Propensity models with logistic regression clarity
Download as PPTX, PDF7 likes4,530 views
The document discusses propensity modeling using logistic regression for various applications such as insurance, banking, and consumer purchases. It describes how propensity models use logistic regression to predict the probability of a binary outcome based on multiple independent variables. The document then provides a specific example of building a propensity model to predict the likelihood of a customer obtaining a new mortgage within 3 months using logistic regression on customer database variables. It evaluates the economic impact of the model by estimating costs, revenues, and 5-year net present value when targeting customer segments identified by the model.
![Click to edit Master text styles [Confidential]
Propensity Models with Logistic Regression—Economic Analysis
Jeffrey Strickland, Ph.D.
January 2015](https://image.slidesharecdn.com/propensitymodelswithlogisticregression-clarity-150103103558-conversion-gate01/85/Propensity-models-with-logistic-regression-clarity-1-320.jpg)
![Click to edit Master text styles [Confidential]
• http://www.amazon.com/Jeffrey-Strickland/e/B00IQ69QZK/
• http://www.lulu.com/spotlight/strickland_jeffrey
1/3/2015 Copyright © 2014 Jeffrey Strickland 11](https://image.slidesharecdn.com/propensitymodelswithlogisticregression-clarity-150103103558-conversion-gate01/85/Propensity-models-with-logistic-regression-clarity-11-320.jpg)
Propensity models with logistic regression clarity
- 1. Click to edit Master text styles [Confidential] Propensity Models with Logistic Regression—Economic Analysis Jeffrey Strickland, Ph.D. January 2015
- 2. Propensity Models • Life Insurance • Auto Insurance • Homeowner’s Insurance • Mortgage • Re-Financing • Credit cards • Personal Loans • Investments • Satellite TV • Cable • Netflix • Online banking • Online money management • Voting • Disease 1/3/2015 Copyright © 2014 Jeffrey Strickland 2 • Propensity to buy • Propensity to use • Propensity to engage • Propensity to contract • Etc.
- 3. Propensity models • One dependent variable • Target variable/Response is binary • Yes or no • 1 or 0 • One to many independent variables • Parametric • Log-linear functional assumption 1/3/2015 Copyright © 2014 Jeffrey Strickland 3 Logistic regression
- 4. Propensity to buy a new mortgage • Target Data • Homeowner data by county • Compiled monthly • Ported to SAS/SPSS/etc. database • Independent variables • 1 to 20 customer databases • Up to 2000 variable • Up to 20 million records 1/3/2015 Copyright © 2014 Jeffrey Strickland 4 • Homebuyer model • Buy a new home within next 3 months • Acquisition window = 3 months • Preprocessing period = 3 months • Profile window = 1 year
- 5. Model Performance 1/3/2015 Copyright © 2014 Jeffrey Strickland 5 1.00 1.50 2.00 2.50 3.00 3.50 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Cumulative Lift Model Lift Actual Lift 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Cumulative % Captured Response Random Model Actual
- 6. Economic Evaluation • 5-Year NPV = $1,500.00 (approximate) • Cost per direct mail = $0.45 • Evaluated by pentile • About 9 million customers 1/3/2015 Copyright © 2014 Jeffrey Strickland 6 Assumptions
- 7. Economic Evaluation Cost of Mail $ 0.45 Direct Mail Net Profit with Model Net Profit without Model NVP $ 1,500.00 Total Mailed Acq Cost per Acq Incremental Revenue due to Model Total Mail Cost with No Model Total Mail Cost with Model PTI Net Profit Cum Net Profit with Model PTI Net Profit Cum Net Profit No Model Net Profit Difference Cum Net Profit Difference Mailed Pop Resp Rate Cum Resp Rate No Model With Model No Model With Model 5% 2.97% 2.97% 150000 1437 4457 $ 46.96 $ 15.14 $ 5,408,837 $ 67,500.00 $ 1,992.71 $ 7,930,990 $ 7,928,998 $ 7,928,998 $ 2,574,361 $ 2,506,861 $ 2,506,861 $ 5,422,136 $ 5,422,136 10% 2.17% 5.14% 150000 1437 3255 $ 46.96 $ 20.74 $ 3,254,726 $ 67,500.00 $ 1,496.75 $ 5,957,071 $ 5,955,574 $13,884,572 $ 2,574,361 $ 2,506,861 $ 5,013,723 $ 3,448,712 $ 8,870,849 15% 1.58% 6.72% 150000 1437 2363 $ 46.96 $ 28.56 $ 1,658,568 $ 67,500.00 $ 1,110.86 $ 4,421,203 $ 4,420,092 $18,304,665 $ 2,574,361 $ 2,506,861 $ 7,520,585 $ 1,913,230 $10,784,079 20% 1.35% 8.07% 150000 1437 2031 $ 46.96 $ 33.24 $ 1,062,658 $ 67,500.00 $ 906.44 $ 3,607,637 $ 3,606,731 $21,911,396 $ 2,574,361 $ 2,506,861 $ 10,027,447 $ 1,099,869 $11,883,949 25% 1.25% 9.32% 150000 1437 1872 $ 46.96 $ 36.06 $ 778,410 $ 67,500.00 $ 803.56 $ 3,198,152 $ 3,197,349 $25,108,745 $ 2,574,361 $ 2,506,861 $ 12,534,309 $ 690,4878 $12,574,436 30% 1.00% 10.31% 150000 1437 1494 $ 46.96 $ 45.19 $ 101,127 $ 67,500.00 $ 730.12 $ 2,905,897 $ 2,905,167 $28,013,913 $ 2,574,361 $ 2,506,861 $ 15,041,171 $ 398,305 $12,972,741 35% 0.99% 11.30% 150000 1437 1481 $ 46.96 $ 45.58 $ 78,195 $ 67,500.00 $ 676.64 $ 2,693,014 $ 2,692,337 $30,706,250 $ 2,574,361 $ 2,506,861 $ 17,548,033 $ 185,475 $13,158,217 40% 0.90% 12.20% 150000 1437 1349 $ 46.96 $ 50.05 $ (158,801) $ 67,500.00 $ 626.62 $ 2,493,949 $ 2,493,322 $33,199,573 $ 2,574,361 $ 2,506,861 $ 20,054,894 $ (13,539) $13,144,678 45% 0.87% 13.07% 150000 1437 1307 $ 46.96 $ 51.65 $ (233,710) $ 67,500.00 $ 598.28 $ 2,381,142 $ 2,380,544 $35,580,117 $ 2,574,361 $ 2,506,861 $ 22,561,756 $ (126,317) $13,018,360 50% 0.81% 13.89% 150000 1437 1222 $ 46.96 $ 55.22 $ (384,977) $ 67,500.00 $ 554.53 $ 2,207,035 $ 2,206,481 $37,786,598 $ 2,574,36 $ 2,506,861 $ 25,068,618 $ (300,380) $12,717,980 55% 0.77% 14.65% 150000 1437 1150 $ 46.96 $ 58.69 $ (514,643) $ 67,500.00 $ 513.98 $ 2,045,626 $ 2,045,112 $39,831,711 $ 2,574,361 $ 2,506,861 $ 27,575,480 $ (461,748) $12,256,231 60% 0.66% 15.32% 150000 1437 992 $ 46.96 $ 68.06 $ (798,125) $ 67,500.00 $ 483.64 $ 1,924,892 $ 1,924,408 $41,756,120 $ 2,574,361 $ 2,506,861 $ 30,082,342 $ (582,453) $11,673,778 65% 0.70% 16.01% 150000 1437 1045 $ 46.96 $ 64.59 $ (702,743) $ 67,500.00 $ 453.90 $ 1,806,534 $ 1,806,080 $43,562,201 $ 2,574,361 $ 2,506,861 $ 32,589,204 $ (700,781) $10,972,996 70% 0.64% 16.65% 150000 1437 963 $ 46.96 $ 70.09 $ (849,502) $ 67,500.00 $ 420.14 $ 1,672,147 $ 1,671,727 $45,233,928 $ 2,574,361 $ 2,506,861 $ 35,096,066 $ (835,134) $10,137,862 75% 0.63% 17.29% 150000 1437 949 $ 46.96 $ 71.15 $ (875,309) $ 67,500.00 $ 386.76 $ 1,539,307 $ 1,538,920 $46,772,849 $ 2,574,361 $ 2,506,861 $ 37,602,928 $ (967,940) $ 9,169,921 80% 0.48% 17.76% 150000 1437 716 $ 46.96 $ 94.33 $(1,292,708) $ 67,500.00 $ 349.63 $ 1,391,542 $ 1,391,193 $48,164,042 $ 2,574,361 $ 2,506,861 $ 40,109,789 $(1,115,668) $ 8,054,252 85% 0.48% 18.25% 150000 1437 724 $ 46.96 $ 93.27 $(1,278,218) $ 67,500.00 $ 312.11 $ 1,242,186 $ 1,241,873 $ 49,405,916 $ 2,574,361 $ 2,506,861 $ 42,616,651 $(1,264,987) $ 6,789,264 90% 0.41% 18.65% 150000 1437 611 $ 46.96 $110.47 $(1,479,972) $ 67,500.00 $ 272.76 $ 1,085,577 $ 1,085,305 $50,491,221 $ 2,574,361 $ 2,506,861 $ 45,123,513 $(1,421,556) $ 5,367,707 95% 0.31% 18.96% 150000 1437 463 $ 46.96 $145.73 $(1,744,789) $ 67,500.00 $ 224.93 $ 895,234. $ 895,009 $51,386,230 $ 2,574,361 $ 2,506,861 $ 47,630,375 $(1,611,852) $ 3,755,855 100% 0.20% 19.17% 150000 1437 304 $ 46.96 $221.68 $(2,029,022) $ 67,500.00 $ 154.29 $ 614,094. $ 613,939 $52,000,170 $ 2,574,361 $ 2,506,861 $ 50,137,237 $(1,892,922) $ 1,862,933 1/3/2015 7Copyright © 2014 Jeffrey Strickland Mail to top 35% or 1st through 7th Pentile
- 8. Economic Evaluation 1/3/2015 Copyright © 2014 Jeffrey Strickland 8 $- $50.00 $100.00 $150.00 $200.00 $250.00 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Cost per Acquisition No Model With Model $- $1,000,000.00 $2,000,000.00 $3,000,000.00 $4,000,000.00 $5,000,000.00 $6,000,000.00 $7,000,000.00 $8,000,000.00 $9,000,000.00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Net Profit Net Profit Net Profit
- 9. Economic Evaluation 1/3/2015 Copyright © 2014 Jeffrey Strickland 9 $(4,000,000.00) $(2,000,000.00) $- $2,000,000.00 $4,000,000.00 $6,000,000.00 $8,000,000.00 $10,000,000.00 $12,000,000.00 $14,000,000.00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Net Profit with Model Net Profit Difference Cum Net Profit Difference $- $10,000,000.00 $20,000,000.00 $30,000,000.00 $40,000,000.00 $50,000,000.00 $60,000,000.00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Cumulative Net Profit Cum Net Profit with Model Cum Net Profit No Model
- 10. Predictive Modeling and Analytics This book is about predictive modeling. Yet, each chapter could easily be handled by an entire volume of its own. So one might think of this as a survey of predictive models, both statistical and machine learning. We define A predictive model as a statistical model or machine learning model used to predict future behavior based on past behavior. In order to use this book, the reader should have a basic understanding of statistics (statistical inference, models, tests, etc.)-this is an advanced book. Every chapter culminates in an example using R. R is a free software environment for statistical computing and graphics. It compiles and runs on a wide variety of Unix platforms, Windows and MacOs. The book is organized so that statistical models are presented first (hopefully in a logical order), followed by machine learning models, and then applications: uplift modeling and time series. One could use this as a textbook with problem solving in R (there are no "by-hand" exercises). 1/3/2015 Copyright © 2014 Jeffrey Strickland 10
- 11. Click to edit Master text styles [Confidential] • http://www.amazon.com/Jeffrey-Strickland/e/B00IQ69QZK/ • http://www.lulu.com/spotlight/strickland_jeffrey 1/3/2015 Copyright © 2014 Jeffrey Strickland 11