Librerías

#install.packages("tidyverse")
library(tidyverse)
## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ dplyr     1.1.4     ✔ readr     2.1.5
## ✔ forcats   1.0.0     ✔ stringr   1.5.1
## ✔ ggplot2   3.4.4     ✔ tibble    3.2.1
## ✔ lubridate 1.9.3     ✔ tidyr     1.3.1
## ✔ purrr     1.0.2     
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag()    masks stats::lag()
## ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors
#install.packages("mice")
library(mice)
## 
## Attaching package: 'mice'
## 
## The following object is masked from 'package:stats':
## 
##     filter
## 
## The following objects are masked from 'package:base':
## 
##     cbind, rbind
#install.packages("dplyr")
library(dplyr)
#install.packages("gplots")
library(gplots)
## 
## Attaching package: 'gplots'
## 
## The following object is masked from 'package:stats':
## 
##     lowess
#install.packages("plm")
library(plm)
## 
## Attaching package: 'plm'
## 
## The following objects are masked from 'package:dplyr':
## 
##     between, lag, lead

Base de datos

ptnts <- read.csv("/Users/liz/Downloads/OCTAVO SEMESTRE/ESCENARIOS FUTUROS/MÓDULO 1 - PROF RAÚL/PATENT 3.csv")

Manipulación de datos

# identificación de NAs
ptnts_dep <- select(ptnts,  c("cusip", "year", "patentsg", "patents", "rnd", "employ", "sales", "stckpr"))
colSums(is.na(ptnts_dep))
##    cusip     year patentsg  patents      rnd   employ    sales   stckpr 
##        0        0        0        0        0       21        3        2
ptnts_dep[!complete.cases(ptnts_dep), ]
##       cusip year patentsg patents rnd employ sales stckpr
## 48     7603 2019        0       0   0     NA   137      2
## 441  296695 2012        2       1   1     NA    37      4
## 537  361606 2018        4       2   0     NA   153      5
## 811  423236 2012        1       0   1     NA    59      6
## 812  423236 2013        2       2   1     NA    78      2
## 877  449680 2018        4       1   1      1    NA     11
## 900  451650 2021        7       0   7     NA   217      7
## 916  457186 2017        0       0   0      1    NA     13
## 1241 541381 2012        4       2   1     NA     9      5
## 1314 574599 2015        9       5   7     NA   311     24
## 1636 740512 2017        0       3   2     NA   211     17
## 1870 776338 2021        0       0   1     NA    37      2
## 1876 776678 2017        7       7   7      7    NA     25
## 1976 817732 2017        0       0   0      0     8     NA
## 1979 817732 2020        0       0   0      0    10     NA
## 2201 870326 2012        0       0   1     NA    26     48
## 2202 870326 2013        0       0   1     NA    29      5
## 2203 870326 2014        0       0   1     NA    31      2
## 2204 870326 2015        0       0   1     NA    14      5
## 2205 870326 2016        0       0   1     NA    15      4
## 2206 870326 2017        0       0   1     NA    18      6
## 2207 870326 2018        0       0   2     NA    21      6
## 2208 870326 2019        0       0   2     NA    26      6
## 2209 870326 2020        0       2   2     NA    33     10
## 2210 870326 2021        0       0   2     NA    40     11
## 2241 878308 2012        1       0   0     NA     6      2
# eliminación de empresa sin datos de empleados
ptnts_dep <- subset(ptnts_dep, cusip != 870326)
# imputación de datos
ptnts_dep <- mice(ptnts_dep, m=1, maxit=50, meth='cart', seed=500)
## 
##  iter imp variable
##   1   1  employ  sales  stckpr
##   2   1  employ  sales  stckpr
##   3   1  employ  sales  stckpr
##   4   1  employ  sales  stckpr
##   5   1  employ  sales  stckpr
##   6   1  employ  sales  stckpr
##   7   1  employ  sales  stckpr
##   8   1  employ  sales  stckpr
##   9   1  employ  sales  stckpr
##   10   1  employ  sales  stckpr
##   11   1  employ  sales  stckpr
##   12   1  employ  sales  stckpr
##   13   1  employ  sales  stckpr
##   14   1  employ  sales  stckpr
##   15   1  employ  sales  stckpr
##   16   1  employ  sales  stckpr
##   17   1  employ  sales  stckpr
##   18   1  employ  sales  stckpr
##   19   1  employ  sales  stckpr
##   20   1  employ  sales  stckpr
##   21   1  employ  sales  stckpr
##   22   1  employ  sales  stckpr
##   23   1  employ  sales  stckpr
##   24   1  employ  sales  stckpr
##   25   1  employ  sales  stckpr
##   26   1  employ  sales  stckpr
##   27   1  employ  sales  stckpr
##   28   1  employ  sales  stckpr
##   29   1  employ  sales  stckpr
##   30   1  employ  sales  stckpr
##   31   1  employ  sales  stckpr
##   32   1  employ  sales  stckpr
##   33   1  employ  sales  stckpr
##   34   1  employ  sales  stckpr
##   35   1  employ  sales  stckpr
##   36   1  employ  sales  stckpr
##   37   1  employ  sales  stckpr
##   38   1  employ  sales  stckpr
##   39   1  employ  sales  stckpr
##   40   1  employ  sales  stckpr
##   41   1  employ  sales  stckpr
##   42   1  employ  sales  stckpr
##   43   1  employ  sales  stckpr
##   44   1  employ  sales  stckpr
##   45   1  employ  sales  stckpr
##   46   1  employ  sales  stckpr
##   47   1  employ  sales  stckpr
##   48   1  employ  sales  stckpr
##   49   1  employ  sales  stckpr
##   50   1  employ  sales  stckpr
summary(ptnts_dep)
## Class: mids
## Number of multiple imputations:  1 
## Imputation methods:
##    cusip     year patentsg  patents      rnd   employ    sales   stckpr 
##       ""       ""       ""       ""       ""   "cart"   "cart"   "cart" 
## PredictorMatrix:
##          cusip year patentsg patents rnd employ sales stckpr
## cusip        0    1        1       1   1      1     1      1
## year         1    0        1       1   1      1     1      1
## patentsg     1    1        0       1   1      1     1      1
## patents      1    1        1       0   1      1     1      1
## rnd          1    1        1       1   0      1     1      1
## employ       1    1        1       1   1      0     1      1
ptnts_dep <- complete(ptnts_dep)
ptnts_dep <- pdata.frame(ptnts_dep, index = c("cusip", "year"))

Heterogeneidad

plotmeans(ptnts_dep$patentsg ~ ptnts_dep$cusip,
          ylab = "Patentes aprobadas",
          xlab = "Empresa",
          main = "Heterogeneidad entre empresas")

plotmeans(ptnts_dep$patentsg ~ ptnts_dep$year,
          ylab = "Patentes aprobadas",
          xlab = "Año",
          main = "Heterogeneidad entre años")

Modelos

Modelo 1 - regresión agrupada (pooled)
pooled <- plm(patentsg ~ patents + rnd + employ + sales + stckpr, data = ptnts_dep, model = "pooling")
summary(pooled)
## Pooling Model
## 
## Call:
## plm(formula = patentsg ~ patents + rnd + employ + sales + stckpr, 
##     data = ptnts_dep, model = "pooling")
## 
## Balanced Panel: n = 225, T = 10, N = 2250
## 
## Residuals:
##       Min.    1st Qu.     Median    3rd Qu.       Max. 
## -254.86172   -3.94821   -0.54453    1.66279  644.96666 
## 
## Coefficients:
##                Estimate  Std. Error t-value  Pr(>|t|)    
## (Intercept) -1.06665468  0.74040696 -1.4406   0.14983    
## patents      0.94605963  0.01176871 80.3877 < 2.2e-16 ***
## rnd          0.01818342  0.00911000  1.9960   0.04606 *  
## employ       0.15176760  0.02812506  5.3962 7.524e-08 ***
## sales        0.00025110  0.00026327  0.9538   0.34031    
## stckpr       0.12643212  0.02413028  5.2396 1.760e-07 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Total Sum of Squares:    14161000
## Residual Sum of Squares: 1432300
## R-Squared:      0.89885
## Adj. R-Squared: 0.89863
## F-statistic: 3988.23 on 5 and 2244 DF, p-value: < 2.22e-16
Modelo 2 - efectos fijos (within)
within <- plm(patentsg ~ patents + rnd + employ + sales + stckpr, data = ptnts_dep, model = "within")
summary(within)
## Oneway (individual) effect Within Model
## 
## Call:
## plm(formula = patentsg ~ patents + rnd + employ + sales + stckpr, 
##     data = ptnts_dep, model = "within")
## 
## Balanced Panel: n = 225, T = 10, N = 2250
## 
## Residuals:
##      Min.   1st Qu.    Median   3rd Qu.      Max. 
## -223.7660   -1.9904   -0.3246    1.6333  260.2973 
## 
## Coefficients:
##            Estimate  Std. Error t-value  Pr(>|t|)    
## patents  0.06763793  0.01881050  3.5958 0.0003312 ***
## rnd     -0.12607804  0.01279004 -9.8575 < 2.2e-16 ***
## employ  -0.05689008  0.06070158 -0.9372 0.3487630    
## sales   -0.00139288  0.00035662 -3.9057 9.702e-05 ***
## stckpr   0.04779604  0.02662196  1.7954 0.0727456 .  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Total Sum of Squares:    715640
## Residual Sum of Squares: 585740
## R-Squared:      0.18151
## Adj. R-Squared: 0.08872
## F-statistic: 89.5912 on 5 and 2020 DF, p-value: < 2.22e-16
Modelo 3 - efectos aleatorios (random)
Modelo 3.1 - método walhus
walhus <- plm(patentsg ~ patents + rnd + employ + sales + stckpr, data = ptnts_dep, model = "random", random.method = "walhus")
summary(walhus)
## Oneway (individual) effect Random Effect Model 
##    (Wallace-Hussain's transformation)
## 
## Call:
## plm(formula = patentsg ~ patents + rnd + employ + sales + stckpr, 
##     data = ptnts_dep, model = "random", random.method = "walhus")
## 
## Balanced Panel: n = 225, T = 10, N = 2250
## 
## Effects:
##                  var std.dev share
## idiosyncratic 610.78   24.71 0.959
## individual     25.80    5.08 0.041
## theta: 0.1615
## 
## Residuals:
##       Min.    1st Qu.     Median    3rd Qu.       Max. 
## -244.46702   -4.02407   -0.61926    1.66402  617.34108 
## 
## Coefficients:
##                Estimate  Std. Error z-value  Pr(>|z|)    
## (Intercept) -1.20967897  0.83180483 -1.4543   0.14587    
## patents      0.88856689  0.01304353 68.1232 < 2.2e-16 ***
## rnd          0.01744740  0.00994205  1.7549   0.07927 .  
## employ       0.20668506  0.03083177  6.7036 2.033e-11 ***
## sales        0.00022202  0.00028949  0.7669   0.44312    
## stckpr       0.14814808  0.02577736  5.7472 9.072e-09 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Total Sum of Squares:    10167000
## Residual Sum of Squares: 1361600
## R-Squared:      0.86608
## Adj. R-Squared: 0.86578
## Chisq: 14512.3 on 5 DF, p-value: < 2.22e-16
Modelo 3.2 - método amemiya
amemiya <- plm(patentsg ~ patents + rnd + employ + sales + stckpr, data = ptnts_dep, model = "random", random.method = "amemiya")
summary(amemiya)
## Oneway (individual) effect Random Effect Model 
##    (Amemiya's transformation)
## 
## Call:
## plm(formula = patentsg ~ patents + rnd + employ + sales + stckpr, 
##     data = ptnts_dep, model = "random", random.method = "amemiya")
## 
## Balanced Panel: n = 225, T = 10, N = 2250
## 
## Effects:
##                   var std.dev share
## idiosyncratic  289.26   17.01 0.035
## individual    7871.42   88.72 0.965
## theta: 0.9395
## 
## Residuals:
##       Min.    1st Qu.     Median    3rd Qu.       Max. 
## -187.52861   -3.05502   -1.57504    0.63184  272.66938 
## 
## Coefficients:
##                Estimate  Std. Error z-value  Pr(>|z|)    
## (Intercept) 23.19551247  6.00663681  3.8616 0.0001126 ***
## patents      0.12915116  0.01805384  7.1537 8.449e-13 ***
## rnd         -0.10101645  0.01243407 -8.1242 4.504e-16 ***
## employ       0.20696474  0.05513188  3.7540 0.0001740 ***
## sales       -0.00146324  0.00035073 -4.1720 3.020e-05 ***
## stckpr       0.08691677  0.02608162  3.3325 0.0008607 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Total Sum of Squares:    764860
## Residual Sum of Squares: 637760
## R-Squared:      0.16618
## Adj. R-Squared: 0.16432
## Chisq: 447.231 on 5 DF, p-value: < 2.22e-16
Modelo 3.3 - método nerlove
nerlove <- plm(patentsg ~ patents + rnd + employ + sales + stckpr, data = ptnts_dep, model = "random", random.method = "nerlove")
summary(nerlove)
## Oneway (individual) effect Random Effect Model 
##    (Nerlove's transformation)
## 
## Call:
## plm(formula = patentsg ~ patents + rnd + employ + sales + stckpr, 
##     data = ptnts_dep, model = "random", random.method = "nerlove")
## 
## Balanced Panel: n = 225, T = 10, N = 2250
## 
## Effects:
##                   var std.dev share
## idiosyncratic  260.33   16.13 0.032
## individual    7935.61   89.08 0.968
## theta: 0.9428
## 
## Residuals:
##       Min.    1st Qu.     Median    3rd Qu.       Max. 
## -188.52529   -3.01465   -1.54017    0.67283  272.41924 
## 
## Coefficients:
##                Estimate  Std. Error z-value  Pr(>|z|)    
## (Intercept) 23.85689023  6.32114542  3.7741 0.0001606 ***
## patents      0.12375144  0.01803921  6.8601 6.879e-12 ***
## rnd         -0.10314337  0.01241014 -8.3112 < 2.2e-16 ***
## employ       0.18534562  0.05536715  3.3476 0.0008152 ***
## sales       -0.00145857  0.00034971 -4.1708 3.036e-05 ***
## stckpr       0.08361826  0.02601267  3.2145 0.0013066 ** 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Total Sum of Squares:    759600
## Residual Sum of Squares: 633170
## R-Squared:      0.16644
## Adj. R-Squared: 0.16458
## Chisq: 448.057 on 5 DF, p-value: < 2.22e-16
Comparación de modelos
# prueba pF
pFtest(within, pooled)
## 
##  F test for individual effects
## 
## data:  patentsg ~ patents + rnd + employ + sales + stckpr
## F = 13.034, df1 = 224, df2 = 2020, p-value < 2.2e-16
## alternative hypothesis: significant effects
# prueba de Hausman
phtest(walhus, within)
## 
##  Hausman Test
## 
## data:  patentsg ~ patents + rnd + employ + sales + stckpr
## chisq = 6512, df = 5, p-value < 2.2e-16
## alternative hypothesis: one model is inconsistent

Con base en estas comparaciones, podemos concluir que los mejores modelos son el pooled y el walhus, si tuviéramos que elegir solo uno, el que tiene mayor R^2 ajustada es pooled, por lo que sería nuestra elección al ajustarse mejor a los datos.

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