5
$\begingroup$

I have the following time series process $y_t $

$$\Delta y_t = \delta + \gamma t + \epsilon_t$$

where $e_t$ is white noise process with the variance of $\sigma^2$.

I guess that whereas $\Delta y_t$ is trend stationary process, $y_t$ is a random walk with a quadratic deterministic trend.

So, by using the process $y_t$ which is defined as a random walk with a quadratic deterministic trend, I want to apply the ADF unit root test. During the testing procedure, which form do we assign to $y_t$ under the null hypothesis?

What I did is as follows:

My solution is that! But my solution is totally wrong. I just want to exhibit my attempt, so I added them as a picture because this solution is too long to type it and it is wrong. (Sorry for not typing )

But, I have no idea other than this solution. Please share your ideas with me. Thank you a lot!

enter image description here

enter image description here

$\endgroup$

1 Answer 1

2
+100
$\begingroup$

Following this reference:

Given: Time series process $y_t $

$$\Delta y_t = \delta + \gamma t + \epsilon_t$$

where $e_t$ is white noise process with the variance of $\sigma^2$.

Question: For the ADF test, which form do we assign to $y_t$ under the null hypothesis?

ADF Test:

For a process of the form,

$$y_t = \beta_0 + \beta_1 t + \phi y_{t-1} + \sum_{j=1}^p \psi_j \Delta y_{t-j} + \epsilon_t$$

Test: $H_o: \phi = 1$

Answer:

For the given process, the higher order lag terms are not present:

Option 1: $y_t = \phi y_{t-1} + \delta + \gamma t + \epsilon_t, \ \ \ H_o: \phi = 1$

Option 2: $\Delta y_t = \pi y_{t-1} + \delta + \gamma t + \epsilon_t, \ \ \ H_o: \pi = 0$

Using the method of least squares with Option 2, fit this equation $\Delta y_t = \beta_0 + \beta_1 y_{t-1} + \beta_2 t + \epsilon_t$ and then test if $H_o: \beta_1 = 0$ using the normal t-test in regression results.

Example:

y <- numeric(100)
y[1] <- 3 + rnorm(1, 0, 2)
for (i in 2:100)
{
  y[i] <- y[i-1] + 3 + 0.5*(i-1) + rnorm(1, 0, 2)
}

test_data <- data.frame(
  dy = diff(y),
  t = 1:99,
  ym1 = y[1:99]
)

lm1 <- lm(dy ~ ym1 + t, data = test_data)
summary(lm1)
```
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.