+ Reply to Thread
Results 1 to 3 of 3

Thread: Panel Data - Pooled OLS vs Fixed Effects vs Random Effects

  1. #1
    Points: 2,872, Level: 32
    Level completed: 82%, Points required for next Level: 28

    Posts
    1
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Panel Data - Pooled OLS vs Fixed Effects vs Random Effects




    Hi forummers,

    I am a finance PhD student and am currently undertaking some statistical analysis of bankruptcy risk. I have data on roughly 500 firms over 5 years - a panel with a large cross-sectional dimension.

    I want to estimate a firmís bankruptcy risk (I have a continuous dependent variable measuring this) as a function of 10 or so accounting variables. A colleague has suggested that because I have only 5 years of data I should pool it and perform an OLS regression with year dummies.

    I suspect that there are unobserved fixed effect specific to individual firms that donít vary over time (e.g. variables relating to the competence of management, geographical proximity to market etc). To that effect I was planning to estimate a fixed effect panel regression in Stata. Equivalently (and I believe this is the same thing!), I was thinking of running a Pooled OLS regression with a dummy variable for each company. I think I might also have unobserved fixed effect specific to individual years that donít vary over firms (eg. market conditions / the availability of credit etc). I think that the fixed effects estimation is only designed to look at firm specific effects not time specific. Is this a correct assumption?

    In addition to the consideration of a fixed effects model, i considered running a random effects panel regression. In fact, I ran both FE and RE and then tried to perform a Hausman test to see which one was more apt (as per some of the econometrics lit I have read). My Hausman test yields a negative Chi-Square test stat and an accompanying error message.

    In summary, is there ever a plausible rationale for favouring a Pooled OLS model to the panel data models? Secondly, in the absence of a Hausman test is there an intuitive way in which I can decide between using fixed and random effects?

    Apologies for the long post and thank you very much in advance for any help / replies.

    Kind regards,
    g1362901

  2. #2
    Points: 288, Level: 5
    Level completed: 76%, Points required for next Level: 12

    Posts
    1
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Panel Data - Pooled OLS vs Fixed Effects vs Random Effects

    If you're going to create dummy variables for all the firms then you're best off doing Fixed Effects. You'll likely lose an insane amount of degrees of freedom doing dummy variables for all the firms, which is why Fixed Effect is used in the first place.

    What program are you running? Supposedly the Fixed Effect regression in STATA spits out an F-statistic which compares whether it would make more sense to Pool the data or run the fixed effect regression, but I'm currently trying to figure out how to interpret the F-stat.

    Basically when I run the fixed effect regression the elasticities I'm getting make zero sense (doing a patent production function and all the elasticities are negative, huh?)

  3. #3
    Devorador de queso
    Points: 95,819, Level: 100
    Level completed: 0%, Points required for next Level: 0
    Awards:
    Posting AwardCommunity AwardDiscussion EnderFrequent Poster
    Dason's Avatar
    Location
    Tampa, FL
    Posts
    12,935
    Thanks
    307
    Thanked 2,629 Times in 2,245 Posts

    Re: Panel Data - Pooled OLS vs Fixed Effects vs Random Effects


    Quote Originally Posted by RonSwanson View Post
    If you're going to create dummy variables for all the firms then you're best off doing Fixed Effects. You'll likely lose an insane amount of degrees of freedom doing dummy variables for all the firms, which is why Fixed Effect is used in the first place.

    What program are you running? Supposedly the Fixed Effect regression in STATA spits out an F-statistic which compares whether it would make more sense to Pool the data or run the fixed effect regression, but I'm currently trying to figure out how to interpret the F-stat.

    Basically when I run the fixed effect regression the elasticities I'm getting make zero sense (doing a patent production function and all the elasticities are negative, huh?)
    RonSwanson. I am your biggest fan. It is an honor to be on the same forum as you.

    I can't believe I'm talking to Ron ****ing Swanson.
    I don't have emotions and sometimes that makes me very sad.

+ Reply to Thread

           




Similar Threads

  1. Fixed Effects Common Panel Regression versus GMM estimation
    By Herbert in forum Regression Analysis
    Replies: 0
    Last Post: 02-18-2011, 07:43 AM
  2. Replies: 1
    Last Post: 07-29-2010, 05:43 PM
  3. Quick Q on Fixed vs. Random effects
    By brownmoses in forum Statistics
    Replies: 0
    Last Post: 10-08-2009, 01:17 PM
  4. Panel data with entity and time fixed effects
    By iamsevenup in forum Statistics
    Replies: 0
    Last Post: 12-06-2008, 01:11 AM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats