# 【评分卡】评分卡入门与创建原则——分箱、WOE、IV、分值分配

## 变量分析

### IV

IV(information value)衡量的是某一个变量的信息量，公式如下：

N为分组的组数；
IV可用来表示一个变量的预测能力。

IV预测能力
<0.03无预测能力
0.03~0.09
0.1~0.29
0.3~0.49
>=0.5极高

1. 组间差异大
2. 组内差异小
3. 每组占比不低于5%
4. 必须有好、坏两种分类

### 举例说明

<185040 l n ( 50 / 330 40 / 220 ) = − 0.182321556793955 ln(\frac{50/330}{40/220}) = -0.182321556793955
18~3010060 l n ( 100 / 330 60 / 220 ) = 0.105360515657826 ln(\frac{100/330}{60/220}) = 0.105360515657826
30~6010080 l n ( 100 / 330 80 / 220 ) = − 0.182321556793955 ln(\frac{100/330}{80/220}) = -0.182321556793955
>608040 l n ( 80 / 330 40 / 220 ) = 0.287682072451781 ln(\frac{80/330}{40/220}) = 0.287682072451781
ALL330220

## 评分卡

### 评分卡计算方法

o d d s = 好 客 户 概 率 坏 客 户 概 率 = p 1 − p odds=\frac{好客户概率}{坏客户概率}=\frac{p}{1-p}

s c o r e 总 = A + B ∗ l n ( o d d s ) score_总=A+B*ln(odds)

{ P 0 = A + B l n ( θ 0 ) P 0 + P D O = A + B l n ( 2 θ 0 ) \begin{cases} P_0 &= A+Bln(\theta_0) \\ P_0+PDO &= A+Bln(2\theta_0) \end{cases}

{ B = P D O l n 2 A = P 0 − B l n ( θ 0 ) \begin{cases} B &= \frac{PDO}{ln2} \\ A &= P_0-Bln(\theta_0) \end{cases}

P 0 P_0 P D O PDO 的值都是已知常数，可以设置 P 0 = 600 P_0 = 600 P D O = 20 PDO = 20

### 分值分配

p = 1 1 + e − θ T x p = \frac{1}{1+e^{-\theta^Tx}}

l n ( p 1 − p ) = θ T x ln(\frac{p}{1-p})=\theta^Tx

l n ( p 1 − p ) = l n ( o d d s ) ln(\frac{p}{1-p})=ln(odds)

l n ( o d d s ) = θ T x = w 0 + w 1 x 1 + ⋅ ⋅ ⋅ + w n x n ln(odds)=\theta^Tx =w_0+w_1x_1+···+w_nx_n

s c o r e 总 = A + B ∗ ( θ T x ) = A + B ∗ ( w 0 + w 1 x 1 + ⋅ ⋅ ⋅ + w n x n ) score_总 = A+B*(\theta^Tx)=A+B*(w_0+w_1x_1+···+w_nx_n)
= ( A + B ∗ w 0 ) + B ∗ w 1 x 1 + ⋅ ⋅ ⋅ + B ∗ w n x n =(A+B*w_0)+B*w_1x_1+···+B*w_nx_n

( A + B ∗ w 0 ) (A+B*w_0) 为基础分数， B ∗ w 1 x 1 , ⋅ ⋅ ⋅ , B ∗ w n x n B*w_1x_1,···,B*w_nx_n 为每个变量对应分配到的分数。

x 1 x_1 1
2

i i
( B ∗ w 1 ) ∗ W O E 11 (B*w_1)*WOE_{11}
( B ∗ w 1 ) ∗ W O E 12 (B*w_1)*WOE_{12}
···
( B ∗ w 1 ) ∗ W O E 1 i (B*w_1)*WOE_{1i}
x 2 x_2 1
2

j j
( B ∗ w 2 ) ∗ W O E 21 (B*w_2)*WOE_{21}
( B ∗ w 2 ) ∗ W O E 22 (B*w_2)*WOE_{22}
···
( B ∗ w 2 ) ∗ W O E 2 j (B*w_2)*WOE_{2j}
·········
x n x_n 1
2

k k
( B ∗ w n ) ∗ W O E n 1 (B*w_n)*WOE_{n1}
( B ∗ w n ) ∗ W O E n 2 (B*w_n)*WOE_{n2}
···
( B ∗ w n ) ∗ W O E n k (B*w_n)*WOE_{nk}

《信用风险评分卡研究》Mamdouh Refaat著
《互联网金融时代消费信贷评分建模与应用》单良著

《统计学习方法》李航著

10-23
04-18 5540

03-02 19万+
11-23 7424
05-15 2377
09-09 164
06-03 1万+
06-06 1052
08-26 5592