Automotive demand models

From DDL Wiki

Jump to: navigation, search

This page contains a summary of select literature on modeling consumer demand for conventional and alternatively-fueled vehicles

Contents

Introduction

The logit (multinomial logit), mixed logit, probit, and variations on these models seem to be the most popular ways of modeling consumer demand for automobiles. The logit model is a discrete choice model in which it is assumed that all other alternatives are uncorrelated over alternatives. In other words, each choice is independent of the others. The mixed logit model is an extension of the normal logit model allowing for unobserved factors to follow any random distribution. The probit discrete choice model differs from the logit and mixed logit in that it accounts for the unobserved factors by distributing them jointly normal <ref>Train, K., 2003, Discrete Choice Methods with Simulation. Cambridge, UK: Cambridge University Press.</ref>.

Another important part of some of these papers is the effect of Corporate Average Fuel Economy (CAFE) regulations on manufacturers to build more fuel efficient cars. This regulation says that companies must achieve a minimum sales- weighted average fuel efficiency of 27.5 mpg. If a company fails to meet this standard, it incurs a penalty of $5 per 1/10 of a gallon that the corporate average fuel economy falls below standards.

The following section summarizes the important points of each paper and what it adds to the progress of the issue of trying to model the demand for automobiles.

Automobile Demand Modeling

Table 1 summarizes some of the major papers related to modeling automobile demand. Not all of these papers include data on alternatively-fueled vehicles, such as electric vehicles, compressed natural gas, or methanol. Some papers obtain data from consumer surveys, while others get their information from standard trade references, and yet others use both. Still others discuss the effects of government regulations on the demand for automobiles as well as other factors. Table 2 is a more concise comparison of the key content in each paper.

Boyd & Mellman (1980)

<ref>Boyd, J. H., and Mellman, R. E., 1980, “The Effect of Fuel Economy Standards on the U.S. Automotive Market: An Hedonic Demand Analysis,” Transp. Res., Part A, 14, pp. 367-368.</ref>

Boyd & Mellman (1980) wrote one of the first papers exploring how government regulations, namely CAFE regulations and rising fuel prices are forcing auto manufacturers to improve fuel economy. The paper considers whether it is worthwhile for a company to abide by the new CAFE regulations. Boyd & Mellman utilize the hedonic demand model, also known as the “random coefficients logit model”, which is an extension of the multinomial logit probability choice model. This model incorporates the variation in consumer tastes and preferences. The paper compares the accuracy of the hedonic model’s predictions of market shares to the accuracy of the logit model’s predictions. It argues that the hedonic model is often a better fit to the data than the logit model, but not necessarily all the time. This study also reveals that consumers do value improvements in fuel economy, but these improvements will also eventually change other vehicle attributes. Boyd & Mellman conclude that the shift in market shares from less to more fuel efficient cars in response to feasible changes in new car prices or gasoline prices are likely to have only modest short-run impacts on fleet average fuel economy [2].

McCarthy (1996)

<ref>McCarthy, P., 1996, "Market Price and Income Elasticities of New Vehicle Demands," The Review of Economics and Statistics, Vol. 78, No. 3. (Aug., 1996), pp. 543-547.</ref>

McCarthy’s 1996 study differed from the other studies considered here in that it focuses more on the market price elasticity of demand for autos. It is somewhat relevant to the study of demand for alternatively fueled vehicles because it utilizes a 1989 nationwide household survey of new vehicle buyers, conducted by J.D. Power & Associates. McCarthy also used a multinomial logit demand model in this study. However, this paper does not go into much detail on how the logit model was used, it focuses more on the results by comparing the different cross price elasticities of demand.

Golob et al. (1997)

<ref>Golob, T.F., Torous, J., Bradley, M., Brownstone, D., Soheila, S.C., Bunch, D.S., 1997, "Commercial Fleet Demand for Alternative-Fuel Vehicles in California," Transportation Research Part A, Vol. 31, No. 3, pp. 219-233.</ref>

The paper collected the fleet demand information of alternative-fuel vehicles (AFVs) from 2000 fleet sites in California. They mainly investigate the preference trade-offs for fuel types and other vehicle attributes. They use binomial probit model and ordered-response probit model to study the responses from fleet managers. The attributes are similar to the personal AFV purchase preferences, including capital cost, range, operating cost, refueling availability, tailpipe emission, etc. The stated preference model results also showed that there were major differences in preferences for fuel types among fleet market segments. They also found private fleet operators have less directly influence than government and school sectors by the environmental factor.

Goldberg (1995)

<ref>Goldberg, P. K., 1995, "Product Differentiation and Oligopoly in International Markets: The Case of the U.S. Automobile Industry," Econometrica, Vol. 63, No. 4, pp. 891-951.</ref>

This paper utilized nested logit model on the supply side to model the consumer demand. On the supply side, the first order game-theoretic equilibrium conditions of the profit maximizing firms are used to estimate the price, cost and margin. The author used ordinary least square (OLS) method for nested logit coefficient regression. The effect of two trade policies, voluntary export restraint (VER) and exchange rate pass-through, on prices change he price changes during 1983-87 has been discussed. [11]

Goldberg (1998)

<ref>Goldberg, P. K., 1998, "The Effects of the Corporate Average Fuel Efficiency Standards in the US," The Journal of Industrial Economics, Vol. 46, No. 1, pp 1-33.</ref>

A later paper that also focuses on the effects of CAFE standards on automobile sales, prices, and fuel consumption is Goldberg (1998). In this study, Goldberg finds a discrete choice model of auto demand and a continuous model of vehicle utilization from the Consumer Expenditure Survey (1984-1990). She then combines this with a model of oligopoly and product differentiation on the supply side in order to better assess the effects of CAFE regulation through simulations. It is assumed that in each period firms maximize profits with respect to two variables: prices p and the fractions of their domestic production α (thus 1-α is classified as imported. The effects of CAFE standards on profits are then compared to the effects of alternative policy instruments, such as an increased gasoline tax. The paper considers the pros and cons of CAFE regulations and their effectiveness. It also argues that nested logit models are better to use for modeling automobile demand than simple multinomial logit models. This is because the nested logit models consider the possibility that the consumer forgoes the purchase and includes information on past purchases. Moreover, this paper includes equations derived to find the profit and penalties associated with not meeting regulations, whatever they may be. The paper explores the possibility of abolishing CAFE regulations or replacing the CAFE regulations with higher gasoline taxes. The data used in these simulations is mostly from 1989. It can be concluded from this method of analysis that CAFE regulations are not enough of an incentive for consumers to purchase more fuel efficient vehicles, since the regulations are a “set of internal taxes (on fuel inefficiency) and subsidies (on fuel efficient vehicles) within each firm” [3]. This paper seems to be very relevant and worth using as a good resource for future studies similar to it.

Brownstone & Train (1999)

<ref>Brownstone, D., and Train, K., 1999, "Forecasting New Product Penetration with Flexible Substitution Patterns," Journal of Econometrics, Vol. 89, pp 109-129.</ref>

In 1996, Brownstone & Train did a study which heavily influenced future studies. This paper discusses how the “independence from irrelevant alternatives” property restricts the logit and nested logit models from being more accurate. In other words, these models do not account for the fact that “the ratio of probabilities for any two alternatives is independent of the existence and attributes of any other alternative” [4]. This paper also explores models for new product forecasting, specifically mixed logit models with various different structures and probit models. Brownstone & Train utilize a survey that was given to a sample of California households, which got about 4654 completed surveys. The surveys asked people to choose one of six alternatives, which included one of 4 fuel types (gas, methanol, compressed natural gas, or electricity), 5 class sizes (mini, subcompact, compact, midsize, or large), and 6 body types (regular car, sports car, truck, van, station wagon, or SUV). This seems to be the most comprehensive and complete survey obtained up until this point in this study of automobile demand for alternatively fuel vehicles. The problem with this survey method is that people do not always buy what they say they will buy. This study was done when there were not many alternative fuel vehicles available on the market, so the results are not necessarily indicative of how consumers will actually react to changes in gas prices and CAFE regulations. The most important contribution of this paper is that it compares the accuracy of various types of mixed logit and probit choice models when they are used to model consumer demand for autos.

Brownstone, Bunch & Train (2000)

<ref>Brownstone, D., Bunch, D., and Train, K., 2000, "Joint mixed logit models of stated and revealed preferences for alternative-fuel vehicles," Transportation Research Part B, Vol. 34, pp 315-338.</ref>==

Building on the 1996 study, in 1999 Brownstone, Bunch, & Train focused on comparing the multinomial logit model to the mixed logit model for data on California households’ revealed and stated preferences for automobiles. Like the Brownstone & Train (1999), this study included cars fuelled by gasoline, electricity, methanol, and compressed natural gas. The paper argues that the mixed logit model is superior to the multinomial logit model in that it fits the data more accurately for this purpose. Most importantly, this paper discusses how critical it is to use both stated and revealed preferences of consumers. The revealed preference data are “critical for obtaining realistic body-type choice and scaling information, but they are plagued by multicollinearity and difficulties with measuring vehicle attributes” [5]. The stated preference data are critical for getting information about attributes not available in the marketplace, but the forecasts from this data can be implausible.

McFadden & Train (2000)

<ref>McFadden, D., and Train, K., 2000, "Mixed MNL Models for Discrete Response," Journal of Applied Economics, Vol 15, Issue 5, pp 447-470.</ref>

A study that delves more deeply into how the mixed multinomial logit model is used can be found in McFadden & Train (2000). This paper is very lengthy and describes in detail how the mixed multinomial logit model works, using as an example the problem of demand for alternative vehicles. For this example, they used the results from the Brownstone et al. (1996) survey of preferences among alternative vehicles, and then used the estimates for the beta coefficients from Brownstone and Train (1999). The proofs of all of the theorems used are also included in the appendix of this paper.

Berry, Levinsohn and Pakes (1995)

<ref>Berry, S., Levinsohn, J., and Pakes, A., 1995, "Automobile Prices in Market Equilibrium," Econometrica, Vol. 63, No. 4, pp. 841-890.</ref>

The major contribution of this classic paper (so called BLP paper afterwards) is not only on the analysis of the demand in the U.S. automobile market, but, most importantly, provide a framework to utilize the existing aggregate consumer-level data and estimate the cost and demand parameters. BLP offered a useful method to deal with endogeneity and move it out of nonlinear choice models into linear regression, such as GMM. In BLP’s utility form, they included a specific unobservable term, which is known to the market participants but not the econometrician. For example, the term may include the aspects of style, prestige, reputation, advertisement and consumer’s past purchase experience. BLP approach is able to perform regression based on market data (revealed preference), not survey data (stated preference). The product attributes in their logit demand model are ratio of horsepower to weight (HP/WT), air conditioning dummy variable, MP$ (miles per dollar), vehicle size, and price (in 1983 dollars). The automobile market data is collected in a 20-years period, 1971-1990, from Automotive News Market Data Book. The substitution behavior of moving from vehicle purchase to outside good, when price increased, is compared under standard logit and BLP’s random coefficient logit model. [12]

Berry, Levinsohn, and Pakes (2004)

<ref>Berry, S., Levinsohn, J., Pakes, A., 2004, “Differentiated Products Demand Systems from a Combination of Micro and Macro Data: The New Car Market,” Journal of Political Economy, Vol. 112, No. 1, pp. 68-105.</ref>

Another influential paper is Berry, Levinsohn, and Pakes (2004), which discusses “an algorithm for estimating characteristic based demand models from alternative data sources, and applies it to new data on the market for passenger vehicles” [8]. They have found that provided that the data is rich enough, the model can rationalize existing results and provide realistic out of sample predictions for future purchases. An essential part to this study is that they utilize not only a consumer’s first choice car (the one purchased), but also the second choice car that the consumer might have purchased. This information proved helpful in determining just how important each characteristic or factor was for each consumer. The study compares the results of using a logit model where only the first choice data is used and one where both the first and second choice data is used. The models used also allow for characteristics to vary as a function of both the observed and unobserved consumer attributes, data was acquired from the 1993 CAMIP Sample by General Motors, which included about 37,500 complete observations (34,500 of which also reported their second choice car choice), and the Current Population Survey was used. This paper is slightly more complicated than the other papers, but it seems to include important ideas for the future of alternatively fuel auto demand.

Sudhir (2001)

<ref>Sudhir, K., 2001, "Structural Analysis of Manufacturer Pricing in the Presence of a Strategic Retailer," Marketing Science, Vol. 20, No. 3, pp. 244-264.</ref>

Basically this paper is a further development of the classical BLP paper (1995) [12] which first presented a framework to estimate demand and cost parameters and take price endogeneity into account. Sudhir (2001) derived the structural pricing equation (cost plus margin) upon first-order optimality condition. The data in the paper are the same as BLP’s U.S. auto market data but within a confined period, 1981-1990. The author normalized the gasoline and vehicle prices with Consumer Price Index (CPI) and calibrated the potential market size. While the demand model form (random-coefficient logit) and regression method (GMM) are the same as BLP (1995), the author more focuses on the competitive interactions in the different market segments. [13]

Dagvik et al. (2002)

<ref>Dagsvik, J. K. Wennemo, T., Wetterwald, D. G., and Aaberge, R., 2002, "Potential demand foralternative fuel vehicles," Transportation Research B, Vol. 36, No. 4, pp. 361-384.</ref>

This paper analyzes the potential demand for alternative fuel vehicles. Treating gasoline vehicle as base model, the alternative fuel vehicles consider are liquid propane gas (LPG) and electric powered vehicles (EV) and hybrid vehicle (HV). The data were obtained from a stated preference survey conducted by Statistics Norway [9]. Each individual has been asked 15 question of ranking 3 three hypothetical vehicles characterized by specific attributes, including price, top speed, driving range, etc. The results indicate the driving range of is an important attribute. The electric vehicles will not be fully competitive in the automobile market unless the technology advances solved the limited driving range issue.

Choo and Mokhtarian (2004)

<ref>Choo, S., and Mokhtarian, P.L., 2004, "What type of vehicle do people drive? The role of attitude and lifestyle in influencing vehicle type choice," Transportation Research Part A: Policy and Practice, Vol. 38, Issue 3, pp. 201-222.</ref>

This unique study related vehicle type to travel attitude, personality, lifestyle, mobility, and demographic variables individually, using one-way analysis of variance (ANOVA) and chi-squared tests. A multinomial logit model for vehicle type choice was then estimated. The MNL model confirmed that the variables analyzed significantly affect an individual's vehicle type choice. The data used in the study are from a 1998 mail-out/mail-back survey of 1904 residents in the San Francisco Bay Area. The dependent variable is classified into nine categories: small, compact, mid-sized, large, luxury, sports, minivan/van, pickup, and sport utility vehicle (SUV) categories.

Santini and Vyas (2005)

<ref>Santini, D., and Vyas, A., 2005, Suggestions for a New Vehicle Choice Model Simulating Advanced Vehicles Introduction Decisions (AVID): Structure and Coefficients. http://www.transportation.anl.gov/pdfs/TA/350.pdf</ref>

The Advanced Vehicles Introduction Decisions (AVID) model is developed by Argonne National Lab (ANL) for evaluating the market penetration of potential new vehicle technologies. AVID uses multinomial logit (a single set of constant decision coefficients that represents all buyers) to predict consumer preferences in a dynamic simulation model (2000-2020) and applies segmentation approach to categorize vehicle buyers as early buyers and late (majority) buyers. The study basically focuses on the market penetration prediction of hybrid electric vehicles (HEV). At low market share, the AVID shows that the early group has significantly lower price sensitivity than the majority buyer group. To be noticed , there is no outside good included in the logit probability of AVID model.

Train and Wilson (2007)

<ref>Train, K.E. and Winston, C., 2007, "Vehicle choice behavior and the declining market share of US automakers," International Economic Review, 48(4), pp. 1469-1496.</ref>

This paper employed mixed logit demand model to study the relation between the consumer choice behavior and market share drops of the U.S. automakers in the past decade. It showed that the loss of U.S. automaker market shares can be explained by the vehicle attributes, such as retail price, power, weight, fuel consumption, body type, transmission type and reliability, where Japanese and European manufacturers have more improvements on attributes than U.S. manufacturers.


A Concise Comparison of Literature Related to Modeling Automotive Demand

Paper Preference Model Fuel Types Type of Data Data years
Boyd & Mellman (1980) MNL, MXL G S, R 1977-1978
Goldberg (1995) NL, MNL G S,R 1989-1990
Berry, Levinsohn, & Pakes (1995) MNL, MXL G R 1971-1990
Brownstone & Train (1999) MNL, MXL, P G, E, CNG, M S 1993
McCarthy (1996) MNL G J 1983-1985
Golob et al. (1997) P E, CNG, M S 1994
Goldberg (1998) NL, MNL G S,R 1993
Brownstone, Bunch, & Train (2000) MNL, MXL G, E, CNG, M S, R 1993
McFadden & Train (2000) MXL G, E, CNG, M S, R 1993
Sudhir (2001) MXL G R 1981-1990
Dagvik et al. (2002) Luce G, E, LPG, HEV S Norway
Choo & Mokhtarian (2004) MNL G S 1998
Berry, Levinsohn & Pakes (2004) MXL G R 1993
Santini & Vyas (2005) MNL G, HEV S
Train and Wilson (2007) MXL G R 2000
Dagsvik and Liu (2009) NL G,AFV S 2001 Shanghai, China

MNL = Multinomial Logit (standard Logit or aggregate logit), NL = Nested Logit (considers not buying, info on past purchases, used vehicles, etc), MXL = Mixed Logit (random coefficient logit), P = Probit, Luce = Luce Model.

G = Gasoline, E = Electric, CNG = Compressed Natural Gas, LPG = Liquid Propane Gas, M = Methanol, HEV = Hybrid, AFV = Alternative Fuel Vehicle.

S = Surveys, R = Standard Trade References, J = Nationwide household survey of new vehicle buyers by J.D. Power & Ass.

Conclusion

Although each of these papers may not be directly useful as a reference for the research of how fuel efficiency and emission policy impact optimal vehicle designs, they very useful learning tools. The most similar paper to the one that Michalek has already written seems to be Goldberg (1998), while the BLP (2004) also has very relevant and current ideas. It is evident that a number of these papers favor the mixed multinomial logit discrete choice model. Overall, each of these studies seems to shed new light on the issue of modeling demand for automobiles.

Acknowledgments

Initial material for this article was taken from a report written by Ali Gitomer, B.S. Candidate, Industrial Engineering & Management Sciences, Northwestern University, e-mail: a-gitomer@northwestern.edu

References

<references/>

Personal tools