot be attributed to advertising, we consider two objections that might be raised by the reader. First, one might object that there is a difference between advertising expenditure and the advertising message, and that the increase in price and sales was due to the content of the advertising message (rather than mere expenditure). Survey evidence collected by one of the Big Three manufacturers (which we will describe in more detail later) asked customers to list their primary reasons for buying in response to the EDP promotion. Every one of the top five reasons was about price. (See Table 7.) This suggests that the advertising message of the EDP promotion—at least what came across to customers— was about price. This is not a message that ordinarily would be expected to increase willingness-to-pay. Second, one might wonder whether extensive press coverage—which the employee discount promotions did receive—could have made up for a decrease in paid advertisements. A review of articles in the USA Today, The Wall Street Journal, and the New York Times around the introduction of the promotions reveals that most simply describe the existence and structure of EDP promotions.17 Towards the end of June, the articles started describing the sales success of these promotions. Reading the coverage, it seems unlikely that it served to increase willingness-to-pay the way that traditional brandbuilding advertising would. 6.3 Financing Costs In defining the final price, we have not considered any borrowing costs incurred by customers who borrow money in order to pay for their new car.18 In this subsection, we investigate whether the introduction of the EDP programs coincided with reductions in these financing costs. If so, although the final price may have increased, the total cost to the customer of purchasing the car, taking financing into account, may have fallen, and that could explain how final price and sales could increase at the same time. We examine three types of financing costs: (1) the annual percentage rates offered for new car purchases (the “loan APR”); (2) the implicit interest rates for new car leases (the “lease APR”); and (3) the residual values offered for leased cars. For the loan and lease APR we restrict attention to the captive lending divisions of the Big Three manufacturers. We find no evidence that manufacturers lowered APRs at the time of the EDP promotions. The GM loan APR rose by 0.6 percentage points (p-value 0.000) and the lease APR by 0.1 percentage point (p-value 0.003) at the time of the GM EDP, while 17 We did not find coverage that unequivocally stated that the EDP prices were very attractive prices relative to past or future prices. 18 The reason we have not accounted for this explicitly is that we only observe financing terms if the buyer finances through the dealer. A buyer who uses a car loan from a bank looks like a cash buyer in our data. As a result we cannot incorporate financing costs into our calculation of the final price while treating dealer-financing and outside-financing customers symmetrically. Page 17 Ford and Chrysler’s loan APRs rose by 0.8 and 1.8 percentage points (p-values 0.000) respectively at the start of the Ford and Chrysler EDP promotions. The lease APR rose by 1 percentage point (p-value 0.000) for Ford but fell by 0.5 percentage point (p-value 0.000) for Chrysler at the advent of their EDP promotions.19 While we do not observe the interest rates for customers who do not finance through the dealer, the prime lending rate had risen monotonically from January 2004 through our sample period; indeed, the prime rate jumped from 6% to 6.25% the same week as the Ford and Chrysler EDP promotions started. The third financial component we investigate is the residual value. When customers lease a car they negotiate a purchase price with the dealer. The difference between that purchase price and the residual value of the car is the amount that the customer must finance over the lifetime of the lease. The higher the residual value, the less the lessee will have to pay per month over the course of the lease. If manufacturers wished to counteract an increase in final price, one way to do so would be to increase residual values. We find that residual values decreased at the advent of the EDP promotion, increasing the amounts customers who leased paid for their cars. At the start of the GM EDP promotions, we estimate that residual values fell by $170 (p-value 0.008) for GM, while the start of the Ford and Chrysler EDP promotions coincided with a decrease in residual values of $1,235 for Ford and $258 for Chrysler.20 These results lead us to conclude that the changes in final prices that we estimated in previous sections were not mitigated by changes in financing terms. If anything, changes in financing terms appear to further increase the wealth outlay of customers who bought at the beginning of the EDP promotions relative to those who bought just before. 6.4 Disutility of Bargaining We know from Zettelmeyer, Scott Morton, and Silva-Risso (2006) that many consumers dislike the bargaining process usually associated with buying a car. Knowing this, one might suggest that consumers may have taken advantage of the EDP promotion as an opportunity to obtain a car at a relatively low price without having to undergo an unpleasant negotiation in order to get that price, even if the price that they could have obtained via haggling in the period before the EDP took effect was lower. Indeed, in the aforementioned survey, the third most common reason cited for purchasing a vehicle under employee pricing was “no need to negotiate prices.”21 If enough buyers prefer 19 Looking just within the sample of cars with statistically significant estimated price increases, the results are similar. GM’s loan APR rises by 0.6 (p-value 0.000) and its lease APR has no statistically significant change. Ford and Chrysler’s loan APRs both rise by 1 percentage point (p-values 0.000) while the lease APR rises by 0.5 percentage point for Ford and falls by 0.5 percentage point for Chrysler (p-values 0.000). 20 If we look only with the sample of cars with statistically significant price increases, we estimate that residual values decreased by $76 for GM, by $820 for Ford, and by $257 for Chrysler at the advent of their respective EDP promotions. 21 Although many customers apparently interpreted the prices offered under the employee discount promotions as non-negotiable, this was not in fact the case. Under the rules of the promotion set by the manufacturer, dealers who were participating in an EDP promotion were not allowed to sell cars above the EDP price, but they could sell cars below the EDP price. The variance of prices does appear to have fallen in conjunction with the EDP promotions (Busse, Simester, and Zettelmeyer, 2007) Page 18 “medium low price without haggling” to “very low price with haggling” then the EDP could be associated with an increase in price and an increase in sales at the same time. To examine whether the perceived “no-haggle” aspect of EDP is capable of explaining the effect on prices and sales, we repeated our earlier analyses of changes in prices and unit sales when restricting attention to “no-haggle” dealerships. There are two categories of no-haggle dealers. One category is Saturn dealerships. Saturn is a nameplate of GM, and its business model dictates that all Saturn dealerships are “no-haggle.” The second category is a major publicly traded national chain (NC) owning about 300 dealerships of many different nameplates. This chain has an explicit policy of not haggling. For sales at dealerships belonging to NC and also at Saturn dealerships, any difference in final prices and sales between the pre and post EDP periods cannot be attributed to a perceived change from “haggle” to “no-haggle” pricing since transactions were “no-haggle” both before and after the EDP. Table 9 reveals that 28.6% of GM models had positive and significant price changes at “no-haggle” dealerships around GM’s introduction of the EDP promotion. In contrast, at non-GM dealerships owned by NC, only 2.9% of models had positive and significant price changes. Similarly, 30% of Ford models and 26.7% of Chrysler models had positive and significant price changes at the start of the Ford and Chrysler EDP promotions, which compares with 2.9% of non-Ford/Chrysler models. Table 10 reports the sales changes around the GM EDP promotion. Of the 8 GM models whose prices increased at no-haggle dealerships, 7 of those models had increased sales at these dealerships. Both a sign test and a signed rank test reject that sales decreases are as likely as or more likely than increases for GM. Error! Reference source not found. presents the same analysis for the Ford/Chrysler EDP. For Ford, there are 6 models with positive and significant price changes at no-haggle dealerships, and 5 of those models experience sales increases. A signed rank test rejects that sales decreases are as likely as or more likely than sales increases for these models (because of the small sample size, the sign test rejects only at a p-value of 0.11). For Chrysler, there are 4 models with significant price increases, all of which experience sales decreases. Overall, the pattern of findings replicates our earlier results. Prices were more likely to increase around the introduction of the EDP promotion for firms involved in the promotion. Moreover, for those GM and Ford cars whose prices increased, sales also tended to increase. This replication suggests that the simultaneous increase in prices and sales cannot be solely the result of perceptions that customers could avoid haggling costs by buying under the EDP. 6.5 Customer Differences If the EDP promotions had greater appeal to some types of customers than others, we might expect that we would see differences in the characteristics of the customers who purchased before and after the introduction of the EDP promotions. Notice that this explanation is best interpreted as a complement to other interpretations, rather than as an alternative explanation. If the promotion prompted some customers to purchase immediately rather than delay, it seems plausible (indeed likely) that these incremental customers would be more price sensitive than customers who would have purchased even Page 19 without the EDP promotion. Similarly, the no-haggle interpretation anticipates an increase in the proportion of customers who do not like to negotiate. It would be surprising if there were no differences in the characteristics of customers who like to negotiate and those who are averse to it. To investigate whether there are differences in the types of customers who purchased before and after the introduction of the EDP promotions, we compared the characteristics of the customers on both demographic measures, and the types of vehicles (if any) that they traded in. In our analysis of the demographic measures we regressed each demographic measure on the EDP indicator variable and car fixed effects. This yielded an estimate of how customers who bought a given car during the EDP promotions differed from those who bought the same car just prior to the EDP period. The regressions were run separately for each of the Big 3 manufacturers. Table 11 reports the findings (for ease of exposition we only report the significant coefficients). For each manufacturer there are a number of demographic measures that vary statistically significantly between the pretest and posttest periods. The most common differences are in age, commute time, median household size, and vehicles per household. Although these differences are statistically significant, the magnitudes of the differences are small – in most cases around 1% of the variable value. For example, in the Ford analysis the largest effect (relative to the mean) is for the percent black measure. The percent blacks falls by just 0.6%, compared to a mean of 8.0%. It seems implausible that differences of this size could on their own be responsible for the 30-40% increases in unit sales that coincided with the EDP promotions.22 We also investigated whether EDP buyers are more likely to use trade-ins than non-EDP buyers. In unreported results, we find that customers who buy a GM car during the GM EDP are 3.6% more likely to use a trade-in under the EDP promotion (54.6% in the posttest period vs. 51% in the pretest period). For Ford, the fraction of buyers using a trade-in increases by 1.7 percentage points at the start of the EDP, while the fraction of Chrysler buyers using trade-ins falls by 8.3 percentage points. 7 Changes in expectations So far, none of the possible explanations we have considered that would explain why prices and quantities would increase at the same time have done so. Some potential explanations are directly contradicted by the data (an overall industry trend, an increase in advertising, a decrease in financing costs), while others do not appear to have big enough effects to explain the entire observed change (avoiding haggling, changes in customer demographics). In this section, we turn to the explanation proposed in the introduction: that the effect of the EDP promotions is attributable to EDP changing customers’ expectations about current versus future prices. For a large durable good purchase such as a car, where prices are known to fluctuate, we would expect customers 22 The small sizes of these changes also indicates that customers who are usually disadvantaged in traditional price negotiations, such as women and blacks (Scott Morton, Zettelmeyer, and Silva-Risso, 2003), are not attracted to buying in much greater numbers under the EDP promotions than are other customers. This might be interpreted as further evidence that avoiding the disutility of bargaining is not the primary appeal of the EDP promotions. Page 20 to trade-off the gains of having the good in hand immediately with the possibility of future gains from discounted prices. If customers do consider this tradeoff, then the fact that we don’t see a large increase in GM sales in May, but we do see a large increase in June (or in June vs. July for the Ford and Chrysler promotions) means that customers believed in May that the prices available to them in May were not particularly low relative to the prices they believed at that time they would be able to obtain in later months, but they must have believed in June that the prices available to them in June were low relative to the prices that they then believed would be available to them in later months. What we know based on our investigation in section 5, however, is that between the month before and the first month of the promotion, prices did not decrease; instead they remained about the same, or even slightly increased for most car models. In this section, we describe a set of scenarios that could reconcile what we know happened to actual prices with the beliefs that customers must have had in order to generate the sales changes that we see. We then discuss evidence that we think supports the change in price expectations explanations as the reason for the EDP to have had the effect that it did. 7.1 Scenarios for price expectations and actual prices In this section we describe three scenarios that are consistent with our observation that actual prices did not fall at the start of the EDP, but in which customers have beliefs about prices that would generate large sales in conjunction with the EDP. For simplicity in explaining the different scenarios, we will consider manufacturers to have two choices for prices, a regular price, P, and a discounted price, PL. We will also consider a single EDP promotion which starts at the beginning of June. 7.1.1 An obfuscating price cue Suppose that in the month before the EDP begins (May) and the first month of the EDP (June), actual prices are in fact at their regular level, P. Suppose that in May, consumers’ beliefs about May prices are correct, and they expect these price to continue in the future. In June, however, when they learn about the EDP, consumers believe that they are being offered a temporary discount. We refer to this as an “obfuscating price cue” and it is depicted in the figure below. Page 21 The key aspect of this story is that customers believe in June that prices are discounted when they are not. Why might customers believe this? First, it would have to be the case that customers are not perfectly informed about price. This is likely to be true for several reasons in the car industry. The price a customer will actually have to pay for a specific car can only be ascertained by engaging in a negotiation, a process that can be timeconsuming and that many customers dislike. Second, the final price depends not just on the price negotiated for the new car, but also the price negotiated for the trade-in (if any), and the size of the available rebates. Third, the prices change over time, especially relative to how frequently customers purchase in this market. Even if customers are not perfectly informed about prices, why would they believe that prices actually fell in June when they did not? First, of course, is the fact that the promotional message seemed designed to convey this. Employee discounts are offered by many firms to their employees, and presumably generally are actually discounts. Second, if customers believed that the car manufacturers do sometimes want to offer discounts, then Summer 2005 would be a reasonable time to believe this: inventories were known to be high and there was widespread discussion of the financial difficulties of GM and Ford. Third, there is ample experimental evidence that customers respond to price cues that seem to suggest lower prices, even when the prices are not lower. For example, Anderson and Simester (2001) performed an experiment with a women’s clothing catalog in which half the experimental subjects received an entry in which an item was offered at a particular price, and the other half received the same entry with the same price offered, but the price was labeled as a sale price. Sales were 50% higher in the latter group. Anderson and Simester (1998) present a theoretical model in which a retail firm posting “sale” signs on its products can post some sale signs on items that are not discounted, and still get customers to respond to sale signs as a credible signal as long as they don’t try to mislead customers too often. Although the model isn’t directly applicable to our situation (Anderson and Simester consider the choice to post a sale indicator or not across a product line), one could imagine a similar intuition holding for our situation, in which firms decide to offer or not offer promotions at different points across time. Firms can sometimes offer promotions that don’t actually lower prices as long as, enough of the time, the promotions they offer do lower prices. A signal that is “noisy” can still be credible, as long as it is not too noisy. While we have focused on the subset of cars for which final price increased, there was also a subset of cars for which the final price decreased. In that sense, as in Anderson and Simester’s model, the signal is accurate for some items, misleading for others. 7.1.2 An attention focusing price cue A second possibility arises if the prices offered during the May and June are both at the discounted level, PL, but customers beliefs in May are that prices are at the regular level, P, and will continue to be at that level in the future. In June, however, the announcement of the EDP alerts customers to the fact that prices are actually at the discounted level, but that they will be returning in future to the regular level. If this were the case, then the EDP announcement would encourage customers to buy in June at the attractive prices they did not realize were available, but which they now realize will not last. Page 22 Why would customers not be aware that prices in May were low? As in the previous subsection, it would have to be the case that customers were not perfectly informed about prices. The same features of the industry that, in the previous scenario might allow customers to believe the prices were discounted when they were really at regular levels would also allow customers to believe that prices were at regular levels when they were really discounted. A second reason that applies to the specific case at hand is that, in the years before the EDP promotions, car manufacturers had been engaging in a price-promotion escalation spiral. After the September 11, 2001 attacks, GM launched an aggressive “Keep America Rolling” promotion in the face of fears of a recession. Since then, rebates had been creeping steadily up. In order to keep pace with rebates, manufacturers had also been edging up MSRPs, eroding how much a $3000 rebate was worth, relative to what it would have been worth on the same model car a year or so before. The manufacturers’ perception was that customers had in turn been becoming less and less responsive to large rebate values. If this were the case, then it could indeed be that in the months prior to the EDP, rebate values were large enough to actually lower transaction prices relative to historical levels, but that customers could have not paid much attention to these rebates, believing them to be just another round of a price-promotion escalation. 7.1.3 Information about the future A third possibility is that the EDP promotion does not cause consumers to change their beliefs about current prices, but about future prices. This would be the case if the EDP promotion is interpreted by customers to mean that prices are going to be higher in the future than they originally believed. For example, suppose that the actual prices offered in both May and June are the discounted price, PL. Suppose also that in May, customers expect that price will continue at that discounted price in June and on into the future. However, in June when customers learn about the EDP promotion, they believe that the low prices will continue only for the duration of the promotion, and then will rise. Page 23 Why might customers believe this? If customers were not aware that the rebates that were offered before the EDP was available had termination dates, or if those dates were not particularly well-publicized, and if then the EDP program was announced with its termination date much better publicized, or if the publicity surrounding the EDP emphasized “limited time” and “hurry in now,” then customers might believe that quasipermanent discounts and been replaced by a temporary discount. 7.1.4 Customers care about contract prices A third possible explanation is that the price that customers care about primarily is the negotiated contract price. As we will show later, although final prices did not generally decrease in conjunction with the EDP, one component of the price—the contract price negotiated for the new car—did fall. If this is the price that customers primarily respond to, then customers would have been correct in their beliefs in May that prices were at their regular level, P, and they also would have been correct that in their belief in June that prices were at their discounted level, PL. Page 24 The chief obstacle to this story is that is requires that customers know what happened to the contract prices—that they fell—but not know what happened to rebate values—they fell by more than the contract prices. The contract prices are individually negotiated with dealers for a specific car, while rebates are widely advertised. It seems unlikely that a customer who was sufficiently well-informed to know that the contract price had changed would not realize that rebates had also changed, and by enough to wipe out any gains from a lower contract price. 7.2 Evidence for Price Expectations Explanations We do not observe what customers’ price expectations were before the EDP promotion, or once it started. As a consequence, we must look to secondary evidence for support that the EDP promotions changed customers’ beliefs about how current prices compared to future prices. 7.2.1 Temporal shift of sales One piece of evidence that the sales surge at the start of the EDP promotions was due to customers’ timing their purchases of a durable good is the decrease in sales that occurred in the later months of the EDP promotions. Figure 1a shows that although all three manufacturers continued their promotions through September, the sales gains relative to 2004 were much smaller in August and September than they were in June and July; in some cases, sales were even down relative to 2004 in the later months of the promotion. For Ford and GM the reduction in sales later in the year actually outweighed the sales increases associated with the EDP plans, so that annual unit sales were over 4% lower for both firms in 2005 compared to 2004. Chrysler did not experience the same sales dip at the end of 2005 and so it reported an annual increase in unit sales of 4.4%.23 This is exactly the pattern we would expect to see for customers of a durable good who are trying to optimally time their purchases. If a particularly low price is offered, or— more the point—if customers believe that a particularly low price is being offered, we should see a large surge in sales followed by a “post-promotion dip” in sales resulting from customers temporally shifting their purchases to take advantage of the low prices they believe are available. 7.2.2 Survey evidence A second piece of evidence on customers’ beliefs and expectations about prices comes from survey evidence supplied by one of the three US manufacturers that participated in the EDP promotions. The firm asked a sample of 200 customers to identify both the most important reasons to purchase a vehicle under Employee Pricing, and the biggest disadvantages of Employee Pricing. A summary of the key findings is provided in Table 7. 23 One possible additional explanation for the reduction in GM and Ford car sales later in the year is the landfall of Hurricane Katrina at the end of August of 2005. We do not think that Katrina contributed much to the sales slump since during September (when Katrina’s effects should have been most severe) all major foreign manufacturers and Chrysler increased sales relative to the previous year. Page 25 The responses are consistent with customers believing that the promotion offered unusually low prices. The most frequently cited reason for purchasing under the program was that the EDP promotion offered the “best pricing ever available to consumers.” Moreover, almost half of the customers listed “limited time offer” as the main disadvantage of the program, suggesting that they expected prices to increase once the promotion was over. 7.2.3 Credibility of EDP as a claim of lower price There were several reasons that it made sense for customers to believe that the prices available under the EDP program would be low relative to future prices. As we have already mentioned, inventory levels were high and Ford and GM had financial difficulties, two good reasons to offer a temporary price discount.24 Additionally, manufacturers did have pre-existing employee purchase programs, with prices that were lower than posted MSRPs. Furthermore, it seems sensible for customers to believe that a manufacturer would not go to the effort of establishing an employee purchase plan, and then not offer below-market prices. Finally, had manufacturers not offered the employee prices, or had they attempted to manipulate the prices upwards in conjunction with the EDP promotions, they would have been exposed to potential lawsuits. Indeed, managers at one of the Big Three manufacturers confirmed that the EDP promotions had attracted the attention of several states’ attorneys general, who wanted to ensure that the programs were not misleading. Consistent with this interpretation, our findings confirm that contract prices fell by $800 or more for all three manufacturers following introduction of the EDP promotion. This was the component of price that the EDP purported to lower, and it did so dramatically. In this sense, the signal contained in the EDP promotion was accurate. 8 Conclusions In this paper, we have investigated the effect of the employee discount pricing promotions offered by the Big Three U.S. manufacturers on prices and sales. We find that prices for about 40% of models actually increased in conjunction with the promotion. For a majority of models within this subset of vehicles, sales also increased. We investigate a variety of explanations for this phenomenon. Our evidence indicates that this was not the result of increased advertising, decreased financing costs, or an overall auto industry trend of rising sales. Our leading hypothesis is that the EDP promotions changed customers’ price expectations, convincing them to purchase immediately rather than delay in anticipation of future discounts. We also consider whether the perceived “no haggle” aspect of the EDP promotions can explain the effect. While we cannot rule out that this played a role, it does not appear to be the entire story. In decomposing the final price of a car into its components, we find that the price increases that occurred at the time of the EDP were caused partly by customers getting paid less for their trade-ins, and partly by decreases in direct-to-customer rebates. 24 On May 5, 2005 Standard & Poor’s had lowered the corporate credit ratings for both GM and Ford to non-investment (junk) grade (Businessweek.com, 2005). Page 26 We conclude that promotional activities can increase sales, even in a market for an expensive purchase where customers engage in extensive price search. This suggests that these activities may play a role in a much broader range of markets than previously thought. We conclude that it is possible for firms to influence their sales by changing customers expectations about current versus future prices, even without changing actual prices very much. Furthermore, this can occur even in markets for high-value goods for which consumers spend a lot of time on information search. Price obfuscation by the firms can counteract information search by consumers, leaving customers susceptible to efforts to manipulate their beliefs and expectations about prices. Page 27 References Anderson, Eric and Duncan Simester (1998), “The Role of Sale Signs” Marketing Science, 17(2), 139-155. Anderson, Eric and Duncan Simester (2001), “Are Sale Signs Less Effective When More Products Have Them?” Marketing Science, 20(2), 121-142. Anderson, Eric and Duncan Simester (2003a), “Effects of $9 Price Endings on Retail Sales: Evidence from Field Experiments,” Quantitative Marketing and Economics, 1(1), 93-110. Anderson, Eric and Duncan Simester (2003b), “Mind Your Pricing Cues” Harvard Business Review, 81(9), September, 96-103. Anderson, Simon P. and Régis Renault (1999), “Pricing, Product Diversity, and Search Costs: a Bertrand-Chamberlin-Diamond model,” RAND Journal of Economics, 30(4), 719-35. Bagwell, Kyle (1987), “Introductory Price as a Signal of Cost in a Model of Repeat Business,” Review of Economic Studies, 54, 365-84. Bayus, Barry L. (1991), “The Consumer Durable Replacement Buyer,” Journal of Marketing, 55, January, 42-51. Brown, Jeffrey R. and Austan Goolsbee (2002), “Does the Internet Make Markets More Competitive? Evidence from the Life Insurance Industry,” Journal of Political Economy, 110(3), 481-507. Bruce, Norris and Desai, Preyas and Staelin, Richard (2005), “The Better They Are, the More They Give: Trade Promotions of Consumer Durables,” Journal of Marketing Research, 42 (1), 54-66. Bruc e, Norris and Desai, Preyas and Staelin, Richard (2006), “Enabling the Willing: Consumer Rebates for Durable Goods,” Marketing Science, Vol. 25 (4), 350-366. Brynjolfsson, Erik and Michael D. Smith (2000), “Frictionless Commerce? A Comparison of Internet and Conventional Retailers,” Management Science, 46, April, 563-85. Businessweek.com (2005). http://www.businessweek.com/investor/content/may2005/pi2005055_7793_pi036.htm Busse, Meghan, Jorge Silva-Risso, and Florian Zettelmeyer (2006), "$1000 Cash Back: The Pass-Through of Auto Manufacturer Promotions," American Economic Review, 96 (4), 1253-70. Busse, Meghan, Duncan Simester, and Florian Zettelmeyer (2007), “’The Best Price You’ll Ever Get’: The 2005 Employee Discount Pricing Promotions in the U.S. Automobile Industry,” NBER Working Paper No. W13140. Copeland, Adam, Wendy Dunn, and George Hall (2005), “Prices, Production and Inventories over the Automotive Model Year," NBER Working Papers 11257. Page 28 Diamond, Peter A. (1971), “A Model of Price Adjustment,” Journal of Economic Theory, 3, 156-68. Dickson, Peter and Alan Sawyer (1990), “The Price Knowledge and Search of Supermarket Shoppers,” Journal of Marketing, 54, July, 42-53. Ellison, Glenn (2005), “A Model of Add-On Pricing,” mimeo, MIT. Ellison, Glenn and Sara Fisher Ellison (2004), “Search, Obfuscation, and Price Elasticities on the Internet,” mimeo, MIT. Hahn, Jinyong, Petra Todd, and Wilbert Van der Klaauw (2001), "Identification and Estimation of Treatment Effects with a Regression Discontinuity Design," Econometrica, 69 (1), 201-09. Imbens, Guido W. and Joshua D. Angrist (1994), "Identification and Estimation of Local Average Treatment Effects," Econometrica, 62 (2), 467-75. Inman, Jeffrey J., Leigh McAlister and Wayne D. Hoyer (1990), "Promotion Signal: Proxy For A Price Cut," Journal of Consumer Research, 17(1), 74-81. Jain, Sanjay and Joydeep Srivastava (2000), “An Experimental and Theoretical Analysis of Price-Matching Refund Policies,” Journal of Marketing Research, 37 (August), 351362. J.D. Power (2000), New Shopper.com Study, J.D. Power and Associates, Westlake Village, CA. J.D. Power (2006), New Autoshopper.com Study, J.D. Power and Associates, Westlake Village, CA. Perloff, Jeffrey M. and Steven C. Salop (1985), Equilibrium with Product Differentiation,” Review of Economic Studies, 52, 107-20. Ratchford, Brian T. and Narasimhan Srinivasan (1993), “An Empirical Investigation of Returns to Search,” Marketing Science, 12(1), 73-87. Scott Morton, Fiona, Florian Zettelmeyer, and Jorge Silva-Risso (2003), “Consumer Information and Discrimination: Does the Internet Affect the Pricing of New Cars to Women and Minorities?” Quantitative Marketing and Economics, 1, 65-92. Schindler, Robert M. and Thomas Kibarian (1996), “Increased Consumer Sales Response Through Use of 99-Ending Prices,” Journal of Retailing, 72(2), 187-199. Simester, Duncan (1995), “Signaling Price Image Using Advertised Prices,” Marketing Science, 14 (2), 166-88. Thompson, Patrick A., and Thomas Noordewier (1992), “Estimating the Effects of Consumer Incentive Programs on Domestic Automobile Sales,” Journal of Business & Economic Statistics, 10(4), 409-417. Wolinsky,Asher (1986), “True Monopolistic Competition as a Result of Imperfect Information,” Quarterly Journal of Economics, 101, 493-511. Page 29 Zettelmeyer, Florian, Fiona Scott Morton and Jorge Silva-Risso (2006a), “Cowboys or Cowards: Why are Internet Car Prices Lower?” mimeo, Haas School of Business, University of California, Berkeley. Zettelmeyer, Florian, Fiona Scott Morton and Jorge Silva-Risso (2006b), “How the Internet Lowers Prices: Evidence from Matched Survey and Automobile Transaction Data,” Journal of Marketing Research, XLIII, May, 168-81. Page 30 Figure 1a: Percent Change in Monthly Unit Sales from 2004 to 2005 Domestic Manufacturers Page 31 Figure 1b: Percent Change in Monthly Unit Sales from 2004 to 2005 Foreign Manufacturers Toyota 25% 20% 15% 10% 5% 0% Jan Feb Mar April May June July Aug Sept Oct Nov Dec Honda 20% 15% 10% 5% 0% -5% -10% Jan Feb Mar April May June July Aug Sept Oct Nov Dec Hyundai 25% 20% 15% 10% 5% 0% -5% -10% Jan Feb Mar April May June July Aug Sept Oct Nov Dec Page 32 Nissan 30% 25% 20% 15% 10% 5% 0% -5% -10% -15% -20% Jan Feb Mar April May June July Aug Sept Oct Nov Dec Volkswagon 30% 20% 10% 0% -10% -20% -30% -40% Jan Feb Mar April May June July Aug Sept Oct Nov Dec Page 33 Figure 2: Percent Change in Monthly Advertising Expenditures, 2004 to 2005 GM Monthly Advertising Expenditure 350 Expenditure ($million) 300 250 200 150 100 50 0 Ja n Fe b M ar Ap r M ay Ju l Au g Se p O ct No v D ec Ju l Au g Se p O ct No v D ec Ford Monthly Advertising Expenditure 250 Expenditure ($million) 200 150 100 50 0 Ja n Fe b M ar Ap r M ay Ju l Au g Se p O ct No v D ec Ju n Ju n Chrysler Monthly Advertising Expenditure Expenditure ($million) 200 150 100 50 0 Ja n Fe b M ar Ap r M ay Ju n Page 34 Table 1: Summary Statistics Variable Contract Price Customer Cash Trade Buyer Loss Final Price GM Ford Chrysler Female Age % Black % Asian % Hispanic % Less High School % College % Professional Occup. % Health Support Occup. % Protective Occup. % Food Occup. % Maintenance Occup. % Housework Occup. % Sales Occup. % Administrative Occup. % Repair Occup. % Construction Occup. % Production Occup. % Transportation Occup. Income Household Size House Value Vehicles / Household % Own House % Vacant Commute Time % Unemployed % Bad English % Poverty Sale on Weekend Model Month 5-13 Model Month 14+ Vehicle Cost Sample Size Mean 28,214 1,462 -410 26,342 0.24 0.14 0.09 0.33 45.52 0.09 0.04 0.13 0.16 0.36 0.22 0.02 0.02 0.04 0.03 0.03 0.12 0.16 0.04 0.05 0.07 0.05 53,130 2.68 161,152 1.78 0.70 0.07 27.29 0.05 0.05 0.09 0.24 0.73 0.22 26,699 290,910 Std. Dev. 10,182 1,614 1,581 10,134 0.43 0.35 0.28 0.47 14.53 0.14 0.07 0.18 0.11 0.16 0.07 0.01 0.01 0.02 0.02 0.01 0.03 0.03 0.02 0.03 0.05 0.03 19,130 0.4 102,025 0.3 0.17 0.07 5.60 0.03 0.07 0.07 0.43 0.44 0.42 9,572 Min. 7,500 0 -18,500 6,731 0 0 0 0 16 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4,444 1 0 0 0 0 3.04 0 0 0 0 0 0 6,370 Max. 163,690 10,000 12,550 163,690 1 1 1 1 105 0.99 0.67 1 1 0.96 0.75 0.27 0.77 0.46 1 1 1 0.56 0.65 1 0.53 1 200,001 8.49 1,000,001 4 1 0.89 123.5 0.83 0.69 0.86 1 1 1 143,274 Page 35 Table 3: Summary of Price Changes Fraction of models with price changes that are… GM Event GM Chrysler Ford Non-domestic models Total without GM Total with GM Test statistic (GM vs. rest) Ford & Chrysler Events GM Ford Chrysler Non-domestic models Total without Ford & Chrysler Total with Ford & Chrysler Ford test statistic Chrysler test statistic 13.8% 23.4% 67.7% 5.9% 8.4% 15.2% 2.65 8.74 * ** Positive & Significant Positive Sample Size (# of models) 38.6% 15.2% 11.1% 6.4% 8.2% 15.4% 6.90 ** 73.9% 63.6% 40.0% 47.8% 48.4% 54.5% 4.19 ** 88 33 45 203 281 369 48.9% 40.4% 80.6% 45.0% 46.3% 48.4% (0.75) 3.64** 94 47 31 202 296 374 Page 36 Table 4: Summary of Sales Changes Models with Price Increases Unweighted GM Event GM Chrysler Ford Non-domestic models Total without GM Total with GM Test statistic (GM vs. rest) Ford & Chrysler Events GM Chrysler Ford Non-domestic models Total without Ford & Chrysler Total with Ford & Chrysler Ford test statistic (t,1) Chrysler test statistic (t,1) 0.0% 52.6% 36.4% 16.7% 8.3% 29.6% 2.11 * Weighted by Pre-Promotion Sales 88.6% 0.0% 29.2% 0.0% 1.0% 53.3% 12.67 ** Sample Size 76.5% 0.0% 20.0% 0.0% 4.2% 46.6% 7.63 ** 34 6 5 13 24 58 0.0% 41.8% 50.0% 9.6% 3.2% 22.8% 3.96 ** 12 19 11 12 24 54 3.60** 3.59** Page 37 Table 5: Estimates of Price Components Final price Trade Buyer Loss Contract Price Customer Cash GM Event GM*EDP Ford*EDP Chrysler*EDP Others*EDP Ford & Chrysler Events GM*EDP Ford*EDP Chrysler*EDP Others*EDP 29+ (15) -28 (18) 578** (24) -36** (10) -20 (16) 277** (20) 433** (25) -17 (11) -28 (19) -802** (23) -1,044** (30) 13 (13) -77** (9) -496** (11) -1,189** (14) 32** (6) 344** (18) -234** (24) 194** (30) -11 (12) 320** (19) 17 (25) -13 (31) 4 (12) -881** (23) 52+ (30) 77** (37) -13 (15) -905** (10) 303** (14) -131** (17) 2 (7) Cars with Positive & Significant Changes in Final Prices GM Event GM*EDP Ford & Chrysler Events Ford*EDP Chrysler*EDP 593** (42) 650** (29) 337** (52) 471** (36) -875** (59) -1,123** (41) -1,131** (26) -1,302** (18) 626** (27) 329** (25) -888** (30) -1,186** (19) Robust SE in parentheses ** significantly different from zero, p<0.01 * significantly different from zero, p<0.05 + significantly different from zero, p<0.10 Page 38 Table 6: Estimates of Price Components No Trade-In Sample Final price Trade Buyer Loss Contract Price Customer Cash GM Event GM*EDP Ford*EDP Chrysler*EDP Others*EDP Ford & Chrysler Events GM*EDP Ford*EDP Chrysler*EDP Others*EDP 22 (21) 30 (27) 150** (39) -26* (13) – – – – -45** (17) -486** (22) -575** (31) -5 (10) -66** (12) -517** (15) -726** (22) 22** (7) 345** (26) -217** (35) 613** (48) -6 (15) – – – – -543** (21) 43 (29) 38 (39) -7 (12) -888** (15) 260** (20) -575** (27) 0 (8.3) Cars with Positive & Significant Changes in Final Prices GM Event GM*EDP Ford & Chrysler Events Ford*EDP Chrysler*EDP 366** (46) 398** (76) – – -512** (33) -470** (55) -879** (30) -868** (49) 693** (46) – -534** (31) -1,227** (34) Robust SE in parentheses ** significantly different from zero, p<0.01 * significantly different from zero, p<0.05 + significantly different from zero, p<0.10 Page 39 Table 7: Survey Responses % of Respondents The most important reasons to purchase a vehicle under Employee Pricing Best pricing ever available to consumers Clear and simple pricing communication No need to negotiate prices Feel valued as a consumer by receiving same price as employees Easy to compare prices across manufacturers The biggest disadvantages of Employee Pricing Limited time offer Certain models are excluded No other incentive is available 47% 44% 24% 67% 47% 31% 28% 27% Summary of survey responses from a sample of 200 customers following the introduction of the EDP promotions. Page 40 Table 8: Advertising Expenditure Before and After the EDP Introduction 2005 GM Spending Around the GM Event Pre-introduction Post-introduction Difference Ford Spending Around the Ford Event Pre-introduction Post-introduction Difference Chrysler Spending Around the Chrysler Event Pre-introduction Post-introduction Difference $ 55,937,300 $ 42,635,300 -24% $ 42,323,900 $ 42,091,200 -1% $ 49,057,300 $ 48,005,100 -2% $ 55,714,500 $ 58,871,900 6% $145,568,700 $114,307,600 -21% $115,046,400 $ 95,328,300 -17% 2004 Page 41 Table 9: Summary of Price Changes No-Haggle Dealers Positive & Significant GM Event GM Chrysler Ford Other models Total without GM Total with GM Test statistic (GM vs. rest) Ford & Chrysler Event GM Ford Chrysler Other models Total without Ford & Chrysler Total with Ford & Chrysler Ford test statistic Chrysler test statistic 6.5% 30.0% 26.7% 1.4% 2.9% 9.4% ** ** Sample size 28.6% 0% 6.7% 2.7% 2.9% 8.3% ** 133 138 Page 42 Table 10: Change in Sales Relative to the Segment Around the GM Event No-Haggle Dealers Models For Which Prices Increased Manufacturer GM Event General Motors Chrysler Ford Ford/Chrysler Event Chrysler Ford General Motors 0 5 0 4 1 2 4 6 2 1 0.109 1 0.966 0.023 0.91 7 0 1 1 8 1 0 0.035 1.000 0.013 0.841 Sales Increased Sales Decreased Total Sign test p-value Signed rank test p-value Page 43 Table 11: Demographic Differences Pretest vs. Posttest Periods Demographic GM Cars Around the GM Event Age Commute time Percent poor English Median house value Percent Asian Percent employed in sales Median household size Percent Hispanic Percent employed in transportation Vehicles per household Ford Cars Around the Ford/Chrysler Event Percent black Percent Asian Percent Hispanic Percent poor English Percent employed in administration Vehicles per household Age Commute time Percent employed in repair Median household size Chrysler Cars Around the Ford/Chrysler Event Commute time Median household size Vehicles per household Percent home ownership Income Age Percent employed in construction * Only statistically significant coefficients are reported. -0.37 -0.0217 -0.0108 -0.0056 -596 0.42 -0.0008 0.08 0.0052 0.0041 0.0022 262 0.21 0.0004 26.9 2.6 1.8 0.72 52,333 45.9 0.06 -0.014 -0.008 -0.006 -0.008 -0.011 0.009 -0.014 -0.0061 -0.0019 -0.0051 -0.0017 -0.0009 0.0074 -0.40 -0.15 0.0004 -0.0081 0.0015 0.0005 0.0018 0.0007 0.0004 0.0030 0.17 0.06 0.0002 0.0040 0.08 0.03 0.12 0.04 0.16 1.81 47.0 26.6 0.04 2.7 -0.076 -0.064 -0.044 -0.046 -0.006 0.004 -0.008 -0.006 0.010 -0.003 -0.64 -0.21 -0.0021 -2,647 -0.0016 -0.0007 -0.0095 -0.0041 0.0007 0.0066 0.15 0.06 0.0006 817 0.0005 0.0003 0.0037 0.0017 0.0003 0.0030 46.3 26.8 0.04 141,585 0.03 0.12 2.7 0.12 0.06 1.8 -0.014 -0.007 -0.053 -0.019 -0.050 -0.006 -0.004 -0.035 0.011 0.004 EDP coefficient Standard error Variable mean Coefficient / mean Page 44