APPLIED ECONOMICS WORKSHOP
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
APPLIED ECONOMICS WORKSHOP Business 33610 Spring Quarter 2009 Fiona Scott Morton Yale School of Management (Jorge Silva-Risso, Florian Zettelmeyer, Victor Bennett, & NBER) "The Interaction of Technology, Organization, and Product Market Strategy: The Case of Auto Dealerships" Wednesday, June 3, 2009 1:30 to 2:50pm Location: HC 3B For any other information regarding the Applied Economics Workshop, please contact Tamara Lingo (AEW Administrator) at 773-702-2474, tammy.lingo@ChicagoBooth.edu, or stop by HC448.
The interaction of technology, organization, and product market strategy: The case of auto dealerships ∗ Fiona Scott Morton Yale University and NBER Jorge Silva-Risso University of California, Riverside Florian Zettelmeyer Northwestern University and NBER Victor Bennett University of California, Berkeley May 2009 Extremely preliminary. Comments welcome. ∗ We are very grateful to seminar participants at Kellogg for influencing the direction of this paper. In particular Tom Hubbard, Shane Greenstein, and Scott Stern provided valuable comments. We also thank Brian Viard for his input. Scott Morton and Zettelmeyer gratefully acknowledge the support of NSF grant SES-0111885. Corresponding author: fiona.scottmorton@yale.edu
The interaction of technology, organization, and product market strategy: The case of auto dealerships Abstract We study the benefit to automobile franchise dealer from adopting website tech- nology. We demonstrate considerably heterogeneity in the strategies adopted by auto dealers. Our evidence shows dealers adopt clusters of policies: product mar- ket strategies, internal organization and incentive policies, and technology adoption policies. A simple model demonstrates that dealers that adopt web technology will tend to have lower margins and customer-friendly organizational practices. Because we show that web capability is intrinsically correlated with other dealership policies through the dealer‘s strategy, we instrument for technology adoption. Our instru- ments include demographics of the county and census block of the dealer, as well as dealer characteristics such as nationality of nameplate. An IV regression shows that web technology significantly predicts larger dealer volume and lower dealer margins on new cars. Our results are stronger when we focus on the combined characteristic: the dealership adopts web technology and also implements appropriate job design for its internet sales staff.
1 Motivation We study the benefit to automobile franchise dealer from adopting website technology. We demonstrate considerably heterogeneity in the approaches adopted by auto dealers in their effort to sell cars effectively. Our evidence shows dealers adopt clusters of policies. These clusters include characteristics of the dealer’s product market strategy, internal organization and incentive policies, and technology adoption policies. We refer to these clusters as “dealer strategies.” For example, a dealer that sells new cars in high volume and at low margins is one type of strategy we see in the data. Another strategy relies on high profits from financing and more sales of used cars. These product market strategies are associated with different human resource and organizational choices, e.g. the large new car dealer does more inhouse training and the financing/used dealer bases compensation more heavily on gross margin. Inputs such as the quality of the dealer‘s building also vary by strategy. And of course, choice of technology is determined in conjunction with product market and other strategies. The high volume-low margin dealership has more website capabilities than the other. Later in the paper we elucidate these choices using a simple model of complementarities among policies. In our model, dealers that adopt web technology will tend to have lower margins and customer-oriented organizational practices. Because we show that web capability is determined jointly with other dealership policies through the dealer‘s strategy, attempting to explain sales or margins with a measure of technology will not generate an estimate of the causal relationship. We will use an instrumental variables strategy to determine the relationships between technology adoption, volume, and margins. Technology and organization A dealer‘s website (technology, in our setting) allows a dealer to be found by a consumer searching with a search engine. Once a consumer arrives at the website, she becomes a “lead,” or a “prospect.” She looks at the information on the website and decides whether or not to visit the dealership or contact it by phone or email. Dealers can put contact information, such as email addresses and phone numbers on their sites; they can allow past customers to schedule repairs; they can allow prospects (who may become customers if they decide to buy) to view dealer inventory of vehicles; they can provide driving directions and hours of operations; and they can allow a prospect to make an appointment with a salesperson. However, to get the full benefit of visits to the site, the dealership must make appropriate organizational change within 2
the dealership. For example, if no one is made responsible and given incentives for answering email in a timely fashion, messages will not be answered, and prospects will buy elsewhere.1 Some dealerships will have a dedicated person to answer email enquiries, some will have an Internet department, and some will have an Internet service manager, or ISM. Additionally, an even bigger productivity boost may occur if this ISM has the optimal skills for the job, for example, the ability to effectively use both email and the phone. One motivation for this paper was the experience of an author who, several years ago, attempted to schedule an appointment at the website of her VW dealership. The website had a place to enter a phone number which, it said, would cause the service department to telephone to schedule a repair. The author never received a return phone call. This experience illustrates that dealerships may not (at least, right away) make the organizational changes needed to take advantage of a new technology. 2 Literature review For years the Economics literature has attempted to elucidate the three-way interdependence of a firm’s organizational form, technology use, and product market strategy. Unfortunately, because of the rarity of matched data on all three, most work has focused on interactions between two of the three. For the interaction of information technology and organization, a recent literature has emerged relying on estimating production functions. The papers in this group have typically discovered a complimentarity between information technology use and organizational form. Bresnahan, Brynjolfsson, and Hitt (2002) carry out a survey of work organization at a sample of large firms by asking senior human resource managers both to provide information about the types of workers at the firm (education, etc) and to rate firm practices in areas like team-building and worker control over work practices. The paper shows positive correlation between work organization and the amount of IT (stock of computers) utilized by the firm. The paper also estimates production functions including interactions between IT and work organization, and finds this interaction has a positive impact on productivity. Bloom, Sadun, and Van Reenen (2007) also consider the question of the effect on productiv- ity of the interaction between organization and IT (PCs per employee). They examine a sample of UK establishments that were acquired either by US, other multinational, or domestic firms in the 1990s. The US-owned establishments showed higher use of IT and more productivity 1 Anecdotally, a response time under 20 minutes for price quote enquiries significantly increases sales. 3
per unit of IT. Another branch of this literature forgoes production function estimation and attempts to measure the causal effect of information technology on organizational form. One example, by Acemoglu, Aghion, Lelarge, Van Reenen, and Zilibotti (2007) models firms experimenting with technology and demonstrates that firms closer to the technological frontier have shallower organizational structures. All three of these papers have the advantage of a great deal of generalizability from their use of multiple industries. This approach, however, has two main drawbacks. The first being that IT stock must be measured at a very general level: for example, the value of the computing stock or the number of PCs per person in the firm. The second limitation is that product market strategy is difficult to measure across such disparate firms, so strategic changes that occur with technological adoption of organizational change cannot be identified or their effects correctly attributed. In contrast, work by George Baker and Tom Hubbard (Baker and Hubbard 2003, Baker and Hubbard 2004, Hubbard 2003) analyze a case in which technology serves as a substitute for certain organizational tasks, like managerial monitoring. In this work the authors study the effects of IT (on-board computers) and argue convincingly that the monitoring function of the technology replaces the incentives created by truck ownership for the truck driver, while the coordination-enhancing abilities of the technology make outsourcing of shipping activities more efficient. This work uses waves of a trucking census to demonstrate that adoption of this type of IT resulted in organizational change: more vertical integration into truck ownership and less vertical integration into shipping. The nature of the truck census data does not allow for an examination of product market strategy. A second literature, studying the interaction between technology and product market strat- egy includes that of Bartel, Ichniowski, and Shaw (2007). This paper examines the impact of the adoption of new technology in a dataset of valve plants. The authors survey plant man- agers to learn about the technology and product market strategy at the time of the survey. They also ask the managers to answer the same questions for the plant five years prior to the survey. This yields a dataset that is similar to a panel, though not quite as clean due to the strong time trends in technology and product market strategy in this industry. Nonetheless, the authors find that plants that adopt advanced computerized machine tools have higher la- bor productivity and shift their product mix toward more customized products, which have become less costly to produce due to the new machines. Though the authors note that higher- 4
technology plants also demand more skill in workers due to their operation of the more complex machines, they do not observe any commensurate change in organizational form, like compen- sation structure, hierarchy, or job design. The authors do not address the causal relationships between technology and product mix and labor. Demand for customized products might lead a firm to purchase machines or purchasing machines might lead a firm to make more customized products. The single industry focus of many of these papers opens the door to more specific and comparable measurement of information technology and product market switches, but has the disadvantage of less heterogeneity in organizational form. For the third pairing, organization and product market strategy, the literature has sug- gested that firms operating in more competitive product markets take on shallower organiza- tional forms. Guadalupe and Wulf (2008) use changing tariff rates to measure increases in competition and demonstrate a flattening of firms. As with the first set of literature, this mea- surement yields a great deal in terms of generalizability, but requires very coarse measurement of organizational form. Furthermore, changes in organization due to technological adoption spurred by competition are attributed directly to competition. Our paper aims to convincingly address the interaction of all three of organization, infor- mation technology, and product market strategy. In order to do so, we focus on one industry, new automobile sales. This narrow focus allows us to measure a very specific form of Informa- tion Technology and therefore, we hope, get something reasonably precise. We analyze only one piece of technology, the dealer‘s web page, and its attributes. Furthermore, the limited variations in organizational form amongst dealerships allows us to divide up types of firms in a reasonable way and link other attributes to variation in form. The third main benefit of this narrow focus is the ability to observe and identify changes in product market strategy while controlling for confounding changes in competition. We can quantify product market strategies such as markups, price discrimination, and sales of complementary products. We refer readers interested in the economics of auto retailing to our own previous work (Scott Morton, Zettelmeyer, and Silva-Risso 2001, Scott Morton, Zettelmeyer, and Silva-Risso 2003, Zettelmeyer, Scott Morton, and Silva-Risso 2006). In these studies we examine the effect of shopping online, race and gender discrimination, and how consumers search and bargain for a new car. These papers also describe the industry and the institutional environment of auto retailing. 5
3 Industry Background 3.1 Production Function The production function for auto retailing is extremely simple. It requires an inventory of cars, a building, and a group of salesmen who match potential customers with cars and bargain over price with the customer. However, within this basic structure, an auto dealer must choose particular policies that create variation across franchises. Charging low margins to attract high volume is one such choice. Others might be to incur the higher selling costs associated with price discrimination, or charge high margins and accept lower sales as a consequence. Earning profits from selling financing and insurance is another area a dealer can emphasize. Financial incentives for salesmen (volume, margin, services), training, and monitoring practices such as tracking leads must be selected by the sales manager. 3.2 Adoption Dealerships choose if they will adopt a website. The adoption of the technology will be made by dealerships that expect to gain the most from it. A website may attract a particular type of customer (low margin, more informed, younger) and may therefore be more or less attractive depending on the strategy of the dealership. The website may also allow for monitoring of leads coming in to the dealership, which may or may not be a match for existing dealership practices concerning leads. Nameplates (e.g. Chevrolet, BMW, Toyota) are an important source of influence on adop- tion of IT. They can endorse a software product by saying it is approved for franchisees. Often, multiple competing products are approved in this way. Or, a nameplate can endorse a set of products and include substantial incentives for dealers to adopt. (Franchise laws prevent name- plates from requiring adoption of any technology policy by a dealer.) These incentives typically require a dealership to have the technology in order for it to attain some special status with the manufacturer, which in turn earns the dealership privileges (e.g. Chrysler 5-star dealer or Lexus “Elite” status). An autogroup of multiple franchises with a single owner may invest in a website platform that it offers at low marginal cost, or requires, for its members. 3.3 The role of the franchise system Franchise territories are strictly governed by the franchise agreements between the auto fran- chisee and the manufacturer (franchisor) and state laws. Franchisees can object to any change 6
the manufacturer might make that would reduce expected profits, such as granting another franchise nearby or changing the location of an existing franchise. Furthermore, US-based automakers established dealer networks when the optimal number of dealers was larger due to higher market shares of US nameplates and lower economies of scale in repair operations. Additionally, populations grow and move (e.g. from the city to the suburbs) and so a dealer network designed for consumers 50 years ago is no longer in the right place. Population growth has varied considerably across the counties in our sample; Pennsylvania has some counties that shrank while the west coast has many very high growth counties. Thus, any configuration of dealerships that was established long ago is no longer optimally designed relative to demand today. One can see evidence of this in the current auto crisis. Bankruptcy is giving American nameplates the chance to terminate dealerships without the high costs of compensating the dealerships with the present discounted value of their rents, and they are taking advantage of that option. Non-American nameplates (e.g. VW, Toyota, Honda) have many fewer dealerships which were located more recently, and therefore we assume are matched more closely to cur- rent demand. We exploit these differences across nameplates and counties in our instrumental variables strategy. 3.4 The Impact of Organization A fundamental choice facing a sales manager is how to configure the organization to handle the leads arriving from the new source, the website, as well as any other channels. The sales manager may simply assign the leads to a showroom floor salesperson he thinks is competent and let him handle them along with his other work. Alternatively, the dealership might set up an Internet department and assign salespeople who have to respond to email leads with email and phone selling. An Internet Sales Manager, or ISM, is the name for a salesperson who heads this department. Because the mix of skills that is desirable for a salesperson to have may be different across the two jobs, the dealership might set up an Internet department and hire new staff with the particular skills needed to work in it, rather than simply deploying floor salesmen to handle Internet customers. The survey asks questions about who is assigned to which jobs and about how leads are tracked. Compensation is also a critically important issue in auto retailing. The usual compensa- tion scheme for a salesperson is a percentage of the gross margin on a vehicle (less a constant subtracted off by the dealership). Internet salespeople may be compensated differently. For example, Autobytel.com requires that salespeople handling leads from its website be compen- 7
sated on volume, not margin. (We have no knowledge of how many dealerships adhere to this rule.) The Internet Sales Manager’s typical compensation is a combination of some percentage of the net sales in his department plus a function of customer satisfaction. 3.5 Managerial input We held conversations with a number of managers at a firm that sells software to dealerships (Cobalt) as well as other industry experts and dealers in order to determine what practices impeded productivity gains for dealerships that had adopted websites. Based on those conver- sations we formed several hypotheses about the sort of organizational changes dealers may need to make to gain maximum productivity out of the software. We developed survey questions that measure the organizational practices at the dealership. We focus on practices we think are particularly productive once leads are coming in to a dealership from its website. Below we describe the organizational problems and then the associated questions. Dealerships typically allocate customers who come on the lot to whichever salesperson greets them first. Thus, a salesperson who is also supposed to be answering email queries via the website faces a conflict between waiting on the front steps in order to sell to walk-in customers and spending time at his desk sending email. Allocating leads from the Internet and creating incentives to respond to them quickly is one organizational issue for dealerships when there is not a dedicated Internet Sales Manager (ISM). A feature of the Internet that is different from walk-in leads and even phone leads is that it is easier to accurately measure website leads and track them. The dealership has an interest in following up with every possible lead because some of those leads will buy today, and others may be ready to buy in the future. These future sales are more likely if the salesman creates a favorable impression and obtains some contact information from the customer. However, the salesman is focused on leads that will convert today, e.g. the ones that walk onto the lot. The interests of the salesman and the dealership do not perfectly coincide because there is considerable turnover among sales staff: 30-50% a year is not unusual. Thus the salesperson is not likely to put as much effort as the dealer would like into prospects that appear less likely to buy today. Additionally, regardless of the source of the lead, salesmen have an incentive to under-report the number of prospects at the dealership because their performance is essen- tially the ratio of sales to prospects. Accurate counting of leads coming in via the website and other channels becomes a measurement tool which can be used to create incentives and compensate salespeople. We ask several questions about the extent and type of lead-tracking 8
at the dealership. The ISM typically contacts the lead and attempts to arrange an appointment with a sales- person at the dealership. Since both the ISM and the salesperson who closes the sale are critical, the right compensation scheme that gives the two parties good incentives must be designed. Creating a compensation scheme both in terms of tasks and in terms of what is measured and rewarded is an important organizational issue for dealerships. The survey asks managers about the basis of compensation for several job types. Finally, the traditional salesperson is skilled at selling to customers who walk into the showroom. Responding to website inquiries requires communicating effectively and persuasively by email and over the phone. Thus the mix of skills that is desirable for a salesperson to have may be different once there is a steady stream of leads coming through the Internet. Alternatively, the type of customers using the Internet may be different from the norm (perhaps more informed) and so the salesperson assigned to them will be more effective if he can address their different needs. Internet sales staff might benefit from training that is particular to the medium they use. The survey asks about job description, training, and compensation. 4 Model An advantage of studying a particular technology is that we can write down a model that reflects the role of that technology in the activities of firms in our industry. Web-based communication between car dealers and potential customers allows customers to do a number of things. For example, customers can obtain information about the dealership, make an appointment, search inventory, or get a price quote on a particular car. These activities increase the efficiency of car-buying for the customer. The customer saves time and disutility of shopping by using the Internet (Zettelmeyer, Scott Morton, and Silva-Risso 2006). The interesting problem for the researcher is that this sort of efficiency gain will not (fully) show up in GDP, nor in measures of revenue or productivity of the dealership. Unmeasured efficiency gains are an important characteristic of the use of the Internet in retailing and other settings (Scott Morton 2005). A consumer clearly gains when it is easier to locate a store, determine opening hours, identify a product of interest etc. Competition from other retailers who also adopt the same technology prevents the retailer from capturing the full benefit in prices or quantities. Of course, an important reason stores deploy web technology is for inventory management and other tasks that raise measured productivity. However, some portion is used to attract customers by 9
making the entire shopping experience more pleasant or lower (transaction) cost. On the other hand, web technology may have some impact on observable performance measures. It is possible that a more informed consumer is ready to buy once she arrives at the dealership and therefore takes less sales time, allowing the salesman to sell more cars per day. A good website may increase matches between customers and the dealership or a car, and in this way save sales staff time. However, the Internet also informs consumers of the accurate market price of a car (Zettelmeyer, Scott Morton, and Silva-Risso 2006), and previous work shows that people who use the Internet pay below average prices for a car. Thus, if output of a dealership is measured as revenue or gross profit, there would potentially be two opposing effects of attracting internet consumers: higher quantity of cars sold per salesman, but lower gross margin per car sold, with an uncertain net effect. With our data, we are able to measure each component of dealer performance, quantity and price, separately. We propose the following simple model: Consumers with mass one can purchase from either of N = 2 firms; the car gives a consumer value V no matter where she purchases it. She must search to find a car to purchase. Search on the Internet is less costly than search offline (s m above marginal cost. The consumer first contacts N dealerships with her search questions. Suppose, for this exercise that the marginal benefit of searching both dealerships, M − m, is greater than the cost of contacting them, N · s. A dealership receiving a question online responds with an answer of quality r(IT, JD). The quality of the response increases in the dealerships level of information technology (IT ), in the dealerships level of job design and incentives (JD), and in their interaction. If all responses are unsatisfactory (r < 0) the consumer spends S to visit a random dealership. Essentially, this is just a cutoff below which the response is too poor to be useful and the consumer reverts to offline shopping. Provided at least one answer is of sufficient quality r > 0, the consumer picks the highest response and purchases a car from that dealership. The mass of consumers who do not have Internet access (1 − α), visit a random dealership to purchase. Because Internet customers arrive with more information and a more specific idea 10
of the car they want, we assume online customers cost the dealership c in sales time, while offline customers cost C > c because it takes the salesman more time to match the consumer with a car. Selling costs are below their respective margins (c < m, C < M ). Quantity sold by a dealer is therefore a function of the local market size, α, and the answer quality function r(IT, JD). 1 qi = αI(r(ITi , JDi ) > r(ITj , JDj )) + (1 − α) 2 Any dealership may purchase incremental IT for f (IT ), where f (·) is invertible and increas- ing monotonically. The dealership is endowed with JD and cannot alter it. We model heterogeneity of orga- nizational practices across dealerships with JDj , where j indexes dealerships. We choose to put the heterogeneity in the job design rather than the IT because of the inherent replicabil- ity of website software. It seems more reasonable to us to assume that HR practices, social norms, incentive schemes, and work rules vary across stores and are harder to change than the technology. Profits take the form, 1 πI = αI(r(ITi , JDi ) > r(ITj , JDj ))(m − c) + (1 − α) (M − C) − f (ITi ) 2 This form exhibits clearly that, in this model, there is no benefit from investing in IT for a firm that will not be the leader in providing responses to customers, and therefore capture the Internet-using population. In this model all equilibria will be characterized by the adopting firm being the one with higher levels of job design, JD. This brings us to our first empirically testable implication. Proposition 1 Firms investing in IT will be those with higher levels of Job Design Sketch of Proof: Suppose, without loss of generality, that JDi > JDj and ITi < ITj . The marginal cost benefit comparison for investing in IT is, ∂πi (IT, JD) −f (IT ) where ri (IT, JD) < rj (IT, JD) or ri (IT, JD) > rj (IT, JD) , = ∂IT α(m − c) − f (IT ) where ri (IT, JD) = rj (IT, JD) 11
Since the value of capturing the Internet market is α(m − c), no firm would ever invest more than that amount in IT . Investing weakly less than the value of the Internet market is worth it only if the investment results in capturing said market. For any answer quality rj and commensurate ITj , firm i is willing to invest up to the marginal cost of capturing the Internet-user market, α(m − c), for IT 0 if the resulting ri0 from doing so is ri0 > rj . For any ri0 < rj , ri would not be willing to invest anything, as its market size would not grow. Given a maximum investment of f −1 (α(m − c)), this means that the maximum answer quality that firm i would be willing to invest to attain is: r̄i = r(f −1 (α(m − c)), JDi ) Similarly, the maximum that firm j would be willing to invest is r̄j = r(f −1 (α(m − c)), JDj ) The increasing differences of r(·) implies: r̄i = r(f −1 (α(m − c)), JDi ) > r̄j = r(f −1 (α(m − c)), JDj ) Thus, for any ITj and commensurate rj∗ in which firm j is willing to invest, firm is dominant strategy is to invest in a level ITi such that the commensurate ri > rj , and firm i captures the entire Internet-using market. Given that firm j cannot capture any Internet-users, investing in IT is a dominated strategy, which contradicts the notion that its IT spending could be higher. QED. The second testable implication is that dealers who invest in IT will have a higher volume of sales than those who do not. Firms that adopt IT capture α + (1 − α)/2. Those who don’t capture (1 − α)/2. The third implication, and one that this data is uniquely suitable to test, is that the average margin of dealerships who invest in IT will be lower than the average margin for non-adopters αm + (1 − α)M < M In section 5 we describe the data we use to test these three implications. 12
5 Data 5.1 Output and operational data Our primary dataset consists of transactions gathered from the internal IT systems of franchised dealers. These are collected by a specialist auto data firm, hereafter MPAD. The MPAD sales data are extremely detailed. A year’s worth of data includes all new and used car transactions at the specified dealerships with a large number of characteristics of the car, the trade-in, and the additional services sold by the dealer such as finance and insurance. These data allow us to summarize the dealer‘s strategy using measures of the volume of new cars sold, the margin on those cars, and the profits earned by the dealer on insurance and financing. We also know the address of each buyer and the dealer, so we can create measures of market structure such as the number of same-nameplate dealers in a geographic radius. The age of the buyer is also known, and from this we calculate a county-level average age of all (new and used) car buyers in that county (in our sample). 5.2 Dealer Survey We begin with the names and address of all Connecticut, California, Pennsylvania, New York, Oregon, and Washington franchised auto dealers. We will use all nameplates with significant sales in the US. Practically speaking, this means all nameplates except boutique marques such as Lamborghini. From the states’ lists we can obtain basic data about each dealership: phone, nameplate, owner, etc. We outsourced the actual implementation of the survey to a market research firm. Em- ployees of the firm phoned Sales Managers of the targeted dealerships. Prior to this call, we introduced it with a letter explaining the academic nature of the project. We stress the con- fidentiality of the project and provide the responder with university-themed golf balls as an indicator of both the academic orientation and our interest in speaking with them. The purpose of the first phone call is to schedule a time to run the survey over the phone. A second item, such as a Yale or Cal hat, is mailed to the manager upon completion of the survey. We ran a pilot test of the survey on dealerships in Utah which is not a state we are using for the survey. The pilot allowed us to adjust many of the incentives and logistics of the survey to improve its design and attractiveness. A copy of the survey may be found in Appendix A. We completed surveys for approximately 500 dealerships. The surveys were conducted during August and September of 2008. We have merged our survey dealerships with a year of 13
the MPAD sales data (the third quarter of 2007 through the second quarter of 2008). Note that the US car market changes drastically during this time. The 2007 quarters are fairly normal, but car sales drop dramatically in 2008. 5.3 Technology adoption: website The website information on the survey is reported by the general manager or sales manager of the dealership. We also collect objective data on the visible attributes of each dealership’s website. We hired research assistants to visit the websites of all our successfully-surveyed dealerships and rate them on what information, buttons, drop-down menus, etc, they had that were useful to a car shopper. This work was done a few months after the survey was completed (March-May 2009) For example, a website with an address and directions could simply contain an address, it could have written directions from various points, or it could have a live link to googlemaps with the address of the dealer already inserted. Likewise, a request for a quote function could simply be instruction to telephone the dealer, a form to fill out which will produce a follow-up call, email options, and fax options. The physical capital inputs used in auto retailing are not standard production function inputs like plant and equipment. The primary capital of the dealer is his inventory; a dealer that has many varieties of model and trim on his lot will find it easier to satisfy consumer demand. Note that dealers do own and take title to the cars on their lot, and therefore inventory has the normal financing costs of other capital expenditure. We use the MPAD data to construct an average inventory held by each nameplate during our time period. Then we create a variables which is the difference between a dealer’s holdings and the average inventory for that nameplate. We also use a measure of capital that is the dealer’s subjective valuation of his own physical facilities relative to competitors. 5.4 Demographics We merge data from the census into our transaction data. In particular, we gather character- istics of the census block in which the dealer is located and the county in which the dealer is located. We collect the 1950 population for the county as well as the population in 2000. 5.5 Summary Statistics Summary statistics are reported in Table 1. On average, our dealerships have had the ability to post hours and directions for 8.4 years, 7.9 for an appointment, 7.8 for a price quote, and 7 to 14
search inventory online. Many dealerships report not offering consumers the ability to purchase online; this is good in the sense that it is not possible to complete the entire process over the web. We interpret the answer of dealerships that claim to have this capability as reflecting a website with a streamlined process to bring consumers close to purchase completion. The mean number of years for this capability is 3. We sum these age variables to create a “web capability” summary measure which is reported in the table. The mean webcap is 33.9 with a standard deviation of 14. We assume that years of capability reflects technological sophistication. The table also shows the scores our websites obtain for both purchase ease, as evaluated by the RAs, and information collection, also evaluated by the RAs. The latter measures how well the website collects information about the consumer through online forms. The former reflects whether there is an easy to use button for purchase, and functions such as payment calculator and appointment request. We report some operational characteristics of our dealerships. Average volume at our dealerships is about 1600 cars over a full year, or about 130 per month. The standard deviation is substantial relative to the mean, at 1516 cars per month. The median volume is 1200, which reflects the existence of many small dealerships. (Recall the discussion of franchise laws above.) Dealership average margins on a new car ranges from -1.98% to 12.49% in our sample. with a mean of 4.37%. Used car margins range from 3 to 62% with a mean of 13%. The percentage of cars sold that are used ranges from 0.1 to 1, with a mean of 0.66. We know how much a dealership profits from sales of ancillary services. Finance profits as a percentage of the car’s price average .027, which is approximately $700 on the average car. Given that the average profit margin on a new car is about $1500, it is clear that some dealerships are earning a substantial portion of profits through financing. The financing markup is the average number of basis points a dealership marks up its own loans (over the rate its contract lender will offer the customer). This ranges from zero to 1.6 percentage points, with a mean of .60. There is heterogeneity in the amount of inventory different stores hold; the mean is -89 (recall we normalized by average inventory of the nameplate) and the standard deviation is 171, which reflect the large number of small dealerships. Census data allows us to learn about the tracts and counties in which our dealerships are located. The report percent with a college education and median household income. The average age of car buyers in the country is calculated by averaging the ages reported in all transactions for a particular county. We calculate the number of same-nameplate competitors within 10 and 25 miles, the total number of auto franchises in the country, the average distance 15
from a dealership to car-buyers in the area (whether or not they actually bought at that dealership), and the growth in county population from 1950 to 2000. Our survey yielded many interesting variables. We construct a training variable that runs from zero to 5, according to answers to survey questions. Zero reflects no training for floor salesmen. The variable increases in value as the dealership reports giving salesmen materials to read, or classes to take, with higher values for internal materials and classes as opposed to external materials and classes. The mean training level in the data is 3.5. We can also measure how floor and internet salespeople are compensated. Compensation may be based on (one or more of) a flat salary, a percentage of gross margin, or a performance matrix including customer service. 83% of dealerships use percent gross as one of their metrics but only 51% use it as the only performance metric. Most dealerships track leads fairly closely (2.8 out of 3 for trackdetail and 3.6 out of 4 for track type). Only 36% have a centralized group that tracks customers known as a “business development center” or bdc. This indicates size and sophistication of the sales process. We know how the dealership assigns leads to the floor staff. Assign equal to one indicates that the staff watches for a customer to arrive on the lot and calls out who gets whom. Higher levels of assign indicate equal rotation (2) and assignment based on performance (3). We track whether the dealership has an Internet Service Manager (ism) and find that 62% of them have this position. The salesforce at our dealerships averages 13, though the range runs from 1 to 85. Facility quality is a self-reported estimate of the physical plant of the dealership. Ad and lead budgets vary widely; many dealerships spend nothing, but the larger dealerships spend significant amounts. 6 Empirical Work In many settings in Industrial Organization the researcher expects there will be heterogeneity across firms but is unable to observe the nature of that heterogeneity when it spans both external and internal characteristics of the firm. Table 2 shows correlations between the size of the dealership and many policy choices. The table provides evidence of economies of scale in auto retailing. Large dealerships, measured by volume or salesforce, invest in business development centers, lead tracking software, CRM software, and advertising. Large dealerships have lower gross margins per car. 16
Table 1: Summary Statistics Variable Mean Std. Dev. N volume 1602.508 1516.922 478 dlrnewmargin 4.367 1.762 463 dlrusedmargin 13.212 4.591 462 fimargin 0.028 0.013 478 finmark 0.596 0.258 478 percnew 0.66 0.137 476 invdiff -89.205 171.1 478 pctcol 0.358 0.186 461 avgcage 45.179 2.531 478 income 56406.625 27016.308 461 comp10 0.442 0.782 430 comp25 2.395 2.791 430 countyfran 11.165 13.816 478 disttocust 51.267 17.405 426 popgrow 3.125 2.217 476 trainlvl 3.507 1.466 477 grossonly 0.517 0.5 478 ism 0.617 0.487 478 assign 1.363 0.515 477 bdc 0.365 0.482 477 trkdetail 2.847 0.449 477 tracktype 3.577 0.773 477 facqual 3.597 1.068 477 adbudget 34800.676 65009.705 444 lbudg 3134.738 6665.904 450 fte 12.79 9.303 476 webcap 33.877 13.897 478 webpurch 1.938 0.935 469 webinfo 1.763 1.106 469 17
Table 2: Correlations between dealership features Variables volume fte percnew dlrnew- invdiff trainlvl gross- bdc crm ism trk- facqual webcap web- web- margin only detail purch info volume 1.000 fte 0.378 1.000 (0.000) percnew 0.085 0.046 1.000 (0.065) (0.319) dlrnewmargin -0.122 -0.170 -0.030 1.000 (0.009) (0.000) (0.517) invdiff 0.431 0.118 0.056 -0.115 1.000 (0.000) (0.010) (0.224) (0.013) trainlvl 0.040 0.049 0.059 0.022 0.017 1.000 (0.388) (0.286) (0.198) (0.637) (0.705) grossonly -0.055 -0.088 0.038 0.078 -0.067 -0.121 1.000 18 (0.228) (0.056) (0.404) (0.094) (0.144) (0.008) bdc 0.116 0.150 0.044 -0.035 -0.011 0.097 -0.027 1.000 (0.011) (0.001) (0.343) (0.458) (0.816) (0.034) (0.556) crm 0.105 0.100 0.085 -0.041 0.114 0.088 -0.043 0.065 1.000 (0.021) (0.030) (0.064) (0.379) (0.013) (0.054) (0.353) (0.157) ism 0.118 0.105 0.033 -0.132 0.026 0.134 -0.004 0.066 0.117 1.000 (0.010) (0.021) (0.469) (0.004) (0.573) (0.003) (0.935) (0.149) (0.010) trkdetail 0.093 0.101 0.083 -0.030 0.047 0.115 -0.067 0.113 0.347 -0.018 1.000 (0.041) (0.028) (0.070) (0.521) (0.305) (0.012) (0.142) (0.014) (0.000) (0.697) facqual 0.112 0.047 0.079 -0.004 0.089 0.082 -0.006 0.147 0.040 0.031 0.090 1.000 (0.015) (0.306) (0.087) (0.924) (0.053) (0.072) (0.892) (0.001) (0.383) (0.495) (0.048) webcap 0.106 0.086 0.034 -0.078 0.088 0.103 -0.070 0.029 0.163 0.110 0.037 0.084 1.000 (0.021) (0.062) (0.457) (0.095) (0.055) (0.025) (0.124) (0.522) (0.000) (0.016) (0.425) (0.066) webpurch 0.208 0.110 0.003 -0.109 0.082 -0.053 0.031 0.124 0.075 0.109 0.027 0.050 0.039 1.000 (0.000) (0.017) (0.948) (0.020) (0.076) (0.255) (0.500) (0.007) (0.104) (0.018) (0.561) (0.285) (0.405) webinfo 0.087 0.100 -0.023 0.025 -0.051 -0.031 0.042 -0.001 0.036 -0.031 0.140 0.091 -0.032 -0.051 1.000 (0.060) (0.031) (0.622) (0.597) (0.273) (0.505) (0.368) (0.977) (0.440) (0.510) (0.002) (0.049) (0.483) (0.267)
Consistent with Ichniowski, Shaw, and Prennushi (1997), Milgrom and Roberts (1980) etc. we see correlation of “good”, or what we might call consumer-oriented, organizational and human resource practices within a dealership. For example, a high score on our training variable is correlated with performance-based compensation, keeping track of leads, and having an internet sales manager. One concern we want to address is the possibility that many of the dealership’s policy choices are driven by the nameplate, both because of constraints from the manufacturer and because of the positioning of the brand in product and demographic space. One might wonder if these correlations remain once nameplate fixed effects are included. Conditional correlations, in the form of regressions of volume on these various characteristics controlling for nameplate fixed effects, demonstrate similar patterns compared to the unconditional correlations. We do not report these regressions as they are numerous and do not materially add to the story. There are two important differences to report. Training and performance compensation are no longer positively correlated with dealership size once nameplate is conditioned on. We run some descriptive regressions explaining the pattern in the new car margin. These are also unreported for space considerations and because they can be easily summarized. In these regressions we put a dealer’s new car margin on the left hand side, control for nameplate fixed effects and either include a competition measure or a county fixed effect. The results generally show that the conditional correlation between the margin on new cars and a dealer’s volume is negative. Additionally, the correlation between new car margin and a compensation policy based on margin is positive. This latter result is reassuring to see because the information on the compensation policy comes from our survey, while the new car margin data come from our transaction-level dataset. The positive relationship gives us additional confidence in our survey. We would like to understand whether these policy choices appear in clusters among our dealerships. Because we have product market strategy variables, variables measuring internal policies, and variables measuring technology adoption, we undertake a factor analysis to orga- nize the data. Table 3 reports the results of a principle components factor analysis, retaining the 5 factors with the largest eigenvalues. What we see is that the most distinguishable strat- egy is “high volume-low margin.” These dealers are very large and have low profit margins on new cars. They sell new cars disproportionately, carry high inventory, track leads vigorously, and have good web technology. These competitive dealerships face many competitors and are located close to potential customers. 19
Table 3: Correlations between factors and variables Variable Competitive Sheltered Finance/used Niche Stereotypical volume 0.6314 0.3212 0.0803 -0.1862 0.3573 fte 0.517 0.3112 -0.0545 -0.0181 0.3678 invdiff 0.5797 0.0991 0.0513 -0.3136 0.0866 dlrnewmargin -0.3311 0.039 0.1339 0.429 0.3553 dlrusedmargin -0.3978 0.1148 0.0573 0.1765 0.6343 fimargin 0.1557 0.367 0.7037 0.1623 -0.021 finmark 0.088 -0.2403 0.6556 0.1483 -0.2439 percnew 0.3008 -0.1513 -0.3845 0.1312 -0.1131 comp10 0.3562 -0.3869 0.2623 0.0337 0.3491 disttocust -0.4632 0.7435 0.0107 -0.1475 -0.1332 avgdisttocomp -0.4525 0.7499 0.0063 -0.2068 -0.1077 franppop2000 -0.4846 0.1531 0.2342 0.0055 0.1298 assign -0.1151 0.0948 -0.3214 -0.129 0.0466 grossonly -0.0817 -0.0794 0.2306 -0.0091 -0.1322 trainlvl 0.1106 0.2493 -0.1541 0.3749 -0.0582 trkdetail 0.3378 0.2821 -0.0474 0.543 -0.2422 tracktype 0.3687 0.2253 -0.1238 0.4215 -0.2457 bdc 0.2908 0.1802 -0.1234 0.1818 0.1326 ism 0.3211 0.1635 0.0121 -0.2181 -0.3376 facqual 0.1949 0.3137 -0.3012 -0.0206 0.1297 webcap 0.2399 0.0644 -0.3 -0.0241 0.0391 webpurch 0.2798 0.1573 0.1924 -0.2988 0.1428 webinfo 0.0183 0.1928 -0.052 0.3851 0.1061 avgdage -0.4047 -0.2773 -0.3991 0.0506 0.1047 20
The strategy listed next we call “sheltered” for lack of a better term because these dealers face low intrabrand competition, and are further from customers. They are also fairly large but their margins on new cars are higher and they earn money from financing and insurance. The internal policies of these dealerships show that they invest in training, lead tracking, physical facilities, and websites. We infer that these investments allow these dealerships to provide high-quality service to customers. The third strategy is one based on profits from financing, profits from interest rate markups, and profits from sales of used cars. These dealerships have incentive pay based on gross margins, as one might expect. They also have poor internal policies for lead assignment, training, lead tracking, and poor facilities. The fourth type of strategy we call a niche strategy. Dealers in this group have high margins and low volume and low levels of inventory. However, they make some money from financing and they invest in their dealerships with high training levels and tracking of customer leads. They appear not to systematically invest in other fixed attributes (website, physical facilities) perhaps due to lack of economies of scale. The last strategy we describe is a stereotypical one of high margins and low service. These dealerships are relatively large and have high margins on new cars and very high margins on used cars despite intrabrand competition. They do not track leads or have an internet department. They have average websites and physical facilities. Customer age is positively correlated with this factor, and negatively correlated with the first three factors. We suspect that the correlation with age arises because older consumers tend not to use the Internet as much and so do not engage in as much price search. Therefore, these high margin dealerships will have a disproportionate share of “uninformed” consumers among their customers. Table 4 lists the correlations between nameplates and our five strategies for the interest of readers who might also be car buyers. Clearly, the type of cars being sold will determine cus- tomer demographics to some extent, as will franchise policies concerning geographic territories, IT use, promotions, etc. However, one can see that most of these nameplates exhibit multiple strategies within the dataset because they are either correlated with more than one franchise strategy, or are uncorrelated with most of them. 6.1 Instruments We need instruments that impact a dealership’s use of IT but yet are not part of the cluster of choices of dealership management. The characteristics of the local population and competitive 21
Table 4: Correlations between nameplates and our five strategies Competitive Sheltered Finance/used Niche Stereotypical Acura ++ – - – Audi - BMW ++ - Buick – Cadillac – Chevrolet - + ++ Chrysler ++ – Dodge ++ – – Ford – ++ ++ GMC – Honda ++ + - ++ Hyundai ++ Infiniti - Jaguar – Jeep ++ – Kia + ++ Lincoln – ++ Mercury – + Mitsubishi – – Nissan ++ ++ + Pontiac – Porsche – Saturn – ++ Scion ++ ++ ++ Toyota ++ ++ VW – Volvo – - LandRover – Mercedes ++ environment determine many aspects of the dealership including IT but clearly are not deter- mined by dealer policies. Of course, dealer location will be optimally chosen at the time the dealership is launched. However, many dealerships were located decades ago when demograph- ics were quite different. Additionally, as one can see from Table 4, all nameplates show variety in the strategies they pursue in terms of pricing and selling cost. Lastly, these dealerships were almost all established before the internet became a major issue in auto retailing. We include percent college and income in the census block in our group of instruments. Our instruments at the county level are average age of the car buyers in the county, population growth since the 1950’s, and total number of auto franchises in the county. The former drives web adoption, the latter are determinants of competition and scale. Other county level variables we use are from the survey of business use of IT detailed in (Forman, Goldfarb, and Greenstein 2009). We use two of their measures: participation, which measures fairly basic internet use by businesses in the county, and enhancement, which indicates more sophisticated use of the 22
internet by businesses in the country. The dealerships in our sample are in counties that have a mean of .83 for participation and .11 for enhancement. We expect that building and maintaining a website in geographic areas with deep and well-developed labor markets of IT professionals and service organizations is lower cost. Thus we expect higher measures of enhancement to proxy for lower costs of creating and maintaining a website. How much competition a dealership faces is also pre-determined relative to web technology adoption and transaction volume, particularly for the US nameplates. We include the number of competitors of the same nameplate within 10 miles, whether the name plate is of US origin, and interactions between number of competitors and population growth and US nameplate and population growth. The component of market structure that is driven by old franchise territories (proxied by population growth) is arguably exogenous to the dealer‘s behavior and policy choices. Additionally, for each dealer, we also calculate the average distance to (all) auto consumers in our dataset and use this an an instrument. These variables vary at the dealership level. We report first stage regressions in Table 5. The explanatory power of our instruments is fairly low, but a number of our instruments attain p-values of below .1. The three columns of Table 6 show an ordered probit regression for the web purchase measure, a poisson regression for web capability, and a probit for Internet Sales Manager. 23
Table 5: First stage of Instrumental variables regressions Ordered Probit Poisson Probit (1) (2) (3) VARIABLES webpurch webcap ism pctcol -0.534 0.171** 0.336 (0.480) (0.073) (0.574) income 5.30E-07 5.00E-07 6.67e-07 (0.000) (0.000) (3.98e-06) comp10 0.361** -0.0375* -0.103 (0.147) (0.022) (0.167) usname 0.0878 -0.142** -0.00167 (0.416) (0.066) (0.491) enhancement 1.008 0.238 0.803 (1.695) (0.263) (2.008) popgrow 0.0395 0.0113* 0.0262 (0.045) (0.007) (0.0581) avgcage 0.0348 0.0129*** -0.0461 (0.028) (0.004) (0.0333) growcomp10 -0.0587 0.00564 -0.0174 (0.049) (0.007) (0.0571) growcomp25 0.00786 -0.00290** 0.000816 (0.008) (0.001) (0.00933) growus -0.0142 -0.0114 0.0392 (0.059) (0.009) (0.0719) pop20 8.65E-08 1.56E-10 -4.86e-07** (0.000) (0.000) (2.41e-07) participation 0.372 -0.131 0.492 (0.890) (0.137) (1.028) distus -0.00455 0.00260** -0.00681 (0.008) (0.001) (0.00949) disttocust 0.00749 -0.00437*** -0.00131 (0.007) (0.001) (0.00826) countyfran -0.00267 0.00215** 0.0151** (0.006) (0.001) (0.00747) Constant -0.245 1.939 Observations 365 370 370 Pseudo R-sq .0181 .0219 .0553 Standard errors in parentheses - Second Stage reported in Table 6 *** p
Table 6: instrumental variables regressions of dealer volume and margin on measures of IT (1) (2) (3) (4) (5) (6) (7) (8) (9) VARIABLES volume ln(volume) margin volume ln(volume) margin volume ln(volume) margin webpurch 1346*** 0.723*** -0.357 (421.8) (0.229) (0.416) webcap -2.289 -0.00310 -0.116*** (23.57) (0.0128) (0.0427) ism 2152*** 1.148*** -2.118*** (632.2) (0.338) (0.760) Constant -968.0 5.749*** 5.018*** 1718** 7.257*** 8.294*** 279.1 6.425*** 5.657*** (822.4) (0.447) (0.811) (809.2) (0.440) (1.461) (408.0) (0.218) (0.487) 25 Observations 365 365 356 370 370 361 370 370 361 Standard errors in parentheses First Stage reported in Table 5 *** p
6.2 Effect of IT We carry out simple correlations and instrumental variables regressions corresponding to our three hypotheses in this section. We analyze three measures of dealer IT policies: two measures of website capability and one measure of organization and website capability, the existence of the ISM. The first web measure was created by examination of the website attributes by RAs, the second by the survey of Sales Managers. We ask if either measure is significantly predictive of volume or new car margins at the dealership level. The IV regression results are reported in Table 6. The first columns show the results from an IV regression of website purchase capability. The estimated coefficient on volume, whether in logs or levels, is positive and statistically significant. The coefficient estimate of .72 implies that a one-standard deviation in crease in web purchase rating results in an increase in quantity of cars sold by 60%. In levels, the estimated coefficient of 1300 implies a one standard deviation increase in the web purchase measure increases cars sold by about 1200 per year (mean 1600). The IV regression of web capability on quantity of cars sold shows no significant effect at all. Turning to the instrumented regressions of new car margins, we find the opposite. The web purchase measure has no impact on margins. However, the web capability measure has an estimated coefficient of -.12. If web capability increased one standard deviation (14), then new car margins would be predicted to decline by 1.5 percentage points. This magnitude seems reasonable given that new car margins range from -2 to +12 in the data, with a mean of 4.4 and a standard deviation of 1.8. find that both measures predict lower margins. We return to the question of whether a dealership that adopts both the website and the accompanying organizational change has a stronger impact on either volume or margin. Our measure of both together is the dummy variable “Internet Sales Manager” or ISM. In the remaining columns of table 6, we use the same instruments to predict the choice of having an Internet Sales Manager. We find that the relationship between ISM and volume is strongly positive and significant, with an estimated coefficient of 1.15. This coefficient suggests that adopting an ISM increases sales by 15%, which is a smaller effect than the one we found using instrumented web purchase ratings. The estimated coefficient in the levels regression (2100) is more similar to the previous result with web purchase. The final column shows the estimated impact on dealer new margin from adopting an ISM. It is negative and significant at -2.11, which is similar in size to the coefficient on web capability. Our final hypothesis is that the dealerships that choose to adopt IT will have higher levels 26
You can also read