In my research im dealing with a longitudinal data set, which consists of a single response variable and a set of several predictor variables, which comprises a mixture of individual specific constants and other time-varying variables. As my data is longitudinal, I have data on several individuals with a varying amount of (unevenly spaced) observations per individual.

For the response variable that I am modelling, it would be very untuitive to assume a mean-reverting process, where the mean is individual specific. I am familiar with the standard Ornstein-Uhlenbeck model and its parameters. However, I was wondering what the best way would be to incorporate the additional predictor variables into my model. The Ornstein-Uhlenbeck process can be summarized by:

$dy_{it} = \kappa(\mu_i-y_{it})dt + \sigma dW_t$,

where $y_{it}$ refers to the measured response variable of individual $i$ at time $t$. I was thinking I could estimate a specific mean $\mu_i$ for every individual in my dataset. However, I am not sure how I can inporate the additional predictor variables into my model. I could set $\mu_i$=$\mathbf{x}_{it}^\intercal\mathbf{\beta} + \alpha_i$, so that I am estimating a contstant per individual, as well as a general parameter vector that links the predictors to the mean of the Ornstein-Uhlenbeck process. However, I think the non-constant predictors mess up this approach, as I dont think that the mean $\mu_i$ is supposed to be time-varying for a specific individual.

What are your thoughts on this? Does anyone has a suggestion how to best incorporate additional explanatory variables in a Ornstein-Uhlenback model (related to longitudinal or panel data)?


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.