Hi all,
I have two questions regarding the use of clustered standard errors in my model. First, I will explain what kind of research I'm doing:
For my thesis I investigate the relationship between bank quality (proxy by credit ratings, variable name: Hqualitybank which is a dummy) and market liquidity (proxy by cash over assets). I use several financial control variables in my model. My data is in panel data format where I have observations on multiple banks( id= gvkey) over the years 2001 untill 2014. I have a static linear panel model with bank fixed effects.
In the beginning of my research I used xtreg for my fixed effects model. As cluster(gvkey) (so cluster on the bank level) or robust did not change the st dev of my coefficients, I thought that all my bank specif fixed effects took account of the potential autocorrelation. However, I am using -areg -now with absorbing the gvkey (bank) and i mentioned today that using clustered errors on the bank level yields different results than using robust st errors. The reason why this is happening, is already answered in a previous topic that I made today.
My problem now: when I use clustered std errors when clustering on the bank level, my F statistic gets very low. Do any of you knows what this implies in terms of my research? Is this normal?
However, as I have data of banks which legal incorporation is in different countries, I can also cluster on the country level by using cluster(ISOINC), but this gives my extremly high F values (which can not be good I guess).
For your information, here the regression results without using clustered standard errors:
Certainly I do not expect one of you to tell me what is going on here from an economic perspective, but I hope one could give me an explanation of what is going on here in a econometric point of view. Furthermore, is it right that clustering on the country level (with 13 clusters) allows the error terms of different banks to be correlated (if they are in the same cluster)? Do you agree that this does not really makes sense?
I hope some of you could provide me with an aswer, as this is one of the last steps in finalizing my thesis.
Thank you in advance and best regards,
Yannick
I have two questions regarding the use of clustered standard errors in my model. First, I will explain what kind of research I'm doing:
For my thesis I investigate the relationship between bank quality (proxy by credit ratings, variable name: Hqualitybank which is a dummy) and market liquidity (proxy by cash over assets). I use several financial control variables in my model. My data is in panel data format where I have observations on multiple banks( id= gvkey) over the years 2001 untill 2014. I have a static linear panel model with bank fixed effects.
In the beginning of my research I used xtreg for my fixed effects model. As cluster(gvkey) (so cluster on the bank level) or robust did not change the st dev of my coefficients, I thought that all my bank specif fixed effects took account of the potential autocorrelation. However, I am using -areg -now with absorbing the gvkey (bank) and i mentioned today that using clustered errors on the bank level yields different results than using robust st errors. The reason why this is happening, is already answered in a previous topic that I made today.
My problem now: when I use clustered std errors when clustering on the bank level, my F statistic gets very low. Do any of you knows what this implies in terms of my research? Is this normal?
Code:
areg mliq L.Hqualitybank Hbank_Crisis size nim racr loanstoassets depositsloans L.Crisisdummy fedfund tedspread, absorb(gvkey) cluster(gvkey) Linear regression, absorbing indicators Number of obs = 3755 F( 10, 128) = 7.31 Prob > F = 0.0000 R-squared = 0.8284 Adj R-squared = 0.8218 Root MSE = 0.0258 (Std. Err. adjusted for 129 clusters in gvkey) ------------------------------------------------------------------------------- | Robust mliq | Coef. Std. Err. t P>|t| [95% Conf. Interval] --------------+---------------------------------------------------------------- Hqualitybank | L1. | -.0048984 .0041785 -1.17 0.243 -.0131663 .0033695 | Hbank_Crisis | .0015327 .0045802 0.33 0.738 -.0075301 .0105955 size | .0037935 .0048815 0.78 0.439 -.0058653 .0134524 nim | -.0034218 .0028936 -1.18 0.239 -.0091473 .0023036 racr | .0016601 .0009384 1.77 0.079 -.0001967 .0035168 loanstoassets | -.046014 .0374986 -1.23 0.222 -.1202114 .0281834 depositsloans | .0278706 .0115182 2.42 0.017 .0050799 .0506613 | Crisisdummy | L1. | .0057142 .0033581 1.70 0.091 -.0009303 .0123586 | fedfund | -.0014767 .0007933 -1.86 0.065 -.0030463 .0000929 tedspread | -.0030272 .0018389 -1.65 0.102 -.0066657 .0006114 _cons | -.002524 .0798735 -0.03 0.975 -.1605675 .1555194 --------------+---------------------------------------------------------------- gvkey | absorbed (129 categories)
Code:
Linear regression, absorbing indicators Number of obs = 3755
F( 10, 14) = 1326070.70
Prob > F = 0.0000
R-squared = 0.8284
Adj R-squared = 0.8218
Root MSE = 0.0258
(Std. Err. adjusted for 15 clusters in ISOINC)
-------------------------------------------------------------------------------
| Robust
mliq | Coef. Std. Err. t P>|t| [95% Conf. Interval]
--------------+----------------------------------------------------------------
Hqualitybank |
L1. | -.0048984 .0014496 -3.38 0.004 -.0080075 -.0017892
|
Hbank_Crisis | .0015327 .0050501 0.30 0.766 -.0092987 .0123641
size | .0037935 .0027065 1.40 0.183 -.0020112 .0095983
nim | -.0034218 .0004105 -8.34 0.000 -.0043023 -.0025414
racr | .0016601 .0004722 3.52 0.003 .0006473 .0026728
loanstoassets | -.046014 .0153696 -2.99 0.010 -.0789786 -.0130495
depositsloans | .0278706 .000891 31.28 0.000 .0259597 .0297815
|
Crisisdummy |
L1. | .0057142 .0006086 9.39 0.000 .0044089 .0070194
|
fedfund | -.0014767 .0003877 -3.81 0.002 -.0023083 -.0006452
tedspread | -.0030272 .0010628 -2.85 0.013 -.0053066 -.0007478
_cons | -.002524 .0406889 -0.06 0.951 -.089793 .0847449
--------------+----------------------------------------------------------------
gvkey | absorbed (129 categories)
Code:
. areg mliq L.Hqualitybank Hbank_Crisis size nim racr loanstoassets depositsloans L.Crisisdummy fedfund tedspread , absorb(gvkey) r
Linear regression, absorbing indicators Number of obs = 3755
F( 10, 3616) = 47.10
Prob > F = 0.0000
R-squared = 0.8284
Adj R-squared = 0.8218
Root MSE = 0.0258
-------------------------------------------------------------------------------
| Robust
mliq | Coef. Std. Err. t P>|t| [95% Conf. Interval]
--------------+----------------------------------------------------------------
Hqualitybank |
L1. | -.0048984 .0016164 -3.03 0.002 -.0080675 -.0017292
|
Hbank_Crisis | .0015327 .0027668 0.55 0.580 -.0038919 .0069573
size | .0037935 .0016172 2.35 0.019 .0006229 .0069642
nim | -.0034218 .0013215 -2.59 0.010 -.0060129 -.0008308
racr | .0016601 .0003776 4.40 0.000 .0009197 .0024004
loanstoassets | -.046014 .0131866 -3.49 0.000 -.071868 -.02016
depositsloans | .0278706 .0039685 7.02 0.000 .02009 .0356513
|
Crisisdummy |
L1. | .0057142 .0017543 3.26 0.001 .0022745 .0091538
|
fedfund | -.0014767 .0002993 -4.93 0.000 -.0020635 -.00089
tedspread | -.0030272 .0011775 -2.57 0.010 -.0053358 -.0007185
_cons | -.002524 .0230106 -0.11 0.913 -.0476391 .0425911
--------------+----------------------------------------------------------------
gvkey | absorbed (129 categories)
I hope some of you could provide me with an aswer, as this is one of the last steps in finalizing my thesis.
Thank you in advance and best regards,
Yannick
Comment