DR.

HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB

DEMAND ESTIMATION After studying this chapter, you should be able to: 1. Discuss how the firms managers use the information about demand for its product to determine correctly its profit-maximizing rate of output and price, or whether to produce a particular product at all. 2. Discuss demand respond to consumer income increase or decrease as a result of an economic expansion or contraction. 3. pecify the components of a regression model that can be used to estimate a demand e!uation. ". #nterpret the regression results $i.e., explain the !uantitati%e impact that changes in the determinants ha%e on the !uantity demanded&. '. (xplain the meaning of )2. *. (%aluate the statistical significance of the regression coefficients using the t-test and the statistical significance of )2 using the +-test. Introduction: ,n important contributor to firm ris- arises from sudden shifts in demand for the product or ser%ice. Demand estimation ser%es two managerial ob.ecti%es/ $1& it pro%ides the insights necessary for effecti%e management of demand, and $2& it aids in forecasting sales and re%enues. The theory SIMP E INEA! !E"!ESSION )elationships, among other things, may ser%e as a basis for estimation and prediction.

Si#ple prediction0when we ta-e the obser%ed %alues of X to estimate or predict corresponding Y %alues. !egression analysis uses simple and multiple predictors to predict Y from X %alues. 1ith respect to similarities and differences of correlation and regression, their relatedness would suggest that beneath many correlation problems is a regression analysis that could pro%ide further insight about the relationship of Y with X. The $asic Model , straight line is fundamentally the best way to model the relationship between two continuous %ariables. !egression coefficients are the intercept and slope coefficients.

Slope %&'(0the change in Y for a 1-unit change in X. 2his is the ratio of change $3& in the rise of the line relati%e to the run or tra%el along the X axis.

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB

Intercept %&)(0one of two regression coefficients, is the %alue for the linear function when it crosses the Y axis or the estimate of Y when X is zero. *oncept Application 4nfortunately, one rarely comes across a data set composed of four paired %alues, a perfect correlation, and an easily drawn line.

, model based on such data is deterministic in that for any %alue of X, there is only one possible corresponding %alue of Y. , probabilistic model also uses a linear function. Error ter# is the de%iations of %alues of Y from the regression line of Y for a particular %alue of X.

Method of east S+uares 2he #ethod of least s+uares is a procedure for finding a regression line that -eeps errors of estimate to a minimum.

1hen we predict the %alues for Y for each Xi the difference between the actual Yi and the predicted Y is the error. 2his error is then s!uared and then summed. !esiduals , residual is the difference between the regression line %alue of Y and the real Y %alue. 1hen standardized, residuals are comparable to Z scores with a mean of 5 and a standard de%iation of 1. #t is important to apply other diagnostics to %erify that the regression assumptions $normality, linearity, e!uality of %ariance and independence of error& are met.

Predictions Prediction and confidence bands are bow-tie shaped confidence inter%al around a predictor. 6onfidence inter%als can be expanded or narrowed.

Testing for "oodness of ,it "oodness of fit is a measure of how well the regression model is able to predict Y.

2he most important test in bi%ariate linear regression is whether the slope,71, is e!ual to zero. 8ero slopes result from %arious conditions/

Y is completely unrelated to X, and no systematic pattern is e%ident.

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB

2here are constant %alues of Y for e%ery %alue of X. 2he data are related but represented by a nonlinear function. The t-Test

2o test whether 71 9 5, we use a two-tailed test. The , Test

2he F test has an o%erall role for the model in multiple regressions. ee F test example for an illustration. *oefficient of Deter#ination

#n predicting the %alues of Y without any -nowledge of X, our best estimate be Y mean. (ach predicted %alue that does not fall on Y contributes to an error estimate.

Multiple !egression Multiple regression0statistical tool used to de%elop a self-weighting estimating e!uation that predicts %alues for a dependent %ariable from the %alues of independent %ariables. :ultiple regression is used as a descripti%e tool in three types of situations/

#t is often used to de%elop a self-weighting estimating e!uation by which to predict %alues for a criterion %ariable $D;& from the %alues for se%eral predictor %ariables $#;s&.

, description application of multiple regression calls for controlling for confounding %ariables to better e%aluate the contribution of other %ariables. :ultiple regression can be also used to test and explain causal theories. 2his approach is referred to as path analysis $e.g., describes, through regression, an entire structure of lin-ages ad%anced by a causal theory&. :ultiple regression is also used as an inference tool to test hypotheses and to estimate population %alues. Method :ultiple regression is an extension of the bi%ariate linear regression discussed in 6hapter 1<. Du##y .ariables0nominal %ariables con%erted for use in multi%ariate statistics. )egression coefficients are stated either in raw score units $the actual X %alues& or standardi/ed coefficients $regression coefficients in standardized form =mean 9 5> used to determine the comparati%e impact of %ariables that come from different scales.

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB
1hen regression coefficients are standardized, they are called beta 0eights %&( $standardized regression coefficients where the size of the number reflects the le%el of influence X exerts on Y&, and their %alues indicate the relati%e importance of the associated X %alues, particularly when the predictors are unrelated. E1a#ple :ost statistical pac-ages pro%ide %arious methods for selecting %ariables for the e!uation.

,or0ard selection0se!uentially adds the %ariable to a regression model that results in the largest R2 increase. $ac20ard eli#ination0se!uentially remo%es the %ariable from a regression model that changes R2 the least. Step0ise selection0a method for se!uentially adding or remo%ing %ariables from a regression model to optimize R2 . *ollinearity0when two independent %ariables are highly correlated. Multicollinearity0when more than two independent %ariables are highly correlated. ?oth of the abo%e can ha%e damaging effects on multiple regression. ,nother difficulty with regression occurs when researchers fail to e%aluate the e!uation with data beyond those used originally to calculate it. , solution to the abo%e problem can be the holdout sa#ple $the portion of the sample excludes for later %alidity testing when the estimating e!uation is first computed&.

?ased on the formula $see chapter&, the coefficient of determination is the ratio of the line of best fits error that incurred by using Y. @ne purpose of testing is to disco%er whether the regression e!uation is a re effecti%e predicti%e de%ice than the mean of the dependent %ariable. 2he coefficient of determination is symbolized by r s!uared. #t has se%eral purposes/

,s an index of fit, it is interpreted as the total proportion of %ariance in Y explained by X. ,s a measure of linear relationship, it tells us how well the regression line fits the data. #t is also an important indicator of the predicti%e accuracy of the e!uation. 2ypically, we would li-e to ha%e an r s!uared that explains A5 percent or more of the %ariation.

I#portant *oncepts: Indi.idual De#and *ur.e the greatest !uantity of a good demanded at each price the consumers are willing to buy, holding other influences constant

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB
The Mar2et De#and *ur.e is the horizontal sum of the indi%idual demand cur%es. The De#and ,unction includes all %ariables that influence the !uantity demanded 3 4 f% P, Ps, Pc, 5, N, 6, PE( 7 7 8 7 8 7 0here: B is price of the good B is the price of substitute goods B6 is the price of related goods C is income, D is population, 1 is wealth, and B( is the expected future price Downward lope to the Demand 6ur%e )easons that price and !uantity are negati%ely related include/ income effect -- as the price of a good declines, the consumer can purchase more of all goods since his or her real income increased. substitution effect -- as the price declines, the good becomes relati%ely cheaper. , rational consumer maximizes satisfaction by reorganizing consumption until the marginal utility in each good per dollar is e!ual. Sign of the esti#ated !egression *oefficients , good regression model should be based on a good economic theory. 2he theory should indicate what sign each estimated coefficient must ta-e. +or example, the coefficient for the price %ariable in a demand e!uation should ha%e negati%e sign, that is, when price increases, demand decreases. 2he income %ariable should ha%e a positi%e sign. #f the signs of estimated coefficients do not agree with the theory, the %alidity of the model should be !uestioned. 9o0 to do De#and Esti#ation8 #n estimating the demand for a particular good or ser%ice, the process will be/ +irst step/ determine all the factors that might influence this demand $i.e the formation of Demand model&. (xample/ uppose we wanted to estimate the demand for pizza by uni%ersity students in :alaysia. 1hat %ariables would most li-ely affect their demand for pizzaE )emember demand theoryE 1e could start to answer this !uestion by using price and all the nonprice determinants F such as income, prices of related goods, taste and preferences, future expectation and number of buyers. ?ut it is not always possible or appropriate to include all these %ariables in a particular demand estimation. 1hyE +actors of the a%ailability of data and the cost of generating new data. 2he two types of data used in regression analysis are cross-sectional and time series. +or the purpose of illustration, let us assume we ha%e obtained cross-sectional data on uni%ersity students $Bublic and Bri%ate 4ni%ersity in :alaysia& by conducting a sur%ey of thirty randomly selected 4ni%ersity during a particular month. econd step/ Data 6ollection uppose we ha%e gathered the following information for each campus from this sur%ey/

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB
$1& $2& $3& $"& $'& a%erage number of slices $!uantity& consumed per month by students, a%erage price of a slice of pizza in places selling pizza annual income $B2BD and +,:,& a%erage price of soft drin- sold in the pizza places, and location of the campus $urban %ersus rural&

2he data obtained from our hypothetical sur%ey are presented in 2able 1.

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB

2able 1.

ample data/ 2he demand for Bizza

BriceGB #ncomeGC B 6omGBc Hoc I" JuantityDD 1 2 3 " ' * K A < 15 11 12 13 1" 1' 1* 1K 1A 1< 25 21 22 23 2" 2' 2* 2K 2A 2< 35 155.55 155.55 <5.55 <'.55 115.55 12'.55 12'.55 1'5.55 A5.55 A5.55 <5.55 155.55 155.55 115.55 12'.55 115.55 1'5.55 155.55 1'5.55 1'5.55 1'5.55 12'.55 12'.55 155.55 K'.55 155.55 115.55 12'.55 1'5.55 1'5.55 1".55 1*.55 A.55 K.55 11.55 '.55 12.55 15.55 1A.55 12.55 *.55 '.55 12.55 15.55 1".55 1'.55 1*.55 12.55 12.55 15.55 13.55 1'.55 1*.55 1K.55 15.55 12.55 *.55 15.55 A.55 15.55 155.55 <'.55 115.55 <5.55 155.55 155.55 12'.55 1'5.55 155.55 <5.55 A5.55 K'.55 155.55 12'.55 135.55 A5.55 <5.55 <'.55 155.55 <5.55 <'.55 155.55 <'.55 155.55 155.55 115.55 12'.55 <5.55 A5.55 <'.55 1.55 1.55 1.55 1.55 .55 .55 1.55 .55 1.55 1.55 1.55 1.55 1.55 .55 .55 1.55 .55 1.55 .55 .55 .55 1.55 1.55 .55 1.55 1.55 .55 .55 .55 .55 15 12 13 1" < A " 3 1' 12 13 1" 12 15 15 12 11 12 15 A < 15 11 12 13 15 < A A A

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB
2hird step/ Data ,nalysis 2o estimate the demand for pizza, we employed the regression function contained in B . 2he result )egression - Demand (stimation/ imple )egression ,nalysis
:ariables Entered;!e#o.ed%b( ;ariables (ntered HocationGI ", 2uitionGI2, BriG6rossG I3, BriceGI1$a& ;ariables )emo%ed

:odel 1

:ethod

(nter

a ,ll re!uested %ariables entered. b Dependent ;ariable/ JuantityGC Model Su##ary ,d.usted ) td. (rror of ) ) !uare !uare the (stimate .A"*$a& .K1K .*K1 1.*"5"A a Bredictors/ $6onstant&, HocationGI", 2uitionGI2, BriG6rossGI3, BriceGI1 :odel 1 ANO:A%b( um of !uares )egressio n )esidual 2otal 1K5.5AK *K.2K<

:odel 1

df " 2'

:ean !uare "2.'22 2.*<1

+ 1'.A51

ig. .555$a&

23K.3*K 2< a Bredictors/ $6onstant&, HocationGI", 2uitionGI2, BriG6rossGI3, BriceGI1 b Dependent ;ariable/ JuantityGC *oefficients%a( 4nstandardized 6oefficients :odel 1 $6onstant& BriceGI1 2uitionGI2 ? 2*.**K -.5AA .13A td. (rror 3.2KA .51A .5AK .51< .AA' tandardized 6oefficients ?eta -.K33 .1K" -."3A -.5<K t A.13' -".A'A 1.'<' -3.<"A -.*1' ig. .555 .555 .123 .551 .'""

BriG6rossG -.5K* I3 HocationGI -.'"" " a Dependent ;ariable/ JuantityGC

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB Fourth Step: Testify the Validity of the independent variables A good o!"#$%&'(g )od"* )$+ (o& %$&'% + $** &," %&$&'%&'#$* &"%&% $(d -.,o*d $** &," -(d"!*+'(g $%%-).&'o(%. R"%"$!#,"!% #$( /'%-$**+ "0$)'(" &," &$1*" o )od"* %&$&'%&'#%, &o d"&"!)'(" ' &," o**o2'(g #!'&"!'$ $!" )"&. 3," #!'&"!'$ /$!+ 2'&, &," (-)1"! o '(d"."(d"(& /$!'$1*"% $% 2"** $% 2'&, &," (-)1"! o o1%"!/$&'o(%. 3," o**o2'(g !-*" o &,-)1 '% 1$%"d o( $ )od"* '(#*-d'(g &,!"" '(d"."(d"(& /$!'$1*"%, &,'!&+ d$&$ .o'(&% $(d &," 455 #o( 'd"(#" *"/"*6 R26 3," /$*-" o R2 $**% 1"&2""( 7 $(d 1. H'g,"! &," /$*-" 1"&&"! '% &," #o!!"*$&'o( 1"&2""( &," '(d"."(d"(& /$!'$1*"% -%"d '( $ )od"*. 3," -%"!% %,o-*d 1" %o(&"(& 2'&, 2,"( &," /$*-" o R2 '% g!"$&"! &,$( 7.4. F8&"%&6 3," #$*#-*$&"d F8/$*-" %,o-*d 1" g!"$&"! &,$( 3. I (o&, &," "%&')$&"d )od"* do"% (o& !".!"%"(& $ good #$-%$* !"*$&'o(%,'. 1"&2""( $(d '(d"."(d"(& /$!'$1*"%. 3,'% $ &"%& o &," o/"!$** %o-(d("%% o $ )od"*. &8&"%&6 3," #$*#-*$&"d &8/$*-"% o! &," !"g!"%%'o( #o" '#'"(&% %,o-*d 1" g!"$&"! &,$( 2 '( $1%o*-&" &"!)%. 3," &8/$*-" )"$%-!"% &," %'g(' $(#" o '(d'/'d-$* !"g!"%%'o( #o" '#'"(&. S&$(d$!d E!!o! o R"g!"%%'o(6 S)$**"! &," %&$(d$!d "!!o! o !"g!"%%'o(, 1"&&"! 2'** 1" &," $##-!$#+ o o!"#$%&%.

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB F' &, S&".6 D$&$ I(&"!.!"&$&'o( F'(d &," .o'(& .!'#" "*$%&'#'&+, &," .o'(& '(#o)" "*$%&'#'&+, $(d &," .o'(& #!o%%8.!'#" "*$%&'#'&+ $& 9:177, ;:14, 9% :117, $(d '( U!1$( A!"$ <Lo# : 1= ' &," d")$(d -(#&'o( 2"!" "%&')$&"d &o 1"6 QD = 26.67 . !6"# $ .%&!"' . 76"#( ) .*++.,o(

I% &," d")$(d o! &,'% .!od-#& <.'>>$= "*$%&'# o! '("*$%&'#? I% '& $ *-0-!+ o! $ ("#"%%'&+? Do"% &,'% .!od-#& ,$/" $ #*o%" %-1%&'&-&" o! #o).*")"(&? F'(d &," .o'(& "*$%&'#'&'"% o d")$(d. L"& -% $%%-)" &," "0.*$($&o!+ /$!'$1*"% ,$/" &," o**o2'(g /$*-"%6 9!'#" o 9'>>$ <9= : 177 <'.".,RM177= I(#o)" <;= : 14 <'." RM14,777= 9!'#" o So & D!'(@ <9#= : 117 <'." RM1.17= Lo#$&'o( <Lo#= : U!1$( A!"$ :1 A(%2"! A F'!%& '(d &," B-$(&'&+ $& &,"%" .!'#"% $(d '(#o)"6 CD : 26.67 . !6"-% . $ .%&!"-%+. . 76"-%% . ) .*++.-%. : 17.848 A 9!'#" "*$%&'#'&+ ED : <CD9=<9DC= : <8 . !6=<177D17.848= : 87.78 2,'#, '% '("*$%&'#

I(#o)" E*$%&'#'&+ E; : <CD ;=<;DC= : < .%&!=<14D17.848= : E.177 2,'#, '% $ (o!)$* good, 1-& $ ("#"%%'&+ C!o%%8.!'#" "*$%&'#'&+ EAB : <CAD 9B=<9B DCA= : < . 76=<117D17.848= : 8.767 2,'#, '% $ #o).*')"(&$!+

Si/ Step: 0on(lusions 0o1bined 2ffe(t of De1and 2lasti(ities 2/a1ple: The fir1 (an use these elasti(ities to fore(ast the de1and for their produ(t -(offee. ne/t year. A A Fir1 3'4 has a pri(e elasti(ity of )2 for (offee Fir1 3'4 have an in(o1e elasti(ity of %.*

17

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB A A The (ross pri(e elasti(ity is $.* Mo%& )$($g"!% '(d &,$& .!'#"% $(d '(#o)" #,$(g" "/"!+ +"$!. 3," #o)1'("d " "#& o %"/"!$* #,$(g"% $!" $dd'&'/". 5DQ = 2D-5 D#. $ 2'-5 D'. $ 23-5 D#6. F A A 2,"!" 9 '% .!'#", ; '% '(#o)", $(d 9R '% &," .!'#" o $ !"*$&"d good.

I +o- @("2 &," .!'#", '(#o)", $(d #!o%% .!'#" "*$%&'#'&'"%, &,"( +o- #$( o!"#$%& &," ."!#"(&$g" #,$(g"% '( B-$(&'&+. G,$& 2'** ,$.."( &o &," B-$(&'&+ %o*d ' +o- !$'%" .!'#" 35, '(#o)" !'%"% 25, $(d .!'#" o %-1%&'&-&" good% !$'%"% '&% .!'#" 15? F 5DC : E9 A 5D9 EE; A 5D; E EH A 5D90 F F F : 82 A 35 E 1.5 A 25 E.57 A 15 : 865 E 35 E .55 5DC : 82.55. G" "0."#& %$*"% &o d"#*'(".

C6 A6

G'** 3o&$* R"/"(-" o! +o-! .!od-#& !'%" o! $**? 3o&$* !"/"(-" 2'** !'%" %*'g,&*+ <$1o-& E .55=, $% &," .!'#" 2"(& -. 35 $(d &," B-$(&'&+ o #o " %o*d 2'** $** 2.55.

A%%"%%)"(& o Mod"* 9"! o!)$(#" I( 1-%'("%% o!"#$%&'(g, $ !"%.o(%" /$!'$1*" '% o &"( d!'/"( 1+ )$(+ o&,"! /$!'$1*"%. A good o!"#$%&'(g )od"* do"% (o& ,$/" &o '(#*-d" $** o &," !"*"/$(& /$!'$1*"%. G,"( $ )od"* $&&$'(% '&% o.&')$* ."! o!)$(#", '(#*-%'o( o $dd'&'o($* /$!'$1*"% %').*+ #o).*'#$&"% &," &$%@ o o!"#$%&'(g. B-& &,"+ do (o& $dd $(+&,'(g &o &," $##-!$#+. I &2o )od"*% +'"*d &," %$)" o!"#$%& $##-!$#+, &," o(" 2,'#, #o(&$'(% "2"! /$!'$1*"% %,o-*d 1" #,o%"(.

11

DR. HUSSIN ABDULLAH SCHOOL OF ECONOMICS, FINANCE AND BANKING, UUM COB
B)@?H(: 3<ESTION ' 1. Lusin dn. ?erhad is the ma-er of a high-!uality 2ong-at ,li. , linear regression model used to estimate the demand function for Lusin 2ong-at ,li yielded the following results/ JD 9 15, "2' F 2,<15 BI M 5.52A , M 11,155B@B $2.AA& $K& $3.13& .(.(. 9 3."

)2 9 5.A1

1here JD 9 !uantity of :ash 2ong-at ,li demanded BI 9 price of :ash 2ong-at ,li , 9 Lusin dn. ?hd. ,d%ertising in dollars B@B 9 percentage of the :alaysia population o%er 21 years of age $i& $ii& $iii& Determine the point price elasticity for prices of ):' and ):15, when , 9 ):1,555,555 and B@B 9 .5'. $15 mar-s& Determine the point ad%ertising elasticity at an ad%ertising le%el of ):2,555,555, if price remain at ):' and B@B 9 .5'. $' mar-s& 2he 2-statistic %alue for each coefficient is gi%en in parentheses. #f you -now that the demand function was estimated using 2' obser%ations, can you re.ect at the <'-percent confidence le%el the hypothesis that there is no relationship between each of the independent %ariables and JDE $' mar-s& 1hat steps are usually in%ol%ed in the estimation of a demand e!uation by regression analysisE $' mar-s&

$i%&

3<ESTION = ;i%ian :a.u egar dn. ?hd. $;: & has hired you as a consultant to analyze the demand for its line of telecommunications de%ices in 3' different mar-et areas. 2he a%ailable data set includes obser%ations on the number of thousands of units sold by ;: per month $J I&, the price per unit charged by ;: $BI&, the a%erage unit price of competing brands $B 8&, monthly ad%ertising expenditures by ;: $,&, and a%erage gross sales $in N1,555& of businesses in the mar-et area $#&. 2he result of a regression analysis $with t-ratios in parenthesis& is gi%en below. JI 9 355 $3.5& - * BI M 2 B8 M 5.5" , M 5.51 # $3.33& $2.'& $1.33& $2.'& .(.(. 9 3.*

)2 9 5.<1 $a& $b&

$c&

(%aluate the statistical significance of the e!uation as a whole and of each of its coefficients. $' mar-s& 2he a%erage %alues of the independent %ariables in the data set used to estimate the e!uation are BI 9 N1<', B8 9 N22', , 9 N11,555, and # 9 N255,555. 6alculate a point estimate of ;: s a%erage sales and a <'O inter%al estimate of sales based on these %alues. $15 mar-s& 1hat steps are usually in%ol%ed in the estimation of a demand e!uation by regression analysisE $15 mar-s&

12