Humboldt-Universität zu Berlin

Transcription

1 Humboldt-Unverstät zu Berln Insttute for Statstcs and Econometrcs A Bnary Logstc Analyss of Car Consumer Behavor n Chna Bachelor Thess Bachelor of Scence Study of Statstcs August, 22, 2006 Presented by: Shen Guan (184728) Tester: Prof. Dr. B. Rönz Emal: cerbrnags@hotmal.com

2 AUTHORSHIP DECLARATION I hereby declare and confrm that ths thess s entrely the result of my own work except where otherwse ndcated. I have acknowledged the supervson and gudance I have receved from Professor Bernd RÖNZ. Shen GUAN 22 /August /2006

3 ACKNOWLEDGMENT I hereby acknowledge Professor Bernd RÖNZ for hs supervson, gudance, avalablty and frendly support durng my work on ths thess. I also apprecate the gudance and help from Szymon Borak. I am very grateful to my famly and my dear frend Sona Boyum for ther encouragement and support. 1

4 Contents 1. Introducton.1 2. Background: Today s Chnese car market General Data Overvew..7 I. Data Source 7 II. Choce of Varables Logt Model Stepwse Backward Elmnaton and Bvarate Analyss The ft of the model Interpretaton of the model Some Dscussons about Chnese Auto market 25 I. Second-Hand Auto market...25 II. Fnancng n Chnese Auto market...27 Appendx.30 References

5 Lst of Fgures Fgure 2.1: Car Unt Proftablty n Chna (n US dollars)...6 Fgure 2.2: Chnese passenger car and lght vehcle sales n perod 1993 to Fgure 3.1: Not Buy vs. Buy 9 Fgure 3.2: Dstrbuton of Gender...10 Fgure 3.3: Dstrbuton of Age..10 Fgure 3.4: Dstrbuton of Educaton Fgure 3.5: Dstrbuton of Income.11 Fgure 3.6: Dstrbuton of Occupaton...11 Fgure 3.7: Dstrbuton of Famly Sze..12 Fgure 5.1: Decson vs. Income.19 Lst of Tables Table 2.1: Market share of Auto makers Market share..5 Table 3.1: Varable Lst...8 Table 5.1 Forward Stepwse wth Wald.17 Table 5.2: Pearson s Ch-Square Sgnfcance Test...18 Table 7.1: Logt Model done wth subcategores Table 8.1: Dstrbuton of Payment.27 Table 8.2 Dvded Payment vs. Full Payment.28 3

6 1. Introducton Chna s entry nto the World Trade Organzaton (WTO) n December 2001 opened the market to foregn competton and trggered a surge n foregn drect nvestment. Consequently, automoble tarffs declned and thus caused a prce drop n the overall Chnese automoble market. It was assumed that the demand for cars would be stmulated as a result, whch would drve the car market to grow rapdly. From December 2001 to 2002, a growth n sales descrbed as gushng sales brought by the accesson nto the WTO can be read from the followng data: total vehcle sales reached 3.38 mllon n 2002 (up 1 mllon from 2001), a surge of 37%. Over 80% of the ndvdual car buyers purchased a car for the frst tme. (Shangha Stock Newspaper, Jan. 23, 2003). There s a small group of Chnese who are younger, better educated, more urbanzed and have a hgher ncome that are typcally consdered as the mddle class segment and leaders n the Chnese consumpton market. Many already possessed a car before Chna s entry nto the WTO. On one hand, although the car prce fell as a result of WTO-accesson, the relatonshp between prvate ncome and car prce s stll to a great degree dsproportonal. It mght stll be a heavy fnancng burden for the mddle class Chnese people to replace ther cars or own a second one. On the other hand, smultaneously, some new benefts other than prce cuttng brought by WTO would generate new characters of the consumer behavor of car buyers. New rules and possbltes affect not only the people who decde to buy a car for the frst tme, but also those who have already owned cars. Varous choces for car brands and modern car desgns, ntegrated after-sale-servces, professonal and advanced fnancng means, etc., would effectvely attract car owners to purchase a second automoble. Ths paper analyzes the probabltes of the car owners of replacng or buyng a second car and the factors that nfluence ther decsons by usng the logt model performed n SPSS

7 2. Background: Today s Chnese car market Chna s the fourth largest and the fastest growng car market n the world. Moreover, hgh unt proftablty of cars and Chnese consumers contnually boomng demand for motor vehcles have made Chna a huge potental car market for many carmakers, especally consderng the 1.3-bllon populaton of ths country. In the last few years, some profound developments, such as Chna s entry nto the WTO, have greatly accelerated the pace of Chnese car market growth. The foregn car producers, who have taken the frst move years earler, stll domnate the market wth consderable market shares (Table 2.1) and make very hgh proft per unt (Fgure 2.1). VW for nstance, enjoyed huge frst mover advantage (approx. 1985) of large market share, sellng close to 500,000 passenger cars n Chna n 2002 at a hgh proft. Source: Chna Assocaton of Automoble Manufacturers. Table 2.1 Market share of Auto makers Market share As showed n Fgure 2.2, Chnese passenger car and lght vehcle sales n the perod from 1993 to 2002 grew rapdly, especally from December 2001, the tme of Chna s entry nto WTO, to For many years, government offcals and 5

8 corporate customers were the man car buyers, but the recent rate of acceleraton can only be explaned by a large surge n purchases by prvate ndvduals. In Shangha for example, prvate vehcle regstratons were 7,000 n 1998, 83,700 n 2001, and 167,000 n June Source: Company data, Goldman Sachs Research estmates. Fgure 2.1 Car Unt Proftablty n Chna (n US dollars) Fgure 2.2 Chnese passenger car and lght vehcle sales n perod 1993 to

9 3. General Data Overvew I. Data Source The data used n ths paper comes from a survey done by Bejng Kangka Controllng Company n cooperaton wth Natonal Bureau of Statstcs of Chna n August 2001, about three months before Chna s entry nto WTO. The entry nto the WTO would defntely ncrease the accessons of foregn enterprses nto Chnese market, whch would strengthen the market competton and thus put enormous pressure on Chnese state-owned enterprses. After a long perod of tme of planned economy, the WTO membershp means more chances and also larger challenge for those enterprses. Ths survey was done to help the Chnese state-owned auto manufacturers to work out a proper marketng strategy to face the market change. It ncluded four parts:. External stuaton for car usng (ar qualty, ol supply, parkng possbltes, etc.);. Self-evaluaton of consumpton behavor and psychology;. Consumer car purchasng behavor; v. Plans for buyng a new car n the next 5 years, whch tred to: analyze the consumer behavor of the current car owners; fnd out the attrbutes of prvate ndvdual demand for cars; fnd out the varous nfluences on car demand; research and defne a proper car prce sutable for Chnese market; forecast the development of the Chnese car market n 5 years. It was done n 38 ctes, each cty ncluded 218 ntervewees and three lmtatons were set on the ntervewees: yearly ncome: over 30,000 RMB 1 ; age: between 20 to 55 years; car ownershp: yes. 1 Change rate n 2001: 1$= 8.277RMB n average. (data source: State admnstraton of foregn exchange, Chna). 7

10 More explanatons are necessary to the frst lmtaton. Accordng to the analyss of nternatonal car market, most ndvduals have the ablty to purchase a new car when the GDP per capta reaches $3000,.e., 3000$ s the boundary to fnancng ablty for purchasng a new car. In 2003, although the GDP per capta n Chna s $900, but n some well developed ctes, especally n some east coast ctes (Appendx 1), the GDP per capta level has already exceeded 3000$,.e., there s a large group of people who have the fnancng ablty to buy cars (Data source: Bejng Auto, The currency and development of Chnese car market, Page20, February 2003). The fnancng boundary to those who want to own a second car may be hgher, so t makes sense to set a hgher ncome lmtaton of 30,000 RMB whch s equal to $3625 on the ntervewees. II. Choce of Varables Only the data from 7 ctes whch cover 1526 ntervewees would be analyzed, snce t s not necessary to do the analyss of all the 38 ctes due to the enormous amount and also the repetton of the data. The varables (Table 3.1) from the last two parts of the survey consumer car purchasng behavor and plans for buyng a new car n the next 5 years are pcked out because of the objectves of ths paper. Table 3.1 Varable Lst 8

11 ). The dependent varable Decson s a bnary varable wth two categores not buy and buy. The queston was: do you plan to replace your car or to buy a second one n next 5 years? The answers were: a) yes, very sure; b) maybe yes; c)maybe not; d) defntely not; e) I do not know yet. In order to satsfy the condtons of Logt Model, the answers are classfed nto two groups: a) yes, I plan to buy a car; b) no, I do not plan to buy a car. The answers a or b are classfed nto the category buy and answers c or d nto the category not buy. Answer e wth the weght n whole sample 10.3% s defned as mssng value (appendx 1). The vald percent for not buy s 54.4% and buy 45.6%. 60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% not buy buy Fgure 3.1 Not Buy vs. Buy ). The ndependent varables Sx varables are chosen from the survey as ndependent varables. Basc descrptons to the characterstc of each varable are showed as followed. Cty: Bejng, Shangha, Tanjn, Nanjng, Hangzhou, Fuzhou, Guangzhou and each cty ncludes 218 ntervewees (Appendx 2). Compared to other ctes n Chna, these 7 ctes are more developed and consdered to be more senstve to market changes; hgher lvng standard enables ther resdents to pay for new desgns and fashons. Addtonally, those ctes are geographcally dstrbuted from north to south across Chna,.e. the analyss by usng the data from these ctes would well 9

12 represent the fluctuatons and dfferentatons of car consumpton n dfferent regons. Gender. 73.8% of the vald responses, whch s 1123 observatons, are male and 26.2% of the vald responses, n amount of 399, are female. The fact that most car owners n Chna are male results n the proporton of male car ownershp about 3 tmes hgher than female (Fgure 3.2). Age. In order to guarantee that every subcategory has enough observatons, ths varable s re-classfed nto 3 categores. The group s 34.2% n the whole sample, whch has 521 vald observatons and the group has 697 observatons wth the weght of 45.8%. The oldest group of 304 observatons s only 20% of the whole sample, snce auto s a relatve modern household consumpton n Chna and therefore more young people own cars than old people (Fgure 3.3). 80.0% 50.0% 60.0% 40.0% 30.0% 40.0% 20.0% 20.0% 10.0% 0.0% female male 0.0% Fgure 3.2 Dstrbuton of Gender Fgure 3.3 Dstrbuton of Age Educaton. Ths varable s re-classfed nto 2 categores attended unversty and dd not attend unversty (n short: not attend un and attend un). The subcategory not attend un has 680 observatons equal to 44.6% of total ntervewees and attended un 844 observatons equal to 55.4% of the whole sample observatons are vald for ths varable. There are enough observatons n each subgroup for analyss. As showed, the proporton of auto ownershp for those people who have a unversty degree s lghtly hgher than those who wthout (Fgure 3.4). 10

13 Income. Ths varable s re-classfed from 12 categores nto 3 categores: RMB, RMB, more than 5000 RMB. The category declned to answer whch has 127 observatons s defned as mssng values n ths paper. So the vald observaton for ths varable s 91.7% of the whole sample. The subcategory RMB has 696 observatons whch s 49.7 % of the vald sample, and RMB 276 observatons equal to 19.8% of the vald sample, RMB has 427 observatons equal to 30.5% (Fgure 3.5). 60.0% 50.0% 50.0% 40.0% 40.0% 30.0% 30.0% 20.0% 20.0% 10.0% 10.0% 0.0% 0.0% not attend un attend un RMB RMB 5001RMB + Fgure 3.4 Dstrbuton of Educaton Fgure 3.5 Dstrbuton of Income Occupaton. Ths varable s re-classfed from 13 categores nto 4 categores. There are 350 observatons n the sub-category Government servce equal to 23% of the whole sample and Employer 444 observatons equal to 29%, Employee 370 equal to 24.3%, Others (ncludng student, retred and other occupatons) 361 observatons equal to 23.7% (Fgure 3.6). 30.0% 25.0% 20.0% 15.0% 10.0% 5.0% 0.0% government servce employer employee others Fgure 3.6 Dstrbuton of Occupaton 11

14 Famly Sze. The queston for ths varable was: what s the number of famly members? Ths varable s re-classfed from 6 categores nto 3 categores: 1 or 2 persons, 3 persons, 4 persons or more. The frst group has 192 observatons equal to 12.6% of the whole sample and the second group 925 observatons equal to 60.8%, the last group 405 observatons equal to 26.6%, respectvely. As seen, the sub-group 3 persons has an overwhelmng majorty due to the Chnese one chld polcy (Fgure 3.7). 70.0% 60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% 1 or 2 persons 3 persons 4 persons or more Fgure 3.7 Dstrbuton of Famly Sze Accumulated Mleage. Ths varable s a numerc varable and the queston was: what s the accumulated mleage of your car. Accordng to the survey, 80% of the prvate owned autos were bought after 1998 and only 17.6% of autos are second hand. Moreover, the mean value of Accumulated Mleage s ca. 95,000 klometers, that s, the average condton of the autos s qute well, whch may lead to the percentage of not buy n the dependent varable s apprecably hgher than that of buy. Addtonally, because a car s stll a luxury tem for most Chnese famles, Accumulated mleage should play a bg role n the decson to replace a car. In next chapters, ths answer would be tred to fnd out f Accumulated Mleage has sgnfcant nfluence on the dependent varable or not. 12

15 4. Logt Model In many applcatons, logstc regresson s a standard method for explanng a bnary dependent varable that has two categores equal to ether 1 or 0 partly because of ts mathematcal convenence. Logstc regresson s dfferent from lnear regresson, but there are many smlartes. Concepts from lnear regresson would be carred over to logstc regresson. Under the assumpton that a set of ndependent varables x are ncluded n the model, then the dependent varable y can be descrbed as a lnear combnaton of the ndependent varables x and the parameters β plus the error term ε, n form as: y = x β + ε = β + β x β x + ε T j j (4.1) The lnear regresson conssts of two parts: the mean value of the outcome varable that can be expressed as a lnear functon of the ndependent (predctor) varable and the error that attempts to descrbe how ndvdual measurements vary around the mean value. It s assumed that ndvdual responses vary around the mean accordng 2 to a normal dstrbuton wth varance σ. Ths model can be expressed as: Structure on the means: E ( Y X ) = β 0 + β1x β j xj (4.2) 2 Error structure: ε ~ N(0, σ ) (4.3) For a bnary response varable, we assume that prob( = 0) = 1 π, n general we then have: Y Y prob( = 1) = π, so we have E( ) = 0 (1 π ) + 1 π = π (4.4) The expresson (4.2) mples that s possble for the quantty E Y X ) to take on Y ( any value as x ranges from to +. If the dependent varable s a bnomal, ths quantty s constraned to [0,1], whch can be seen n (4.4). The bnary nature of the response also creates dffcultes n how we vew the varablty of ndvdual values around the mean. The varance of a bnary response 13

16 s a functon of the probabltyπ, whch s: Var ( Y ) = π (1 π ) (4.5) The quantty Var( Y ) s the functon of π,.e., the assumpton constant varance 2 σ s volated. So the lnear regresson s not proper for a bnary response varable. The quantty π = E Y X ) s used for logstc dstrbuton n order to smplfy ( notaton, and we have: π = F ( y ) = 1 e + y e y (4.6) Wth y = x β = β + β x β x T j j There s a very mportant property about functon (4.6) that makes t proper to buld up a model for bnary dependent varable, that s, (4.6) s bounded between 0 and 1. Ths wll elmnate the possblty of gettng nonsenscal predctons of proportons or probabltes. From (4.6) we get: 1 π π = e y (4.7) 1 π π s called odds whch represents the relatonshp between the probabltes that dependent varable Y takes the outcome of 1 comparng to the outcome of 0, and odds can also take any postve value and therefore have no celng restrcton. Moreover, there s a lnear model hdden n (4.6) that can be revealed wth a proper transformaton of the response. Ths transformaton called logt transformaton that converts the probablty nto a contnuous varable that s lnear wth respect to 14

17 the explanatory (ndependent) varables (McCullagh and Nelder 1991) s defned n terms of π as followed: y = π... T log( ) = X β = β0 + β1x β jxj (4.8) 1 π From (8) we can see, y s exact the logarthms of 1 π π n form as π log( ) 1 π also called as logt or log-odds. In lnear regresson, the least squares method (LS) s most often use to estmate parameters. But the least-squares regresson approach s plagued wth many statstcal problems for logt model, so the maxmum-lkelhood (ML) fttng procedure s most frequently used (Hosmer and Lemeshow 1989). In general, the ML technque s used to maxmze the log-lkelhood functon, whch ndcates how lkely t s to obtan the observed values of Y, gven the values of the ndependent varables and parameters (Menard 1995). 15

18 5. Stepwse Backward Elmnaton and Bvarate Analyss Ths chapter s about testng of a statstcal hypothess to determne whether the ndependent varables ncluded n the model are sgnfcantly assocated wth the response varable. Stepwse logstc regresson, whch offers a fast and effectve means of screenng a large number of varables, and smultaneously ft a number of logstc regresson equatons, s most often used n stuatons were the mportant ndependent varables are not known and assocatons wth the outcome not well understood (Hosmer and Lemeshow 1989). There are two basc forms of stepwse logstc regresson: forward ncluson and backward elmnaton. Backward elmnaton, more exact, backward elmnaton of Wald s used n ths paper. In backward stepwse elmnaton, the analyss begns wth a model that contans all of the explanatory varables. At each step, the sgnfcance of the explanatory varable beng removed s tested usng the Wald test 2 (Hosmer and Lemeshow 2000; Duncan and Chapman 2003). If a varables p- value s equal to or greater than the sgnfcant level, t wll be elmnated from the model, otherwse, t remans n the model. As a result, the fnal model conssts entrely of varables that are statstcally sgnfcant (Hosmer and Lemeshow 2000). Table (5.1) shows the SPSS output of forward stepwse wth Wald. The model ncludes all the explanatory varables at the frst step and we can see, the p-value of Famly Sze s that s much large than the sgnfcant level, whch ndcates 2 The Wald statstc W = ˆ j β σˆ ( ˆ β ) j follow a ch-square dstrbuton and n ths case, wth one degree of freedom. 16

19 that Famly should be removed from the model, whch s also showed n step 2. Varables n the Equaton B S.E. Wald df Sg. Exp(B) cty gender age educaton ncome occupaton famly mleage Constant cty gender age educaton ncome occupaton mleage Constant a. Varable(s) entered on step 1: cty, gender, age, educaton, ncome, occupaton, famly, mleage. Step 1 a Step 2 a Table 5.1 Forward Stepwse wth Wald But, the model should not totally be based on the results of the Wald test. Hauck and Donner (1977) found out that the Wald test behaves n an aberrant manner, often falng to reject the null hypothess when the coeffcent s sgnfcant. They suggested that the lkelhood rato test should be used for the logt model. For large sample sze, there s practcally no dfference between the results of the lkelhood rato test and Pearson's ch-square test (Rönz 2000). In the next step, Pearson's chsquare test would be used to fnd out f there any more varable except Famly should be excluded from the model. Pearson's ch-square s by far the most common type of ch-square sgnfcance test. Ths statstc s used to test the hypothess of no assocaton of columns and rows n contngency table. In the bnary analyss n ths paper, the assocaton between response varable Decson and every other varable (except Famly) s tred to fnd out based on bnary contngency tables. Pearson's ch-square test s not approprate 17

20 for the contnuous varable, out of convenence, Accumulated Mleage s turned from a contnue varable nto a categorcal one by dvdng t nto 2 groups: under and over the average value 95,000 klometers. Snce ths classfcaton has ensured that each subgroup has enough samples to satsfy the condton of Pearson's chsquare test, other classfcatons would not be tred (Appendx 3). If the p-value of the tested varable s larger than the sgnfcant level, the hypothess would be rejected, that s, ths varable should be elmnated from the model. Table (5.2) ndcates that at 5% sgnfcant level, every ndependent varable s related to the response varable except Gender, meanwhle, Cty, Educaton and Income are statstcally the most sgnfcant ones wth p=0. Accordng to the suggeston of Hosmer und Lemeshow (1989), f the p-value n Pearson's ch-square sgnfcance test s smaller than 0.25,.e., the varable may have n some sense weak effect on the outcome varable, and then t should not be elmnated from the model. So, although the Gender s not sgnfcant, t stll remans n the model. Table 5.2 Pearson s Ch-Square Sgnfcance Test From Table (5.2) we can see that Income s the one of the varables that has s strongly assocated wth the outcome varable Decson. In Chna, more and more ndvduals plan to buy a car, but the dsproportonately hgh cost of new cars compared to ndvdual wealth makes t dffcult for consumers to afford, especally 18

21 a the second purchase. Income s assumed to be the key factor that nfluents the decson of buy or not buy. Fgure (5.3) shows us that for the group RMB, the percentage for not buy s hgher than that of buy, for the group RMB, the percentage for buy precedes that of not buy, even though the dfference s not that sgnfcant. For the thrd group 5001RMB+, ths dfference s even greater wth a bgger margn. Ths result also confrms the presumpton that the correlaton between Income and Decson should be strong (Fgure 5.1). 60.0% decson not buy buy 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% RMB RMB 5001RMB + Fgure 5.1 Decson vs. Income 19

22 6. The ft of the model The ft of the model was tested after the elmnaton to ensure that the model adequately fts the data. As mentoned earler, the Maxmum-lkelhood-method would be used to estmate the parameters β n the logstc regresson. For the bnomal dstrbuted response varable y we have the lkelhood functon as followed: n L( y π y n y π ; y ) = π (1 ) (6.1) Takng the log of (6.1), we get: π n l( π ; y ) = y log + n log(1 π ) + log (6.2) 1 π y So the jont lkelhood-functon s: π n l( π ; y) = l( π ; y ) = y log + n log(1 π ) + log (6.3) 1 π y Devance n notaton of D s an mportant statstc n some approaches to assessment of goodness-of-ft, defned as: D l( ˆ; π y) 2 [ l( ˆ π ; y) l( ˆ; π y) ] ~ χ = 2 log = 2 sat ( j+ 1) l( ˆ π sat; y) (6.4) l( ˆ; π y) : lkelhood of the current model l( ˆ π ; y) : lkelhood of the saturated model 3 sat ' y = ( y1, y2,..., ' x = ( x1, x2,..., y x j ) ) The role of devance n logstc regresson s as same as SSE (resdual sum-ofsquares) n lnear regresson. Actually, devance s exactly equal to SSE when 3 Saturated model s one that contans as many parameters as there are data ponts. The log-lkelhood-functon y y n for the saturated model s: + l(π ˆ sat ; y) = y log ( n y )log 1 + log n n y 20

23 computed for lnear regresson. Now we assume s the devance for the model M whch has a set of D0 0 ' ndependent varable x = ( x1, x2,..., x j ), analog, D1 s the devance for the model M 1 M1 ' whch has a set of ndependent varable x, x,..., ), then M and are nested models. The statstc x = ( 1 2 x j + p 0 Δ D, also refer to as the lkelhood rato test, would be appled to answer the queston Does the model wth the varable gve us more nformaton about the response varable than that wthout the varable. It s a close analogue to the F statstc for lnear regresson. [ l( ˆ π ; y) l ( ˆ; π y) ] 2[ l( π ; y) l ( ˆ; π y) ] = 2[ l ( ˆ; π y) l ( ˆ; π )] Δ D = D (6.5) 0 D1 = 2 sat 0 sat y The degrees of freedom of Δ D s: M 0 1 (degrees of freedom of ) (degrees of freedom of M ) = [ ( j + 1) ] [ ( j + p + 1) ] = p 2 Snce and D are both χ -dstrbuted, D0 1 2 Δ D s also χ -dstrbuted wth p degrees of freedom. If the p-value exceeds the sgnfcant level, t s concluded that the reduced model s as good as the full model. The SPSS statstcal package presents not the log-lkelhood tself but the loglkelhood multpled by 2 (SPSS Inc. 1998). Output from SPSS denotes loglkelhood multpled by -2 as -2 Log Lkelhood (Appendx 4). By multplyng the log-lkelhood by -2 t approxmates a ch-square dstrbuton (Menard 1995). As showed n chapter 5, Famly Sze should be excluded from the model. Now Δ D would be calculated to compare the two models wth and wthout Famly Sze to see the goodness-of-ft of the reduced model: ΔD = 2[lkelhood of full model - lkelhood of reduced model] = = (6.6) In ths case, the freedom of degree of Δ D s 1, snce only one varable s removed from the model. The statstc value (6.6) s larger than the crtcal value at 10% sgnfcant level equal to 2.71, demonstratng that Famly Sze adds lttle to the model once the other varables have been taken nto the model. 21

24 7. Interpretaton of the model The ndependent varables, except Accumulated Mleage, are categorcal varables. These varables would be analyzed as a seres of ndcator varables to correctly evaluate ther mportance n the model. For the varable wth k categores, k 1 ndcator varables must be constructed. In the data, for example, Educaton has two categores, not attend un and attend un, and needed only one ndcator varable. Meanwhle, Income has 3 categores and therefore two ndcator varables are necessary for our analyss of ths ndependent varable. Here, every frst category of the categorcal dependent varables s taken as the reference category (Appendx 5). Table 7.1 shows us how mportant the other categores of the categorcal varables are to the outcome varable. Snce Gender s not sgnfcant to 5% wth p-value 0.113, the program proceeds to the second step. Gender s removed from the model n the second step, whch s consstent wth the result of Pearson's ch-square sgnfcance test (see Table 5.2). In the second step we can see, Cty has sgnfcant nfluence to response varable Decson at 5% sgnfcance level, but n the subcategores, only Shangha and Fuzhou are sgnfcant. The ndependent varable has no nfluence on the response varable f the odds rato, whch s presented n SPSS as Exp (B),equals to 1. The larger the dfference s between the observed odds rato and 1.0, the stronger the relatonshp s. Here, the odds rato of 1.6 ndcates a moderate relatonshp. Accordng to the odds ratos, we also can say, for those who lve n Shangha, the probablty of plannng to purchase a second car n next 5 years s about 1.6 tmes hgher than n Bejng. So, only based on the decson to purchase a second car, the automoble market n Shangha has more potental compared to Bejng. Meanwhle, those n Fuzhou are 33% less lkely to plan to buy another car than those n Bejng. The p-value of Age s 0.013, whch results to the concluson that t s strongly assocated wth the outcome varable Decson. Broadly speakng, younger people 22

25 show stronger desre to purchase a second auto compared to older people. The age group s not sgnfcant, wth a p-value equal to 0.180, whch exceeds the crtcal p-value ( α = 0. 1). The age group 45+ s about 58.6% less lkely to answer yes to do you plan to renew your car n next 5 years when compared to the age group Educaton shows also sgnfcance to the response varable wth a p-value equal to The group wth a unversty degree s about 1.35 tmes more lkely to plan a second auto purchase than the group wthout one. Income plays a very mportant role n consumer decsons. It s also confrmed to be true accordng to the analyss that the p-value of Income s 0, that s, Income s one of the two varables that are most sgnfcant to the outcome varable. In ths varable, both subgroups and are strongly assocated wth Decson. The result shows us that the probablty for the group wth ncome of RMB s 1.38 hgher than the group wth lower ncome of RMB consderng the second auto purchase. And the probablty for group 5001+RMB s 1.82 tmes hgher than that for group RMB. The assocaton between Occupaton and Decson s strong as t can be seen from the p-value equal to 0.014, but only one subcategory employer of ths varable s sgnfcant. From the value of Exp(B), t s concluded that the probablty for employer to replace ther cars s 1.32 tmes more than the people who work for government. As showed, the p-value of Accumulated Mleage equals to 0.006, whch confrms the presumpton that ths varable should have sgnfcant nfluence on the dependent varable. The coeffcent of Accumulated Mleage s 0.015, the odds rato Exp (b) s equvalently equal to as showed n the output. An odds rato above 1 ndcates an ncrease, n ths case, the odds rato equals to 1.015, t s sad that when the Accumulated Mleage ncreases one unt, the odds that the dependent = 1 ncrease by a factor of when other varables are controlled,.e., the probablty to purchase a second car would ncrease 1.5%. 23

26 The analyss shows that the varables Cty, Age, Educaton, Income, Occupaton and Accumulated Mleage have nfluence on the dependent varable. General speakng, those who are younger, better educated and have hgher ncome are more lkely to have a second car n next 5 years. Step 1 a Step 2 a cty Shangha Tanjn Fuzhou Hangzhou Nanjng Guangzhou gender(male) age educaton(attend un) ncome occupaton employer employee others mleage Constant cty Shangha Tanjn Fuzhou Hangzhou Nanjng Guangzhou age educaton(attend un) ncome occupaton employer employee others mleage Constant Varables n the Equaton B S.E. Wald df Sg. Exp(b) a. Varable(s) entered on step 1: cty, gender, age, educaton, ncome, occupaton, mleage. Varables not n the Equaton Step 2 a Varables gender(male) Score df Sg. Overall Statstcs a. Varable(s) removed on step 2: gender. Table 7.1: Logt Model done wth subcategores 24

27 8. Some Dscussons about Chnese Auto market As mentoned n last chapter, the desre for younger people to purchase a second car s stronger than older people. Accordng to the bnary cross tabulaton between Income and Age, 59.1% of group s labeled wth the monthly ncome above 3500RMB whle 64.1% of group 45+ earns no more than 3500RMB a month, so the ncome level may be one obstacle for older people to renew ther cars (Appendx 6). And also, the hgher educated people have hgher probablty to purchase a second car or to renew ther cars than the lower educated ones. One of the explanatons for such a dfference may be found out from the bnary analyss of Income and Educaton: 62% of the people who dd not go to unversty belong to the ncome level of RMB whle the ones ownng a unversty degree have much hgher probablty to earn more, more explctly, 60.2% of the group who enter unversty earn more than 3500RMB n month (Appendx 7). Hgher ncome enables the hgher educated people have more purchasng power to replace ther cars. In Chna, the monthly ncome of the government servce people s relatvely low compared to employers: 54.5% of government servce people belong to the earnng level of RMB whle 64.9% of employers earn more than 3500RMB (Appendx 8). Obvously, the ncome level of the government servce people reduces the probablty to renew ther cars. Addtonally, In order to know more about the Chnese auto market whch s reflected drectly by auto consumer behavor, two varables from the thrd part of the survey Consumer car purchasng behavor are chosen. I. Second-Hand Auto market. Ths varable has two categores: second-hand car and new car. The queston for ths varable was: s the car that you bought a new car or a second-hand car? Accordng to the data of ths survey, 80% of the cars owned by the Chnese ctzens were purchased after the year of 1998, n whch 82.4% were 25

28 brand-new when purchased. Only 17.6% of the car owners chose to buy secondhand cars (Fgure 8.1). Therefore, the average accumulated runnng dstance of the cars was less than 100,000 klometers when the cars were under nvestgaton, n another word, the cars were n good condton. Based on ths pont, presently most of the car owners have no plan of changng ther cars and an estmated clmax of second round car purchase would occur n 5 years due to car owners desre to replacng the old cars. Ths fact also results n the hgher percentage of car owners who do not plan to buy a new car % 80.0% 60.0% 40.0% 20.0% 0.0% old car new car Fgure 8.1 Old Car or New Car As t can be concluded from the data shown above, once a decson of purchasng a car s made, most of the customers are more wllng to buy a new car. Obvously we must see that a prvate car s stll a very new tem n the household budget for a normal Chnese famly. On one hand, the ones who can afford prvate cars are the people who frst lead affluent lves n Chna; those people are often n possesson of consderable wealth and car purchasng s not much of a burden for them. So t s very reasonable that they lean more towards a brand-new self-owned car; on the other hand, because the car market s relatvely young n Chna, second-hand car market lags behnd because there are not many used cars yet. That lmts the sale and purchase of second-hand cars. The potental second-hand car users have no used car to buy: the supply s just much smaller than the demand. In the future ths stuaton wll change tself thanks to the boomng of the car market and used car 26

29 market, as another mportant opton for the potental car buyer, wll also start to develop on an approprate and sold foundaton. II. Fnancng n Chnese Auto market The queston for ths varable was: whch s your payment for your car? 410 of the ntervewees took the answer of dvded payment and 1116 of full payment. In Chna, the car fnancng market s qute underdeveloped, wth only 15% of cars beng fnanced, n comparson to more than 80% n USA (Data source: Goldman Sachs, 2003), whch s also proved by the data of ths survey. Focusng on the means of how to pay the bll, a pont s clear: most of the customers prefer more to clear the car bll all at once, payng the debt to the bank every month s not ther frst choce. 79.4% of the car owners chose to pay for ther cars n full (Table 8.1). In recent years, a tendency emerges that monthly or yearly nstallments nterest more and more customers and more people choose to pay for the cars by these means. Wth the possblty to fnance a car purchase, many famles have the opportunty to own prvate cars earler than they ever thought. Accordng to the data from nvestgaton, more than 20% of the customers pay for ther cars by nstallments and therefore sooner than estmated have ther own prvate cars (Table 8.1). payment Vald dvded payment full payment Total Cumulatve Frequency Percent Vald Percent Percent Table 8.1: Dstrbuton of Payment In Table 8.2 we can see, the percentage dstrbuton of dvded payment and full payment for the under category Bejng of Cty are 17% and 83% whch are about equal to those of Shangha and Fuzhou. The ntervewees out from Tnajn, Nanjn and Guangzhou are more lkely to pay for ther cars through dvded payments. 27

30 Table 8.2 Dvded Payment vs. Full Payment The dfference of ths percentage dstrbuton for the under categores of other varables s not that sgnfcant. Accordng to the results above, generally speakng, male consumers prefer to pay n full more than female consumers, young consumers more than old consumers, hgh-educated consumers more than low-educated consumers and consumers wth hgher ncomes more than those wth lower ncomes, although the dfference s qute slght. Some new fnancng polces have been ssued by Chnese government n recent years n order to encourage the potental car buyers to fnance ther purchase by combnng cash and loans. Based on the Chnese government s effort gven to auto fnancng market and Chnese people s strong desre for cars, we may say that the Chnese car fnancng market would develop nto a new stage n next several years. 28

31 Appendx: 1. Frequency dstrbuton of Decson Vald Mssng Total no buy buy Total System decson Cumulatve Frequency Percent Vald Percent Percent The wealth s concentrated n urban and eastern coast areas. Bejng, Tanjn and Shangha: drectly under the jursdcton of the central government Nanjng: the captal cty of Jangsu provnce Hangzhou: the captal cty of Zhejang provnce Fuzhou: the captal cty of Fujan provnce Guangzhou: the captal cty of Guangdong provnce Source: Natonal Bureau of Statstcs of Chna. Chna Statstcal Yearbook:

32 3. Accumulated Mleage under and over Average Accumulated Mleage Under and Over Average Vald Mssng Total under over Total System Cumulatve Frequency Percent Vald Percent Percent Log- lkelhood values of full model and reduced model Full model: Step 1 Model Summary -2 Log Cox & Snell Nagelkerke lkelhood R Square R Square a a. Estmaton termnated at teraton number 3 because parameter estmates changed by less than.001. Reduced model: Step 1 Model Summary -2 Log Cox & Snell Nagelkerke lkelhood R Square R Square a a. Estmaton termnated at teraton number 3 because parameter estmates changed by less than

33 5. The categorcal varables Codng Categorcal Varables Codngs cty occupaton age ncome educaton gender Bejng Shangha Tanjn Fuzhou Hangzhou Nanjng Guangzhou government servce employer employee others RMB RMB 5001RMB + not attend un attend un female male Parameter codng Frequency (1) (2) (3) (4) (5) (6) Cross tabulaton between Income and Age ncome * age Crosstabulaton % wthn age ncome RMB RMB 5001RMB + Total age Total 40.9% 49.8% 64.1% 49.7% 22.7% 19.1% 16.5% 19.8% 36.4% 31.1% 19.4% 30.5% 100.0% 100.0% 100.0% 100.0% 7. Cross tabulaton between Income and Educaton ncome * educaton Crosstabulaton % wthn educaton ncome RMB RMB 5001RMB + Total educaton not attend un attend un Total 62.0% 39.8% 49.8% 16.1% 22.7% 19.7% 21.9% 37.5% 30.5% 100.0% 100.0% 100.0% 31

34 8. Income vs. Occupaton ncome * occupaton Crosstabulaton % wthn occupaton ncome RMB RMB 5001RMB + Total occupaton government servce employer employee others Total 54.5% 35.1% 56.4% 55.8% 49.7% 18.8% 22.2% 19.1% 18.4% 19.7% 26.8% 42.7% 24.6% 25.9% 30.5% 100.0% 100.0% 100.0% 100.0% 100.0% 32

35 References Asa Case Research Centre (2005), Chna s Automotve Industry, the unversty of Hongkong: 3-7 Backhaus, K., Erchson, B., Plnke, W. und Weber, R. (2005). Multvarate Analyse methoden. Berln: Sprnger Chna Daly, (2002), Prce Cuts Boost Auto Sales, Aprl 15. Global Insght, (2003), Asan Automotve Industry Forecast Report: Hayes, K., Warburton, M., Lapdus, G., Shohara, K., Chang, Y., McKenna, S. (February, 2003). The Chnese Auto Industry, Global Automobles, Goldman Sachs Equty Research: 1-24 Hosmer, D. und Lemeshow, S. (1989). Appled Logstc Regresson. New York: John Wley & Sons Laurence W. Carstensen Jr., Char, Campbell, J., Shfflett, C. (1999). Predctve Probablty Model for Amercan Cvl War Fortfcatons Usng A Geographc Informaton System: L.L. Qu, L. Turner, L. Smyrk, A Study of Changes n the Chnese Automotve Market Resultng from WTO Entry: Vctora Unversty, AUSTRALIA Long, J.S. (1997). Regresson models for categorcal and lmted dependent varables. Thousand Oaks, CA: Sage 33

36 N, B. (2005). The Development of Chnese Auto Market. Shangha Auto, September, 2005 Rönz, B. (2001). Computer-aded Statstcs II ( Rönz, B. (2001). General Lnear Model ( Wang, XT., Guan SF. (2003). The Current Stuaton and Development of Chnese Auto Market. Bejng Auto, February,