I saw some solutions for people who wanted to jump straight into an R package to do the work, but I am trying to do some of the process piece-by-piece. The model that I refer to is

$$Y_t = m_t + s_t + X_t,$$

where $Y_t$ are the observations, $m_t$ is the trend component, $s_t$ is the seasonal component, and $X_t$ is the rough component.

I'm using R and the dataset is AirPassengers. What I can understand is fitting a Loess or polynomial trend separately using either loess() or lm(...poly()). Also, I understand creating a series of indicator variables in a matrix and using lm() to find the coefficients for the seasonality component. This would in essence give me values for $s_t$ of the seasonality and $m_t$ of the trend separately, but I am not sure that I can simply add these two values together to find the rough. I tried looking at the residuals for this setup and they don't appear like white noise.

In the case of polynomial degree 3 trend, $m_t = \beta_0 + \beta_1t+\beta_2t^2+\beta_3t^3.$ Also, with a seasonal component using 12 months, there is $s_t = \beta_4I_{t,1} + \beta_5I_{t,2} + \cdots + \beta_{14}I_{t,11},$ where $$ I_{t,1} \begin{cases} 1 & \text{if time t is Jan} \\ -1 & \text{if time t is Dec}\\ 0 & \text{otherwise} \end{cases}, I_{t,2} \begin{cases} 1 & \text{if time t is Feb} \\ -1 & \text{if time t is Dec}\\ 0 & \text{otherwise} \end{cases} ,\cdots, I_{t,11} \begin{cases} 1 & \text{if time t is Nov} \\ -1 & \text{if time t is Dec}\\ 0 & \text{otherwise} \end{cases}. $$

So my question is how do I find these components at the same time? The reason is that I would like to ultimately test different transformations of Box-Cox using $R^2$, but the improper looking rough makes me hesitant togo on further.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.