low coefficient Value of exploratory variables

[amin khan]

Respected Members,
I am dealing with panel data consisting 300 observations, eight exploratory variables and one dependent. after Hausman test it is observed that Fixed effect is appropriate model for my study. however after apply fixed effect model, i observed very low coefficient values of exploratory variables. i already check different assumption of Fixed effect such normality, autocorrlation, heteroskedasticity etc etc. kindly guide me about the acceptable range of coefficient for panel data and also provide, if any, test to increase these values. detail of my Fixed effect model in givel below;..... thanks


xtreg te npl car ppl lr clar bs ms roa, fe

Fixed-effects (within) regression Number of obs = 300
Group variable: pid Number of groups = 30

R-sq: within = 0.1577 Obs per group: min = 10
between = 0.2765 avg = 10.0
overall = 0.2203 max = 10

F(8,262) = 6.13
corr(u_i, Xb) = -0.5405 Prob > F = 0.0000

------------------------------------------------------------------------------
te | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
npl | .0003412 .0001714 1.99 0.048 3.66e-06 .0006787
car | .0006233 .000161 3.87 0.000 .0003063 .0009404
ppl | -.0008087 .0002908 -2.78 0.006 -.0013813 -.0002362
lr | .0006671 .0003297 2.02 0.044 .0000179 .0013162
clar | -.0010263 .0004606 -2.23 0.027 -.0019332 -.0001193
bs | .0057359 .0017689 3.24 0.001 .0022528 .009219
ms | -.0076023 .0022863 -3.33 0.001 -.0121042 -.0031005
roa | .0003093 .0007325 0.42 0.673 -.0011331 .0017517
_cons | .8587813 .0330103 26.02 0.000 .7937821 .9237804
-------------+----------------------------------------------------------------
sigma_u | .0262539
sigma_e | .01828363
rho | .67340302 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(29, 262) = 9.92 Prob > F = 0.0000

[Clyde Schechter]

Well, there's nothing wrong with small coefficients. They are what they are.

Now, I can only speculate here, because you tell us nothing about what these variables are. But perhaps the coefficients came out very small because you choose to measure npl, car, etc. in units where a change of 1 unit is actually a very large, perhaps even unattainably large, change. If you change the units of the predictor variables so they are measured in smaller units (say from km to m, kg to g, whatever) then the corresponding coefficients will scale up accordingly. This won't change anything substantive in your model: the inferences will all be the same. But the coefficients will be more like "normal" numbers--an aesthetic consideration.

By the way, in the future, please post code and Stata output between code delimiters. Read FAQ #12 for the details of how to do that. It will make your posts more readable, as things will align properly.

[amin khan]

thanks you Slyde Schechter for sharing your knowledge. actually i am examine the impact of different variables such Non-Performing Loan (npl), Capital Adequacy Ratio (car), Provision for Loan Losses (ppl)on efficiency. The data was collected from annual reports based PKR (Pakistani Rupees) and all the variables are measured in Percentage. however after running my model i found very beta values of exploratory variables. so i need your suggestion about either these value are acceptable or recommend me you some test to enhance these values.

[Carlo Lazzaro]

Amin:
Clyde already gave you sound advice.
As far as I can see your query, nobody on the list (and many more off the list, I suspect) can give you a threshold value about the "plausibility" of your values.
By the way: what does the literature in your research field say about the "usual" magnitude of those coefficients?
As an aside, as Clyde (and FAQ) wisely recommended, please post what you typed and what Stata gave you back via CODE delimiters. Thanks.

[Clyde Schechter]

So your predictor variables are probably denominated in small currency units such as dollars, euros, yen, yuan, pounds, etc. If you re-scale them so they are denominated in thousands or tens of thousands of currency units, the variables themselves will scale down and the coefficients will scale up.

Again, I want to emphasize that this is purely an esthetic matter. It makes no difference whether your effect is .002 per dollar or 2 per $1,000 dollars:it means the same thing. Carlo's advice to see what the literature has to say about how these variables are usually handled and the kind of results usually obtained is excellent.