Forum: help

RE: SUR-Estimation in R, error:"LU.dgC(a):cs_lu(A) failed: near-singular A (...)" [ Reply ]
By: Arne Henningsen on 2017-06-17 18:25

[forum:45200]

You can estimate your model both by SUR and by OLS and then use, e.g., a likelihood ratio test to test whether the likelihood value of the SUR model (which allows for cross-equation dependence by allowing for non-zero covariances between error terms of different equations) is significantly larger than the likelihood value of the OLS model (which does not allow for cross-equation dependence by restricting all covariances between error terms of different equations to be zero).

RE: SUR-Estimation in R, error:"LU.dgC(a):cs_lu(A) failed: near-singular A (...)" [ Reply ]
By: Gürcan Özden on 2017-06-13 18:52

[forum:45197]

Dear Arne,

thank you very much for your response. Recently I found the problem...There were two companies which had missing values ("NAs") in the time series of the return data..That was the reason for the output of the error (so yes, it was due to insufficient observations). And indeed the variance-covariance matrix is huge and takes really long time to compute.

My model looks like this: R(i,t) = alpha(i)+beta(i)*Rm(i,t) + gamma(i)*EventDummy+ error(i,t)
R stands for daily return in time t for company i (for each company (i) I have 261 observations); Rm stands for the daily return of my benchmark index, which is the total market index of the respective country of the company ( I have companies of over 6 countries); the EventDummy takes the value of 0 or 1: it is 1 on specific dates, which probably have an effect on the returns of the companies.

If you allowed, I would like to ask something else. Is there a way to test for the cross-equation dependence ? I know that you can actually test this with the lm test of breusch pagan (1980), do you know with which command I can test this in R ? And are there any other tests I can use to test the assumptions of the SUR-model ?

Thank you in advance

Gürcan

RE: SUR-Estimation in R, error:"LU.dgC(a):cs_lu(A) failed: near-singular A (...)" [ Reply ]
By: Arne Henningsen on 2017-06-10 05:27

[forum:45194]

It seems that the error message occurs, because your model parameters cannot be identified (theoretically), e.g. because of (nearly) perfect multicollinearity or insufficient degrees of freedom. What is exactly your model specification? Please note that estimating one equation for each of the N companies and allowing for a non-zero correlation between each pair of the companies (as done in a SUR estimation) requires the estimation of an NxN covariance matrix with N*(N+1)/2 independent elements. With N = 145 companies, you have to estimate 145 * 146 / 2 = 10,585 independent (co)variance parameters plus the other model parameters, which likely cannot be identified due to insufficient observations and/or too high 'multicollinearity' within these 10,585 (co)variance parameters and between these (co)variance parameters and the other model parameters.

SUR-Estimation in R, error:"LU.dgC(a):cs_lu(A) failed: near-singular A (...)" [ Reply ]
By: Gürcan Özden on 2017-06-02 09:23

[forum:45174]

Hello everybody, I'm new to this forum!
I have a problem with conducting a SUR-Estimation in R. I would appreciate it very much if someone could help me out...

Background information:
I have the daily returns over a time period of 276 days of 145 European companies (from Germany, Netherlands, France, ...) ...and I have the daily returns of the same time period of European market equity ndices (from Germany, Netherlands, France, ...)...finally I have Eventdummies, i.e. a list of zeros and ones over the time period....Now I want to regress the daily returns of the companies on the daily returns of the European market equity indices and the eventdummies...

When I do this in R with the plm and the systemfit package, there comes following error message and I dont have a clue what this means "Fehler in LU.dgC(a) : cs_lu(A) fehlgeschlagen: near-Singuläres A (oder außerhalb des Speichers)" (German); in English it would be: "the equivalent is "LU.dgC(a) : cs_lu(A) failed: near-singular A (or out of memory)"

I really would appreaciate any help, thanks in advance !!!