Carga de base de datos:

series<-uschange
autoplot(uschange[,c(2,5)])

Gráfica de serie de tiempo de las variables seleccionadas

ts.plot(series[,c(2,5)], xlab="Tiempo",col=c(2,5))

Búsqueda y selección de parámetros

a <- VARselect(uschange[,c(2,5)],lag.max=15,type="const")
a$selection
## AIC(n)  HQ(n)  SC(n) FPE(n) 
##      8      1      1      8

Se obtiene un estimado automático de los mejores parametros para la creación del modelo VAR. (8 y 1),

Creación de modelo

modelo1<-VAR(uschange[,c(2,5)],p=5,type=c("const"))

Datos del modelo por variable seleccionada

- Consumo

summary(modelo1,equation="Consumption")
## Warning in summary.varest(modelo1, equation = "Consumption"): 
## Invalid variable name(s) supplied, using first variable.
## 
## VAR Estimation Results:
## ========================= 
## Endogenous variables: Income, Unemployment 
## Deterministic variables: const 
## Sample size: 182 
## Log Likelihood: -262.292 
## Roots of the characteristic polynomial:
## 0.7776 0.7776 0.7661 0.7661 0.7488 0.7488 0.672 0.672 0.6364 0.5646
## Call:
## VAR(y = uschange[, c(2, 5)], p = 5, type = c("const"))
## 
## 
## Estimation results for equation Income: 
## ======================================= 
## Income = Income.l1 + Unemployment.l1 + Income.l2 + Unemployment.l2 + Income.l3 + Unemployment.l3 + Income.l4 + Unemployment.l4 + Income.l5 + Unemployment.l5 + const 
## 
##                 Estimate Std. Error t value Pr(>|t|)    
## Income.l1       -0.13069    0.07628  -1.713  0.08848 .  
## Unemployment.l1 -0.47812    0.22736  -2.103  0.03693 *  
## Income.l2        0.08289    0.07790   1.064  0.28883    
## Unemployment.l2  0.39251    0.23868   1.644  0.10191    
## Income.l3        0.03524    0.07602   0.464  0.64351    
## Unemployment.l3 -0.65814    0.24414  -2.696  0.00773 ** 
## Income.l4       -0.09359    0.07556  -1.239  0.21717    
## Unemployment.l4 -0.14117    0.24531  -0.575  0.56571    
## Income.l5       -0.12761    0.07471  -1.708  0.08944 .  
## Unemployment.l5  0.39764    0.22214   1.790  0.07522 .  
## const            0.87038    0.14889   5.846 2.49e-08 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## 
## Residual standard error: 0.8936 on 171 degrees of freedom
## Multiple R-Squared: 0.1347,  Adjusted R-squared: 0.08413 
## F-statistic: 2.663 on 10 and 171 DF,  p-value: 0.004781 
## 
## 
## 
## Covariance matrix of residuals:
##               Income Unemployment
## Income        0.7985     -0.05280
## Unemployment -0.0528      0.09033
## 
## Correlation matrix of residuals:
##               Income Unemployment
## Income        1.0000      -0.1966
## Unemployment -0.1966       1.0000

- Ingreso

summary(modelo1,equation="Income")
## 
## VAR Estimation Results:
## ========================= 
## Endogenous variables: Income, Unemployment 
## Deterministic variables: const 
## Sample size: 182 
## Log Likelihood: -262.292 
## Roots of the characteristic polynomial:
## 0.7776 0.7776 0.7661 0.7661 0.7488 0.7488 0.672 0.672 0.6364 0.5646
## Call:
## VAR(y = uschange[, c(2, 5)], p = 5, type = c("const"))
## 
## 
## Estimation results for equation Income: 
## ======================================= 
## Income = Income.l1 + Unemployment.l1 + Income.l2 + Unemployment.l2 + Income.l3 + Unemployment.l3 + Income.l4 + Unemployment.l4 + Income.l5 + Unemployment.l5 + const 
## 
##                 Estimate Std. Error t value Pr(>|t|)    
## Income.l1       -0.13069    0.07628  -1.713  0.08848 .  
## Unemployment.l1 -0.47812    0.22736  -2.103  0.03693 *  
## Income.l2        0.08289    0.07790   1.064  0.28883    
## Unemployment.l2  0.39251    0.23868   1.644  0.10191    
## Income.l3        0.03524    0.07602   0.464  0.64351    
## Unemployment.l3 -0.65814    0.24414  -2.696  0.00773 ** 
## Income.l4       -0.09359    0.07556  -1.239  0.21717    
## Unemployment.l4 -0.14117    0.24531  -0.575  0.56571    
## Income.l5       -0.12761    0.07471  -1.708  0.08944 .  
## Unemployment.l5  0.39764    0.22214   1.790  0.07522 .  
## const            0.87038    0.14889   5.846 2.49e-08 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## 
## Residual standard error: 0.8936 on 171 degrees of freedom
## Multiple R-Squared: 0.1347,  Adjusted R-squared: 0.08413 
## F-statistic: 2.663 on 10 and 171 DF,  p-value: 0.004781 
## 
## 
## 
## Covariance matrix of residuals:
##               Income Unemployment
## Income        0.7985     -0.05280
## Unemployment -0.0528      0.09033
## 
## Correlation matrix of residuals:
##               Income Unemployment
## Income        1.0000      -0.1966
## Unemployment -0.1966       1.0000

Validación del modelo

Prueba de Autocorrelación

#>PortManteu Test > 0.05 Autocorrelación
serial.test(modelo1, lags.pt=11, type="PT.asymptotic")
## 
##  Portmanteau Test (asymptotic)
## 
## data:  Residuals of VAR object modelo1
## Chi-squared = 35.653, df = 24, p-value = 0.05929

Con esta prueba se obtuvo un p-value de 0.05929 > 0.05, lo cual indica que existe autocorrelación.

Prueba de estacionariedad

#Raíz unitaria < 1
roots(modelo1)
##  [1] 0.7775660 0.7775660 0.7660914 0.7660914 0.7488061 0.7488061 0.6720236
##  [8] 0.6720236 0.6364359 0.5646411

Ya que los valores obtenidos no son mayores a 1, se confirma que el modelo creado es estacionario.

Prueba de Jarque Bera (Normalidad)

normality.test(modelo1, multivariate.only=FALSE)
## $Income
## 
##  JB-Test (univariate)
## 
## data:  Residual of Income equation
## Chi-squared = 195.51, df = 2, p-value < 2.2e-16
## 
## 
## $Unemployment
## 
##  JB-Test (univariate)
## 
## data:  Residual of Unemployment equation
## Chi-squared = 33.197, df = 2, p-value = 6.184e-08
## 
## 
## $JB
## 
##  JB-Test (multivariate)
## 
## data:  Residuals of VAR object modelo1
## Chi-squared = 216.79, df = 4, p-value < 2.2e-16
## 
## 
## $Skewness
## 
##  Skewness only (multivariate)
## 
## data:  Residuals of VAR object modelo1
## Chi-squared = 7.8916, df = 2, p-value = 0.01934
## 
## 
## $Kurtosis
## 
##  Kurtosis only (multivariate)
## 
## data:  Residuals of VAR object modelo1
## Chi-squared = 208.9, df = 2, p-value < 2.2e-16

Con esta prueba se obtuvo un p-value de 2.2e-16 < 0.05, lo cual indica que existe normalidad en el comportamiento de los residuos del modelo.

Prueba de Heteroscedasticidad

arch<-arch.test(modelo1, lags.multi = 12, multivariate.only = FALSE)
arch
## $Income
## 
##  ARCH test (univariate)
## 
## data:  Residual of Income equation
## Chi-squared = 11.621, df = 16, p-value = 0.7697
## 
## 
## $Unemployment
## 
##  ARCH test (univariate)
## 
## data:  Residual of Unemployment equation
## Chi-squared = 21.391, df = 16, p-value = 0.164
## 
## 
## 
##  ARCH (multivariate)
## 
## data:  Residuals of VAR object modelo1
## Chi-squared = 121.69, df = 108, p-value = 0.1737

Con esta prueba se obtiene un p-value menor a 0.05, lo que significa que no hay heteroscedasticidad en el modelo. Además, este resultado confirma que la varianza de los residuos del modelo es constante.

Prueba de choques estructurales

stab<-stability(modelo1, type = "OLS-CUSUM")
par(mar=c(1,1,1,1))

Gráficas de OLS-CUSUM

plot(stab)

Con la gráfica es posible observar que los valores no sobrepasan los límites de tolerancia, por lo que tienen estabilidad.

Pruba de Granger (Causalidad)

Ingreso

GrangerIncome <-causality(modelo1, cause = 'Income')
GrangerIncome
## $Granger
## 
##  Granger causality H0: Income do not Granger-cause Unemployment
## 
## data:  VAR object modelo1
## F-Test = 2.8057, df1 = 5, df2 = 342, p-value = 0.01687
## 
## 
## $Instant
## 
##  H0: No instantaneous causality between: Income and Unemployment
## 
## data:  VAR object modelo1
## Chi-squared = 6.7734, df = 1, p-value = 0.009253

Se obtuvo un p-value de 0.01687 (menor a 0.05) por lo que se determina que existe causalidad de la variable Income a la variable de Unemployment.

Desempleo

GrangerIncome <-causality(modelo1, cause = 'Unemployment')
GrangerIncome
## $Granger
## 
##  Granger causality H0: Unemployment do not Granger-cause Income
## 
## data:  VAR object modelo1
## F-Test = 3.7264, df1 = 5, df2 = 342, p-value = 0.002681
## 
## 
## $Instant
## 
##  H0: No instantaneous causality between: Unemployment and Income
## 
## data:  VAR object modelo1
## Chi-squared = 6.7734, df = 1, p-value = 0.009253

Se obtuvo un p-value de 0.002681 (menor a 0.05) por lo que se determina que existe causalidad de la variable Unemployment a la variable Income.

Evaluación del modelo ante un impulso

Income

IncomeIRF <- irf(modelo1,  impulse = "Unemployment", response="Income", n.ahead = 20, boot = T )
plot(IncomeIRF, ylab = "Income", main = "Shock desde Unemployment")

Unemployment

UnemploymentIRF <- irf(modelo1,  impulse = "Income", response="Unemployment", n.ahead = 20, boot = T )
plot(IncomeIRF, ylab = "Unemployment", main = "Shock desde Income")

En ambas gráficas es posible observar que al recibir un shock en los valores, en el largo plazo, las variables evaluadas vuelven a correlacionarse.

Pronóstico

fore<-predict(modelo1, n.ahead = 10, ci=0.95)
fanchart(fore)

Gráfica del pronóstico

autoplot(forecast(modelo1))