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

  • Difference between Control variables and Fixed effects

    Dear all,

    I have to do my dissertation about the effect of Bank's liquidity risk on syndicated loan pricing only in the US loan and US Banks. (covering 12-year period)
    My dependent variable(y) is loan pricing (interest rate).
    My independent variable(x) is bank liquidity risk. (my interest variable)

    As I followed some literature, they use Pool OLS with robust standard errors and also added year fix effect as dummy variables.
    Because borrower characteristics and loan purpose can affect loan pricing (y), I saw many papers also stated that they control for borrower characteristics and loan purpose.
    (borrower characteristics such as assets, profitability, rating and loan purpose is dummy variables)
    I am not sure whether my code in Stata for controlling these characteristics is only add all of these as independent variables or not?

    My code is
    " reg Loanpricing Liquidityrisk borrowercharacteristics loanpurpose i.year, vce(robust)"

    Additionally, I am not sure whether I should cluster SE in BankID or add more control or fixed effects.

    Please advise and Thank you in advance

  • #2
    Everything looks fine in what you are saying. Yes, you should cluster your standard errors by BankID, the decisions within banks are probably highly correlated, as they probably follow bank-wide policy.


    • #3
      One problem with the pooled model you present is that you will obtain inconsistent parameter estimates if your regressors are correlated with the error term. This is the reason for adding "control" variables so that you might mitigate omitted variable bias. However, if there are unobserved effects that are specific to each bank that also affect the loan pricing, then the pooled model will still be correlated with the error term. I learned econometrics from Wooldridge's textbook (2010), and he provides examples like ability as an example of unobserved heterogeneity specific to each individual (bank in your case). Hence, this would be the reason for using a panel approach rather than a pooled approach.

      In other words, if you believe there are unobserved effects specific to each bank that also affect your dependent variable, then you should try including firm fixed effects as well in your model.

      Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press.