How to choose the right way of estimation

[Doraemon]

Hello!
I estimate the dependency of export,gdp and human capital. If choosing the linear method, I got this:

Dependent Variable: EXPORTS
Method: Least Squares
Date: 05/23/15 Time: 18:20
Sample: 1960 2011
Included observations: 52

Variable Coefficient Std. Error t-Statistic Prob.

C 2.63E+10 1.38E+10 1.911506 0.0618
HC -1.36E+10 6.08E+09 -2.233089 0.0301
GDP 2903680. 192313.2 15.09870 0.0000

R-squared 0.967407 Mean dependent var 1.90E+10
Adjusted R-squared 0.966076 S.D. dependent var 2.22E+10
S.E. of regression 4.08E+09 Akaike info criterion 47.15324
Sum squared resid 8.16E+20 Schwarz criterion 47.26581
Log likelihood -1222.984 Hannan-Quinn criter. 47.19640
F-statistic 727.1844 Durbin-Watson stat 0.745562
Prob(F-statistic) 0.000000

The sign of HC coefficient is negative, which is against the theory.I have tried logarithmic, exponential forms, but I still get negative results for HC.
I wonder what is the way to estimate it right.
Thank you in advance.

[startz]

Your problem is probably that the variables are all nonstationary. A least squares regression doesn't generally work in these circumstances.