Tento notebook vysvetľuje dva dôležité diagnostické testy v ekonometrii:
Uvažujme lineárny regresný model:
\[ y_i = \beta_0 + \beta_1 x_{1i} + \cdots + \beta_k x_{ki} + u_i \]
Za predpokladu homoskedasticity platí:
\[ \mathrm{Var}(u_i) = \sigma^2 \]
Test skúma, či variancia závisí od vysvetľujúcich premenných.
\[ H_0: \mathrm{Var}(u_i) = \sigma^2 \]
\[ H_1: \mathrm{Var}(u_i) \neq \sigma^2 \]
\[ \hat{u}_i \]
\[ \hat{u}_i^2 = \alpha_0 + \alpha_1 x_{1i} + \cdots + \alpha_m x_{mi} + v_i \]
\[ LM = nR^2 \sim \chi^2(m) \]
Testuje autokoreláciu rezíduí:
\[ u_t = \rho_1 u_{t-1} + \cdots + \rho_p u_{t-p} + \varepsilon_t \]
\[ H_0: \rho_1 = \cdots = \rho_p = 0 \]
\[ H_1: \exists \rho_j \neq 0 \]
Pomocná regresia:
\[ \hat{u}_t = \gamma_0 + \gamma_1 x_{1t} + \cdots + \gamma_k x_{kt} + \rho_1 \hat{u}_{t-1} + \cdots + \rho_p \hat{u}_{t-p} + v_t \]
\[ LM = nR^2 \sim \chi^2(p) \]
library(lmtest)
library(sandwich)
set.seed(123)
n <- 200
x <- runif(n, 1, 10)
u <- rnorm(n, sd = 0.5 * x)
y <- 2 + 3*x + u
m <- lm(y ~ x)
summary(m)
##
## Call:
## lm(formula = y ~ x)
##
## Residuals:
## Min 1Q Median 3Q Max
## -7.3828 -1.5748 -0.1864 1.3994 9.6110
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 2.16353 0.46233 4.68 5.32e-06 ***
## x 2.96535 0.07607 38.98 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.646 on 198 degrees of freedom
## Multiple R-squared: 0.8847, Adjusted R-squared: 0.8841
## F-statistic: 1520 on 1 and 198 DF, p-value: < 2.2e-16
bptest(m)
##
## studentized Breusch-Pagan test
##
## data: m
## BP = 24.741, df = 1, p-value = 6.557e-07
set.seed(123)
n <- 200
x <- rnorm(n)
u <- numeric(n)
eps <- rnorm(n)
rho <- 0.7
u[1] <- eps[1]
for (t in 2:n) {
u[t] <- rho * u[t-1] + eps[t]
}
y <- 1 + 2*x + u
m2 <- lm(y ~ x)
summary(m2)
##
## Call:
## lm(formula = y ~ x)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.7937 -0.9739 -0.0469 0.8434 3.3399
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 1.15262 0.09073 12.70 <2e-16 ***
## x 2.06205 0.09644 21.38 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.283 on 198 degrees of freedom
## Multiple R-squared: 0.6978, Adjusted R-squared: 0.6963
## F-statistic: 457.2 on 1 and 198 DF, p-value: < 2.2e-16
bgtest(m2, order = 1)
##
## Breusch-Godfrey test for serial correlation of order up to 1
##
## data: m2
## LM test = 79.796, df = 1, p-value < 2.2e-16
odhaduje štandardné chyby koeficientov, ktoré sú robustné voči heteroskedasticite
coeftest(m, vcov = vcovHC(m, type = "HC1"))
##
## t test of coefficients:
##
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 2.16353 0.37526 5.7655 3.088e-08 ***
## x 2.96535 0.08092 36.6455 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
sú robustné aj v prípade heteroskedasticity a autokorelácie rezíduí
coeftest(m2, vcov = NeweyWest(m2))
##
## t test of coefficients:
##
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 1.15262 0.19178 6.0101 8.768e-09 ***
## x 2.06205 0.10274 20.0713 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1