1、 EfficientFrontierAnalyticTeamJuly 2010 SettingWeightsonObjectiveFunctionforLongChannelClient Question Whataretheweights Whyweneedsetweights Howwesetweights WhatareWeights WeightedRevenue 1 Metric1 2 Metric2 It stheproportionbetweenthemetricsthatmatters nottheweights WhyweneedtosettheWeights Thelast
2、twokeywordsshouldnotbetreatedasrevenue0Theyshouldbetreateddifferently TwoWaystoGo Client sWay Evaluatethevalueofeverymetric OurWay Catchtherelationshipbetweeneachmetricinthechanneltolifttheperformanceoftail RelyhighlyonintuitionIncludeeverymetric Howwelifttailperformancebysettingweights Weightsthatb
3、estpredictrevenue Wegetestimationsoftherevenueoftwotailterms Weareabletotreatthemdifferently Howdowecomeupwiththebestweights ShouldEveryMetricbeIncluded Everymetriccontainsvalueofothermetrics co linearproblemEverymetrichassomevolatility Somoremetrics morenoise LARSAlgorithm LessisMore Step1 Revenue
4、0 Metric1 0 Metric2 3 Metric3 0 Metric4PredictiveError 0 85 Step2 Revenue 1 Metric1 0 Metric2 3 Metric3 0 Metric4PredictiveError 0 75 Step3 Revenue 1 Metric1 2 Metric2 3 Metric3 0 Metric4PredictiveError 0 78 Step0 Revenue 0 Metric1 0 Metric2 0 Metric3 0 Metric4PredictiveError 1 Step4 Revenue 1 Metri
5、c1 2 Metric2 3 Metric3 4 Metric4PredictiveError 0 81 Step2givesthebestpredictionofrevenuebasedonothermetrics The Ideal Solution EstimatedRevenue 1 Lead 3 Approved LimitonUI OneObjectiveFunctionacrossPortfolio CompromiseSolution Revenue 1 Lead 3 Approved WeightedRevenue 1 Revenue 1 Lead 3 Approved At
6、radeoffbetweenaccuracyfortheheadvs anestimateforthetail WellsFargoCaseStudy AutoTDandUW TDhavepositivecorrelationwithTotal Apps VariableSelection LARSAlgorithm Step2givesbestpredictionofTotalLoanwithlowestpredictiveerrorbasedonothermetrics RecommendedObjectiveFunction Total AppscapturesthevolumeofloanapplicationsApprovedcapturesthevolumeofqualifiedapplicationsTotalLoancapturesthefinalconfirmedloan
