Announcement

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

  • #16
    Dear All,
    I have sorted the second doubt of mine pertaining to centering of variables, but I need to be sure about the code that I have used is it correct or not? Can someone please help me out in this.
    Since, this helps in deciding weather I should proceed with an IV approach or use FE estimation technique, it is important for me to know the validity of the command. Here is the code and results:
    Code:
    xtivreg2 ROTA (l.Lev1 l.Lev1_Square l.Int_Lev1_HHI l.Int_Lev1_M l.Int_Lev1_D =  l.Tang1 l.Tang1_Square l.Int_Tang1_HHI l.Int_Tang1_M l.Int_Tang1_D) l.HHI_Sales l.M l.D l.Log_TA l.Real_INV, fe robust endog(l.Lev1)
    In the above code Lev1 is endogenous and it has been incorporated with Interaction terms (3 variables), this is instrumented using Tang1 and alteration of interaction terms accordingly. All the interaction (with endogenous and with its instrument) have been centered around mean. And the following is the result:

    Code:
    Warning - singleton groups detected.  72 observation(s) not used.
    
    FIXED EFFECTS ESTIMATION
    ------------------------
    Number of groups =      2039                    Obs per group: min =         2
                                                                   avg =       8.9
                                                                   max =        16
    
    IV (2SLS) estimation
    --------------------
    
    Estimates efficient for homoskedasticity only
    Statistics robust to heteroskedasticity
    
                                                          Number of obs =    18141
                                                          F( 10, 16092) =     4.39
                                                          Prob > F      =   0.0000
    Total (centered) SS     =  966849.1269                Centered R2   = -11.3403
    Total (uncentered) SS   =  966849.1269                Uncentered R2 = -11.3403
    Residual SS             =  11931186.54                Root MSE      =    27.22
    
    ------------------------------------------------------------------------------
                 |               Robust
            ROTA |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
    -------------+----------------------------------------------------------------
            Lev1 |
             L1. |  -5.239194   5.194879    -1.01   0.313    -15.42097    4.942582
                 |
     Lev1_Square |
             L1. |  -.0696026   .0409295    -1.70   0.089     -.149823    .0106179
                 |
    Int_Lev1_HHI |
             L1. |  -43.99888   54.42539    -0.81   0.419    -150.6707    62.67292
                 |
      Int_Lev1_M |
             L1. |  -5.048674   35.73508    -0.14   0.888    -75.08815     64.9908
                 |
      Int_Lev1_D |
             L1. |  -162.2579   102.3338    -1.59   0.113    -362.8285    38.31259
                 |
       HHI_Sales |
             L1. |  -13.31835   4.854366    -2.74   0.006    -22.83273   -3.803964
                 |
               M |
             L1. |    8.29546   2.488203     3.33   0.001     3.418673    13.17225
                 |
               D |
             L1. |    -10.352   15.21114    -0.68   0.496    -40.16529    19.46128
                 |
          Log_TA |
             L1. |  -2.134837   .7679852    -2.78   0.005    -3.640061    -.629614
                 |
        Real_INV |
             L1. |   .5034356   .3268634     1.54   0.124    -.1372048    1.144076
    ------------------------------------------------------------------------------
    Underidentification test (Kleibergen-Paap rk LM statistic):              1.481
                                                       Chi-sq(1) P-val =    0.2236
    ------------------------------------------------------------------------------
    Weak identification test (Kleibergen-Paap rk Wald F statistic):          0.304
    Stock-Yogo weak ID test critical values:                       <not available>
    ------------------------------------------------------------------------------
    Hansen J statistic (overidentification test of all instruments):         0.000
                                                     (equation exactly identified)
    -endog- option:
    Endogeneity test of endogenous regressors:                               0.107
                                                       Chi-sq(1) P-val =    0.7442
    Regressors tested:    L.Lev1
    ------------------------------------------------------------------------------
    Instrumented:         L.Lev1 L.Lev1_Square L.Int_Lev1_HHI L.Int_Lev1_M
                          L.Int_Lev1_D
    Included instruments: L.HHI_Sales L.M L.D L.Log_TA L.Real_INV
    Excluded instruments: L.Tang1 L.Tang1_Square L.Int_Tang1_HHI L.Int_Tang1_M
                          L.Int_Tang1_D
    ------------------------------------------------------------------------------
    Following #3 I could deduce that with the above set of results, I fail to reject the exogeneity of Lev1 and hence, can conclude that it is not endogenous in nature.

    Can anyone please suggest me if the code is fine or not.

    regards,
    Mohina

    Comment


    • #17
      What to do when the variable to identify endogeneity is omitted from the model due to collinearity?
      My original model is: xtreg IPDMEduca FPM_POP tx_analfabeto desp_aluno_percap, fe
      qui areg and create variable ab3. And when I run the model again the result is:

      Fixed-effects (within) regression Number of obs = 4,788
      Group variable: id Number of groups = 399

      R-sq: Obs per group:
      within = 0.0566 min = 12
      between = 0.2230 avg = 12.0
      overall = 0.1510 max = 12

      F(3,4386) = 87.67
      corr(u_i, Xb) = 0.2504 Prob > F = 0.0000


      IPDMEduca Coef. Std. Err. t P>t [95% Conf. Interval]

      FPM_POP .0000279 5.61e-06 4.98 0.000 .0000169 .0000389
      tx_analfabeto -.004081 .0008497 -4.80 0.000 -.0057469 -.0024151
      desp_aluno_percap 1.56e-07 6.53e-08 2.39 0.017 2.77e-08 2.84e-07
      ab3 0 (omitted)
      _cons .7177454 .0122465 58.61 0.000 .693736 .7417547

      sigma_u .09090718
      sigma_e .05838362
      rho .70798287 (fraction of variance due to u_i)

      F test that all u_i=0: F(398, 4386) = 27.01 Prob > F = 0.0000

      Comment


      • #18
        Dear All,

        I have a question about Durbin-Wu-Hausman test. By going through old FAQ about this test some questioons came up. First of all, I am trying to test whether my independent variable (agricultural export) is endogenous. What i did, I run a first regression when agricultural export was the independent variable. In this regressions I used the other control variables of the main regression plus other variables. After that, I collect the residual and I insert in the main equation(with agricultural export left out). I did not assume any IVs. Is my approach correct? Thank you a lot before hand

        Comment


        • #19
          Originally posted by Jeff Wooldridge View Post

          Krissy: Your case is a bit harder, but doable in Stata. I cover it in Section 11.2 in my MIT Press book, 2010, 2e. I think you have to do it "by hand" as xtreg2 does not support RE and xtregress does not have the endog test available.

          Let y1 be the dependent variable, y2 the endogenous explanatory variable, z1, ... zL the exogenous variables, with z1, ..., zM, M < L, included in the model.


          Code:
          reg y2 z1 z2 ... zL
          predict v2hat, resid
          xtreg y1 y2 v2hat z1 z2 ... zM, re vce(cluster id)
          The t statistic on v2hat is the test of the null that y2 is exogenous. If you reject, you conclude IV is needed. The test is fully robust to serial correlation and heteroskedasticity.

          A word of caution: You are requiring pretty strong exogeneity of your instrument. It must be uncorrelated with the heterogeneity in the structural equation, as well as the shocks. If your explanatory variable and instruments change over time, FEIV will be more convincing.

          JW

          Jeff: Can I apply the same procedure if I have two endogenous explanatory variables (y2 and y3) and two instruments?

          Code:
          reg y2 z1 z2 ... zL
          predict v2hat2, resid
          reg y3 z1 z2 ... zL
          predict v2hat3, resid
          xtreg y1 y2 v2hat2 y3 v2hat3 z1 z2 ... zM, re vce(cluster id)
          Now is the t statistic on v2hat2 the test of the null that y2 is exogenous, and the t statistic on v2hat3 the test of the null that y3 is exogenous?

          Kind regards,
          Mehrzad

          Comment


          • #20
            Originally posted by Jeff Wooldridge View Post

            Marco: The null hypothesis is that diff is exogenous, and you are failing to reject that. That's different from what you said, although the response is the same in that you don't appear to need an instrument. You should do usual fixed effects (without IV) to see how the point estimate changes. My guess is not by much. You have interpreted the weak instrument test correctly: your instrument appears to be very strong (which, of course, says nothing about whether it is exogenous).
            Dear Jeff,

            I am conducting my research about CHINESE FDI IN AFRICA, my main purpose of the study is: Does Chinese FDI bring some technological spillover into African countries? thus my data consist of N=32 African countries in a period of time t=14 years. My 1st dependant variable is y1=TFP (Total Factor Productivity) change of each country, and the main independent variables are x1= Chinese FDI stock, x2= technological difference, x3=human capital, and some control variable as x4=resource rent by an African country, x5=industries output / GDP, x6= economic openness.

            after running:

            Code:
             xtreg y1 x1 x2 x3 x4 x5 x6 i.year i.COUNTRY, fe
            (NB: for x1 x3 and x4, I lagged them for 1 year in this regression in order to attend my expected P value, at the same time the lagged 1 year value makes more sens in economic point of view for these three variable)

            So my question is, which kind of code should Iuse to check if any of my independant variable is or are exogenous?

            Comment


            • #21
              Duplicate post #20. see tempative answer at: https://www.statalist.org/forums/for...obustness-test
              Kind regards,
              Carlo
              (StataNow 18.5)

              Comment


              • #22
                Originally posted by Carlo Lazzaro View Post
                Duplicate post #20. see tempative answer at: https://www.statalist.org/forums/for...obustness-test
                Dear Carlo,

                Please forgive me if I made a mistake,

                After reviewing the PDF file in the link that you shared: I found that in the material in the 19th slide: "4. Models with heterogeneity and endogeneity". It assumes that y_it2 is a vector of endogenous variables.

                And my question is: after building my model (ex: y_it1 = d + ax_it1 + bx_it2 .... + c_i + z_t + e_it), is there any code that I can run to define which of these variables in my model is ENDOGENOUS? or should I just start with assuming that some x_it is exogenous and then run the estimation by using some IV, and then test with - estate endog- if that x_it is actually exogenous or not?

                Comment


                • #23
                  Rakotondrazaka:
                  as far as I know, there are no Stata built-in test like -estat endogenous- for panel data regression.
                  See the community-contributed module -xtivreg2- though.
                  Kind regards,
                  Carlo
                  (StataNow 18.5)

                  Comment


                  • #24
                    Rakotondrazaka:
                    what above was also suggested in some previous replies to the original query that started this thread.
                    Kind regards,
                    Carlo
                    (StataNow 18.5)

                    Comment


                    • #25
                      Originally posted by Carlo Lazzaro View Post
                      Marco:
                      the slide-set dowloadable at http://www.ifs.org.uk/docs/wooldridge%20session%204.pdf looks interesting.
                      Dear Carlo, or any other friend who can help me;

                      I am trying to locate the document Marco is referring to, but I am unable to do so as it appears that the page does not exist. Does anyone have it available or can you tell me the name of the document?

                      Thanks a lot

                      Comment


                      • #26
                        Jonathan:
                        welcome to this forum.
                        The link I mentioned years ago seems to be boroken.
                        You may find Jeffrey M. Wooldridge | Institute for Fiscal Studies (ifs.org.uk) helpful.
                        Kind regards,
                        Carlo
                        (StataNow 18.5)

                        Comment


                        • #27
                          Hi Everyone.
                          i have a panel data set and i have used fixed effects model, as the research is being conducted in corporate finance. i want to test the robustness of the results, can i use the random effects model to do so? one of my friends has suggested me that i use Heckman's estimation to check the robustness, because he has also done the same but his arguments were not convincing enough.

                          Comment


                          • #28
                            Abid:
                            for the future, please start a new thread with an informative title (as per FAQ). Thanks.
                            That said, if the -fe- estimator is the way to go, the -re- is inconsistent.
                            You may want to consider different specidications of the right-hand side of your -fe- regression instead.
                            Kind regards,
                            Carlo
                            (StataNow 18.5)

                            Comment


                            • #29
                              Thank you, Carlo.
                              and I apologise for posting here rather than starting a new thread. and let me try to share more details about my research with you. I am working on a topic that aims to identify and measure the impact of Intellectual capital and Corporate Governance on the enterprise value of the firm. I have used Pulic's VAIC model as a measure of the intellectual capital of the firm. For measuring corporate governance I have taken the ESG Governance score of firms, which gives the percentage of corporate governance efficiency of the firm. I also hypothesize that there is an interaction between intellectual capital and corporate governance that impacts the enterprise value of the firm. to test the relationship I have used a nested model where I gradually add corporate governance and then the interaction between intellectual capital and corporate governance to the model of intellectual capital and Enterprise Value alone. I have taken the total assets of the firm as a control variable to control the variation in the size of the firms. I have also taken financial leverage as another control variable. the data is collected for four different countries, and i have done the regression on both separate country-wise data and collectively for all the firms in all the company as a single data set. the data has been collected for a 12 year period. now, using the Fixed effects estimator i reject all the null hypotheses. i have also divided the data for each country into 7 industries and the results are mostly consistent throughout all the industries except for a few industries, like the financial industry in the case of India. and real estate industry in the case of China etc.

                              now some of my friends told me that i should check the robustness of my results, and this thing is driving me crazy. i am not able to find any suitable text that would help me go through this process. i would really appreciate your valuable suggestions and any advice that you may have to offer.

                              Comment


                              • #30
                                Abid:
                                I'm not familiar with your research field.
                                However, yiou may investigate what happens when you remove -i.timevar- from the right-hand side of your regression equation.
                                See the literature in your research field (unfortunately, colleagues' advice cannot be quoted in a paper ) and act on accordingly.
                                Kind regards,
                                Carlo
                                (StataNow 18.5)

                                Comment

                                Working...
                                X