# DEA模型简介

## 技术效率的概念

A210.50.625
B320.6670.533
C430.750.938
D540.81.00
E520.40.5

v = v 1 x 1 + v 2 x 2 + . . . + v n x n v = v_1x_1+v_2x_2+...+v_nx_n
u = u 1 y 1 + u 2 y 2 + . . . + u n y n u=u_1y_1+u_2y_2+...+u_ny_n

## 径向距离模型

T = { ( x , y ) : y   a r e   p r o d u c i b l e   b y   x } T = \{(x,y):y\ are\ producible\ by\ x\}

P ( x ) = { y : ( x , y ) ∈ T } P(x) = \{y:(x,y) \in T\}

P ( y ) = { x : ( x , y ) ∈ T } P(y) = \{x:(x,y) \in T\}

F k o ( y k , x k , y , x ∣ C R S ) = m a x θ s . t . ∑ k = 1 K z k y k m ≥ y k m θ m , m = 1 , . . . , M ∑ k = 1 K z k x k n ≤ x k n θ n , n = 1 , . . . , N z k ≥ 0 F_k^o(y_k,x_k,y,x|CRS)=max\theta \\ s.t. \sum_{k=1}^Kz_ky_{km} \geq y_{km}\theta_m,m=1,...,M \\ \sum_{k=1}^Kz_kx_{kn} \leq x_{kn}\theta_n,n=1,...,N \\ z_k \geq 0

## 非径向效率模型

Färe, Grosskopf, and Lovell 1994a）法。则对于以产出为导向的非径向量度定义为：
R M k o ( y K , x K , y , x ∣ C R S ) = m a x { M − 1 ∑ m = 1 M θ m : ( θ 1 y k 1 , . . . , θ m y k m ∈ P ( x ) , θ m ≥ 0 , m = 1 , . . . M ) } RM_k^o(y_K,x_K,y,x|CRS)=max \{M^{-1}\sum_{m=1}^M \theta_m: (\theta_1y_{k1},...,\theta_my_{km}\in P(x),\theta_m\geq0,m=1,...M)\}

R M k o ( y K , x K , y , x ∣ C R S ) = M − 1 m a x ∑ m = 1 M θ m s . t . ∑ k = 1 K z k y k m ≥ y k m θ m , m = 1 , . . . , M ∑ k = 1 K z k x k n ≤ x k n θ n , n = 1 , . . . , N z k ≥ 0 RM_k^o(y_K,x_K,y,x|CRS)=M^{-1}max\sum_{m=1}^M \theta_m \\ s.t. \sum_{k=1}^Kz_ky_{km} \geq y_{km}\theta_m,m=1,...,M \\ \sum_{k=1}^Kz_kx_{kn} \leq x_{kn}\theta_n,n=1,...,N \\ z_k \geq 0

## 对径向模型使用Boostrap进行假设检验

T e s t ≠ 1 : H 0 : T   i s   g l o b a l l y   C R S H 1 : T   i s   V R S Test \neq 1: H_0: T\ is\ globally\ CRS \\ H_1: T\ is\ VRS

T e s t ≠ 2 : H 0 ′ : T   i s   g l o b a l l y   N I R S H 1 : T   i s   V R S Test \neq 2: H_0': T\ is\ globally\ NIRS \\ H_1: T\ is\ VRS

# stata命令的实现

tenonradial使用非径向模型RM估计技术效率，语法详情：
tenonradial outputs = inputs [(ref outputs = ref inputs)] [if] [in] [,rts(rtsassumption) base(basetype) reference(varname) tename(newvar) noprint]

output是产出变量
input是投入变量
ref outputs是产出变量的个数
ref inputs是投入变量的个数
rts(rtsassumption)指定规模收益假设，有CRS,VRS,NIRS三种
base(basetype) 设置最优化的方向，即面向产出base(output)，面向投入base(input)
reference(varname)设定技术参考集
tename(newvar)产生newvar，其包含非径向测量的技术效率。
noprint 取消估算详细信息，数据描述和参考集。

teradialbc命令使用径向模型估计技术效率，语法详情：
teradialbc outputs = inputs [(ref outputs = ref inputs)] [if] [in] [,rts(rtsassumption)base(basetype) reference(varname) subsampling kappa(#) smoothed heterogeneous reps(#) level(#) tename(newvar) tebc(newvar) biasboot(newvar) varboot(newvar) biassqva(newvar)telower(newvar) teupper(newvar) noprint nodots]

stata需要已下载kdens bw()mm quantile()

## nptestind命令

nptestind进行独立性检验

## nptestrts命令

nptestrts进行规模收益检验

# 实例应用

set seed 717117
use ccr81
generate dref = x5 != 10
teradial y1 y2 y3 = x1 x2 x3 x4 x5, rts(crs) base(output) reference(dref) tename(TErdCRSo)
teradial y1 y2 y3 = x1 x2 x3 x4 x5, rts(nirs) base(output) reference(dref) tename(TErdNRSo) noprint
teradial y1 y2 y3 = x1 x2 x3 x4 x5, rts(vrs) base(output) reference(dref) tename(TErdVRSo) noprint
tenonradial y1 y2 y3 = x1 x2 x3 x4 x5, rts(crs) base(output) reference(dref) tename(TEnrCRSo) noprint
tenonradial y1 y2 y3 = x1 x2 x3 x4 x5, rts(nirs) base(output) reference(dref) tename(TEnrNRSo) noprint
tenonradial y1 y2 y3 = x1 x2 x3 x4 x5, rts(vrs) base(output) reference(dref) tename(TEnrVRSo) noprint
list TErdCRSo TErdNRSo TErdVRSo TEnrCRSo TEnrNRSo TEnrVRSo in 1/7


matrix testsindpv = J(2, 3, .)
matrix colnames testsindpv = CRS NiRS VRS
matrix rownames testsindpv = output-based input-based
nptestind y1 y2 y3 = x1 x2 x3 x4 x5, rts(crs) base(output) reps(999) alpha(0.05)
matrix testsindpv[1,1] = e(pvalue)
nptestind y1 y2 y3 = x1 x2 x3 x4 x5, rts(nirs) base(output) reps(999) alpha(0.05) noprint
matrix testsindpv[1,2] = e(pvalue)
nptestind y1 y2 y3 = x1 x2 x3 x4 x5, rts(vrs) base(output) reps(999) alpha(0.05) noprint
matrix testsindpv[1,3] = e(pvalue)
nptestind y1 y2 y3 = x1 x2 x3 x4 x5, rts(crs) base(input) reps(999) alpha(0.05) noprint
matrix testsindpv[2,1] = e(pvalue)
nptestind y1 y2 y3 = x1 x2 x3 x4 x5, rts(nirs) base(input) reps(999) alpha(0.05) noprint
matrix testsindpv[2,2] = e(pvalue)
nptestind y1 y2 y3 = x1 x2 x3 x4 x5, rts(vrs) base(input) reps(999) alpha(0.05) noprint
matrix testsindpv[2,3] = e(pvalue)
matrix list testsindpv


P值结果如下：


CRS       NiRS        VRS
output-based  .06906907  .25625626  .04804805
input-based  .03703704    .001001  .23323323


## Malmquist指数

Malmquist指数的计算与分解公式此处不再列出，只是用程序来举例

use pwt56, clear
reshape wide y k l, i(nu country) j(year)
teradial y1965 = k1965 l1965 (y1965 = k1965 l1965), rts(crs) base(output) tename(F11) noprint
teradial y1990 = k1990 l1990 (y1965 = k1965 l1965), rts(crs) base(output) tename(F21) noprint
teradial y1965 = k1965 l1965 (y1990 = k1990 l1990), rts(crs) base(output) tename(F12) noprint
teradial y1990 = k1990 l1990 (y1990 = k1990 l1990), rts(crs) base(output) tename(F22) noprint
generate mpi = sqrt(F12 / F22 * F22 / F21)
generate effch = F11 / F22
generate techch = mpi / effch


PS：文章中所用到的学习资料为《数据包络分析方法与MaxDea软件》与文章“Nonparametric frontier analysis using Stata
”，此资料可在我的公众号“肖夕木的自习室”中回复dea获取。至于文中的数据文件可以自行在stata上下载，当然，在我的dea学习资料中也有包含。

09-06
05-12
11-08
02-26 1万+
©️2020 CSDN 皮肤主题: 游动-白 设计师:白松林