CHAPTER 10 DETERMINING HOW COSTS BEHAVE 10-1

1. 2. The two assumptions are: Variations in total costs are explained by variations in the level of a single activity related to those costs (the cost driver). ost behavior is approximated by a linear cost function within the relevant range. ! linear cost function is a cost function where" within the relevant range" the graph of total costs versus the level of a single activity forms a straight line. Three alternative linear cost functions are: Variable cost function##a cost function in which total costs change in proportion to the changes in the level of activity in the relevant range. $ixed cost function##a cost function in which total costs do not change with changes in the level of activity in the relevant range. &ixed cost function##a cost function that has both variable and fixed elements. Total costs change but not in proportion to the changes in the level of activity in the relevant range.

10-2
1. 2. %.

10-3

! linear cost function is a cost function where" within the relevant range" the graph of total costs versus the level of a single activity related to that cost is a straight line. !n example of a linear cost function is a cost function for use of a telephone line where the terms are a fixed charge of '1("((( per year plus a '2 per minute charge for phone use. ! nonlinear cost function is a cost function where" within the relevant range" the graph of total costs versus the level of a single activity related to that cost is not a straight line. )xamples include economies of scale in advertising where an agency can double the number of advertisements for less than twice the costs" step*function costs" and learning*curve*based costs.

10-4

+o. ,igh correlation merely indicates that the two variables move together in the data examined. -t is essential to also consider economic plausibility before ma.ing inferences about cause and effect. /ithout any economic plausibility for a relationship" it is less li.ely that a high level of correlation observed in one set of data will be similarly found in other sets of data.

10-5
1. 2. %. 0.

$our approaches to estimating a cost function are: -ndustrial engineering method. onference method. !ccount analysis method. 1uantitative analysis of current or past cost relationships.

10-6

The conference method estimates cost functions on the basis of analysis and opinions about costs and their drivers gathered from various departments of a company (purchasing" process engineering" manufacturing" employee relations" etc.). !dvantages of the conference method include: 1. The speed with which cost estimates can be developed. 2. The pooling of .nowledge from experts across functional areas. %. The improved credibility of the cost function to all personnel.

1(*1

10-7

The account analysis method estimates cost functions by classifying cost accounts in the subsidiary ledger as variable" fixed" or mixed with respect to the identified level of activity. Typically" managers use 2ualitative" rather than 2uantitative" analysis when ma.ing these cost* classification decisions.

10-8
1. 2. %. 0. 4. 5.

The six steps are: hoose the dependent variable (the variable to be predicted" which is some type of cost). -dentify the independent variable or cost driver. ollect data on the dependent variable and the cost driver. 3lot the data. )stimate the cost function. )valuate the cost driver of the estimated cost function.

6tep % typically is the most difficult for a cost analyst.

10-9

ausality in a cost function runs from the cost driver to the dependent variable. Thus" choosing the highest observation and the lowest observation of the cost driver is appropriate in the high*low method.

10-10
1. 2. %.

riteria important when choosing among alternative cost functions are: )conomic plausibility. 7oodness of fit. 6lope of the regression line.

10-11 ! learning curve is a function that measures how labor*hours per unit decline as units
of production increase because wor.ers are learning and becoming better at their 8obs. Two models used to capture different forms of learning are: 1. 2. umulative average*time learning model. The cumulative average time per unit declines by a constant percentage each time the cumulative 2uantity of units produced doubles. -ncremental unit*time learning model. The incremental time needed to produce the last unit declines by a constant percentage each time the cumulative 2uantity of units produced doubles.

10-12 $re2uently encountered problems when collecting cost data on variables included in a
cost function are: 1. 2. %. 0. 4. 5. :. The time period used to measure the dependent variable is not properly matched with the time period used to measure the cost driver(s). $ixed costs are allocated as if they are variable. 9ata are either not available for all observations or are not uniformly reliable. )xtreme values of observations occur. ! homogeneous relationship between the individual cost items in the dependent variable cost pool and the cost driver(s) does not exist. The relationship between the cost and the cost driver is not stationary. -nflation has occurred in a dependent variable" a cost driver" or both.

1(*2

10-13 $our .ey assumptions examined in specification analysis are:

1. 2. %. 0. ;inearity between the dependent variable and the independent variable within the relevant range. onstant variance of residuals for all values of the independent variable. <esiduals are independent of each other. <esiduals are normally distributed.

10-14 +o. ! cost driver is any factor whose change causes a change in the total cost of a
related cost ob8ect. ! cause*and*effect relationship underlies selection of a cost driver. 6ome users of regression analysis include numerous independent variables in a regression model in an attempt to maximi=e goodness of fit" irrespective of the economic plausibility of the independent variables included. 6ome of the independent variables included may not be cost drivers.

10-15 +o. &ulticollinearity exists when two or more independent variables are highly
correlated with each other.

10-16 (1( min.) Est !"t #$ " %&st '(#%t &#

1. 6lope coefficient > ) ) ) '(.%( per machine*hour onstant > Total cost # (6lope coefficient 1uantity of cost driver) > '%"?(( # ('(.%( :"((() > '1"@(( > '%"((( # ('(.%( 0"((() > '1"@(( The cost function based on the two observations is: &aintenance costs > '1"@(( A '(.%( &achine*hours 2. The cost function in re2uirement 1 is an estimate of how costs behave within the relevant range" not at cost levels outside the relevant range. -f there are no months with =ero machine* hours represented in the maintenance account" data in that account cannot be used to estimate the fixed costs at the =ero machine*hours level. <ather" the constant component of the cost function provides the best available starting point for a straight line that approximates how a cost behaves within the relevant range.

1(*%

10-17 (14 min.) I*+#t ', #$ -". "/0+-1 ' 2+*-1 "#* ! 2+*-%&st '(#%t &#s3
1. 2. 6ee 6olution )xhibit 1(*1:. ontract 1: y > '4( ontract 2: y > '%( A '(.2(X ontract %: y > '1X where X is the number of miles traveled in the day. %. C&#t."%t C&st 4(#%t &# 1 $ixed 2 &ixed % Variable SO56TION E7HIBIT 10-17 3lots of ar <ental ontracts Bffered by 3acific orp.

1(*0

10-18 (2( min.) V". &(s %&st-/+8"- &. 9"tt+.#s3

1. 2. %. 0. 4. 5. :. @. ?. C D 7 E ; $ C

+ote that ! is incorrect because" although the cost per pound eventually e2uals a constant at '?.2(" the total dollars of cost increases linearly from that point onward. The total costs will be the same regardless of the volume level. This is a classic step*cost function.

10-19 (%( min.) M"t%8 #$ $."98s : t8 *+s%. 9t &#s &' %&st "#* .+-+#(+ /+8"- &.3
a. b. c. d. e. f. g. h. (1) (5) ! step*cost function. (?) (2) (@) (1() -t is data plotted on a scatter diagram" showing a linear variable cost function with constant variance of residuals. The constant variance of residuals implies that there is a uniform dispersion of the data points about the regression line. (%) (@)

1(*4

10-20 (14 min.) A%%&(#t "#"0,s s !+t8&*3

13 Variable costs: ar wash labor 6oap" cloth" and supplies /ater )lectric power to move conveyor belt Total variable costs $ixed costs: 9epreciation 6alaries Total fixed costs '20("((( %2"((( 2@"((( :2"((( '%:2"((( ' 50"((( 05"((( '11("(((

osts are classified as variable because the total costs in these categories change in proportion to the number of cars washed in ;oren=oFs operation. osts are classified as fixed because the total costs in these categories do not vary with the number of cars washed. -f the conveyor belt moves regardless of the number of cars on it" the electricity costs to power the conveyor belt would be a fixed cost. 23 Variable costs per car >
'%:2"((( > '0.54 per car @("(((

Total costs estimated for ?("((( cars > '11("((( A ('0.54 G ?("((() > '42@"4((

10-21 (%( min.) A%%&(#t "#"0,s s !+t8&*3

1. &anufacturing cost classification for 2((0:
= &' T&t"0 C&sts T8"t Is V". "/0+ 4 2+* V". "/0+ C&sts C&sts ;2< ;3< ) ;1< ;2< ;4< ) ;1< > ;3< 6# t V". "/0+ C&sts ;5<);3< ? 751000

A%%&(#t

T&t"0 C&sts ;1<

9irect materials '%(("((( 9irect manufacturing labor 224"((( 3ower %:"4(( 6upervision labor 45"24( &aterials*handling labor 5("((( &aintenance labor :4"((( 9epreciation ?4"((( <ent" property taxes" admin 1(("((( Total '?0@":4(

1((H 1(( 1(( 2( 4( 0( ( (

'%(("((( 224"((( %:"4(( 11"24( %("((( %("((( ( ( '5%%":4(

( ( ( 04"((( %("((( 04"((( ?4"((( 1(("((( '%14"(((

'

'0.(( %.(( (.4( (.14 (.0( (.0( ( ( '@.04

Total manufacturing cost for 2((0 > '?0@":4(

1(*5

10-21 ( ontFd.) Variable costs in 2((4:

6# t V". "/0+ C&st '&. 2004 ;6< I#%.+"s+ # V". "/0+ P+.%+#t"$+ C&sts 6# t V". "/0+ I#%.+"s+ 9+. 6# t C&st '&. 2005 ;7< ;8< ) ;6< ;7< ;9< ) ;6< @ ;8< T&t"0 V". "/0+ C&sts '&. 2005 ;10< ) ;9< 801000

A%%&(#t

9irect materials 9irect manufacturing labor 3ower 6upervision labor &aterials*handling labor &aintenance labor 9epreciation <ent" property taxes" admin. Total $ixed and total costs in 2((4:

'0.(( %.(( (.4( (.14 (.0( (.0( ( ( '@.04

4H 1( ( ( ( ( ( (

'(.2( (.%( ( ( ( ( ( ( '(.4(

D&00". I#%.+"s+ # 4 2+* C&sts ;13< ) ;11< ;12<

'0.2( %.%( (.4( (.14 (.0( (.0( ( ( '@.?4

'%%5"((( 250"((( 0("((( 12"((( %2"((( %2"((( ( ( ':15"(((

A%%&(#t

4 2+* C&sts '&. 2004 ;11<

P+.%+#t"$+ I#%.+"s+ ;12<

4 2+* C&sts '&. 2005 ;14< ) ;11< @ ;13<

V". "/0+ C&sts '&. 2005 ;15<

T&t"0 C&sts ;16< ) ;14< @ ;15<

9irect materials ' ( 9irect manufacturing labor ( 3ower ( 6upervision labor 04"((( &aterials*handling labor %("((( &aintenance labor 04"((( 9epreciation ?4"((( <ent" property taxes" admin. 1(("((( Total '%14"(((

(H ( ( ( ( ( 4 :

'

( ( ( ( ( ( 0":4( :"((( '11":4(

'

( ( ( 04"((( %("((( 04"((( ??":4( 1(:"((( '%25":4(

'%%5"((( ' %%5"((( 250"((( 250"((( 0("((( 0("((( 12"((( 4:"((( %2"((( 52"((( %2"((( ::"((( ( ??":4( ( 1(:"((( ':15"((( '1"(02":4(

Total manufacturing costs for 2((4 > '1"(02":4( 2. Total cost per unit" 2((0 Total cost per unit" 2((4
'?0@":4( > '12.54 :4"((( '1"(02":4( > > '1%.(% @("(((

>

%. ost classification into variable and fixed costs is based on 2ualitative" rather than 2uantitative" analysis. ,ow good the classifications are depends on the .nowledge of individual managers who classify the costs. 7ower may want to underta.e 2uantitative analysis of costs" using regression analysis on time*series or cross*sectional data to better estimate the fixed and variable components of costs. Detter .nowledge of fixed and variable costs will help 7ower to better price his products" .now when he is getting a positive contribution margin" and to better manage costs.

1(*:

10-22 (2( min.) Est !"t #$ " %&st '(#%t &#1 8 $8-0&: !+t8&*3
1. 6ee 6olution )xhibit 1(*22. There is a positive relationship between the number of service reports (a cost driver) and the customer*service department costs. This relationship is economically plausible. 2. ,ighest observation of cost driver ;owest observation of cost driver 9ifference N(!/+. &' S+.- %+ R+9&.ts 0%5 122 %10 C(st&!+.-S+.- %+ D+9".t!+#t C&sts '21"@?( 12"?01 ' @"?0?

ustomer*service department costs > a A b (number of service reports) 6lope coefficient (b) onstant (a) ustomer*service department costs %. >
'@"?0? ) '2@.4( per service report %10

> '21"@?( # '2@.4( (0%5) > '?"050 > '12"?01 # '2@.4( (122) > '?"050 > '?"050 A '2@.4( (number of service reports)

Bther possible cost drivers of customer*service department costs are: a. +umber of products replaced with a new product (and the dollar value of the new products charged to the customer*service department). b. +umber of products repaired and the time and cost of repairs.

SO56TION E7HIBIT 10-22 3lot of +umber of 6ervice <eports versus ustomer*6ervice 9ept. osts for apitol 3roducts
C(st&!+.-S+.- %+ D+9".t!+#t C&sts '24"((( 2("((( 14"((( 1("((( 4"((( '( ( 1(( 2(( %(( 0(( 4(( N(!/+. &' S+.- %+ R+9&.ts

1(*@

10-23 (%(#0( min.) 5 #+". %&st "99.&2 !"t &#3

1. 6lope coefficient (b) > ) onstant (a) ost function
'42?"((( - 0(("((( ) '0%.(( :"(((- 0"(((

> '42?"((( # ('0%.(( G :"((() > '22@"((( > '22@"((( A '0%.(( (professional labor*hours)

The linear cost function is plotted in 6olution )xhibit 1(*2%. +o" the constant component of the cost function does not represent the fixed overhead cost of the &emphis 7roup. The relevant range of professional labor*hours is from %"((( to @"(((. The constant component provides the best available starting point for a straight line that approximates how a cost behaves within the %"((( to @"((( relevant range. 2. ! comparison at various levels of professional labor*hours follows. The linear cost function is based on the formula of '22@"((( per month plus '0%.(( per professional labor*hour. Total overhead cost behavior: M&#t8 1 M&#t8 2 M&#t8 3 M&#t8 4 M&#t8 5 M&#t8 6 !ctual total overhead costs ;inear approximation !ctual minus linear approximation 3rofessional labor*hours '%0("((( '0(("((( '0%4"((( '0::"((( '42?"((( '4@:"((( %4:"((( 0(("((( 00%"((( 0@5"((( 42?"((( 4:2"((( '(1:"((() ' ( ' (@"((() ' (?"((() ' ( ' 14"((( %"((( 0"((( 4"((( 5"((( :"((( @"(((

The data are shown in 6olution )xhibit 1(*2%. The linear cost function overstates costs by '@"((( at the 4"(((*hour level and understates costs by '14"((( at the @"(((*hour level. %. ontribution before deducting incremental overhead -ncremental overhead ontribution after incremental overhead The total contribution margin actually forgone is '%"(((. B"s+* &# A%t("0 '%@"((( %4"((( ' %"((( B"s+* &# 5 #+". C&st 4(#%t &# '%@"((( 0%"((( ' (4"((()

1(*?

10-23 ( ontFd.) SO56TION E7HIBIT 10-23 ;inear ost $unction 3lot of 3rofessional ;abor*,ours on Total Bverhead osts for &emphis onsulting 7roup ':(("((( T&t"0 O-+.8+"* C&sts 5(("((( 4(("((( 0(("((( %(("((( 2(("((( 1(("(((
( (

1"((( 2"((( %"((( 0"((( 4"((( 5"((( :"((( P.&'+ss &#"0 5"/&.-H&(.s B 00+*

@"((( ?"(((

10-24 (2( min.) C&st--&0(!+-9.&' t "#* .+$.+ss &# "#"0,s s.

1a. !verage cost of manufacturing > >
Total manufacturing costs +umber of bicycle frames '?(("((( > '%( per frame %("(((

This cost is greater than the '2@.4( per frame that <yan has 2uoted. 1b. 7arvin cannot ta.e the average manufacturing cost in 2((2 of '%( per frame and multiply it by %5"((( bicycle frames to determine the total cost of manufacturing %5"((( bicycle frames. The reason is that some of the '?(("((( (or e2uivalently the '%( cost per frame) are fixed costs and some are variable costs. /ithout distinguishing fixed from variable costs" 7arvin cannot determine the cost of manufacturing %5"((( frames. $or example" if all costs are fixed" the manufacturing costs of %5"((( frames will continue to be '?(("(((. -f" however" all costs are variable" the cost of manufacturing %5"((( frames would be '%( %5"((( > '1"(@("(((. -f some costs are fixed and some are variable" the cost of manufacturing %5"((( frames will be somewhere between '?(("((( and '1"(@("(((.

1(*1(

10-24 ( ontFd.) 6ome students could argue that another reason for not being able to determine the cost of manufacturing %5"((( bicycle frames is that not all costs are output unit*level costs. -f some costs are" for example" batch*level costs" more information would be needed on the number of batches in which the %5"((( bicycle frames would be produced" in order to determine the cost of manufacturing %5"((( bicycle frames. 2.
)xpected cost to ma.e %5"((( bicycle frames

> '0%2"((( A '14 %5"((( > '0%2"((( A '40("((( > '?:2"(((

3urchasing bicycle frames from <yan will cost '2@.4( %5"((( > '1"(25"(((. ,ence it will cost 7arvin '1"(25"((( '?:2"((( > '40"((( more to purchase the frames from 7arvin rather than manufacture them in*house. %. 7arvin would need to consider several factors before being confident that the e2uation in re2uirement 2 accurately predicts the cost of manufacturing bicycle frames. a. -s the relationship between total manufacturing costs and 2uantity of bicycle frames economically plausibleI $or example" is the 2uantity of bicycles made the only cost driver or are there other cost*drivers (for example batch*level costs of setups" production*orders or material handling) that affect manufacturing costsI b. ,ow good is the goodness of fitI That is" how well does the estimated line fit the dataI c. -s the relationship between the number of bicycle frames produced and total manufacturing costs linearI d. 9oes the slope of the regression line indicate that a strong relationship exists between manufacturing costs and the number of bicycle frames producedI e. !re there any data problems such as" for example" errors in measuring costs" trends in prices of materials" labor or overheads that might affect variable or fixed costs over time" extreme values of observations" or a nonstationary relationship over time between total manufacturing costs and the 2uantity of bicycles producedI f. ,ow is inflation expected to affect costsI g. /ill <yan supply high*2uality bicycle frames on timeI

1(*11

10-25 (24 min.)

R+$.+ss &# "#"0,s s1 s+.- %+ %&!9"#,3

1. 6olution )xhibit 1(*24 plots the relationship between labor*hours and overhead costs and shows the regression line. y > '0@"2:1 A '%.?% X Economic plausibility. ;abor*hours appears to be an economically plausible driver of overhead costs for a catering company. Bverhead costs such as scheduling" hiring and training of wor.ers" and managing the wor.force are largely incurred to support labor. Goodness of fit. The vertical differences between actual and predicted costs are extremely small" indicating a very good fit. The good fit indicates a strong relationship between the labor* hour cost driver and overhead costs. Slope of regression line. The regression line has a reasonably steep slope from left to right. 7iven the small scatter of the observations around the line" the positive slope indicates that" on average" overhead costs increase as labor*hours increase. 2. The regression analysis indicates that" within the relevant range of 2"4(( to :"4(( labor* hours" the variable cost per person for a coc.tail party e2uals: $ood and beverages ;abor ((.4 hrs. '1( per hour) Variable overhead ((.4 hrs '%.?% per labor*hour) Total variable cost per person '14.(( 4.(( 1.?: '21.?:

%. To earn a positive contribution margin" the minimum bid for a 2((*person coc.tail party would be any amount greater than '0"%?0. This amount is calculated by multiplying the variable cost per person of '21.?: by the 2(( people. !t a price above the variable costs of '0"%?0" Dob Eones will be earning a contribution margin toward coverage of his fixed costs. Bf course" Dob Eones will consider other factors in developing his bid including (a) an analysis of the competition##vigorous competition will limit EonesJs ability to obtain a higher price (b) a determination of whether or not his bid will set a precedent for lower prices##overall" the prices Dob Eones charges should generate enough contribution to cover fixed costs and earn a reasonable profit" and (c) a 8udgment of how representative past historical data (used in the regression analysis) is about future costs.

1(*12

10-25 ( ontFd.) SO56TION E7HIBIT 10-25 <egression ;ine of ;abor*,ours on Bverhead osts for Dob EonesJs atering ompany

'?("((( @("((( :("(((

O-+.8+"* C&sts

5("((( 4("((( 0("((( %("((( 2("((( 1("((( ( ( 1"((( 2"((( %"((( 0"((( 4"((( 5"((( :"((( @"(((

C&st D. -+.A 5"/&.-H&(.s

10-25 E2%+0 A990 %"t &#

Determining How Costs Behave Bob Jones Catering Company
Original Data Month Labor-Hours January February March April May June July August epte!ber "ct#ber %#&e!ber 'ece!ber (#tal 2,500 2,700 3,000 4,200 7,500 5,500 6,500 4,500 7,000 4,500 3,100 6,500 57,500 Overhead Costs $55,000 59,000 60,000 64,000 77,000 71,000 74,000 67,000 75,000 6$,000 62,000 73,000 $$05,000

1(*1%

10-25 ( ontFd.) Relationship Between Overhead Costs and LaborHours

$90,000 $$0,000 Overhead Costs $70,000 $60,000 $50,000 $40,000 $30,000 $20,000 $10,000 $0 0 1000 2000 3000 4000 5000 6000 7000 $000 Labor-Hours

1(*10

10-25 ( ontFd.)
Regression Output )MMA*+ ")(,)( Regression Statistics Multiple * * .uare A/0uste/ * .uare tan/ar/ 1rr#r "bser&ati#ns

0-9$0995276 0-962351731 0-95$5$6904 1447-996$47 12

A%"2A df *egressi#n *esi/ual (#tal 1 10 11 Significance F 53594971$ 53594971 255-616459 1-$9126130$ 2096694$-69 2096694-$ 556916666-7 t Stat P-value Lower 95% Upper 95% Lower 95 !% Upper 95 !% 3$-65619 3-203561312 454$$-30459 51052-93166 454$$-30459 51052-93166 15-9$$00 1-$9126130$ 3-37$973$09 4-4732$9931 3-37$973$09 4-4732$9931 SS MS F

4ntercept 5 2ariable 1

Coefficients Standard Error 4$270-61$12 124$-7161$$ 3-926131$7 0-245567266

1(*14

10-26 (%( min.) R+$.+ss &# "#"0,s s1 "%t - t,-/"s+* %&st #$1 %8&&s #$ %&st *. -+.s3
1a. 6olution )xhibit 1(*25! presents the plots and regression line of number of pac.aged units moved on distribution costs. SO56TION E7HIBIT 10-26A 3lots and <egression ;ine of +umber of 3ac.aged Knits &oved on 9istribution osts
'04"((( 0("((( %4"((( %("((( 24"((( 2("((( 14"((( 1("((( 4"((( ( ( 2("((( 0("((( 5("((( @("(((

D st. /(t &# C&sts

N(!/+. &' P"%B"$+* 6# ts M&-+*

1b.

6olution )xhibit 1(*25D presents the plots and regression line of number of shipments made on distribution costs.

SO56TION E7HIBIT 10-26B 3lots and <egression ;ine of +umber of 6hipments &ade on 9istribution osts
'04"((( 0("((( D st. /(t &# C&sts %4"((( %("((( 24"((( 2("((( 14"((( 1("((( 4"((( ( ( 1(( 2(( %(( 0(( 4(( N(!/+. &' S8 9!+#ts M"*+

1(*15

10-26 ( ontFd.) +umber of pac.aged units moved appears to be a better cost driver of distribution costs for the following reasons: (i) Economic plausibility. Doth number of pac.aged units moved and number of shipments are economically plausible cost drivers. Decause the product is heavy" however" costs of freight are li.ely to be a si=able component of distribution costs. Thus" number of pac.aged units moved will affect distribution costs significantly because freight costs are largely a function of the number of units transported. Goodness of fit. ompare 6olution )xhibits 1(*25! and 1(*25D. +umber of pac.aged units moved has a better goodness of fit with distribution costs than do number of shipments made. That is" the vertical differences between actual and predicted distribution costs are smaller for the number of pac.aged units moved regression than for the number of shipments made regression.

(ii)

(iii) Slope of regression line. !gain" compare 6olution )xhibits 1(*25! and 1(*25D. The number of pac.aged units moved regression line has a relatively steep slope and a small scatter of observations around the regression lines indicating a strong relationship between number of pac.aged units moved and distribution costs. Bn average" distribution costs increase with the number of pac.aged units moved. The number of shipments made regression line is flatter and has more scatter of observations about the line indicating a wea. relationship between the number of shipments made and distribution costs. Bn average" the number of shipments made has a smaller effect on distribution costs. 2. Ksing the preferred cost function" 9istribution costs > '1"%0? A ('(.0?5 +umber of pac.aged units moved)" $laherty would budget distribution costs of '1"%0? A ('(.0?5 0("((() > '1"%0? A '1?"@0( > '21"1@?

10-27 (2( min.) 5+".# #$ %(.-+1 %(!(0"t -+ "-+."$+-t !+ 0+".# #$ !&*+03

The direct manufacturing labor*hours (9&;,) re2uired to produce the first 2" 0" and @ units given the assumption of a cumulative average*time learning curve of ?(H" is as follows: C(!(0"t -+ N(!/+. &' 6# ts ;1< 1 2 0 @ C(!(0"t -+ A-+."$+-T !+ 9+. 6# t ;2< %"((( 2":(( (%"((( (.?() 2"0%( (2":(( (.?() 2"1@: (2"0%( (.?() C(!(0"t -+ T&t"0 T !+ ;3< ) ;1< ;2< %"((( 4"0(( ?":2( 1:"0?5

!lternatively" to compute the values in column (2) we could use the formula

1(*1:

10-27 ( ontFd.) y = aXb where a > %"(((" X > 2" 0" or @" and b > # (.142(" which gives when X > 2" y > %"((( 2# (.142( > 2":(( when X > 0" y > %"((( 0# (.142( > 2"0%( when X > @" y > %"((( @# (.142( > 2"1@: V". "/0+ C&sts &' P.&*(% #$ 2 6# ts 4 6# ts 8 6# ts '15("((( '%2("((( ' 50("((( 1%4"((( @1"((( '%:5"((( 20%"((( 104"@(( ':(@"@(( 0%:"0(( 252"00( '1"%%?"@0(

9irect materials '@("((( 2L 0L @ 9irect manufacturing labor '24 4"0((L ?":2(L 1:"0?5 Variable manufacturing overhead '14 4"0((L ?":2(L 1:"0?5 Total variable costs

10-28 (2( min.) 5+".# #$ %(.-+1 #%.+!+#t"0 (# t-t !+ 0+".# #$ !&*+03

1. The direct manufacturing labor*hours (9&;,) re2uired to produce the first 2" %" and 0 units" given the assumption of an incremental unit*time learning curve of ?(H" is as follows: C(!(0"t -+ N(!/+. &' 6# ts ;1< 1 2 % 0 I#* - *("0 6# t T !+ '&. 7t8 6# t ;2< %"((( 2":(( (%"((( (.?() 2"4%? 2"0%( (2":(( (.?()

C(!(0"t -+ T&t"0 T !+ ;3< %"((( 4":(( @"2%? 1("55?

Values in column 2 are calculated using the formula y = aXb where a > %"(((" X > 2" %" or 0" and b > # (.142(" which gives when X > 2" y > %"((( 2# (.142( > 2":(( when X > %" y > %"((( %# (.142( > 2"4%? when X > 0" y > %"((( 0# (.142( > 2"0%(

1(*1@

10-28 ( ontFd.) 9irect materials '@("((( 2L %L 0 9irect manufacturing labor '24 4":((L @"2%?L 1("55? Variable manufacturing overhead '14 4":((L @"2%?L 1("55? Total variable costs 2. -ncremental unit*time learning model (from re2uirement 1) umulative average*time learning model (from )xercise 1(*2:) 9ifference V". "/0+ C&sts &' P.&*(% #$ 2 6# ts 3 6# ts 4 6# ts '15("((( '20("((( ' %2("((( 102"4(( @4"4(( '%@@"((( 2(4"?:4 12%"4@4 '45?"45( 255":24 15("(%4 ':05":5(

V". "/0+ C&sts &' P.&*(% #$ 2 6# ts 4 6# ts '%@@"((( %:5"((( ' 12"((( ':05":5( :(@"@(( ' %:"?5(

Total variable costs for manufacturing 2 and 0 units are lower under the cumulative average*time learning curve relative to the incremental unit*time learning curve. 9irect manufacturing labor*hours re2uired to ma.e additional units decline more slowly in the incremental unit*time learning curve relative to the cumulative average*time learning curve when the same ?(H factor is used for both curves. The reason is that" in the incremental unit*time learning curve" as the number of units double" only the last unit produced has a cost of ?(H of the initial cost. -n the cumulative average*time learning model" doubling the number of units causes the average cost of all the additional units produced (not 8ust the last unit) to be ?(H of the initial cost.

10-29 (24 min.) H $8-0&: !+t8&*3

1. ,ighest observation of cost driver ;owest observation of cost driver 9ifference &aintenance costs onstant (a) or onstant (a) &aintenance costs
'@("((( > '2 0("(((

M"%8 #+-H&(.s 124"((( @4"((( 0("(((

M" #t+#"#%+ C&sts '24("((( 1:("((( ' @("(((

> a A b (&achine*hours) > > > > > '24("((( # ('2 G 124"((() '24("((( # '24("((( > '( '1:("((( # ('2 G @4"((() '1:("((( # '1:("((( > '( '2 G &achine*hours

6lope coefficient (b) >

1(*1?

10-29 ( ontFd.) 2. SO56TION E7HIBIT 10-29 3lot and ,igh*;ow ;ine of &achine*,ours on &aintenance osts
'25("((( 20("((( M" #t+#"#%+ C&sts 22("((( 2(("((( 1@("((( 15("((( 10("((( 12("((( 1(("((( @("((( ?("((( 1(("((( 11("((( 12("((( 1%("((( M"%8 #+-H&(.s

6olution )xhibit 1(*2? presents the high*low line. Economic plausibility. The cost function shows a positive economically plausible relationship between machine*hours and maintenance costs. There is a clear*cut engineering relationship of higher machine*hours and maintenance costs. Goodness of fit. The high*low line appears to MfitN the data well. The vertical differences between the actual and predicted costs appear to be 2uite small. Slope of high-low line. The slope of the line appears to be reasonably steep indicating that" on average" maintenance costs in a 2uarter varies with machine*hours used. %. Ksing the cost function estimated in 1" predicted maintenance costs would be '2 G ?("((( > '1@("(((. ,oward should budget '1@("((( in 2uarter 1% because the relationship between machine* hours and maintenance costs in 6olution )xhibit 1(*2? is economically plausible" has an excellent goodness of fit" and indicates that an increase in machine*hours in a 2uarter causes maintenance costs to increase in the 2uarter.

1(*2(

10-30 (%(0( min.) H $8-0&: !+t8&* -+.s(s .+$.+ss &# "#"0,s s3

1. 6olution )xhibit 1(*%( presents the plots of advertising costs on revenues. SO56TION E7HIBIT 10-30 3lot and <egression ;ine of !dvertising osts on <evenues
'?("((( @("((( :("((( 5("(((

R+-+#(+s

4("((( 0("((( %("((( 2("((( 1("((( ( '( '1"((( '2"((( '%"((( '0"((( '4"(((

A*-+.t s #$ C&sts

2. 6olution )xhibit 1(*%( also shows the regression line of advertising costs on revenues. /e evaluate the estimated regression e2uation using the criteria of economic plausibility" goodness of fit" and slope of the regression line. Economic plausibility. !dvertising costs appears to be a plausible cost driver of revenues. <estaurants fre2uently use newspaper advertising to promote their restaurants and increase their patronage. Goodness of fit. The vertical differences between actual and predicted revenues appears to be reasonably small. This indicates that advertising costs are related to restaurant revenues. Slope of regression line. The slope of the regression line appears to be relatively steep. 7iven the small scatter of the observations around the line" the steep slope indicates that" on average" restaurant revenues increase with newspaper advertising.

1(*21

10-30 ( ontFd.) %. The high*low method would estimate the cost function as follows: A*-+.t s #$ C&sts ,ighest observation of cost driver '0"((( ;owest observation of cost driver 1"((( 9ifference '%"((( <evenues > a A (b advertising costs) 6lope coefficient (b) onstant (a) =
'24"((( > @.%%% '%"(((

R+-+#(+s '@("((( 44"((( '24"(((

> '@("((( ('0"((( @.%%%) > '@("((( %%"%%2 > '05"55@ or onstant (a) > '44"((( ('1"((( @.%%%) > '44"((( @"%%% > '05"55: <evenues > '05"55: A (@.%%% !dvertising costs) 0. The increase in revenues for each '1"((( spent on advertising within the relevant range is a. Ksing the regression e2uation" @.:2% '1"((( > '@":2% b. Ksing the high*low e2uation" @.%%% '1"((( > '@"%%%

The high*low e2uation does fairly well in estimating the relationship between advertising costs and revenues. ,owever" &artine= and Drown should use the regression e2uation. The reason is that the regression e2uation uses information from all observations whereas the high* low method relies only on the observations that have the highest and lowest values of the cost driver. These observations are generally not representative of all the data.

10-31 (%(#%4 min.)

R+$.+ss &# "#"0,s s1 "%t - t,-/"s+* %&st #$1 %8&&s #$ %&st *. -+.s3

1. 6olution )xhibit 1(*%1! presents the plots and regression line of machine#hours on support overhead. 6olution )xhibit 1(*%1D presents the plots and regression line of number of batches on support overhead. !s described below" using the three criteria of economic plausibility" goodness of fit" and slope of regression line" hu should choose number of batches as the cost driver of support overhead costs. Economic plausibility. +umber of batches appears to be a more plausible cost driver of support overhead costs than machine*hours. 6upport staff indicate that they spend a good portion of their time at the start of each batch ensuring that the e2uipment is set up correctly and chec.ing that the first units of production in each batch are of good 2uality. Bnce the machine is wor.ing properly" support staff are not needed to supervise the actual running of the machines. onse2uently" support staff resources are more li.ely to vary with the number of batches rather than the total number of machine*hours wor.ed.

1(*22

10-31 ( ontFd.) Goodness of fit. ompare 6olution )xhibits 1(*%1! and 1(*%1D. The vertical differences between actual and predicted costs are much smaller for number of batches than for machine* hours. This indicates that number of batches has a better fit and a stronger relationship with support overhead costs. Slope of regression line. !gain" compare 6olution )xhibits 1(*%1! and 1(*%1D. The slope of the regression line of number of batches on support overhead is relatively steep with less scatter of observations about the regression line while the regression line of machine*hours on support overhead is relatively flat (small slope) with more scatter of observations about the regression line. ! relatively steep regression line with less scatter for number of batches indicates that" on average" support overhead costs increase as number of batches increase. Bn the other hand" the relatively flat regression line for machine*hours with more scatter indicates a wea. or no relationship between support overhead costs and machine hours##on average" changes in machine*hours appear to have a minimal effect on support overhead costs. 2. !s described in re2uirement 1" number of batches is the preferred cost driver. Ksing this cost driver and the regression e2uation y > '15"(%1 A '1?:.%( number of batches" hu should budget the following support overhead costs for the %(( batches that will be run next month: y > '15"(%1 A '1?:.%( %(( > '15"(%1 A '4?"1?( > ':4"221. %a. Ksing machine*hours as the cost driver and the regression e2uation y > '2@"(@? A '1(.2% machine*hours" hu would budget support overhead costs for the 2"5(( machine*hours that will be wor.ed next month as: y > '2@"(@? A '1(.2% 2"5(( > '2@"(@? A '25"4?@ > '40"5@: B(*$+t+* R+-+#(+s "#* C&sts '&. N+2t M&#t8 6s #$ N(!/+. &' B"t%8+s M"%8 #+-H&(.s "s "s t8+ C&st D. -+. t8+ C&st D. -+. osts other than support overhead 6upport overhead costs Total costs !dd margin of 2(H of total costs Target revenues '124"((( :4"221 2(("221 0("(00 '20("254 '124"((( 40"5@: 1:?"5@: %4"?%: '214"520

3ic.ing machine*hours rather than the number of batches as the cost driver will cause hu to underestimate costs and choose lower target revenues and prices. 6upport overhead costs" however" will vary with number of batches rather than machine*hours. Ksing information from the preceding table" actual costs will be closer to '2(("221 against target revenues of '214"520. Target profitability is unli.ely to be met. /ith better cost driver information hu would probably have priced products higher and earned greater revenues" assuming" of course" that customers are willing to pay the higher prices.

1(*2%

10-31 ( ontFd.) hoosing the OwrongO cost driver and estimating the incorrect cost function will also have repercussions for cost management and cost control. 6uppose <ohan 3lastics budgets support overhead costs of '40"5@: for next month using the machine*hour regression. 6uppose actual support overhead costs" driven by number of batches" are ':0"((( next month. &anagement of <ohan 3lastics would regard this as unsatisfactory performance and begin to explore ways to cut costs to bring them more in line with budgeted support overhead costs. -n fact" on the basis of the preferred cost driver" number of batches" the plantJs actual costs are lower than the predicted amount" ':4"221##a performance that management should see. to replicate rather than change. Ksing OwrongO cost drivers misleads management in cost planning" cost management" and cost control besides contributing to inappropriate product pricing decisions. SO56TION E7HIBIT 10-31A <egression ;ine of &achine*,ours on 6upport Bverhead osts for <ohan 3lastics
'?("( (( @("(( (

S(99&.t O-+.8+"* C&sts

:("(( ( 5("(( ( 4("(( ( 0("(( ( %("(( ( 2("(( ( 1("(( ( ( ( 4(( 1"((( 1"4(( 2"((( 2"4(( %"((( %"4(( 0"(((

C&st D. -+.A M"%8 #+-H&(.s

1(*20

10-31 ( ontFd.) SO56TION E7HIBIT 10-31B <egression ;ine of +umber of Datches on 6upport Bverhead osts for <ohan 3lastics
$90,000

S(99&.t O-+.8+"* C&sts

80,000 70,000 60,000 50,000 40,000 30,000 20,000 10,000 0 0 50 100 150 200 250 300 350

C&st D. -+.A N(!/+. &' B"t%8+s

10-32 (1(*14 min.) T !+ 0"$ %&#s *+."t &# # #t+.9.+t #$ .+$.+ss &# .+s(0ts3
The relationship between machine*hours and maintenance costs is negative because !=a ompany schedules maintenance during slow production periods. There is a time lag between heavy machine usage during pea. production months" and maintenance in later months when the production volume is low. To correctly understand the relationship between machine*hours and maintenance costs" !=a should estimate the regression e2uation of maintenance costs on lagged machine*hours (that is" machine*hours in prior months). The explanation for the relationship between sales revenue and advertising costs is that the result of advertising is often not instantaneous. 7enerally advertising generates increased sales revenue in subse2uent month(s)" so !=a should estimate the relationship between advertising costs in a particular period and sales in future periods. !nother explanation is that !=a increases its advertising costs during periods of declining sales in an attempt to increase sales volume of the clothing lines that will soon be out of season. $or example" !=a may increase its advertising on winter clothing during late winter before the customers start purchasing spring clothes.

1(*24

10-33 (%(#0( min.) C&st +st !"t &#1 %(!(0"t -+ "-+."$+-t !+ 0+".# #$ %(.-+3
1. ost to 3roduce the 6econd through the )ighth Troop 9eployment Doats: 9irect materials" : '1(("((( 9irect manufacturing labor" %?"1%(P '%( Variable manufacturing overhead" %?"1%( '2( Bther manufacturing overhead" 24H of '1"1:%"?(( Total costs ' :(("((( 1"1:%"?(( :@2"5(( 2?%"0:4 '2"?0?"?:4

PThe direct manufacturing labor*hours to produce the second to eighth boats can be calculated in several ways" given the assumption of a cumulative average*time learning curve of @4H:

a.

Kse of Table $ormat: C(!(0"t -+ N(!/+. &' 6# ts 1 2 0 @ C(!(0"t -+ A-+."$+-T !+ 9+. 6# t 1("(((.(( @"4((.(( (1("((( (.@4) :"224.(( (@"4(( (.@4) 5"101.24 (:"224 (.@4) C(!(0"t -+ T&t"0 T !+ 1("((( 1:"((( 2@"?(( 0?"1%(

The direct labor*hours re2uired to produce the second through the eighth boats is 0?"1%( # 1("((( > %?"1%( hours. b. Kse of $ormula: b y > aX where a > 1("(((" X > @" and b > # (.2%04 y > 1("((( @# (.2%04 > 5"101 hours (rounded) The total direct labor*hours for @ units is 5"101 @ > 0?"12@ hours The direct labor*hours re2uired to produce the second through the eighth boats is 0?"12@ # 1("((( > %?"12@ hours. (Dy ta.ing the b factor to 5 decimal digits" an estimate of 0?"1%( hours would result.) Note: 6ome students will debate the exclusion of the tooling cost. The 2uestion specifies that the tooling Ocost was assigned to the first boat.O !lthough +autilus may well see. to ensure its total revenue covers the ':24"((( cost of the first boat" the concern in this 2uestion is only with the cost of producing seven more 3T1(?s.

1(*25

10-33 ( ontFd.) 2. ost to 3roduce the 6econd through the )ighth Doats !ssuming ;inear $unction for 9irect ;abor*,ours and Knits 3roduced: 9irect materials" : '1(("((( 9irect manufacturing labor" : 1("((( hours '%( Variable manufacturing overhead" : 1("((( hours '2( Bther manufacturing overhead" 24H of '2"1(("((( Total costs The difference in predicted costs is: 3redicted cost in re2uirement 2 (based on linear cost function) 3redicted cost in re2uirement 1 (based on an @4H learning curve) 9ifference '0":24"((( 2"?0?"?:4 '1"::4"(24 ' :(("((( 2"1(("((( 1"0(("((( 424"((( '0":24"(((

10-34 (2(#%( min.) C&st +st !"t &#1 #%.+!+#t"0 (# t-t !+ 0+".# #$ !&*+03
1. ost to 3roduce the 6econd through the )ighth Doats: 9irect materials" : '1(("((( 9irect manufacturing labor" 0?"%45P '%( Variable overhead" 0?"%45 '2( Bther overhead" 24H of '1"0@("5@( Total costs ' :(("((( 1"0@("5@( ?@:"12( %:("1:( '%"4%:"?:(

PThe direct labor hours to produce the second through the eighth boats can be calculated via a table format" given the assumption of an incremental unit*time learning curve of @4H: C(!(0"t -+ N(!/+. &' 6# ts 1 2 % 0 4 5 : @ I#* - *("0 6# t T !+ '&. Xt8 6# t ;m<C 1("((( @"4(( :":2? :"224 5"@45 5"45? 5"%%5 5"101 C(!(0"t -+ T&t"0 T !+ 1("((( 1@"4(( 25"22? %%"040 0("%1( 05"@:? 4%"214 4?"%45

q P alculated as m > pX where p > 1("(((" q > # (.2%04" and X > 1" 2" %". . ." @.

1(*2:

10-34 ( ontFd.) The direct manufacturing labor*hours to produce the second through the eighth boat is 4?"%45 # 1("((( > 0?"%45 hours. 2. 9ifference in total costs to manufacture the second through the eighth boat under the incremental unit*time learning model and the cumulative average*time learning model is '%"4%:"?:( (calculated in re2uirement 1 of this problem) # '2"?0?"?:4 (from re2uirement 1 of 3roblem 1(*%%) > '4@:"??4. The incremental unit*time learning curve has a slower decline in the reduction in time re2uired to produce successive units than does the cumulative average*time learning curve (see 3roblem 1(*%%" re2uirement 1). !ssuming the same @4H factor is used for both curves: C(!(0"t -+ N(!/+. &' 6# ts 1 2 0 @ Est !"t+* C(!(0"t -+ D .+%t M"#('"%t(. #$ 5"/&.-H&(.s C(!(0"t -+ A-+."$+I#%.+!+#t"0 6# t-T !+ T !+ 5+".# #$ M&*+0 5+".# #$ M&*+0 1("((( 1:"((( 2@"?(( 0?"1%( 1("((( 1@"4(( %%"040 4?"%45

The reason is that" in the incremental unit*time learning model" as the number of units double" only the last unit produced has a cost of @4H of the initial cost. -n the cumulative average*time learning model" doubling the number of units causes the average cost of all the additional units produced (not 8ust the last unit) to be @4H of the initial cost. +autilus should examine its own internal records on past 8obs and see. information from engineers" plant managers" and wor.ers when deciding which learning curve better describes the behavior of direct manufacturing labor*hours on the production of the 3T1(? boats.

1(*2@

10-35 (%(#0( min.) E-"0("t #$ "0t+.#"t -+ .+$.+ss &# !&*+0s1 #&#9.&' t3

1a. 1b. 6olution )xhibit 1(*%4! plots the relationship between number of academic programs and overhead costs. 6olution )xhibit 1(*%4D plots the relationship between number of enrolled students and overhead costs.

2. 6olution )xhibit 1(*%4 compares the two simple regression models estimated by ,an.s. Doth regression models appear to perform well when estimating overhead costs. ost function 1 using number of academic programs as the independent variable appears to perform slightly better than cost function 2 which uses number of enrolled students as the independent variable. ost function 1 has a high r and goodness of fit" a high t*value indicating a significant relationship between number of academic programs and overhead costs" and meets all the specification assumptions for ordinary least s2uares regression. ost function 2 has a lower r than cost function 1 and exhibits positive autocorrelation among the residuals" as indicated by a low 9urbin*/atson statistic. %. The analysis indicates that overhead costs are related to the number of academic programs and the number of enrolled students. -f 6outhwestern has pressures to reduce and control overhead costs" it may need to loo. hard at closing down some of its academic programs and reducing its inta.e of students. <educing enrolled students may cut down on overhead costs but it also cuts down on revenues (tuition payments)" hurts the reputation of the school" and reduces its alumni base" which is a future source of funds. $or these reasons" 6outhwestern may prefer to downsi=e its academic programs" particularly those programs that attract few students. Bf course" 6outhwestern should continue to reduce costs by improving the efficiency of the delivery of its programs. SO56TION E7HIBIT 10-35A 3lot of +umber of !cademic 3rograms versus Bverhead osts (in thousands)
'0("((( %4"((( O-+.8+"* C&sts %("((( 24"((( 2("((( 14"((( 1("((( 4"(((
( ( 1( 2( %( 0( 4( 5( :( @( ?( 1(( 11( N (!/ +. &' A%"* +! % P.&$."!s

1(*2?

10-35 ( ontFd.) SO56TION E7HIBIT 10-35B 3lot of +umber of )nrolled 6tudents versus Bverhead osts (in thousands)

'0("((( %4"((( %("((( 24"((( O-+.8+"* C&sts 2("((( 14"((( 1("((( 4"(((
( ( 1"((( 2"((( %"((( 0"((( 4"((( 5"((( :"((( @"(((

N(!/ +. &' E#.&00+* St( *+ #ts

1(*%(

10-35 ( ontFd.) SO56TION E7HIBIT 10-35C omparison of !lternative ost $unctions for Bverhead <egression for 6outhwestern Kniversity C&st 4(#%t &# 1A N(!/+. &' A%"*+! % P.&$."!s "s I#*+9+#*+#t V". "/0+ ! positive relationship between overhead costs and number of academic programs is economically plausible at 6outhwestern Kniversity. r2 > (.:2. )xcellent goodness of fit. osts )stimated with 6imple C&st 4(#%t &# 2A N(!/+. &' E#.&00+* St(*+#ts "s I#*+9+#*+#t V". "/0+ ! positive relationship between overhead costs and number of enrolled students is economically plausible at 6outhwestern Kniversity. r2 > (.44. 7ood goodness of fit but not as good as for number of academic programs. t-value of %.4% is significant. 3lot of the data indicates that assumptions of linearity" constant variance" and normality of residuals hold" but inferences drawn from only 12 observations are not reliableL the 9urbin*/atson statistic > (.:: indicates that independence of residuals does not hold.

C. t+. &# 1. )conomic 3lausibility

2. 7oodness of $it

%. 6ignificance of -ndependent Variable(s) 0. 6pecification !nalysis of )stimation !ssumptions

t-value of 4.(@ is significant. 3lot of the data indicates that assumptions of linearity" constant variance" independence of residuals (9urbin*/atson statistic > 1.@1)" and normality of residuals hold" but inferences drawn from only 12 observations are not reliable.

10-36 (%( min.) E-"0("t #$ !(0t 90+ .+$.+ss &# !&*+0s1 #&#9.&' t
;%&#t #("t &# &' P.&/0+! 10-35<3 1. -t is economically plausible that the correct form of the model of overhead costs includes both number of academic programs and number of enrolled students as cost drivers. The findings in 3roblem 1(*%4 indicate that each of the independent variables affects overhead costs. ()ach regression has a significant r2 and t-value on the independent variable.) ,an.s could choose to divide overhead costs into two cost pools" (i) those overhead costs that are more closely related to number of academic program and (ii) those overhead costs more closely related to number of enrolled students" and rerun the simple regression analysis on each overhead cost pool. !lternatively" ,an.s could run a multiple regression analysis with total overhead costs as the dependent variable and the number of academic programs and number of enrolled students as the two independent variables.

1(*%1

10-36 ( ontFd.) 2. 6olution )xhibit 1(*%5! evaluates the multiple regression model using the format of )xhibit 1(*1?. ,an.s should use the multiple regression model over the two simple regression models of 3roblem 1(*%4. The multiple regression model appears economically plausible" and the regression model performs very well when estimating overhead costs. -t has an excellent goodness of fit" significant t-values on both independent variables" and meets all the specification assumptions for ordinary least*s2uares regression. There is some correlation between the two independent variables but multicollinearity does not appear to be a problem here. The significance of both independent variables (despite some correlation between them) suggests that each variable is a driver of overhead cost. Bf course" as the chapter describes" even if the independent variables exhibited multicollinearity" ,an.s should still prefer to use the multiple regression model over the simple regression models of 3roblem 1(*%4. Bmitting any one of the variables will cause the estimated coefficient of the independent variable" included in the model" to be biased away from its true value. %. 3ossible uses for the multiple regression results include: a. 3lanning and budgeting at 6outhwestern Kniversity. The regression analysis indicates the variables (number of academic programs and number of enrolled students) that help predict changes in overhead costs. ost control and performance evaluation. ,an.s could compare actual performance with budgeted or expected numbers and see. ways to improve the efficiency of the Kniversity operations" and evaluate the performance of managers responsible for controlling overhead costs. ost management. -f cost pressures increase" the Kniversity might save costs by closing down academic programs that have few students enrolled.

b.

c.

1(*%2

10-36 ( ontFd.) SO56TION E7HIBIT 10-36A )valuation of ost $unction for Bverhead 6outhwestern Kniversity C. t+. &# 1. )conomic 3lausibility osts )stimated with &ultiple <egression for

N(!/+. &' A%"*+! % P.&$."!s "#* N(!/+. &' E#.&00+* St(*+#ts "s I#*+9+#*+#t V". "/0+s ! positive relationship between overhead costs and number of academic programs and number of enrolled students is economically plausible at 6outhwestern Kniversity. r > (.@1. )xcellent goodness of fit.

2. 7oodness of $it

%. 6ignificance of -ndependent Variable(s) t-values of %.05 on number of academic programs and 2.(% on number of enrolled students are both significant. 0. 6pecification !nalysis of )stimation !ssumptions The assumptions of linearity" constant variance" and normality of residuals hold" but inferences drawn from only 12 observations are not reliableL the 9urbin*/atson statistic > 1.@0 indicates that independence of residuals holds.

1(*%%

10-37 (0(#4( min.) P(.%8"s #$ D+9".t!+#t %&st *. -+.s1 "%t - t,-/"s+* %&st #$1 s !90+ .+$.+ss &# "#"0,s s3
The problem reports the exact t*values from the computer runs of the data. Decause the coefficients and standard errors given in the problem are rounded to three decimal places" dividing the coefficient by the standard error may yield slightly different t*values. 1. 3lots of the data used in <egressions 1 to % are in 6olution )xhibit 1(*%:!. 6ee 6olution )xhibit 1(*%:D for a comparison of the three regression models. 2. Doth <egressions 2 and % are well*specified regression models. The slope coefficients on their respective independent variables are significantly different from =ero. These results support the outure $abricsJs presentation in which the number of purchase orders and the number of suppliers were reported to be drivers of purchasing department costs. -n designing an activity*based cost system" $ashion $lair should use number of purchase orders and number of suppliers as cost drivers of purchasing department costs. !s the chapter appendix describes" $ashion $lair can either (a) estimate a multiple regression e2uation for purchasing department costs with number of purchase orders and number of suppliers as cost drivers" or (b) divide purchasing department costs into two separate cost pools" one for costs related to purchase orders and another for costs related to suppliers" and estimate a separate relationship for each cost pool. %. 7uidelines 1 and 2 presented in the chapter could be used to gain additional evidence on cost drivers of purchasing department costs. Guideline !: Kse physical relationships or engineering relationships to establish cause*and* effect lin.s. ;ee could observe the purchasing department operations to gain insight into how costs are driven. Guideline : Kse .nowledge of operations. ;ee could interview operating personnel in the purchasing department to obtain their insight on cost drivers.

1(*%0

10-37 ( ontFd.) SO56TION E7HIBIT 10-37A <egression ;ines of Various ost 9rivers on 3urchasing 9ept. osts for $ashion $lair
P(.%8"s #$ D +9".t!+#t C&sts '2"4(("((( 2"((("((( 1"4(("((( 1"((("((( 4(("((( ( ( 4( 1(( 14(

D&0 0". V"0( + &' M+.%8"# * s+ P(.%8 "s+* ; # ! 00 &#s< '2"4(("((( P( .%8"s #$ D+9".t!+#t C&s ts 2"((("((( 1"4(("((( 1"((("((( 4(("((( ( ( 2"((( 0"((( 5"((( @"(((

N(!/+. &' P( .%8"s+ O.*+.s '2"4(("((( P(.%8"s #$ D+9".t !+#t C&s ts 2"((("((( 1"4(("((( 1"((("((( 4(("((( ( ( 1(( 2(( %((

N(!/+. &' S( 990 +.s

1(*%4

10-37 ( ontJd.) SO56TION E7HIBIT 10-37B omparison of !lternative ost $unctions for 3urchasing 9epartment osts )stimated with 6imple <egression for $ashion $lair R+$.+ss &# 1 R+$.+ss &# 2 R+$.+ss &# 3

C. t+. &# 1. )conomic 3lausibility

PDC ) a @ ;b MPD<

PDC ) a @ ;b E &' POs<

PDC ) a @ ;b E &' Ss<

<esult presented at seminar by outure $abrics found little support for &3' as a driver. 3urchasing personnel at the &iami store believe &3' is not a significant cost driver.

)conomically plausible. The higher the number of purchase orders" the more tas.s underta.en.

)conomically plausible. -ncreasing the number of suppliers increases the costs of certifying vendors and managing the $ashion $lair* supplier relationship. r2 > (.%?. <easonable goodness of fit. t*value on Q of 6s of 2.2@ is significant.

2. 7oodness of fit r2 > (.(@. 3oor goodness of fit. %. 6ignificance of -ndependent Variables 0. 6pecification !nalysis !. ;inearity within the relevant range D. onstant variance of residuals . -ndependence of residuals t*value on &3' of (.@0 is insignificant.

r2 > (.02. <easonable goodness of fit. t*value on Q of 3Bs of 2.0% is significant.

!ppears 2uestionable !ppears reasonable. but no strong evidence against linearity. !ppears 2uestionable" !ppears reasonable. but no strong evidence against constant variance. 9urbin*/atson 6tatistic > 2.01 !ssumption of independence is not re8ected. 9urbin*/atson 6tatistic > 1.?@ !ssumption of independence is not re8ected.

!ppears reasonable.

!ppears reasonable.

9urbin*/atson 6tatistic > 1.?: !ssumption of independence is not re8ected.

9. +ormality of residuals

9ata base too small to 9ata base too small to 9ata base too small to ma.e reliable ma.e reliable inferences. ma.e reliable inferences. inferences.

1(*%5

10-38 (%(#0( min.) P(.%8"s #$ D+9".t!+#t %&st *. -+.s1 !(0t 90+ .+$.+ss &# "#"0,s s ;C&#t #("t &# &' 10-37<3
The problem reports the exact t*values from the computer runs of the data. Decause the coefficients and standard errors given in the problem are rounded to three decimal places" dividing the coefficient by the standard error may yield slightly different t*values. 1. <egression 0 is a well*specified regression model:

Economic plausibility: Doth independent variables are plausible and are supported by the findings of the outure $abrics study. Goodness of fit: The r of (.5% indicates an excellent goodness of fit. Significance of independent "ariables: The t-value on Q of 3Bs is 2.10 while the t*value on Q of 6s is 2.((. These t*values are either significant or border on significance. Specification analysis: <esults are available to examine the independence of residuals assumption. The 9urbin*/atson statistic of 1.?( indicates that the assumption of independence is not re8ected. <egression 0 is consistent with the findings in 3roblem 1(*%: that both the number of purchase orders and the number of suppliers are drivers of purchasing department costs. <egressions 2" %" and 0 all satisfy the four criteria outlined in the text. <egression 0 has the best goodness of fit ((.5% for <egression 0 compared to (.02 and (.%? for <egressions 2 and %" respectively). &ost importantly" it is economically plausible that both the number of purchase orders and the number of suppliers drive purchasing department costs. /e would recommend that ;ee use <egression 0 over <egressions 2 and %. 2. <egression 4 adds an additional independent variable (&3') to the two independent variables in <egression 0. This additional variable (&3') has a t-value of #(.(:" implying its slope coefficient is insignificantly different from =ero. The r in <egression 4 ((.5%) is the same as that in <egression 0 ((.5%)" implying the addition of this third independent variable adds close to =ero explanatory power. -n summary" <egression 4 adds very little to <egression 0. /e would recommend that ;ee use <egression 0 over <egression 4. %. Dudgeted purchasing department costs for the Daltimore store next year are: '0@4"%@0 A ('12%.22 %"?(() A ('2"?42 11() > '1"2?("552 0. &ulticollinearity is a fre2uently encountered problem in cost accountingL it does not arise in simple regression because there is only one independent variable in a simple regression. Bne conse2uence of multicollinearity is an increase in the standard errors of the coefficients of the individual variables. This fre2uently shows up in reduced t-values for the independent variables in the multiple regression relative to their t*values in the simple regression:

1(*%:

10-38 ( ontFd.) t--"0(+ # M(0t 90+ R+$.+ss &# 2.10 2.(( 1.?4 1.@0 #(.(: t--"0(+ '.&! S !90+ R+$.+ss &#s # P.&/0+! 10-37 2.0% 2.2@ 2.0% 2.2@ (.@0

V". "/0+s #egression $: Q of 3Bs Q of 6s #egression %: Q of 3Bs Q of 6s &3'

The decline in the t*values in the multiple regressions is consistent with some (but not very high) collinearity among the independent variables. 3airwise correlations between the independent variables are: C&..+0"t &# Q of 3Bs R Q of 6s Q of 3Bs R &3' Q of 6s R &3'

(.2? (.2: (.%0

There is no evidence of difficulties due to multicollinearity in <egressions 0 and 4. 4. 9ecisions in which the regression results in 3roblems 1(*%: and 1(*%@ could be useful are:

&ost management decisions: $ashion $lair could restructure relationships with the suppliers so that fewer separate purchase orders are made. !lternatively" it may aggressively reduce the number of existing suppliers. 'urchasing policy decisions: $ashion $lair could set up an internal charge system for individual retail departments within each store. 6eparate charges to each department could be made for each purchase order and each new supplier added to the existing ones. These internal charges would signal to each department ways in which their own decisions affect the total costs of $ashion $lair. (ccounting system design decisions: $ashion $lair may want to discontinue allocating purchasing department costs on the basis of the dollar value of merchandise purchased. !llocation bases better capturing cause*and*effect relations at $ashion $lair are the number of purchase orders and the number of suppliers.

1(*%@

10-39 (%4 min.) R+$.+ss &# %&!9(t"t &#s1 +t8 %s3

1. Ksing the formulas given in the appendix" a=
( T)( S 2 ) ( S )(ST ) n (S 2 ) ( S )(S )

and b >

n ( ST ) ( S )( T ) n ( S 2 ) ( S )(S )

where n > 0 S > sum of the given S values (units produced) S2 > T > ST > > a >

(:2("((( 0(5"((("((() ( 0("((( :"25?"((("((() 0(0(5"((("((() ( 0("((( 0("((() 0(:"25?"((("((() ( 0("((()(:2("((() b > > 0( 0(5"((("((() ( 0("((()( 0("((()

> ?"((( A 1("((( A ?"((( A 12"((( > 0("((( 2 sum of s2uares of 6 values > (?"((() A (1("((()2 A (?"((()2 (12"((()2 > 0(5"((("((( sum of the given T values (manuf. labor costs) > '1:5"((( A '1:0"((( A '154"((( '2(4"((( > ':2("((( (?"((( 1:5"((() A (1("((( 1:0"((() A (?"((( 154"((() A (12"((( 2(4"((() :"25?"((("((( > 11.4( 54"(((

The regression e2uation is y > '54"((( A ('11.4( units produced) 2. !llison ,artJs benchmar. for 2uarter 4 is '54"((( A ('11.4 12"((() > '2(%"(((

%. 3eter 6mithJs benchmar. differs from !llison ,artJs benchmar. because 6mith considers all manufacturing labor costs as variable at '1@ per motor (':2("((( 0("(((). ,art recogni=es that some manufacturing labor costs are fixed and other manufacturing labor costs are variable. The cost function that ,art estimates separates out '54"((( as the fixed component of costs within the relevant range and '11.4( as the variable cost per motor. leveland )ngineering produces a large 2uantity of motors in 2uarter 4 (12"(((). 6mithJs benchmar. is high because it assumes a proportionate increase in manufacturing labor costs at '1@ per motor. ,artJs benchmar. is lower because fixed manufacturing labor costs will not change even though the volume of production is high. Bnly the variable component of manufacturing labor costs (e2ual to '11.4( per motor) will increase. ,artJs benchmar. is preferred because it recogni=es the appropriate cost*behavior patterns of manufacturing labor costs. 0. ,art should explain to 6mith why the benchmar. is lower than what 6mith had calculated. 6he should also indicate to 6mith her concern about ad8usting the numbers. 6uch behavior would violate the M6tandards of )thical conduct for &anagement !ccountsN described in hapter 1. !d8usting the numbers would violate the standards of competence" integrity" and ob8ectivity re2uired of management accountants and would be unethical. -f 6mith still insists on reporting a higher benchmar." ,art should raise the matter with 6mithJs superior. -f" after ta.ing all these steps" there is continued pressure to overstate the benchmar." ,art should consider resigning from the company rather than engaging in unethical behavior.

1(*%?

10-40

(0( min.)

H $8-0&: !+t8&*1 "0t+.#"t -+ .+$.+ss &# '(#%t &#s1 "%%.("0 "%%&(#t #$ "*F(st!+#ts3

1. 6olution )xhibit 1(*0(! presents the two data plots. The plot of engineering support reported costs and machine*hours shows two separate groups of data" each of which may be approximated by a separate cost function. The problem arises because the plant records materials and parts costs on an Oas purchased"O rather than an Oas used"O basis. The plot of engineering support restated costs and machine*hours shows a high positive correlation between the two variables (the coefficient of determination is (.?0)L a single linear cost function provides a good fit to the data. Detter estimates of the cost relation result because Cennedy ad8usts the materials and parts costs to an accrual accounting basis. 2. ,ighest observation of cost driver (!ugust) ;owest observation of cost driver (6eptember) 9ifference 6lope coefficient" b > > > #'@.%1 per machine*hour onstant (at highest observation of cost driver) onstant (at lowest observation of cost driver) The estimated cost function is y > '1"220 # '@.%1X C&st D. -+. M"%8 #+-H&(.s :% 1? 40 R+st"t+* E#$ #++. #$ S(99&.t C&sts '?55 %:( '4?5 > ' 51: # (#'@.%1 :%) > '1"220 > '1"(55 # (#'@.%1 1?) > '1"220 C&st D. -+. M"%8 #+-H&(.s :% 1? 40 R+9&.t+* E#$ #++. #$ S(99&.t C&sts ' 51: 1"(55 ' (00?)

,ighest observation of cost driver (!ugust) ;owest observation of cost driver (6eptember) 9ifference 6lope coefficient" b >

> > '11.(0 per machine*hour

1(*0(

10-40 ( ontFd.) onstant (at highest observation of cost driver) onstant (at lowest observation of cost driver) The estimated cost function is y > '15( A '11.(0 X %. The cost function estimated with engineering support restated costs better approximates the regression analysis assumptions. 6ee 6olution )xhibit 1(*0(D for a comparison of the two regressions. 0. Bf all the cost functions estimated in re2uirements 2 and %" Cennedy should choose <egression 2 using engineering support restated costs as best representing the relationship between engineering support costs and machine*hours. The cost functions estimated using engineering support reported costs are mis*specified and not*economically plausible because materials and parts costs are reported on an Oas*purchased"O rather than on an Oas*used"O basis. /ith respect to engineering support restated costs" the high*low and regression approaches yield roughly similar estimates. The regression approach is technically superior because it determines the line that best fits all observations. -n contrast" the high*low method considers only two points (observations with the highest and lowest cost drivers) when estimating the cost function. 6olution )xhibit 1(*0(D shows that the cost function estimated using the regression approach has excellent goodness of fit (r > (.?0) and appears to be well specified. 4. 3roblems Cennedy might encounter include: a. ! perpetual inventory system may not be used in this caseL the amounts re2uisitioned li.ely will not permit an accurate matching of costs with the independent variable on a month*by*month basis. b. 1uality of the source records for usage by engineers may be relatively lowL e.g." engineers may re2uisition materials and parts in batches" but not use them immediately. c. <ecords may not distinguish materials and parts for maintenance from materials and parts used for repairs and brea.downsL separate cost functions may be appropriate for the two categories of materials and parts. d. Tear*end accounting ad8ustments to inventory may mas. errors that gradually accumulate month*by*month. > > ' ?55 # ('11.(0 :%) > '15( ' %:( # ('11.(0 1?) > '15(

5. 3ic.ing the correct cost function is important for cost prediction" cost management" and performance evaluation. $or example" had Knited 3ac.aging used <egression 1 (engineering support reported costs) to estimate the cost function" it would erroneously conclude that engineering support costs decrease with machine*hours. -n a month with 5( machine*hours" <egression 1 would predict costs of '1"%?%.2( # ('10.2% 5() > '4%?.0(. -f actual costs turn out to be '@((" management would conclude that changes should be made to reduce costs. -n fact" on the basis of the preferred <egression 2" support overhead costs are lower than the predicted amount of '1:5.%@ A ('11.00 5() > '@52.:@##a performance that management should see. to replicate" not change.

1(*01

10-40 ( ontFd.) Bn the other hand" if machine*hours wor.ed in a month were low" say 24 hours" <egression 1 would erroneously predict support overhead costs of '1"%?%.2( # ('10.2% 24) > '1"(%:.04. -f actual costs are ':((" management would conclude that its performance has been very good. -n fact" compared to the costs predicted by the preferred <egression 2 of '1:5.%@ A ('11.00 24) > '052.%@" the actual performance is rather poor. Ksing <egression 1" management may feel costs are being managed very well when in fact they are much higher than what they should be and need to be managed Odown.O SO56TION E7HIBIT 10-40A 3lots and <egression ;ines for )ngineering 6upport <eported osts and )ngineering 6upport <estated osts
E#$ #++. #$ S(99&.t R+9&.t+* C&sts '1"0(( 1"2(( 1"((( 5(( @(( 0(( 2(( ( ( 1( 2( %( 0( 4( 5( :( @(

M"%8 #+-H&(.s '1"2(( 1"((( @(( 5(( 0(( 2(( ( ( 1( 2( %( 0( M"%8 #+-H&(.s 4( 5( :( @(

E#$ #++. #$ S(99&.t R+st"t+* C&sts

1(*02

10-40 ( ontFd.) SO56TION E7HIBIT 10-40B omparison of !lternative ost $unctions for )ngineering 6upport osts at Knited 3ac.aging R+$.+ss &# 1 D+9+#*+#t V". "/0+A E#$ #++. #$ S(99&.t R+9&.t+* C&sts +egative slope relationship is economically implausible over the long run. r2 > (.0%. &oderate goodness of fit. t*statistic on machine*hours is statistically significant (t > #2.%1)" albeit economically implausible. ;inearity does not describe data very well. !ppears 2uestionable" although 12 observations do not facilitate the drawing of reliable inferences. 9urbin*/atson > 2.25. <esiduals serially uncorrelated. 9atabase too small to ma.e reliable inferences. R+$.+ss &# 2 D+9+#*+#t V". "/0+A E#$ #++. #$ S(99&.t R+st"t+* C&sts 3ositive slope relationship is economically plausible. r2 > (.?0. )xcellent goodness of fit. t*statistic on machine*hours is highly statistically significant (t>1(.4?).

CRITERION 1. )conomic 3lausibility

2. 7oodness of $it

%. 6ignificance of -ndependent Variables

0. 6pecification !nalysis: !. ;inearity D. onstant variance of residuals

;inearity describes data very well. !ppears reasonable" although 12 observations do not facilitate the drawing of reliable inferences. 9urbin*/atson > 1.%1. 6ome evidence of serial correlation in the residuals. 9atabase too small to ma.e reliable inferences.

. -ndependence of residuals 9. +ormality of residuals

1(*0%

C8"9t+. 10 I#t+.#+t E2+.% s+

)he *nternet e+ercise is a"ailable to students only on the 'rentice ,all &ompanion -ebsite www.prenhall.com.horngren. Students can clic/ on &ost (ccounting0 !! th ed.0 and access the *nternet E+ercise for the chapter0 which lin/s to the -eb site of a company or organi1ation. )he *nternet E+ercise on the -eb will be updated periodically so that it is current with the latest information a"ailable on the sub2ect organi1ation3s -eb site. ( printout copy of the *nternet e+ercise for this chapter as of early 44 appears below. )he solution to the *nternet e+ercise0 which will also be updated periodically0 is a"ailable to instructors from the &ompanion -ebsite3s faculty "iew. )o access the solution0 clic/ on &ost (ccounting0 !!th ed.0 5aculty lin/0 and then register once to obtain your password through the online form. (fter the initial registration0 you will ha"e a personal login *6 and password to use to log in. ( printout of the solution to the *nternet e+ercise for this chapter as of early 44 follows. )he e+ercise and solution pro"ide instructors with an idea of the content of the *nternet e+ercise for this chapter. I#t+.#+t E2+.% s+ 6outhwest !irlines is the nationJs fifth largest domestic carrier. -t serves 4: cities with a fleet of %42 Doeing :%:s. 6outhwest 8ust mar.ed its 2@th consecutive year of profitability and en8oys the distinction of having the lowest operating cost structure in the domestic airline industry. -n this exercise you will investigate possible cost drivers for allocating salary" wages" and benefits expense. 7o to http:RRwww.iflyswa.comR" and clic. on the O!bout 6/!O lin." followed by the O-nvestor <elationsO lin.. $rom here you can access 6outhwestJs 2((( annual report. 1. The &anagement 9iscussion and !nalysis section of 6outhwestJs annual report includes charts that provide information on revenue passenger miles and the number of passengers carried over the past five years. Kse these figures to estimate the following regression e2uations: Salaries0 -ages0 and 7enefits E+pense = a 8 b9#e"enue 'assenger :iles; Salaries0 -ages0 and 7enefits E+pense = a 8 b9#e"enue 'assengers &arried; 6alaries" /ages" and Denefits expenses for 1??: and 1??5 were '1"1%5"402"((( and '???":1?"((( respectively. 6ee the 2((( income statement for 1??@*2((( expenses. 2. /hich cost driver is more closely related to 6alaries" /ages" and Denefits expenseI %. ,ow do you interpret the regression interceptI 9oes it represent a fixed costI 0. ,ow do you interpret the coefficient on the independent variableI /hat is the variable labor cost for flying 1"((( passenger milesI

1(*00

I#t+.#+t E2+.% s+ ( ontFd.) S&0(t &# t& I#t+.#+t E2+.% s+ 1a. <egression models N6MBER O4 PASSENGERS #egression Statistics < 62uare

(.?4%%:@ &oefficients *1.1)A(? 00.:1 Standard Error %.10)A(@ 4.:1 t-"alue *%.51:@@? :.@%2004 '-"alue (.(%5%(1 (.((0%%0

-ntercept +umber of passengers REVEN6E PASSENGER MI5ES #egression Statistics < 62uare

(.?:@0:5 &oefficients *?.0)A(: (.(020: Standard Error 1.22)A(@ (.((%5%: t-"alue *(.:5?42 11.5:@(? '-"alue (.0?:521 (.((1%0?

-ntercept 3assenger miles 2.

<evenue passenger miles are more closely related to salaries" wages" and benefits. The < 2 of .?:@ indicates that revenue passenger miles explain ?:.@H of the variation in salary and wage expense. /hile the coefficient on the intercept is a constant in the regression e2uation" it does not represent a fixed cost. This is because it is outside the relevant range. -n addition" a negative fixed cost does not ma.e economic sense. The coefficient on the independent variable predicts the relationship between the independent variable revenue passenger miles and the dependent variable salaries" wages and benefits expense. The regression e2uation suggests that the average wages" salary" and benefit expense associated with flying a passenger 1"((( miles is '02.0: > .(020: 1"(((.

%.

0.

1(*04

C8"9t+. 10 C"s+
63S3 BREWING IND6STRGA COST ESTIMATION 1. (a) Ksing the high*low method: ,ighest Darrels 6old ;owest Darrels 6old 9ifference slope coefficient (b) constant (a)
t

> 1:.(%: ost of 6ales > > 4.@00 ost of 6ales > 11.1?% > '42?.(?5 11.1?% > '0:.2:( > '5?5.(%? # ('0:.2:( 1:.(%:) > #' 1(?.%(( > #'1(?.%(( A ('0:.2:( Vt)

'5?5.(%? '155.?0% '42?.(?5

! potential ambiguity in the high*low method is whether high*low is defined in terms of the dependent or independent variable. -n this case" the choice is important as the highest cost of sales figure is 2(((" whereas the highest volume figure is 1??5. The independent variable is used when determining the high and low observations as the causality runs from barrels produces to costs incurred. (b) The results are reported in <egression Q1 in ase )xhibit 1Usee also ase $igure 1. )valuate the results: Bne approach is to use the following categori=ation: )conomic -nterpretation * Doth high*low and B;6 imply the expected positive relationship between costs and volume. 6tatistical 6pecification * The ma8or concern is with significant positive serial correlation in the residuals (the 9urbin*/atson 6tatistic > (.21?). The conse2uence is that the estimated standard errors are underestimates of the underlying population values. 7oodness of $it # The ad8usted <2 of (.@50 for the B;6 model indicates a reasonably good fit. ! plot of the data" however" reveals that the model does not closely fit the data in 1??? and 2((( when cost of sales increased while barrels sold decreased # see ase $igure 1. B;6 should yield more efficient (using a mean s2uare error criterion) estimates than high*low as the techni2ue is constructed to use all the data rather than two extreme points and to find the minimum mean s2uare error set of estimates. oncentrating on the two extreme observations may also result in Mtrue outliersN affecting the estimated cost* volume relationship.

1(*05

C"s+ ( ontFd.) 2. The results are reported in <egression Q2 in ase )xhibit 1. ! plot of the data is in ase $igure 2. <elative to <egression Q1" deflating by the /3-D results in several improvements in model specification: * * * t*value of the slope coefficient (b) increases from 1(.:0? to 4(.?11" <2 increases from (.@50 to (.??%" and 9urbin*/atson 6tatistic moves closer to 2 (from (.21? to (.510) +ote" however" there is still evidence of significant positive autocorrelation in the residuals ** e.g." 9urbin*/atson 6tatistic > (.510. 3roblems include: (a) The /holesale 3rice -ndex of Deer (/3-D) is an output*based index for the whole industry that need not be representative of the specific wholesale price changes for !ce. Bne could build a firm specific index to overcome this problem. (b) The /3-D may not be a good proxy for changes in input prices to either the industry or !ce. The M ost of 6alesN series is an aggregate of the costs of labor" raw materials (malt" corn" barley" hops" etc.)" depreciation" excise taxes" mar.eting" etc. 9eflating by the /3-D assumes that changes in the prices of the inputs can be well approximated by changes in the /3-D. !n alternative approach is to build a firm*specific index based on changes in prices of !ceFs inputs. (c) /ith any time*series index" issues of structural change arise. The /3-D may reflect a nonconstant mix of beer products over time (malt" light" premium" super*premium" etc.). &oreover" even if the mix of the output was constant" the mix of the inputs may have changed in the 1?@2*2((( period" e.g." an increase in the capitalRlabor ratio through the construction of mechani=ed" high*volume breweries. (d) 9ata availability problems may arise due to the delay in publishing aggregate industry indexes. %. Dac.ground information: 6erial correlation in the residuals exists when there is a systematic pattern in the residuals such that .nowledge of the residual at time t conveys information about the residuals at time tA1" tA2" etc. ! variety of reasons could cause serial correlation in the residuals of a linear model" e.g." (i) Knderlying model is nonlinearUtwo possible rationales for a nonlinear model are: * costs are Mstic.yN downwardsUi.e." when volume decreases by 1(H" costs do not decrease as much as the linear model predicts. $ixed capacity costs could potentially be important for !ce in the 1??5 to 2((( period when volume declined from 1:.(%: to 14.(?1 million barrels. * Mexperience curveN phenomenon could result in costs increasing less than the linear model predicts. Knderlying model is linear but includes more than one variable and the omitted variable results in the MresidualsN being serially correlated. Bmitted variables could include the number of employees and the number of advertisements placed in various media.

(ii)

1(*0:

C"s+ ( ontFd.) (iii) !ctivity levels over time have high percentage of commonality. -t is a common finding that models estimated in levels exhibit serial correlation * * see . /. E. 7ranger and 3. +ewbold" M6purious <egressions in )conometrics"N Eournal of )conometrics (+o. 2" 1?:0). 6erial correlation (in general) affects the efficiency (but not the unbiasedness) of the B;6 regression estimates of a and b. /ith positive (negative) correlation" the estimates of the standard errors will be understated (overstated) relative to the underlying population standard errors. Thus" one may infer that the parameter estimates are more (less) precise than they actually are. /ays to detect serial correlation in the residuals include: (i) Visual inspection of the residuals. (ii) ompute the 9urbin*/atson statistic which tests for first*order serial correlation. The relevant values for this statistic" using a 4H significance level for one independent variable" are: +o. of 3ositive !utocorrelation Bbsvns 9/ lower 9/ upper 1@ 1? 1.(% 1.(5 1.25 1.2@ +egative !utocorrelation 9/ lower 9/ upper 2.:0 2.:2 2.?: 2.?0

The results are reported in <egression Q% in ase )xhibit 1. There is substantial evidence that the serial correlation problem found in 2uestion Q2 has been considerably reduced" e.g." the 9urbin*/atson deviates much less from 2 than previously (2.421 in Q% versus (.510 in Q2 and (.21? in Q1). 0. The best choice is regression Q%. The 9urbin*/atson statistic goes up from .21? in regression Q1" and .510 in regression Q2 to 2.421 in regression Q%. This indicates that the serial correlation problem found in regressions Q1 and Q2 has been reduced. The t*value of b in regression Q% is lower than that of regression Q2" but higher than that of regression Q1. The <2 in regression Q% falls between that of Q1 and Q2" indicating that there is a reasonably good fit for this model. Ta.en together" the significantly improved 9urbin* /atson statistic of regression Q%" combined with a reasonable t*value for b and the <2" ma.es this the preferred model.

1(*0@

V *+& C"s+ ( ontFd.) Case E hibit ! Regression 617 8t 9 :;2t< 627 a =267-10 t-value =3-91$ =2-2$0 b 59-52$ 55-360 tvalue 10-749 50-910 R" 0-$64 0-993 Durbin#atson 0-219 0-614

&t 9 :;2t< =30-517 't &t &t 1 = :;2t = 2t3 637 't 't 1
1

<

4-660

1-453

47-034

12-9$9

0-90$

2-521

C"s+ 4 $(.+ 1 P0&t &' C&st-V&0(!+ D"t"

1(*0?

C"s+ ( ontFd.)

C"s+ 4 $(.+ 2 P0&t &' D+'0"t+* C&st V+.s(s V&0(!+

1(*4(