关于frontier4.01软件！讨论者有奖励！问题时刻更新！

刚才又看了一下SFA中的成本模型，里面的确是要包含四个基本的变量！（where Ci, Qi, R_i and W_i are cost, output, capital price and labour price！ ）而我给的数据的确不符合数据的基本要求！

现在的我主要是使用SFA中的成本思想，

ln(C_i/W_i) = b₀ + b₁ln(Q_i) + b₂ln(R_i/W_i) +。。。。。。。+ (V_i + U_i), ln(C_i/W_i) 整体代表一个变量（EXCEL中的产出数据），ln(Q_i) 整体代表一个变量（EXCEL中的投入数据），ln(R_i/W_i)又整体表一个变量（EXCEL中的投入数据），省略号代表一系列变量！

其目的在于求出V_i 和 U_i 对ln(C_i/W_i) 的解释程度！

[此贴子已经被作者于2006-1-14 22:42:33编辑过]

今天自己使用了一些数据，原始数据和结果数据如下(模型是第一个）：

原始数据：

1 1 9.125087 8.060856 1.203973
2 1 8.942435 8.035279 1.203973
3 1 8.587652 7.431300 5.516690
4 1 8.376229 6.993015 5.302558
5 1 8.470206 7.311218 6.187072
6 1 8.496705 7.700295 6.050441
7 1 8.407378 7.325808 5.415611
8 1 8.432746 7.373374 5.892279
9 1 9.299413 8.677610 1.203973
10 1 8.785417 7.847763 6.105864
11 1 9.039315 7.936303 1.203973
12 1 8.530030 7.569928 7.187740
13 1 8.833434 8.161375 6.635710
14 1 8.459691 7.400621 6.605826
15 1 8.667164 7.690743 6.997111
16 1 8.419007 7.226936 6.649956
17 1 8.558872 7.437206 5.900692
18 1 8.668265 7.573017 6.704929
19 1 9.118872 7.952263 1.203973
20 1 8.633998 7.285507 7.401628
21 1 8.582663 7.629490 1.203973
22 1 8.682029 7.208600 5.827415
23 1 8.608477 7.383989 6.888878
24 1 8.503905 6.986566 6.702918
25 1 8.728863 7.541683 7.044940
26 1 8.445504 7.073270 6.281706
27 1 8.406306 6.858565 5.412226
28 1 8.456041 7.098376 1.203973
29 1 8.405792 7.167809 2.972464
30 1 8.579191 7.541683 1.203973

结果数据：

Output from the program FRONTIER (Version 4.1c)

instruction file = terminal
data file = eg1.dta

Error Components Frontier (see B&C 1992)
The model is a production function
The dependent variable is logged

the ols estimates are :

coefficient standard-error t-ratio

beta 0 0.51051052E+01 0.47208582E+00 0.10813935E+02
beta 1 0.48249259E+00 0.59662713E-01 0.80870040E+01
beta 2 -0.18546524E-01 0.10177004E-01 -0.18223953E+01
sigma-squared 0.13809479E-01

log likelihood function = 0.23248253E+02

the estimates after the grid search were :

beta 0 0.51253195E+01
beta 1 0.48249259E+00
beta 2 -0.18546524E-01
sigma-squared 0.12837151E-01
gamma 0.50000000E-01
mu is restricted to be zero
eta is restricted to be zero


iteration = 0 func evals = 20 llf = 0.23246066E+02
0.51253195E+01 0.48249259E+00-0.18546524E-01 0.12837151E-01 0.50000000E-01
gradient step
iteration = 5 func evals = 47 llf = 0.23247087E+02
0.51210078E+01 0.48248853E+00-0.18553826E-01 0.12664364E-01 0.29403108E-01
iteration = 10 func evals = 135 llf = 0.23248167E+02
0.51115837E+01 0.48251636E+00-0.18552168E-01 0.12472735E-01 0.53219701E-02
iteration = 15 func evals = 223 llf = 0.23248235E+02
0.51090383E+01 0.48252058E+00-0.18547698E-01 0.12447338E-01 0.21366470E-02
iteration = 20 func evals = 330 llf = 0.23248245E+02
0.51080881E+01 0.48250617E+00-0.18546881E-01 0.12438801E-01 0.11830941E-02
iteration = 25 func evals = 437 llf = 0.23248250E+02
0.51072250E+01 0.48248738E+00-0.18546320E-01 0.12432449E-01 0.53924715E-03
iteration = 30 func evals = 547 llf = 0.23248252E+02
0.51067374E+01 0.48249090E+00-0.18546803E-01 0.12431025E-01 0.32633570E-03
iteration = 35 func evals = 655 llf = 0.23248252E+02
0.51064405E+01 0.48248922E+00-0.18546533E-01 0.12430009E-01 0.21426392E-03
iteration = 40 func evals = 765 llf = 0.23248253E+02
0.51060339E+01 0.48249262E+00-0.18546620E-01 0.12429392E-01 0.10799047E-03
iteration = 45 func evals = 859 llf = 0.23248253E+02
0.51058665E+01 0.48249245E+00-0.18546854E-01 0.12429112E-01 0.72238791E-04
iteration = 49 func evals = 914 llf = 0.23248253E+02
0.51057744E+01 0.48249264E+00-0.18546520E-01 0.12428998E-01 0.56289222E-04

the final mle estimates are :

coefficient standard-error t-ratio

beta 0 0.51057744E+01 0.57673215E+00 0.88529388E+01
beta 1 0.48249264E+00 0.56050136E-01 0.86082332E+01
beta 2 -0.18546520E-01 0.93463359E-02 -0.19843627E+01
sigma-squared 0.12428998E-01 0.32026377E-02 0.38808629E+01
gamma 0.56289222E-04 0.66304876E-01 0.84894543E-03
mu is restricted to be zero
eta is restricted to be zero

log likelihood function = 0.23248253E+02

the likelihood value is less than that obtained
using ols! - try again using different starting values

number of iterations = 49

(maximum number of iterations set at : 100)

number of cross-sections = 30

number of time periods = 1

total number of observations = 30

thus there are: 0 obsns not in the panel

covariance matrix :

0.33261997E+00 -0.24791495E-01 -0.20314894E-02 0.27037102E-03 0.24543252E-01
-0.24791495E-01 0.31416177E-02 0.21841877E-03 -0.13417796E-04 -0.21126258E-04
-0.20314894E-02 0.21841877E-03 0.87353995E-04 -0.11359346E-05 0.67804083E-05
0.27037102E-03 -0.13417796E-04 -0.11359346E-05 0.10256888E-04 0.29959413E-04
0.24543252E-01 -0.21126258E-04 0.67804083E-05 0.29959413E-04 0.43963366E-02

technical efficiency estimates :

firm eff.-est.

1 0.99933610E+00
2 0.99933262E+00
3 0.99933296E+00
4 0.99933288E+00
5 0.99933200E+00
6 0.99932864E+00
7 0.99933028E+00
8 0.99933051E+00
9 0.99933359E+00
10 0.99933312E+00
11 0.99933557E+00
12 0.99933105E+00
13 0.99933121E+00
14 0.99933106E+00
15 0.99933259E+00
16 0.99933196E+00
17 0.99933246E+00
18 0.99933366E+00
19 0.99933703E+00
20 0.99933605E+00
21 0.99932926E+00
22 0.99933719E+00
23 0.99933437E+00
24 0.99933608E+00
25 0.99933534E+00
26 0.99933388E+00
27 0.99933486E+00
28 0.99933192E+00
29 0.99933088E+00
30 0.99933006E+00

mean efficiency = 0.99933297E+00

存在的问题如下：为什么计算结果是一致的？很奇怪！

另外一组数据（主要体现在红色地方）

1 1 9.125087 8.060856 -1.203973
2 1 8.942435 8.035279 -1.203973
3 1 8.587652 7.431300 5.516690
4 1 8.376229 6.993015 5.302558
5 1 8.470206 7.311218 6.187072
6 1 8.496705 7.700295 6.050441
7 1 8.407378 7.325808 5.415611
8 1 8.432746 7.373374 5.892279
9 1 9.299413 8.677610 -1.203973
10 1 8.785417 7.847763 6.105864
11 1 9.039315 7.936303 -1.203973
12 1 8.530030 7.569928 7.187740
13 1 8.833434 8.161375 6.635710
14 1 8.459691 7.400621 6.605826
15 1 8.667164 7.690743 6.997111
16 1 8.419007 7.226936 6.649956
17 1 8.558872 7.437206 5.900692
18 1 8.668265 7.573017 6.704929
19 1 9.118872 7.952263 -1.203973
20 1 8.633998 7.285507 7.401628
21 1 8.582663 7.629490 -1.203973
22 1 8.682029 7.208600 5.827415
23 1 8.608477 7.383989 6.888878
24 1 8.503905 6.986566 6.702918
25 1 8.728863 7.541683 7.044940
26 1 8.445504 7.073270 6.281706
27 1 8.406306 6.858565 5.412226
28 1 8.456041 7.098376 -1.203973
29 1 8.405792 7.167809 2.972464
30 1 8.579191 7.541683 -1.203973

结果数据如下：

Output from the program FRONTIER (Version 4.1c)

instruction file = terminal
data file = eg1.dta

Error Components Frontier (see B&C 1992)
The model is a production function
The dependent variable is logged

the ols estimates are :

coefficient standard-error t-ratio

beta 0 0.51435992E+01 0.46904812E+00 0.10966037E+02
beta 1 0.47333440E+00 0.60376935E-01 0.78396560E+01
beta 2 -0.14160271E-01 0.71610594E-02 -0.19773990E+01
sigma-squared 0.13546344E-01

log likelihood function = 0.23536831E+02

the estimates after the grid search were :

beta 0 0.51636200E+01
beta 1 0.47333440E+00
beta 2 -0.14160271E-01
sigma-squared 0.12592543E-01
gamma 0.50000000E-01
mu is restricted to be zero
eta is restricted to be zero


iteration = 0 func evals = 20 llf = 0.23535011E+02
0.51636200E+01 0.47333440E+00-0.14160271E-01 0.12592543E-01 0.50000000E-01
gradient step
iteration = 5 func evals = 47 llf = 0.23535824E+02
0.51601869E+01 0.47325552E+00-0.14169835E-01 0.12432954E-01 0.30614967E-01
iteration = 10 func evals = 134 llf = 0.23536640E+02
0.51511848E+01 0.47353459E+00-0.14149079E-01 0.12272011E-01 0.10283794E-01
iteration = 15 func evals = 222 llf = 0.23536773E+02
0.51510907E+01 0.47316861E+00-0.14170655E-01 0.12229288E-01 0.48109891E-02
iteration = 20 func evals = 330 llf = 0.23536815E+02
0.51473837E+01 0.47340084E+00-0.14156401E-01 0.12210190E-01 0.23574759E-02
iteration = 25 func evals = 435 llf = 0.23536825E+02
0.51465559E+01 0.47335376E+00-0.14159165E-01 0.12201338E-01 0.12240058E-02
iteration = 30 func evals = 529 llf = 0.23536830E+02
0.51454715E+01 0.47333306E+00-0.14160379E-01 0.12195315E-01 0.44290417E-03
iteration = 35 func evals = 635 llf = 0.23536830E+02
0.51450727E+01 0.47333998E+00-0.14159923E-01 0.12193898E-01 0.29598221E-03
iteration = 40 func evals = 745 llf = 0.23536831E+02
0.51446819E+01 0.47334022E+00-0.14159896E-01 0.12192858E-01 0.16334020E-03
iteration = 45 func evals = 854 llf = 0.23536831E+02
0.51445000E+01 0.47333526E+00-0.14160201E-01 0.12192394E-01 0.10525408E-03
iteration = 50 func evals = 966 llf = 0.23536831E+02
0.51443533E+01 0.47333243E+00-0.14160401E-01 0.12192270E-01 0.69796301E-04
iteration = 52 func evals = 997 llf = 0.23536831E+02
0.51443168E+01 0.47333161E+00-0.14160454E-01 0.12192235E-01 0.61875638E-04

the final mle estimates are :

coefficient standard-error t-ratio

beta 0 0.51443168E+01 0.51631819E+00 0.99634622E+01
beta 1 0.47333161E+00 0.54772217E-01 0.86418195E+01
beta 2 -0.14160454E-01 0.67119649E-02 -0.21097330E+01
sigma-squared 0.12192235E-01 0.31315540E-02 0.38933497E+01
gamma 0.61875638E-04 0.58145288E-01 0.10641557E-02
mu is restricted to be zero
eta is restricted to be zero

log likelihood function = 0.23536831E+02

the likelihood value is less than that obtained
using ols! - try again using different starting values

number of iterations = 52

(maximum number of iterations set at : 100)

number of cross-sections = 30

number of time periods = 1

total number of observations = 30

thus there are: 0 obsns not in the panel

covariance matrix :

0.26658448E+00 -0.22408687E-01 -0.13668244E-02 0.10170043E-03 0.17612587E-01
-0.22408687E-01 0.29999958E-02 0.16384744E-03 0.33419806E-05 0.84870061E-04
-0.13668244E-02 0.16384744E-03 0.45050473E-04 0.32647406E-06 0.55293134E-05
0.10170043E-03 0.33419806E-05 0.32647406E-06 0.98066305E-05 0.24805249E-04
0.17612587E-01 0.84870061E-04 0.55293134E-05 0.24805249E-04 0.33808746E-02

technical efficiency estimates :

firm eff.-est.

1 0.99931071E+00
2 0.99930688E+00
3 0.99930747E+00
4 0.99930732E+00
5 0.99930633E+00
6 0.99930272E+00
7 0.99930451E+00
8 0.99930472E+00
9 0.99930807E+00
10 0.99930768E+00
11 0.99931011E+00
12 0.99930523E+00
13 0.99930559E+00
14 0.99930527E+00
15 0.99930697E+00
16 0.99930622E+00
17 0.99930689E+00
18 0.99930816E+00
19 0.99931171E+00
20 0.99931066E+00
21 0.99930310E+00
22 0.99931204E+00
23 0.99930888E+00
24 0.99931069E+00
25 0.99930995E+00
26 0.99930833E+00
27 0.99930945E+00
28 0.99930592E+00
29 0.99930538E+00
30 0.99930396E+00

mean efficiency = 0.99930736E+00

怎么结果差别不大？

[此贴子已经被作者于2006-1-15 10:49:00编辑过]

运用成本模型的思想应用于解释基本可以计算，但是对于计算结果是否正确还需要进一步的检验！

希望hkswen能够给点意见！

怎么在成本模型中有这种结果出现？

firm eff.-est.

1 0.15974617E+35
2 0.10245519+123
3 0.66801478+204
4 0.27150308+177
5 0.12408204+126
6 0.44231182+173
7 0.12192151E+01
8 Inf
9 0.42886593+118
10 Inf
11 Inf
12 Inf
13 0.10403066+104
14 0.51664846+233
15 Inf
16 0.53619299E+04
17 Inf
18 Inf
19 Inf
20 0.10116491+233
21 0.38255825+114
22 0.14181282E+33

mean efficiency = Inf

我的做法是直接使用ln(C_i/W_i)整体 是产出，ln(Q_i) 是第一投入，ln(R_i/W_i) 是第二投入。具体的原始数据是 ：

ln(C_i/W_i) ln(Q_i) ln(R_i/W_i)

1 1 248.81 5365 1688
2 1 200.85 4342.6 1089
3 1 486.42 4770.5 1497
4 1 424.3 4898.6 2209
5 1 224.89 4480 1519
6 1 362.23 4595.1 1593
7 1 448.48 6538.2 2560
8 1 1323.11 5064.6 1939
9 1 761.82 6859.8 3503
10 1 739.39 4720.6 1637
11 1 1093.47 5809 2188
12 1 772.75 4532.4 1376
13 1 365.29 5212.8 1698
14 1 816.42 5815.4 1945
15 1 1638.65 5619.5 1459
16 1 339.48 5896 1351
17 1 981.3 5477.9 1610
18 1 814.78 4934 1082
19 1 1147.04 6178.7 1885
20 1 534.7 4654.1 1180
21 1 224.13 4475.2 952
22 1 19.54 4472.9 1297

直接使用，没有需要经过数据的再次变换。在运行时将该类数据当作已经经过SHAZAM软件处理过！

注：hkswen，如果有什么不对的请原谅！如果是您现在比较忙未有时间上坛子，也没有什么关系！等您那天有时间和心情再解决也不迟！

全世界的经济总量取对数也没有超过100，假设的话，也要建立在合理的经济及数学的基础上，超过了计算机的位数容量，当然给你无限大（inf)。

已发邮件至您邮箱中！请查收！

DEA和SFA是前沿生产率分析的两种不同方法，如过楼主主要用DEA，那先多花精力搞它就够了，而且DEA似乎比SFA技术上简单一些，至少对概率论没有很高的要求，国内介绍DEA的书也比较多，包括我上次推荐的那一本。还有就是Fare等人1995年写的那本书也很好。

以下是引用lustboy在2006-1-17 14:32:00的发言：
DEA和SFA是前沿生产率分析的两种不同方法，如过楼主主要用DEA，那先多花精力搞它就够了，而且DEA似乎比SFA技术上简单一些，至少对概率论没有很高的要求，国内介绍DEA的书也比较多，包括我上次推荐的那一本。还有就是Fare等人1995年写的那本书也很好。

谢谢！的确是这样的！

现在主要是以DEA为主，而SFA只是起到辅助作用！

今天抽空来看看！