Announcement

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

  • Help with explanation of low R-Square

    Hello. I am always being grateful for posting my queries here as I receive helpful comments and suggestions.
    Once again I am looking here experts especially Carlo Lazzaro, Nick Cox and Fei Wang for your kind suggestions.
    My query is:
    I have estimated fixed effect model for predicting stock returns using financial variables such as the market return, interest rate, oil prices and oil price volatility.
    The issue is that R-square is very low ( maximum R-square is 0.12) and my supervisor asked me to change the technique or provide explanation for this low R-square.
    The comment from my supervisor is:
    Code:
    R-square is very low and t-stats for the coefficients are very high. What is the explanation for this?
    I am posting the screen from my FE regression. Please help me to sort out this as soon as possible. thank you very much again.
    Code:
     xtreg Rs Spread Rm rExr zmv rBrent vBrentttt i.date, fe vce(cluster compid)
    note: 691.date omitted because of collinearity
    note: 692.date omitted because of collinearity
    note: 693.date omitted because of collinearity
    note: 694.date omitted because of collinearity
    note: 695.date omitted because of collinearity
    
    Fixed-effects (within) regression               Number of obs     =     24,605
    Group variable: compid                          Number of groups  =        259
    
    R-sq:                                           Obs per group:
         within  = 0.1525                                         min =         95
         between = 0.0032                                         avg =       95.0
         overall = 0.0584                                         max =         95
    
                                                    F(95,258)         =      33.61
    corr(u_i, Xb)  = -0.7842                        Prob > F          =     0.0000
    
                                   (Std. Err. adjusted for 259 clusters in compid)
    ------------------------------------------------------------------------------
                 |               Robust
              Rs |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
    -------------+----------------------------------------------------------------
          Spread |   -.066479   .0073067    -9.10   0.000    -.0808674   -.0520906
              Rm |   1.884146   .1969506     9.57   0.000     1.496311    2.271982
            rExr |  -.5437886   .2824927    -1.92   0.055    -1.100074    .0124964
             zmv |   .0402099   .0041168     9.77   0.000     .0321032    .0483166
          rBrent |  -1.726282   .2543047    -6.79   0.000     -2.22706   -1.225505
       vBrentttt |  -1.858098   .3769667    -4.93   0.000    -2.600422   -1.115775
                 |
            date |
            602  |  -.2007925   .0252852    -7.94   0.000    -.2505842   -.1510008
            603  |  -.0575308   .0189211    -3.04   0.003    -.0947902   -.0202715
            604  |   -.313948   .0425363    -7.38   0.000    -.3977106   -.2301854
            605  |  -.2572561   .0343938    -7.48   0.000    -.3249845   -.1895277
            606  |  -.1102228   .0270083    -4.08   0.000    -.1634075   -.0570381
    --more--

  • #2
    Zulfiqar:
    exception made for -i.date- your predictors seem to have a small within-panel variation (standard errors are, on averare, very low when compared to point estimates; thus causes high t and low p-value, respectively).
    The small within-panel variation is also captured by the very low within R-sq.
    It my well be that -fe- is not the way to go here and/or your model is misspecified.
    As an aside, it is really weird that your supervisor cannot give you a guidance about these issues (which are fairly common in panel data econometrics).
    In addition, please do not urge potentially interested listers to reply to your query ASAP, as this is one of the best way to make your queries unreplied on purpose. Like you, we all have to look after our own deadlines and Stata forum is neither a helpline, nor a call center for quants. Thanks.
    Kind regards,
    Carlo
    (StataNow 18.5)

    Comment


    • #3
      Thank your for your time and helpful suggestions Carlo Lazzaro sir.

      Comment


      • #4
        Zulfiqar:
        Carlo is enough, thanks.
        Kind regards,
        Carlo
        (StataNow 18.5)

        Comment

        Working...
        X