Announcement

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

  • Panel data: impact of a count variable on bank lending (functional form and interpretation)

    I have a panel dataset on lending of six banks and an explanatory variable let's say x1 which is a macro variable and is of count nature that stays the same for all the banks. My questions are

    1. Which panel data method would be most suitable?

    2. What functional form of data would be better suited to reflect the impact of increase in x1 on bank lending?

    Regards,

    Mohsin

  • #2
    Your description is unclear and incomplete. When you say that x1 "stays the same for all banks" do you mean that in any given time period, all banks have the same value for x1? Or do you mean that each bank has its own value of x1, but that value does not change over time?

    How many time periods are covered in your data set?

    Do you expect the effect of changes in x1 on lending to be additive, multiplicative, or something more complicated? Do you expect the effect of a given change in x1 to be the same for all banks, or does it vary among the banks?

    Aren't there other variables that should be taken into account?

    Comment


    • #3
      Thank you for your kind reply. Yes, x1 stays the same for all banks in any given time period; for example, it is the same for all banks in 2010q2.
      2. The data are quarterly and since 2007q2 until 2024q2 thus making the total observations for a single bank as 68 and for the whole panel of six banks as 68x6=408
      3. The effect of changes is such that a sudden substantial and abrupt increase in x1 may have huge impact on an otherwise additive effect.
      4. The anticipated impact is not the same across banks: it does vary
      5. other variables may also be taken into account

      Comment


      • #4
        In light of your responses, I think -xtgls- is your best bet here, because T >> N. With an additive effect, I would avoid log-transforming the lending variable. And with the effect varying across banks, I would tend to include an i.bank#i.x1 interaction. As you will thereby estimate separate x1 effects for each bank, you will probably find it easier to get the marginal x1 effects at each bank from -margins, rather than a series of -lincom- commands that will be tedious and error-prone to write.

        Comment

        Working...
        X