Forum: help

RE: CES Estimation problem [ Reply ]
By: Arne Henningsen on 2018-03-25 07:03

Some suggestions:

1) Take a look at the "message" from the optimisation algorithm to see whether it "thinks" that it converged to a (local) optimum or not.

2) Conduct your estimation with several different optimisation algorithms and check whether you get very similar estimates or rather different estimates.

3) Conduct your estimations with many different sets of starting values and check whether you get very similar estimates or rather different estimates.

4) If you get different estimates (for the same data set and model specification) with different optimisation algorithms and/or different sets of starting values, only the one estimate with the lowest sum of squared residuals is potentially the global optimum, whereas all other estimates are either local optima or no optima at all.

5) Take a close look at the things that we have done in: http://doi.org/10.1016/j.eneco.2017.12.019

RE: CES Estimation problem [ Reply ] By: doga alkan on 2018-03-24 13:29	[forum:45782]
Dear Arne Thank you for your reply. So if I get an estimate of 0.8 with my dataset is it then meaningless since it the estimated rho seems to have the same value as the starting point, which then biases the result towards 0.8?. And is there a way to see from the estimation output if the optimum is a global or a local one? Thanks in advance and kind regards Doga

RE: CES Estimation problem [ Reply ]
By: Arne Henningsen on 2018-03-20 21:57

[forum:45774]

The elasticity of substitution equals to 0.8 at the *starting value* rho = 0.25 so that the elasticity of substitution equals 1 / ( 1 + rho ) = 1 / 1.25 = 0.8. The *estimate* of rho depends on the data. In most empirical applications, the *estimate* of rho should be different from 0.8 so that elasticity of substitution is different from 0.8. Different optimisation algorithms (e.g., Levenberg-Marquardt, BFGS, Nelder-Mead, ...) are usually differently successful in finding the global optimum of the objective function. Therefore, the estimates may depend on the optimisation algorithm. However, estimates that are obtained by an optimisation algorithm that did not find the global optimum of the objective function need to be disregarded.

RE: CES Estimation problem [ Reply ]
By: doga alkan on 2018-02-26 17:20

[forum:45715]

Dear Arne

Thanks a lot I managed to do it finally. But I came up with another question regarding the size of the estimated elasticity. As you wrote in your paper if the ρ coefficient is estimated its starting value is set to 0.25 the value of δ to 0.5 and the value of ν is 1. Now what I didn't understand is how is it possible to generally say that the elasticity of substitution equals to 0.8. Doesn't it depend on the data too? Indeed I received for my estimations with some of the estimation methods as Kmenta/Levenberg-Marquardt, BFGS, but some others such as the Newton f.e. I received 0.6, or the Nelder Mead gave me a 0.4 as an output. Is there a logic behind this or are these just the differences in the way the algorithms work?

RE: CES Estimation problem [ Reply ] By: Arne Henningsen on 2017-12-31 06:25	[forum:45532]
Dear Doga The cesCalc() command that you posted seems to be incomplete, because it does not specify the (dependent) 'output' variable. I guess that the nls() commands that you posted do not work, because your data set "Popular_Indicators_2_" does not contain any variables with names "y2" or "GDP" or the variables with these names are non-numeric variables. Best regards, Arne

CES Estimation problem [ Reply ]
By: doga alkan on 2017-11-30 12:21

[forum:45482]

Dear Prof. Henningsen

I am quite new to R and I am currently writing a paper utilizing your package and I came across a problem. I am trying to estimate the CES production function for the United States at the aggregate level. My inputs are labour force for L and I used Gross Capital Formation for K. Now I tried to estimate this with these codes:

ceslan <- cesCalc( xNames = c( "GrossCapitalFormation", "LaborForce" ), data = Popular_Indicators_2_, coef = c( gamma = 1, delta = 0.6, rho = 0.5, nu = 1.1 ) )

Afterwards I tried to do the non linear estimation with the code:

cesulan <- nls(y2 ~ gamma * (delta * GrossCapitalFormation^(-rho) + (1-delta)*LaborForce^(-rho))^(-phi / rho), data = Popular_Indicators_2_, start = c(gamma = 0.5, delta = 0.5, rho = 0.25, phi = 1) )

Now this is where my Problem is: I dont know what variable is meant to be y2 in my dataset. I can see in the formula that y2 ~ gamma *... so ist plotted against the rest of the term, but I dont know what kind of value I need to plug in there. I thought of using GDP, but when used it in the code:

cesulan <- nls(GDP ~ gamma * (delta * GrossCapitalFormation^(-rho) + (1-delta)*LaborForce^(-rho))^(-phi / rho), data = Popular_Indicators_2_, start = c(gamma = 0.5, delta = 0.5, rho = 0.25, phi = 1) )

But I get the error I get the error non non-numeric argument for a binary operator. I really don't know how to proceed and would be very glad if you could help me.

Thanks in advance and kind regards

Doga Alkan