SAS评分卡建模—PSI

66号学苑 2022-09-08

The following article is from 风控建模

今天小编就介绍一下PSI这个指标。PSI也属于KL散度的一种，它的计算公式和IV的计算公式一样，同样是衡量两个样本的独立程度，值越大，独立程度越大，所以两个样本就发生了变化，这里注意和IV是反的。

话不多说了，毕竟小编的表达能力很不好，上SAS代码吧：

%macro psi(moddata=,indata=,scorename=,psi=);
/*
moddata : 参照数据集（这里是模型数据集）
indata：比较数据集（这里是验证数据集）
同时要注意两个数据集需要比较的特征名要一样，这里是验证分数稳定性，因此模型分数名称要一样
分数要自己分割
*/
proc sql;
 create table temp001 as select min(&scorename) as min,max(&scorename) as max
 from &moddata;
quit;

data temp002;
 set &moddata;
 if &scorename<=300 then score_level=1;
 else if &scorename<=350 then score_level=2;
 else if &scorename<=400 then score_level=3;
 else if &scorename<=450 then score_level=4;
 else if &scorename<=500 then score_level=5;
 else if &scorename<=550 then score_level=6;
 else if &scorename<=600 then score_level=7;
 else if &scorename<=650 then score_level=8;
 else if &scorename<=700 then score_level=9;
 else score_level=10;
run;

proc sql;
 create table temp003 as select score_level,count(*) as count
 from temp002
 group by score_level;
quit;

proc sql;
 select count(*) into:sum from &moddata;
quit;

data modbase;
 set temp003;
 model_ratio=count/&sum.;
run;

proc sql;
 create table temp001 as select min(&scorename) as min,max(&scorename) as max
 from &indata;
quit;

data temp002;
 set &indata;
 if &scorename<=300 then score_level=1;
 else if &scorename<=350 then score_level=2;
 else if &scorename<=400 then score_level=3;
 else if &scorename<=450 then score_level=4;
 else if &scorename<=500 then score_level=5;
 else if &scorename<=550 then score_level=6;
 else if &scorename<=600 then score_level=7;
 else if &scorename<=650 then score_level=8;
 else if &scorename<=700 then score_level=9;
 else score_level=10;
run;

proc sql;
 create table temp003 as select score_level,count(*) as count
 from temp002
 group by score_level;
quit;

proc sql noprint;
 select count(*) into:sum from &indata;
quit;

data temp004;
 set temp003;
 ratio=count/&sum.;
run;

proc sql;
 create table psi as select a.score_level,a.model_ratio,b.ratio
 from modbase a
 full join temp004 b
 on a.score_level=b.score_level;
quit;

data psi;
 set psi;
 if model_ratio=. then model_ratio=0;
 if ratio=. then ratio=0;
 psi=(model_ratio-ratio)*log(model_ratio/ratio);
run;

proc sql;
 create table &psi. as select sum(psi) as psi
 from psi;
quit;

%mend ;