Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Correlated Random Effects: Year Dummy variables multicollinearity

    Dear List,

    I have some panel data and I apply the CRE model as stated by the very helpful paper of Schunck (2013). I analyze the effect of various firm-specific characteristics on the debt interest rate and include macroeconomic variables (GDP growth rate, ECB main refinancing rate, term spread) as well as year dummies (nine years: 2009-2013) in my analysis (besides additional firm-specific owner and industry dummies).

    The macro variables are constant across all entities in each year but vary over time. The year dummies are generated manually and are of dichotomous (0/1) nature.

    For some reason, the last four year dummies are omitted due to collinearity when using the following code:
    Code:
    xtreg r x1 x2 c m_x1 m_x2 GDP ECB_refi TERM year_dummy*, re cluster(id)
    Conversely, the macroeconomic variables are omitted with the year dummies put before the macro variables, i.e., the following code:
    Code:
    xtreg r x1 x2 c m_x1 m_x2 year_dummy* GDP ECB_refi TERM , re cluster(id)
    What could be a reason for this multicollinearity problem? Are the omitted year dummies potentially biasing the analysis?

    I am grateful for any help!


    Source:
    Schunck, Reinhard (2013): Within and between estimates in random-effects models: Advantages and drawbacks of correlated random effects and hybrid models. In: Stata Journal 13 (1), S. 65–76.

  • #2
    Why aren't you using one of the estimation seizures provided by Schunck - xthybrid or mundlak?

    If you include year dummies, then you won't get parameter estimates on anything that is constant across panels but varies by year. They will be collinear with the dummies. It appears that they are collinear and depending on the order of the specification in the equation Stata chooses which one to actually identify as collinear.

    Comment

    Working...
    X