EViews.com

first i need to describe the model a little. I am trying to estimate the values of players.

i got these data: value,age,primary stats,secondary stats, form, potential.
the problem is that most of the data is rounded and both primary stats and secundary stats are the sum of two rounded variables i can't observe, but the most important is that this means that if i get a primary stat of 20 it could be between 19-21 in reality. Also the potential isn't exactly observable, but i know in which interval it is. Currently i just use the middle of this interval as value for potential. The form and age are observed correctly. Furthermore, the equation is not really linear. I solved this by taking the log of the value which solves most problems but doesn't really take the incorrect observations into account.

i want to estimate a model like:

log(value)=c(1) + c(2)*primR + c(3)*secR +c(4)*potentialR+ C(4)*form + c(5)*age
primR=prim + error
secR=sec + error
potentialR= potential + error

i tried to give eviews a system of equations, but it just doesn't seem to work. I got about 15000 observations so that is not the problem if you are worried about estimating. Maybe i am making this to difficult since i estimated a decent model just using the first equation without errors, but i know there is a perfect relation between the value and the other parameters so i am trying to get an almost 100% fit.
Any help is appreciated.

You appear to be trying to estimate an errors-in-variables model. You can't do this by simply writing out all the equations, because the true measurements are unobserved...you don't have any data to give EViews.

In principle, you can do an estimation that takes account of the rounding process as a way to estimate the rounding error variance. But it would be a lot of work to figure out.

You might try the following: Make up data that you think is close to the truth (15,000 observations). Then round the variables. Then run your regression on the made-up, rounded data and see if the estimated coefficients are approximately correct.

Thanks for your quick response. I couldn't find the correct term for the model, but now i know that at least..
I have thought about selecting the correct data, but this would be a very time consuming process, its also very hard to do. I was hoping 15000 observations would somehow make my program converge to the truth, but i am still a little bit from it although the error seems small. I got an R^2 of 0,9946. Also i know some values
for example prim=4*sec for sure and the output of the model gives exactly that.. so it seems there are some things correct, but i have a few mistakes although i am still not sure where. Maybe the measurement errors are not the problem. At least i think they should disappear almost with so many observations. Knowing that the parameters for primaries and secondaries seem to have the right value i might need to look into the remaining variables, but they seem to have such a small weight compared to the estimates for primaries and secondaries. Still sometimes i make large errors in predictions. I am trying to predict secondaries/primaries. For example i know a certain player got 40 primaries and i need to know secondaries. In a test case when i know both secondaries and primaries are 40 and 40 my model predicts 40/35 which is a big difference with the real value. This could only be explained in case my model is not correct (not talking about parameters estimates). I am thinking hard here, but i can't think of another solution that i am capable of using other then the solution you suggested, finding true observations and estimate the model with those.