I am trying to fit a model where there is a sequence of exogenous "shocks", $X_1, X_2, ..., X_T$, and a AR(1) of these shocks explain $Y_1, Y_2, ..., Y_T$. Specifically,

Data (known):

$X_1, X_2, ..., X_T$

$Y_1, Y_2, ..., Y_T$

Priors (known):

$\phi = 2 \phi^* - 1$, where $\phi^* \sim Beta(\alpha, \beta)$, "persistence" parameter, between $(-1, 1)$

$\beta \sim N(\mu, v)$, "shock strength" parameter


$E_t = \phi E_{t-1} + \beta X_t$

$Y_t = E_t + \epsilon_t$, $\epsilon_t \sim N(0, 1)$

Unknown parameters:

$\phi$, $\beta$, $E_0$

My question is, what is the easiest way to sample $\beta$ and $\phi$?

It seems that if I am allowed to use $E_{t-1}$ when sampling from $\phi$, it becomes a bit easier, as:

$Y_t = \phi E_{t-1} + \beta X_t + \epsilon$

$\implies \phi E_{t-1} = Y_t - \beta X_t - \epsilon$

$\implies \phi \sim N(\frac{Y_t - \beta X_t}{E_{t-1}}, \frac{1}{E_{t-1}^2})$

However, since $E_t$ is known deterministically from $\phi$, it seems that if I am conditioning on it I must choose a deterministic value for $\phi$.

So I have two questions:

How do I best fit this model using MCMC?

How do I condition on a deterministic auxiliary variable within a Gibbs sampling step?


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.