Announcement

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

  • Fixed Effects on a weakly balanced data-set of cross-border M&A transactions

    Dear Statalist users,

    I am a complete beginner using Stata, so please bear with me if my question is too obvious.

    I am researching the impact of corruption and governance on cross-border M&A performance. My dataset includes 600+ cross-border M&A announcements between 2008 and 2017. This is a case of a weakly balanced panel data as a country and firm can appear only once or several times during the 10-year period. In some cases, the same country and firm can show up more than once in the same year. The data looks something like this:
    Year Action ID Country Firm CAR X1 X2 X3
    2008 1 France a -0.06
    2008 2 France b 0.02
    2008 3 Croatia c -0.20
    2009 4 England d -0.01
    2009 5 France a 0.04
    2010 6 Belgium f 0.01
    2011 7 Belgium g -0.03
    2011 8 France a 0.02
    2011 9 Belgium f 0.04
    2012 10 Croatia c 0.08
    2012 11 Croatia j -0.09
    I am interested in the model: CAR = a + x1 +x2 +x3 + u

    My question is can I use a fixed or random effect model with this type of data? - if so which variable should I use as cross-section?

    I have tried the following:
    1) xtset country year, yearly
    (which gives me error 451, which indicates that there are repeated values)

    2) xtset ActionID year, yearly

    this one gives me the following result:
    panel variable: actionid (weakly balanced)
    time variable: year, 2008 to 2017
    delta: 1 year

    However, when I run my regression: xtreg car x1 x2 x3 i.year, fe - I get all omitted variables due to collinearity.

    I'd really appreciate any help I can get.

    Many thanks,
    Henry

  • #2
    Henry:
    the way you -xtset- your data in code 2) is troublesome, as -ActionID-, in the excerpt you provided the list with, does not show repeated values ( as a -panelid- should).
    That said:
    - I would focus on the identification of the real -panelid- that you need for your analysis (I would say: country or firm);
    - then, if you do not plan to use time-series commands, such as lags and leads, you can safely -xtset- your panel dataset with the -panelid- only, and get rid of the -timevar-;
    - you show preference for -fe- specification. This impies that you have tested beforehand via -hausman- which specification (-fe- or -re-) fits your data better;
    - as an aside, things would be easier for those interested in helping you out if the outcome of your regression was reported as well (via CODE delimiters, please; see the FAQ on this). Thanks.
    Last edited by Carlo Lazzaro; 15 Jul 2018, 12:01.
    Kind regards,
    Carlo
    (Stata 18.0 SE)

    Comment


    • #3
      Dear Carlo,

      Thanks very much for your response and guidance. Based on your comment, I think I should use country as the -panelid- because firms do not repeat very often in the data-set but countries do. After all, most of the data in my research contains variables at country level.

      As suggested, after getting rid of the -timevar-, I executed a hausman test, but I am not clear about how to interpret the results, which are:

      chi2 (3) = 4.12
      Prob>chi2 = 0.2488

      Based on what I have read/seen online, this result fails to reject the null hypothesis. In other words, random effects should be the appropriate model. Is that the correct interpretation?

      The hausman test also brought up another potential issue about my data. Along with the results, Stata gave me the following note: "the rank of the differenced variance matrix (16) does not equal the number of coefficients being tested (22); be sure this is what you expect, or there may be problems computing the test. Examine the output of your estimators for anything unexpected and possibly consider scaling your variables so that the coefficients are on a similar scale"

      In fact, my data contains variables in different scales (e.g. indexes ranking from 0 to 100, percentages, millions of USD). Would this become a problem if I try to run a multiple regression? - if so, what would be the best way to transform all my explanatory variables into the same scale?

      I'm pasting an excerpt of my data below, which includes a few variables illustrating their different scales:

      Code:
      * Example generated by -dataex-. To install: ssc install dataex
      clear
      input int year long actionid str12 acquirercountry str16 targetcountry str50 acquirername str45 targetname float(car tarcorruption tartaxrate) double targdp byte cashdummy
      2008  30248193 "U.S."         "Japan"       "Walmart Inc"                          "Seiyu GK"                      .006094572  85.43689 51.8 4.51526e+12 0
      2008  32136566 "U.K."         "Philippines" "Ashmore Group PLC"                    "Petron Corp"                   .012289982   27.6699 49.1  1.4936e+11 1
      2009  37468663 "U.S."         "Germany"     "BorgWarner Inc"                       "Beru AG"                        .06451058  93.20388 49.4 3.75237e+12 1
      2009  36352999 "Japan"        "Germany"     "TDK Corp"                             "Epcos AG"                     -.020245453  93.20388 49.4 3.75237e+12 1
      2010  50228427 "Hungary"      "Croatia"     "MOL Hungarian Oil & Gas PLC"          "INA Industrija Nafte DD"       .027952485  57.89474 20.8 62703143057 1
      2011  52107352 "Switzerland"  "Germany"     "Clariant AG"                          "Sued-Chemie AG"                 -.0610842  93.33334   47 3.41709e+12 1
      2011  55502653 "India"        "Germany"     "Suzlon Energy Ltd"                    "Senvion SE"                     .15572175  93.33334   47 3.41709e+12 1
      2012  61227199 "U.K."         "Canada"      "Rio Tinto PLC"                        "Turquoise Hill Resources Ltd"   .03867084  95.26067 20.7 1.78865e+12 1
      2012  66004736 "France"       "Russia"      "Danone SA"                            "Unimilk Co OJSC"               .015166337 15.165876 46.8 2.05166e+12 1
      2012  68997319 "U.S."         "Belgium"     "Liberty Global PLC"                   "Telenet Group Holding NV"        .0403212  91.94313 57.7 5.27008e+11 1
      2014  92837853 "South Africa" "Australia"   "Woolworths Holdings Ltd/South Africa" "Country Road Group Pty Ltd"   -.017820526  93.83886 47.3  1.5737e+12 1
      2014  99102226 "Sweden"       "Finland"     "SSAB AB"                              "Rautaruukki OYJ"              -.037084136  98.10426 39.9  2.6998e+11 1
      2014  99631582 "U.K."         "U.S."        "Amec Foster Wheeler PLC"              "Foster Wheeler AG"            -.002426952  87.67773 43.8 1.66915e+13 0
      2015 112534809 "U.K."         "Germany"     "Sky PLC"                              "Sky Deutschland AG"            .018293347  94.71154 48.8 3.89061e+12 1
      2015 107239675 "Japan"        "South Korea" "Toray Industries Inc"                 "Toray Chemical Korea Inc"      .000480968  70.19231 33.2 1.41133e+12 1
      2015 110138435 "Switzerland"  "France"      "LafargeHolcim Ltd"                    "Lafarge SA"                     .03104728  88.94231 70.8 2.85217e+12 0
      2016 127255046 "Japan"        "Netherlands" "Recruit Holdings Co Ltd"              "USG People BV"                 .007311615  94.71154   41 7.57999e+11 1
      2016 131527731 "Finland"      "France"      "Nokia OYJ"                            "Alcatel Lucent SA"            .0000147832  88.94231 61.9 2.43821e+12 1
      2016 135639431 "Japan"        "France"      "Toyota Tsusho Corp"                   "CFAO SA"                       .014685898  88.94231 61.9 2.43821e+12 1
      2017 150384703 "Netherlands"  "France"      "ALTICE EUROPE NV"                     "Altice France SA/France"       -.06690646  90.38461   64 2.46513e+12 1
      end
      Again thanks so much for your time. Any help will be much appreciated.

      Best regards,
      Henry

      Comment


      • #4
        Henry:
        - as far as the post-hausman- note is concerned, please see: https://www.statalist.org/forums/for...in-stata-13-1;
        - your interpretation about -hausman- outcome is correct (ie, go -re-);
        - just to be sure, I would re-run the test after scaling.
        Kind regards,
        Carlo
        (Stata 18.0 SE)

        Comment


        • #5
          Carlo,

          I'm getting an invalid Page URL error with the link provided. Can you please re-share it?

          Also, any resources on scaling would be appreciated.

          Thanks a lot for your help.

          Henry

          Comment


          • #6
            Henry:
            - the working link is: - unfortunately, I have no stuff about scaling, but the following thread (about -xtlogit-, but the same logic should hold): https://www.statalist.org/forums/for...an-for-xtlogit.
            Kind regards,
            Carlo
            (Stata 18.0 SE)

            Comment

            Working...
            X