<<

Robert Heeter Econ 499H Senior Honors Thesis May 1, 2008

The Value of Automotive Factors between Segments

The goal of this thesis be to research, analyze, and present the

influence of attributes on their price in the market. A number of studies exist

that involve hedonic pricing of automobiles in industrial organization studies,

from Berry, S., J. Levinsohn, and A. Pakes to Triplett or the original Court, A. but

I have not seen a study yet focusing on the comparisons of coefficients of the

value derived by attributes across different segments. In some cases, the

hedonic studies are done to compare the increase in automotive prices to price

indexes or to determine if colluding had occurred. Instead of determining the

change in quality and price over years, I will focus on the change of coefficients

for effects of attributes on price across market segments.

BACKGROUND INFORMATION

I have for a long while have been intrigued by , since middle school, and have been familiar with automotive terms and the industry atmosphere.

However, in order to prepare some unfamiliar readers, I ought to go into a little background information on the subject. In my procedure area, I have included brief descriptions of the segments

and variables, but it is important to also understand the

basics. There are several EPA classes for cars, ranging from minicompact to

large, with subcompact, compact, and midsize in-between. These are grouped by

size generally and would not have leant to very interesting regressions without

dummies for all their standard features, as much of the past studies relied on

physical dimensions and performance/feature specification and segmenting by

EPA class would lose half of that. The government on fueleconomy.gov also

breaks down classes by market class, which includes sporty/sports cars and

luxury/upscale. This sort of segmenting would help group cars of similar quality

and features while describing their variance using physical dimensions as well as

other specifications. For objectively defining performance, I chose to use a

calculation of hp/ton and then chose to use my own definition of luxury in order

to include the luxury wagons that the fueleconomy.gov separated as its own

category with other passenger car station wagons.

PAST RESEARCH

Before I begin discussing my approach, let me review the studies I have

read and have taken into consideration. In the study of the 1955 Price War,

Bresnahan [1987] modeled that consumers minimize the price of a good that will satisfy their desire for quality of a certain product, as Pj – vxj. Applying this to vehicular specifications, a consumer seeking a group of features in a car would choose the car that fulfilled his needs for the least amount. If there were multiple products of varying quality and different prices, then at different quality levels, a consumer would select other alternatives. He would be indifferent if vh

= (Pi – Ph / Xi – Xh) as paying an unit more for a unit more of quality makes him

just as satisfied. If v increases, then the consumer is more fond of an expensive

car. The slope is Ph – Vxh. The cost to produce would be the fixed cost plus

marginal cost to increase quality. The max profit comes from the decision to stick

to the minimum of what will make the costumer satisfied. The focus of this research was the more competitive segments of smaller cars where multiple makes existed. Even today, many people have a horse in the subcompact or compact categories where larger may be skewed. His variables included weight, length, horsepower, number of cylinders, and a as well as make dummy variables. These will help guide my selection as well as a preliminary regression to see how modern cars are differentiated. I am as well using

Automotive News for my data.

Triplett’s study [1969] looked into whether or not the price index was upwardly biased by not accounting for increases in quality. If the cars price rose and xh did as well by the same amount (borrowing terms from Bresnahan), then the real price of the car and its quality remained neutral. He based quality on factors of horsepower, width, length, and then dummy variables for V8, hardtop, automatic, power steering/brakes, and compact. Comparing adjacent years, any residual not accounted for by these variables was chalked up to the years and real price increases. In the years of my study, the majority of cars now have steering/brakes standard, but I will have a variable to help describe differences between a traditional car and a hatchback which are typically associated with cheaper models, but perhaps controlling other factors it may not.

In Boyle and Hogarty’s research [1975] the coefficients can represent cost of producing characteristics if industry is perfectly competitive and have the same processes. They relied on indexes based on MSRP which I have done. (It would have been great to get the type of data Agarwal and Ratchford found along with mystery shopping to get real dealership pricing but for the simplicity,

I have assumed MSRP are representative enough). The principal characteristics of quality desired by car buyers are comfort, durability, economy, maneuverability, performance, safety, and style. All of the statistics were significant except for economy. (But this could have been because when weight, cylinders, and horsepower are fixed it may have collinearity). B&H based their regression upon front room, horsepower, and weight. The Information

Disclosure Act helped pricing become more transparent to consumers and make assumptions of complete information more plausible (although not as much so as today with the internet and comprehensive guides available).

Agarwal and Ratchford [1980] expanded on Rosen’s model [1974] which look into price elasticities of price attributes with attention paid to the varying preferences of the consumer and their demographics, life cycles, etc. Instead of simply finding evidence that the price one is willing to pay is sensitive to attributes, this model built upon it by including whether the car would be used for long distance travel for example. And also add dummy variables in order to capture if manufacturers had preferences, or were capable of providing X attribute easily or cheaply compared to the competition.

Both the consumer and producer were assumed to be making decisions in order to build the market interactions. A&R had objectives of physical specifications and Consumer Reports as well as 27 respondents which answered on 34 constraint ratings that gave perceptual insight. Some attributes included were displacement, luggage capacity, rear leg room, 45-65 passing speed, handling, ride from Consumer Reports ratings, and 2dr or 4dr. But there were issues present with how to rate styling objectively, and the affect of coefficients may have been overstated because of explaining other elements remaining in u but correlated, such as in my first regression which had a great negative coefficient for number of cylinders with a given horsepower rating which could have brought in influence from an unaccounted for fuel mileage. When factoring in curb weight and several characteristics of the engine, fuel mileage should be somewhat included in the model.

Agarwal and Ratchford’s econometric study also included consumers and how many storerooms they visited, the number of cars looked at, the type of driving they were planning, how often they were getting a new car, their amount of do-it-yourself work, income level, education, occupation, family size, number of cars, life-cycle stage, and home/garage ownership. For the producers and dealership activity, a dummy was used. This model assumed consumer

had complete information, there were no economies of scale in models or trims, and that the industry was perfectly competitive. The hedonic equation was the natural log of price = .0349 displacement + .1492 handling + .2391 ride + .2664

(1/passing time) + .0334 luggage capacity + .2674 rear leg room. R squared was .684. When there was only a primary car or one for long distance travel being shopped for, people preferred larger, more comfortable cars but when it was a short trip secondary car, it was smaller. They also calculated the resulting utility taking the hedonic price of the car and subtracting it from the actual.

Griliches and Ohta [1986] researched the adjustment of buying preferences after the increase in gas prices. The hypothesis was that Americans shifted towards smaller cars and drove their prices up as larger cars decreased in numbers from the producers, and mpg was a more attractive attribute while engine size and weight were undesired. This model assumed that automotive prices are a function of the characteristics of it. Utility from the automobile was a combination of its speed, room, comfort, and handling. Gas mileage was not necessarily a part of utility in this discussion as it was only part of the budget constraint. The regression used displacement, cylinder number, weight, x width, less than four doors or more than four, Automatic, power steering, and A/C. In my primary regression, a dummy variable for hatchback is sometimes significant, and probably should not be grouped with in which during this era may have been much more expensive while hatches are economical. Griliches and Ohta described speed with the engine description, comfort with the size and weight, doors and A/C for ride quality, and Automatic and PS for ease of drive. Although these seem like great extrapolations to be making, Taylor in 1994 simply focused horsepower, luggage room, and gas mileage to represent interests.

Two final studies I have looked out were Bajic [1988] and Bordley [1993].

Bajic includes helpful suggestions of excluding small-production expensive luxury cars and using one representative model when multiple trims are present. Both these steps may help to make the regression unbiased towards outliers or several similar cars. I however, decided to leave in the outliers as they can only weigh in on the particular segment they are a part of. If a or was grouped with an , it may not have made much sense, but controlling for a , it seems fine. He also included the trouble index of

Consumer Reports which would help tell about the durability index, but may be more objective than wanted. Variables included wheelbase, weight, height, width, and then a dummy each for , years, and rwd. I chose to include a awd dummy as well in its growth of popularity for safety and handling. Bordley’s main contribution to my evaluation of my study is to look into dummies for different segments. Bresnahan noticed differences between segments and perhaps there will be differing co-efficients across classes.

One great influence for some variables used has be to BLP. The research may have dealt with statistics that made it seem as if some consumers dislike fuel economy while some dislike higher values of HP/weight. Instead of

measuring the mean marginal utility of mile per dollar or HP/weight, I plan try to

divide cars into different category as consumers would consider them and

provide a significantly statistical and economic relation of such utilities within the

market segments. I cannot easily predict if I will be able to show solid figures

before running with the data but that is a prime example of my goal in this

thesis: To show how factors hold different value in different segments. Additional

data could in future studies aid in analysis that determines the effect of reliability

or safety across different categories from economy or luxury. Specifically

identifying the return of another star in crash test ratings between an economy

car and a high end premium may be very helpful to automotive manufacturers as

well as consumers. Each consumer may have different utilities for attributes but

an analysis of the “best value” in their desirable attributes may highlight vehicles

they did not see as preferred before. I do not have the resources or timeline to

be able to account for buyer profiles like Agarwal and Ratchford but hedonic

value can still be found and coefficients across segments can be found which

may contain in them different driver personalities (small economical compact vs.

large luxury model).

The focus of this research is to extend from previous studies which broke cars down into their attributes, and going further to study the difference in co-efficients across market segments. The goal is to show different segments yield varying returns to prices with changes to the automobile factors. PROCEDURE AND DATA

I used data primarily from Automotive News for specifications and prices from 2005, 2006, and 2007. In addition, I found inflation data between for 2005 and 2006 and then also MSRP pricing when missing from Automotive News from

MSN Auto. Automotive News included many specifications for each automobile which I analyzed and picked up the appropriate and significant factors from.

Dimensions, engine specs, and features like occupants and fuel tank size were included to break a car down into its factors. Unfortunately, including standard equipment like sunroof, airbags, navigation, or power items just was not feasible, but in the end seemed to filter out. With a greater pool of years, there could very well be significant differences in equipment level, especially with ABS and airbags and power options through the nineties.

I chose to only work with car data, and not , trucks, or SUVs.

These are very different in specification from passenger cars, so they could likely not be included in one general regression, and there was plenty of variety in segmenting with just cars. There are fewer segments in dealing with these three other groups, with few “sporty” trucks and SUVs. It would be interesting to see a comparison of price changes within SUVs, especially with gas mileage, towing capacity, and standard features included, but that is for another study at another time. With three years of automotive data and on average just over 230 cars per year, I was given 767 observations.

Segments

Performance – defined in general as a car with more than 150 hp/ton. This gave a group of cars that for the most part fit into the average person’s classification of “performance”

This yielded 232 observations. These were not exclusive of luxury.

Luxury – defined subjectively with mostly adhering to traditionally classified luxury brands such as , , , , BMW, , ,

Jaguar, , Lincoln, Maybach, Mercedes, , Saab, Volvo, along with VW

Phaeton.

This yielded 400 observations. These were not exclusive of performance.

Economy – defined as not luxury and having less than 106 hp/ton.

This yielded 216 observations.

Family – the unspecified remainder after Performance, Luxury, and Economy

cars taken out is general family cars. These are defined as cars with between

106 hp/ton and 150 hp/ton, while not being a luxury car.

This yielded 123 observations.

Summary of segment means on following page. Take care to notice trends in

nearly every attribute.

Variables

Dependent:

Price2007dollars is the price adjusted to 2007 dollars taking into consideration inflation.

Logprice2007 is the log of Price2007dollars.

Independent: (*=dummy variable)

Doors: Number of doors on the car. 2 doors for a , several which are high-end luxury models, 3 doors for a hatchback, 4 doors for a , and 5 doors for a wagon.

Interaction terms perfdoors, luxdoor, and econdoors for respective performance*doors, luxury*doors, economy*doors.

Hatchback*: If car is a hatchback or not. (meaning rear window opens up with ), with respective perf, lux, econ dummy variables interacted with variable hatchback.

Note: There are no performance so this term is dropped from regressions.

Convertible*: If car has removable top and rear window, with interaction for segments.

Wagon*: If car is stretched with an integrated rear hatch/window, with interaction terms.

Wheelbasein: Car’s wheel base (length between the centerpoint of the front and rear tires, in inches). Has interaction terms with each segment perf, lux, and econ.

Overalwidthin: Car’s width in inches, with interaction terms for segments.

Overallengthin: Car’s length in inches, with interaction terms for segments. Given a specific wheelbase, length in addition is “overhang” and besides space for engines and luggage, traditionally is not highly vaued.

Curb weight: Weight of car in pounds, with interaction terms for segment. Without standard feature dummy variables, in some segments this may denote luxury features that increase weight

Rear track: Measurement of span between rear tires, with interaction terms. Halfnocyl: ½ the number of cylinders, so counts pairs of cylinders, with interaction terms perfcyl for performance*halfnocycl, luxcyl for luxury*halfnocyl, etc.

Bore: Bore in inches of engine, or the width of the cylinder bore, with interaction terms. Bore is traditionally associated with horsepower rather than torque. Boring out an engine relates much more to high-end horsepower than overall increase in torque.

Stroke: A measure of distance travel in engine, in inches, with interaction terms. Stroke is typically associated with torque rather than horsepower. Stroking an engine will make it have more low-end torque in response usually.

Displacement: Overall size of engine in liters, with interaction terms. A calculation based on bore, stroke, and number of cylinders.

Hp: Horsepower of base engine for specific automobile, with interaction terms.

Ftlbs: Torque measured in foot-pounds, with interaction terms for segment.

Fueltanksize: Number of gallons the fuel tank holds, with interaction terms.

Gears: Number of gears car has, with interaction terms. CVT, or continuously variable transmissions were entered as 99 gears, which may be apparent in the economy segment.

Auto*: If standard equipped with automatic, and segment dummy interaction terms. Base models are usually associated with manuals which are less expensive, with automatics as optional.

AWD*: If base model is equipped with all-wheel-drive, and interaction terms. Usually seen as safe for all-season travel.

RWD*: If base model is equipped with rear-wheel-drive, and interaction terms. This is usually seen as better handling or more fun than FWD or AWD configurations.

The remaining cars would then be FWD, or front-wheel-drive.

Numberofoccupants: Number of occupants the car is made to hold, with interaction term. This variable may be attractive for family cars, but it will likely be associated negatively in high-end luxury or performance cars that do not need to make room for others. Procedure

In my exploratory regression, I included 26 variables and only 14 were

significantly statistically. Number of doors, hatchback and dummies,

height, weight, rear track, # of cylinders, number of valves, bore, stroke,

displacement, horsepower, torque, fuel tank size, and number of occupants all

were significant at the 5% level. Some of the additional information I thought I

should have included turned out to be non-influential such as wagon (often

priced the same, even though length, weight, luggage capacity might have

thrown a curve ball), wheelbase, width, gears, and drive types. I do plan on

attempting a wheelbase x width variable like Griliches. [This turned out not to be

needed and hurt interpretation so it was later removed.] R-squared value was

.7865. When I cut down the regression to only these significant variables, doors was no longer but many probability values dropped to 0.000 or 0.002 otherwise.

I do not plan on including a gas mileage variable although it may already be

explained in relation to number of cylinders, horsepower, and weight. Additional

dummies could have been added for traction control and stability control but

most likely they are going to be included in the expense of high-end luxury and risk-adverse wealthy individuals. Other factors could head/leg room, luggage capacity, and turning circle but these are likely correlated already with size of vehicle and its specifications. Given more time for data discover and entry, the inclusion of variables for airbags and safety ratings would be my goal, as these could be important, but then again, could be controlled by competitive forces. When it came time to return and run more final regressions, I began with

running OLS and using all the variables including interaction terms of the dummy

variables. The goal was to develop and explore the model for all cars in general

and establish the significant variables. I discovered that valves per cylinder has

grown in the past decade to be nearly standard as 4 valves per cylinder and the

only variance was being explained by Mercedes-Benz using 3 on some models

which skewed the model as a whole. If the pooled data went back longer, this would be a useful variable but not for these three years. Overall height turned out to be an insignificant variable, as well as hybrid as there were not enough to draw any conclusions – or they did not alter the price that much given their other factors. Compression ratio also was insignificant, whereas other engine factors could be with some interactions.

The next step after removing the unneeded variables was to determine my optimal form of the independent variable. The price in 2007 dollars could be level so that an increase of X variable would yield a increase of BX dollars, or I could use log price in order to have the increase of BX% in price.

I ran using 79 variables, all the useful terms and their interactions with dummies, and tested these models for heteroskedasticity and looked at the

residuals in histograms. It was clear that log price would be better for this study.

Level hettest:

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Ho: Constant variance Variables: doors perfdoors luxdoor econdoors hatchback econhatchback luxhatch convertible perfvert luxvert econvert wagon perfwag luxwagon econwagon wheelbasein perfWB luxWB econWH overalwidthin perfwide luxwide ecowide curbweight perfCW luxCW ecoCW reartrack perftrack luxtrack ecotrack halfnocyl perfcyl luxcyl ecocyl bore perfbore luxbore ecobore stroke perfstroke luxstroke ecostroke displacement perfdisplace luxdisplace ecodisplace hp perfhp luxhp econhp ftlbs perftq luxtq econtq fueltanksize perffuel luxfuel ecofuel gears perfgears luxgears ecogears auto perfauto luxauto ecoauto awd perfawd luxawd ecoawd rwd perfrwd luxrwd ecorwd numberoccupants perfoccupant luxoccupant econoccupant

F( 79 , 677) = 4.84 Prob > F = 0.0000 3 4.0e-05 3.0e-05 2 Density Density 2.0e-05 1 1.0e-05 0 0 -100000 0 100000 200000 -.5 0 .5 1 1.5 2 Residuals Residuals

Level Log

Log hettest:

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Ho: Constant variance Variables: doors perfdoors luxdoor econdoors hatchback econhatchback luxhatch convertible perfvert luxvert econvert wagon perfwag luxwagon econwagon wheelbasein perfWB luxWB econWH overalwidthin perfwide luxwide ecowide curbweight perfCW luxCW ecoCW reartrack perftrack luxtrack ecotrack halfnocyl perfcyl luxcyl ecocyl bore perfbore luxbore ecobore stroke perfstroke luxstroke ecostroke displacement perfdisplace luxdisplace ecodisplace hp perfhp luxhp econhp ftlbs perftq luxtq econtq fueltanksize perffuel luxfuel ecofuel gears perfgears luxgears ecogears auto perfauto luxauto ecoauto awd perfawd luxawd ecoawd rwd perfrwd luxrwd ecorwd numberoccupants perfoccupant luxoccupant econoccupant

F( 79 , 677) = 0.88 Prob > F = 0.7568

The residuals were much more normal when using log price and there

was not enough evidence to reject the null hypothesis of constant variance when

using log. The complete regression of all interaction terms is as follows: Source SS df MS Number of obs = 757 F( 79, 677) = 134.81 Model 414.945523 79 5.25247498 Prob > F = 0.0000 Residual 26.3776589 677 .038962569 R-squared = 0.9402 Adj R-squared = 0.9333 Total 441.323182 756 .583760823 Root MSE = .19739

logprice2007 Coef. Std. Err. t P>|t| [95% Conf. Interval]

doors -.0106991 .0269796 -0.40 0.692 -.0636728 .0422746 perfdoors .0255309 .0345835 0.74 0.461 -.0423728 .0934346 luxdoor -.0881062 .0369934 -2.38 0.018 -.1607418 -.0154705 econdoors -.0090418 .0323937 -0.28 0.780 -.072646 .0545625 hatchback .0439651 .0673253 0.65 0.514 -.0882264 .1761567 econhatchb~k .0408383 .0780286 0.52 0.601 -.1123688 .1940454 luxhatch .0093955 .1081421 0.09 0.931 -.2029388 .2217297 convertible .030005 .0814349 0.37 0.713 -.1298903 .1899004 perfvert -.0549278 .0727729 -0.75 0.451 -.1978155 .0879599 luxvert .0453909 .0839935 0.54 0.589 -.1195282 .2103101 econvert -.0099106 .1068943 -0.09 0.926 -.2197948 .1999737 wagon .0902213 .0680338 1.33 0.185 -.0433614 .2238039 perfwag -.0644359 .0985197 -0.65 0.513 -.2578768 .1290051 luxwagon .0651338 .0813917 0.80 0.424 -.0946767 .2249442 econwagon .0837235 .0858865 0.97 0.330 -.0849124 .2523595 wheelbasein -.0065325 .0070768 -0.92 0.356 -.0204276 .0073626 perfWB -.0035046 .0061822 -0.57 0.571 -.0156432 .008634 luxWB .0145119 .0087716 1.65 0.099 -.002711 .0317347 econWH .0044327 .0073205 0.61 0.545 -.0099409 .0188064 overalwidt~n .0388259 .0189037 2.05 0.040 .0017089 .075943 perfwide .0211283 .0090901 2.32 0.020 .0032802 .0389764 luxwide -.0296078 .0199955 -1.48 0.139 -.0688684 .0096528 ecowide -.0328445 .0196093 -1.67 0.094 -.0713469 .0056578 curbweight -.0000399 .0001203 -0.33 0.740 -.0002762 .0001963 perfCW -.0000708 .000086 -0.82 0.411 -.0002396 .0000981 luxCW .0001253 .0001192 1.05 0.294 -.0001088 .0003594 ecoCW .0003229 .0001719 1.88 0.061 -.0000146 .0006604 reartrack -.0103234 .023242 -0.44 0.657 -.0559584 .0353117 perftrack -.007367 .018464 -0.40 0.690 -.0436206 .0288866 luxtrack -.0073411 .026998 -0.27 0.786 -.060351 .0456687 ecotrack .0226595 .0280933 0.81 0.420 -.0325011 .07782 halfnocyl .1310686 .127482 1.03 0.304 -.1192389 .3813762 perfcyl .0369352 .090533 0.41 0.683 -.140824 .2146944 luxcyl -.2758806 .127658 -2.16 0.031 -.5265338 -.0252274 ecocyl -.2934953 .183307 -1.60 0.110 -.6534139 .0664233 bore -.3247704 .1929408 -1.68 0.093 -.7036047 .0540639 perfbore .0721147 .1478782 0.49 0.626 -.2182403 .3624696 luxbore .1524844 .1902705 0.80 0.423 -.2211068 .5260755 ecobore .1774672 .2361525 0.75 0.453 -.2862121 .6411464 stroke -.0320964 .1168253 -0.27 0.784 -.2614799 .197287 perfstroke -.286757 .1239992 -2.31 0.021 -.5302262 -.0432878 luxstroke -.0313975 .137404 -0.23 0.819 -.3011866 .2383917 ecostroke -.1665734 .1433904 -1.16 0.246 -.4481167 .11497 displacement -.0661324 .112915 -0.59 0.558 -.287838 .1555733 perfdisplace .0241318 .0795213 0.30 0.762 -.1320062 .1802698 luxdisplace .2107074 .1018121 2.07 0.039 .010802 .4106127 ecodisplace .0981208 .1741687 0.56 0.573 -.2438551 .4400966 hp .0053981 .0010401 5.19 0.000 .0033559 .0074404 perfhp .0005126 .000722 0.71 0.478 -.0009049 .0019302 luxhp -.0038451 .000986 -3.90 0.000 -.0057811 -.0019091 econhp -.0073321 .0018632 -3.94 0.000 -.0109905 -.0036738 ftlbs -.0023846 .0012191 -1.96 0.051 -.0047783 9.14e-06 perftq -.000304 .0005996 -0.51 0.612 -.0014813 .0008733 luxtq .003653 .0012425 2.94 0.003 .0012134 .0060925 econtq .0040611 .0020026 2.03 0.043 .0001291 .0079932 fueltanksize .011203 .0154368 0.73 0.468 -.0191067 .0415128 perffuel .0251294 .0131474 1.91 0.056 -.0006852 .0509441 luxfuel .0444217 .0166721 2.66 0.008 .0116865 .0771569 ecofuel .0370545 .0228033 1.62 0.105 -.0077191 .0818281 gears .0017818 .0022368 0.80 0.426 -.0026101 .0061738 perfgears -.0275985 .0249108 -1.11 0.268 -.0765103 .0213133 luxgears -.0010633 .0024634 -0.43 0.666 -.0059002 .0037735 ecogears .0016502 .0026914 0.61 0.540 -.0036344 .0069347 auto .1286875 .0663856 1.94 0.053 -.0016589 .2590339 perfauto -.2529409 .052772 -4.79 0.000 -.3565573 -.1493245 luxauto -.1703201 .0704186 -2.42 0.016 -.3085852 -.0320551 ecoauto -.024852 .0809714 -0.31 0.759 -.1838372 .1341332 awd .2523387 .0973328 2.59 0.010 .0612282 .4434492 perfawd .052952 .1034301 0.51 0.609 -.1501304 .2560344 luxawd -.1281383 .1067091 -1.20 0.230 -.3376589 .0813824 ecoawd -.1557614 .1512275 -1.03 0.303 -.4526926 .1411698 rwd .0088715 .0692954 0.13 0.898 -.1271882 .1449313 perfrwd -.0542079 .0894755 -0.61 0.545 -.2298907 .121475 luxrwd .1476983 .0808337 1.83 0.068 -.0110166 .3064133 ecorwd .1136291 .1290607 0.88 0.379 -.1397782 .3670364 numberoccu~s -.0712189 .0380245 -1.87 0.062 -.1458789 .0034412 perfoccupant -.0652261 .0325408 -2.00 0.045 -.1291192 -.001333 luxoccupant .0456422 .0428828 1.06 0.288 -.038557 .1298414 econoccupant -.0378204 .0519729 -0.73 0.467 -.1398679 .064227 _cons 9.343494 .7131037 13.10 0.000 7.943334 10.74366

In order to give the full regression in an easier to read format, here are two tables, one of the general coefficients and the segment premiums, and then the segment coefficients. Not all coefficients are significant but paint a better picture than later when only significant variables are used for the segments.

The full model however is still clustered and confusing and has many

more variables than that are significant. It is also difficult to see how a basic car

is broken down into its attributes. To take a step back, I ran a regression for a

general model of all cars without interaction terms. This would serve as a

starting point and a point of reference. Because some variables may be skewed

by segment weight and their coefficients, I have run robust standard errors. Linear regression Number of obs = 757 F( 14, 742) = 468.46 Prob > F = 0.0000 R-squared = 0.9116 Root MSE = .2293

Robust logprice2007 Coef. Std. Err. t P>|t| [95% Conf. Interval]

wagon .1002219 .0328182 3.05 0.002 .0357944 .1646495 halfnocyl -.1122873 .0423469 -2.65 0.008 -.1954212 -.0291534 bore -.3397317 .0809178 -4.20 0.000 -.4985869 -.1808765 stroke -.2513957 .0501784 -5.01 0.000 -.3499043 -.1528872 displacement .1187327 .041744 2.84 0.005 .0367823 .200683 hp .0027759 .0003668 7.57 0.000 .0020557 .003496 ftlbs .0007768 .0002915 2.67 0.008 .0002046 .001349 gears .0019142 .0008184 2.34 0.020 .0003076 .0035208 rwd .0774465 .0290069 2.67 0.008 .0205012 .1343919 numberoccu~s -.1134541 .0099921 -11.35 0.000 -.1330703 -.093838 fueltanksize .0695993 .0049816 13.97 0.000 .0598195 .079379 performance -.0711746 .0356159 -2.00 0.046 -.1410945 -.0012548 luxury .3198506 .0281998 11.34 0.000 .2644898 .3752114 economy .1183076 .031645 3.74 0.000 .0561832 .180432 _cons 10.69406 .3952478 27.06 0.000 9.918126 11.47

This general model is very similar to past studies, with addition of bore, stroke, torque, and displacement to the typical horsepower and other factors. Bore and

Stroke must be taken into consideration that displacement will change with them and instead of the effects solely must be taken into account with that. But, they do give insight into what is preferred in the segments as along with horsepower and torque, they have different returns to price. Overall, attitudes towards preferred grunt can be seen if interpreted carefully. Some of these added aspects can come in handy later as horsepower is not the all-important number in all segments.

For my actual results, I ran the full model and removed insignificant variables until a tidy regression came forward, and then afterwards ran a regression for each segment.

RESULTS

General regression with interaction terms

Source SS df MS Number of obs = 757 F( 24, 732) = 420.69 Model 411.490324 24 17.1454302 Prob > F = 0.0000 Residual 29.8328582 732 .040755271 R-squared = 0.9324 Adj R-squared = 0.9302 Total 441.323182 756 .583760823 Root MSE = .20188

logprice2007 Coef. Std. Err. t P>|t| [95% Conf. Interval]

luxdoor -.099173 .0160426 -6.18 0.000 -.130668 -.0676779 wagon .1114844 .027111 4.11 0.000 .0582599 .1647089 luxWB .0040659 .0015528 2.62 0.009 .0010174 .0071144 overalwidt~n .007946 .0030423 2.61 0.009 .0019735 .0139186 perfwide .018564 .0034981 5.31 0.000 .0116965 .0254316 ecoCW .0004813 .0000678 7.10 0.000 .0003483 .0006144 luxcyl -.1561885 .0262646 -5.95 0.000 -.2077514 -.1046257 bore -.2571945 .0501017 -5.13 0.000 -.3555548 -.1588343 perfstroke -.3373191 .0715633 -4.71 0.000 -.4778129 -.1968254 luxstroke -.1915809 .0407728 -4.70 0.000 -.2716264 -.1115354 ecostroke -.1903499 .033859 -5.62 0.000 -.2568221 -.1238776 luxdisplace .201085 .0270939 7.42 0.000 .147894 .254276 hp .003567 .0003205 11.13 0.000 .0029378 .0041962 luxhp -.001655 .0004001 -4.14 0.000 -.0024404 -.0008695 econhp -.0052912 .0012152 -4.35 0.000 -.0076769 -.0029055 luxtq .0013229 .0002326 5.69 0.000 .0008663 .0017795 fueltanksize .03851 .0071559 5.38 0.000 .0244614 .0525586 luxfuel .0295415 .0092152 3.21 0.001 .0114502 .0476329 perfauto -.1739167 .0424253 -4.10 0.000 -.2572064 -.090627 luxauto -.070653 .0330987 -2.13 0.033 -.1356328 -.0056733 awd .0876818 .0249387 3.52 0.000 .0387219 .1366417 numberoccu~s -.1087682 .0150817 -7.21 0.000 -.1383767 -.0791597 perfoccupant -.0428889 .0187899 -2.28 0.023 -.0797774 -.0060004 luxoccupant .0722453 .0225029 3.21 0.001 .0280673 .1164233 _cons 9.545721 .2710573 35.22 0.000 9.013579 10.07786

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Ho: Constant variance Variables: luxdoor wagon luxWB overalwidthin perfwide ecoCW luxcyl bore perfstroke luxstroke ecostroke luxdisplace hp luxhp econhp luxtq fueltanksize luxfuel perfauto luxauto awd numberoccupants perfoccupant luxoccupant

F( 24 , 732) = 1.15 Prob > F = 0.2867

Table of general regression with interaction terms, showing only significant variables

Performance segment

Linear regression Number of obs = 231 F( 22, 208) = 169.29 Prob > F = 0.0000 R-squared = 0.9067 Root MSE = .21058

Robust logprice2007 Coef. Std. Err. t P>|t| [95% Conf. Interval]

doors -.0732162 .0243898 -3.00 0.003 -.1212992 -.0251333 overalwidt~n .2706238 .042999 6.29 0.000 .1858541 .3553934 luxwide -.236925 .043225 -5.48 0.000 -.3221402 -.1517098 halfnocyl -1.439924 .2882978 -4.99 0.000 -2.008284 -.8715637 luxcyl 1.396615 .2892034 4.83 0.000 .8264696 1.966761 bore -3.614487 .6412246 -5.64 0.000 -4.878619 -2.350355 luxbore 3.734329 .6416456 5.82 0.000 2.469367 4.999291 stroke -1.591273 .3901723 -4.08 0.000 -2.360472 -.8220739 luxstroke 1.338838 .3912947 3.42 0.001 .5674266 2.11025 displacement 1.195209 .2318959 5.15 0.000 .7380418 1.652377 luxdisplace -1.05398 .238505 -4.42 0.000 -1.524177 -.5837829 hp .0065043 .0016507 3.94 0.000 .0032499 .0097586 luxhp -.004696 .0017457 -2.69 0.008 -.0081376 -.0012544 ftlbs -.0081168 .0020441 -3.97 0.000 -.0121465 -.0040871 luxtq .0090954 .0020811 4.37 0.000 .0049927 .0131981 fueltanksize .0786792 .010376 7.58 0.000 .0582236 .0991348 gears .3572059 .1442966 2.48 0.014 .0727346 .6416772 luxgears -.4084517 .1472039 -2.77 0.006 -.6986546 -.1182487 auto -.2849984 .0436394 -6.53 0.000 -.3710306 -.1989661 awd 1.409058 .2548566 5.53 0.000 .9066253 1.911492 luxawd -1.328399 .2607775 -5.09 0.000 -1.842505 -.8142929 numberoccu~s -.0720605 .017915 -4.02 0.000 -.1073788 -.0367422 _cons 7.152659 .8497332 8.42 0.000 5.477466 8.827853

Luxury segment

Linear regression Number of obs = 400 F( 21, 378) = 320.66 Prob > F = 0.0000 R-squared = 0.9335 Root MSE = .17246

Robust logprice2007 Coef. Std. Err. t P>|t| [95% Conf. Interval]

doors -.089798 .0164401 -5.46 0.000 -.1221236 -.0574724 convertible .0532805 .0341292 1.56 0.119 -.0138263 .1203874 wagon .1375983 .0276609 4.97 0.000 .0832098 .1919867 wheelbasein .0054593 .003345 1.63 0.103 -.0011179 .0120364 overalwidt~n .0205194 .0029715 6.91 0.000 .0146767 .0263621 reartrack -.0343582 .0090089 -3.81 0.000 -.0520721 -.0166443 halfnocyl -.1131464 .0283996 -3.98 0.000 -.1689874 -.0573053 bore -.3119521 .0852168 -3.66 0.000 -.4795105 -.1443937 perfbore .2273179 .0878725 2.59 0.010 .0545377 .4000982 perfstroke -.3689997 .071193 -5.18 0.000 -.5089836 -.2290158 displacement .160271 .0374177 4.28 0.000 .0866981 .2338439 hp .0019186 .0003111 6.17 0.000 .0013069 .0025303 ftlbs .0011808 .000227 5.20 0.000 .0007346 .0016271 fueltanksize .0580179 .0069199 8.38 0.000 .0444116 .0716241 perffuel .0392352 .0087662 4.48 0.000 .0219985 .0564719 perfauto -.3027941 .0384011 -7.89 0.000 -.3783006 -.2272876 awd .1110196 .0264954 4.19 0.000 .0589227 .1631164 rwd .1530086 .0274602 5.57 0.000 .0990146 .2070025 perfrwd -.1199267 .0559918 -2.14 0.033 -.2300211 -.0098323 numberoccu~s -.0370469 .0155057 -2.39 0.017 -.0675352 -.0065587 perfoccupant -.0360265 .0170903 -2.11 0.036 -.0696305 -.0024224 _cons 10.17522 .5460636 18.63 0.000 9.101519 11.24892

Economy Segment

Source SS df MS Number of obs = 211 F( 8, 202) = 22.46 Model 11.2970258 8 1.41212823 Prob > F = 0.0000 Residual 12.7012148 202 .062877301 R-squared = 0.4707 Adj R-squared = 0.4498 Total 23.9982406 210 .114277336 Root MSE = .25075

logprice2007 Coef. Std. Err. t P>|t| [95% Conf. Interval]

hatchback .0813434 .0478584 1.70 0.091 -.0130228 .1757096 wagon .1675942 .0600647 2.79 0.006 .04916 .2860283 curbweight .0002585 .0000927 2.79 0.006 .0000758 .0004412 stroke -.1354751 .0721828 -1.88 0.062 -.2778036 .0068534 fueltanksize .0517965 .0173788 2.98 0.003 .0175293 .0860637 gears .0044087 .0016306 2.70 0.007 .0011935 .0076239 auto .1069931 .052043 2.06 0.041 .0043759 .2096103 numberoccu~s -.1576027 .0342539 -4.60 0.000 -.2251436 -.0900617 _cons 9.441569 .33021 28.59 0.000 8.790468 10.09267

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Ho: Constant variance Variables: hatchback wagon curbweight stroke fueltanksize gears auto numberoccupants

F( 8 , 202) = 1.18 Prob > F = 0.3136

Family Segment

Source SS df MS Number of obs = 120 F( 9, 110) = 40.83 Model 4.19255179 9 .465839087 Prob > F = 0.0000 Residual 1.25516833 110 .011410621 R-squared = 0.7696 Adj R-squared = 0.7507 Total 5.44772011 119 .045779161 Root MSE = .10682

logprice2007 Coef. Std. Err. t P>|t| [95% Conf. Interval]

curbweight .0002141 .0000791 2.71 0.008 .0000573 .0003709 reartrack -.0179504 .0091716 -1.96 0.053 -.0361263 .0002255 bore -.4660357 .0701309 -6.65 0.000 -.6050187 -.3270528 stroke -.0997267 .0527161 -1.89 0.061 -.2041976 .0047441 displacement .0756707 .0288651 2.62 0.010 .0184668 .1328747 hp .0026884 .0005796 4.64 0.000 .0015398 .0038369 fueltanksize .0173636 .0089889 1.93 0.056 -.0004503 .0351776 awd .1974938 .0428354 4.61 0.000 .112604 .2823836 numberoccu~s -.090944 .0135031 -6.74 0.000 -.1177041 -.064184 _cons 11.78777 .5840543 20.18 0.000 10.63031 12.94523

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Ho: Constant variance Variables: curbweight reartrack bore stroke displacement hp fueltanksize awd numberoccupants

F( 9 , 110) = 0.60 Prob > F = 0.7929

ANALYSIS

The regressions seemed to explain the data fairly well. The R-squared

value for the general, luxury, and performance regressions is all above .90. For

the family and economy segments, there may be many features not accounted

for by the model. This may make sense as performance and luxury cars are very

competitive in their features and many things may come as standard compared

to the average or economy car. In the family and economy cars,

curbweight has a small premium so perhaps in this are some of the standard features which determine price within those two other segments. Economy cars can only have half their price changes by the variables I have used, and family cars three-fourths. In further research, these values may be able to be increased

by including more dummy variables for features, safety, and other attributes.

The results include a lot of information and for the most part can seem

logical reasons for relationships to be as such. It is quite interesting to see the

disparity across segments when dealing with the same change in attributes. In

either the full regression (with insignificant terms) or the first general regression

with interaction terms (and all are significant) one segment may be going in the opposite direction of other segments, and magnitudes have a lot of variety. It is reassuring to know that even with the insignificant terms removed from the general regression, many of the segment coefficients remain relatively close.

Another door still corresponds with a 9.9% price decrease, performance cars still have a negative price change with an increase of stroke (while keeping displacement and horsepower the same), and luxury cars enjoy displacement, and economy and performance cost less when they seat more people with very similar percentages for price change.

For the most part, the numbers given by the regressions seem to meet my expectations. High-end coupes correlate in luxury to make more doors reduce the price, give a premium among luxury cars but considering all are cheaper, controlling for weight, in performance as they are most common there, and wagons are a premium which is reflected in most brand’s model break-down. Performance cars enjoy more cylinders, when for luxury and economy more effective cylinders are worth more. For performance and economy, stroke is less desired as this is related to an engine’s torque and does not help engine output and efficiency as much as bore, so either the horsepower-focused performance or efficiency-focused economy devalue this attribute in comparison to general cars. Automatic standard is a negative quality to have for a consumer looking to sport in a performance car, but a premium in an economy car which are generally manual to save on gas and cost. All-wheel- drive’s cost and safety has a premium in all classes, although much higher for performance, as this may be correlated with a few higher priced cars given similar specification which are the minority in the class. In performance and economy, it is cheaper to have a car fit to hold more people much more so than with luxury cars.

In the performance segment regression, the focus on horsepower is apparent and so is its preference over torque. However, for the luxury sub- group, torque is attractive comparably. An additional gear for better highway cruising or for better low-speed high-ratio acceleration is positive for both performance and the luxury sub-group. The engine specifications and AWD dummy coefficients seem quite inappropriate in the non-luxury performance but

AWD may be clouded by the small minority of performance cars with it, and cylinders, bore, stroke, and displacement are to be interpreted when being horsepower and torque is being controlled for. If the number of cylinders is increased and horsepower remains the same, the lower cost of this should be understandable. The other regressions and tables follow in much the same suit, with reasonable coefficients and plausible interpretations. The coefficients are still much the same as the general regression with interaction terms, like the auto dummy coefficient for economy cars of 10.70% compared to 10.38%.

CONCLUSION

The goal of this thesis was to explore the relationship of automotive factors across different segments and the effect on price. I have discovered that the model I used can be applied to the luxury and performance segments fairly well. It can be used as well to compare returns of certain qualities in the other two segments but there is much more going on in determining price with family and economy cars. I have reservations to think the model really explains the change in prices as well as it may seem, but competition may filter out many variances between the cars. Also, the selection of years to pool may limit the variety in car models whereas in the decades in the middle of last century cars were modified each year, and they may only do so every four years today. Even so, the model seems to give useful and informative relationships that compare how different segments price different changes in factors.

It would be interesting to add several more years, from 1990 to 2004 into the data pool along with dummies for standard features and safety and reliability ratings. Whereas many things are considered standard today, they were just coming into being in the 90s when passive restraints were just introduced and ABS was not yet as prevalent. Perhaps variation would not be as extreme within segments as across them, but the variety of specifications could be much more troublesome to model across such a pool. Since safety equipment and other features have changed so much in the last two decades, a year dummy may have to be used which would make things complicated, especially if it were to be used to test the price level changes given quality being controlled for.

However, the most interesting possibility of such research may in fact be a

possible “closing of the premium” on things such as airbags, side airbags, ABS, or other features that have grown to be more common today than yesteryear because of competition. This alone could be reason enough for another interesting study. Perhaps as well a study of change in buyer preferences for horsepower, gas mileage, or the like could be intriguing over a similar time period.

The study of pricing in the automotive industry is nothing new and my research has been just another step in the analysis of how companies price automobiles. It is hard to determine whether these coefficients are the result of manufacturing costs, being associated with certain models, or represent a popularity and demand which increases the attributes pricing. Without industry information from the auto makers concerning price to produce, or surveys of the consumer, this is hard to accomplish. At least at this time, it is clear to see that the segments have good reason to be separated by the market and players.

Works Cited

Agarwal, Manoj K.; Ratchford, Brian T. Estimating Demand Functions for Product

Characteristics: The Case of Automobiles. The Journal of Consumer Research,

Vol. 7, No. 3. (Dec., 1980), pp. 249-262.

Arguea, N.M.; Hsiao, C., Taylor, A. Estimating Consumer Preferences Using Market

Data—An Application to US Automobile Demand. Journal of Applied

Econometrics, Vol. 9, No. 1 (Jan. – Mar. 1994), pp. 1-18.

Automotive News. 2005 Market Data Book. Accessed online at www.autonews.com

from March-May 2008

Automotive News. 2006 Market Data Book. Accessed online at www.autonews.com

from March-May 2008

Automotive News. 2007 Market Data Book. Accessed online at www.autonews.com

from March-May 2008

Bajic, Vladimir. Market Shares and Price-Quality Relationships: An Econometric

Investigation of the U.S. Automobile Market. Southern Economic Journal,

Vol. 54, No. 4. (Apr., 1988), pp. 888-900.

Bresnahan, Timothy F. Competition and Collusion in the American Automobile Industry:

The 1955 Price War. The Journal of Industrial Economics, Vol. 35, No. 4,

The Empirical Renaissance in Industrial Economics. (Jun., 1987), pp. 457-482.

Bordley, Robert F. Estimating Automotive Elasticities from Segment Elasticities and

First Choice/Second Choice Data. The Review of Economics and Statistics,

Vol. 75, No. 3. (Aug., 1993), pp. 455-462.

Boyle, Stanley E.; Hogarty, Thomas F. Pricing Behavior in the American Automobile

Industry, 1957-71. The Journal of Industrial Economics, Vol. 24, No. 2.

(Dec., 1975), pp. 81-95.

McMahon, Tim “Current Inflation.” Financial Trend Forecaster. www.inflationdata.com

Accessed March-May 2008.

Ohta, Makoto; Griliches, Zvi. Automobile Prices and Quality: Did the Gasoline Price

Increases Change Consumer Tastes in the U.S.? Journal of Business and

Economic Statistics, Vol. 4, No. 2. (Apr., 1986), pp. 187-198.

“Search by Class for Fuel Efficient Cars.” U.S. Department of Energy.

www.fueleconomy.gov/feg/byclass.htm, accessed March-May 2008

Triplett, Jack E. Automobiles and Hedonic Quality Measurement. The Journal of

Political Economy, Vol. 77, No. 3. (May – Jun., 1969), pp. 408-417.

“Used Car Information.” Microsoft. http://autos.msn.com/home/used_research.aspx

Accessed March-May 2008.