Part 4 Modeling Profitability Instead of Default.txt

Modeling Profitability Instead of Default Modeling Profitability Level as a Continuous Output (Instead of Binary Classification Default/No Default) Introduction Bot your o!n "odel and te forecast based on #ggertopia scores are binary classifications classifications$ $ tey forecast forecast one of %ust t!o outco"es$ outco"es$ � Default� or � No Default& Default&� 'our boss is interested in te idea tat it "igt be preferable instead to "odel and forecast profits and losses as continuous values using a a "ultivariate linear regression "odel on te sa"e si input variables& *is idea as arisen because te ban+ as been revie!ing individual profit and loss nu"bers for eac custo"er over te tree,year period and as "ade an interesting discovery$ so"e defaulting custo"ers carried so "uc debt for so long and paid so "uc interest on it tat tey !ere profitable for te ban+ even toug tey defaulted- Many custo"ers !o see" to ave ris+y spending beaviors are also a"ong te "ost profitable for a lending business& .nd at te opposite etre"ecusto"ers !o al!ays paid off teir cards in full eac "ont never defaulted but !ere not very profitable$ te ban+ barely barely bro+e bro+e even even or even even lost lost "oney "oney on its� safest safest� borro!e borro!ers& rs& 'our boss as+s as+s you to forecast forecast eac applicant applicant� s epected profitabi profitability lity in dollarsbefore deciding !eter or not to issue te" a credit card& e !ants to +no! o! reliable tis type of forecast !ould be$ !at is te range above and belo! te point esti"ate tat !ill be correct 012 of te ti"e3 .ltoug it "igt be possible to co"bine te si inputs in oter !ays in te interests of ti"e and focusing on te +ey learning ob%ectives !e !ill use only a si"ple linear co"bination of te si input variables for Part 4 of tis Pro%ect& ('ou sould not include te #ggertopia 5cores as an input variable)& 6uestion 7 is about te coefficie coefficients nts or or � betas� used to co"bine co"bine te standard standardi8ed i8ed inputs to get te best,fit,line on standardi8ed outputs on te *raining 5et& 9e ten use tose fied betas to "easure te observed residual error of te "odel on te *est 5et& 6uestions : troug ; concern te forecasts on te *est 5et& 6uestions < troug 77 loo+ at te *raining 5et results so tat tey can be co"pared (for possible over,fitting) against te *est 5et =esults& 6uestions 7: troug 74 are about te uncertainty tat re"ains in a ne! individual forecast of profitability& >se te te #cel #cel � Linest� function function on te si si inputs inputs and and profitabil profitability ity output output on te te :11 *raining *raining 5et applican applicants ts to calculate calculate te te coefficient coefficients s (te (te � betas� ) tat tat result result in te best,fit line& 6uestion$ Do you feel prepared to ta+e tis ?ui83 'es

r

6uestion$ 6uestion$ 9at 9at are your values for eac .ge

�

beta� on te te *raining *raining 5et3

'ears at current e"ployer 'ears at current address Inco"e over te past year Current credit card debt Current auto"obile debt &17 &70 ,&1< &;4 ,&1; 1

r

@or tis ?uestion use te Liner =egression @orecasting eplanation and #cel spreadseet& 6uestion$ 9at is te root,"ean,s?uare residual (te standard deviation of "odel error) on 5tandardi8ed output for te *est 5et3 &A:1  1&;<1 r

@or tis ?uestion use te Linear =egression @orecasting #planation and 5preadseet& 6uestion$ 9at is te observed correlation = on te *est 5et3 &<01  &
@or tis ?uestion use te Linear =egression @orecasting eplanation and #cel spreadseet& 6uestion$ 9at is te 5tandard deviation of "odel error in Dollars for te *est 5et3 AA<0&A; A&74

 r

@or tis ?uestion use te Linear =egression @orecasting eplanation and #cel spreadseet$ 6uestion$ 9at is te 012 confidence interval in dollars for te *est 5et3 & above te point esti"ate and & belo! te point esti"ate ;<4&7< above te point esti"ate and ;<4&7< belo! te point esti"ate ;77&07 above te point esti"ate and ;77&07 belo! te point esti"ate

  

9at is te Percentage Infor"ation Eain (P&I&E&) on te *est 5et3 A<&:2 7&02 :;&42 :<&<2

   r

@or tis ?uestion use te Linear =egression @orecasting eplanation and #cel

spreadseet$ 6uestion$ 9at is te Correlation = of your "odel on te *raining 5et3 &<1 &<1 &10

  r

@or tis ?uestion use te Linear =egression @orecasting eplanation and #cel spreadseet$ 'ou need to ?uantify te uncertainty in a regression "odel forecast of applicants� future profitability& .ssu"e tat bot te forecast profits and te errors ave a Eaussian distribution& 'ou !ill calculate te standard deviation of "odel error on standardi8ed data te standard deviation in dollars of te "odel error and te 012 confidence interval for profitability esti"ates& 6uestion$ 9at is te standard deviation of your "odel error on te standardi8ed *raining 5et output3 &<

r

@or tis ?uestion use te Linear =egression @orecasting eplanation and #cel spreadseet& 6uestion$ 9at is te standard deviation of "odel error in dollars on te *raining 5et3 FF*is "ay see" si"ilar to ?uestion  but 6 refers to te *est 5et& 4A7:&07 AA<0&A;

 r

@or tis ?uestion use te Linear =egression @orecasting eplanation and #cel spreadseet& 6uestion$ 9at is te 012 confidence interval in dollars on te *raining 5et3 FF*is "ay see" si"ilar to ?uestion ; but 6; refers to te *est 5et& A:&0A &

 r

@or tis ?uestion use te Linear =egression @orecasting eplanation and #cel spreadseet& 6uestion$ 9at is te Percentage Infor"ation Eain (P&I&E&) on te *raining 5et3 FF*is "ay see" si"ilar to ?uestion < but 6< refers to te *est 5et& 47&42 A:&42

 

A;&2 A<&2

 r

6uestions 7A troug 7 use te sa"e ea"ple applicant& *e follo!ing data are +no!n about te sa"ple applicant$ .ge$ 4:&11 'ears at #"ployer$ 7:&44 'ears at .ddress$ 1&0 Inco"e$ 7:7411 CC debt$ ,A4:: .uto debt$ ,:A477 *o convert above inputs to standardi8ed for" locate te *raining 5et 5preadseet (first botto" tab of !or+boo+) in te Data for @inal Pro%ect 9or+boo+& DataGforG@inalGPro%ect&ls >se te input "eans HCells C:1<$:1< and standard deviations HCells C:10$:10& >se te *raining 5et profitability "ean H701&7 and standard deviation H<&07 fro" te Profit and Loss (last botto" tab) 5preadseet& >se te *est 5et standard deviation of error on standardi8ed outputs of &;<1 6uestion$ 9at is te point esti"ate of profitability in dollars3 ,71;A&;7 71;A&;7

 r

*e follo!ing data are +no!n about te sa"ple applicant$ .ge$ 4:&11 'ears at #"ployer$ 7:&44 'ears at .ddress$ 1&0 Inco"e$ 7:7411 CC debt$ ,A4:: .uto debt$ ,:A477 *o convert above inputs to standardi8ed for" locate te *raining 5et 5preadseet (first botto" tab) in te Data for @inal Pro%ect 9or+boo+& >se tose "eans HCells C:1<$:1< and standard deviations HCells C:10$:10&

>se te *raining 5et profitability "ean H701&7 and standard deviation H<&07 fro" te Profit and Loss (last tab on botto") 5preadseet >se te *est 5et standard deviation of error on standardi8ed outputs of &;<1 6uestion$ 9it 12 confidence !at is te range of profitability3 =ange fro" 7:0;:&;7 to 71;A&;7 =ange fro" 77:A&: to 04A&04 =ange fro" 7AA14&7; to 1;A&1;&

  r

*e follo!ing data are +no!n about te sa"ple applicant$ .ge$ 4:&11 'ears at #"ployer$ 7:&44 'ears at .ddress$ 1&0 Inco"e$ 7:7411 CC debt$ ,A4:: .uto debt$ ,:A477 *o convert above inputs to standardi8ed for" locate te *raining 5et 5preadseet (botto" tab) in te Data for @inal Pro%ect 9or+boo+& >se tose "eans HCells C:1<$:1< and standard deviations HCells C:10$:10& >se te *raining 5et profitability "ean H701&7 and standard deviation H<&07 fro" te Profit and Loss (botto" tab) 5preadseet >se te *est 5et standard deviation of error on standardi8ed outputs of &;<1 & 6uestion$ 9it 002 confidence !at is te range of profitability3 =ange fro" :1;07&A: to ;<&01&

r

Co"paring *est 5et and *raining 5et Perfor"ance 6uestion 7$ Bet!een te *raining 5et and te *est 5et te dollar value of te standard deviation of "odel error� Decreased by about 72 !ic suggests a very strong "odel on *est 5et data& Increased by "ore tan :2 !ic suggests possible "odel over,fitting&  Increased by less tan :12 !ic suggests "ini"al "odel over,fitting& r Increased by "ore tan 12 !ic leads to te conclusion of "odel over,fitting& 

