Forum: help

RE: nlsystemfit: about instrument variable and stratal [ Reply ]
By: Arne Henningsen on 2018-11-17 09:52

The FAQ says: "Can the systemfit package estimate a system of non-linear equations?
The systemfit package provides the function nlsystemfit, which can estimate systems of non-linear equations. However, while the code for estimating systems of linear equations (function systemfit) has been very mature and reliable for many years, the estimation of systems of non-linear equations (function nlsystemfit) is still under development and often has convergence problems. Therefore, we cannot recommend using function nlsystemfit for serious applications." My co-author Jeff Hamann (https://www.linkedin.com/in/jeff-hamann-phd/) has developed nlsystemfit(). I suggest that you contact him directly.

nlsystemfit: about instrument variable and stratal [ Reply ]
By: julie wang on 2018-10-24 06:38

[forum:46365]

Hi, I have two equations like the following:

>earn.formula = earnings ~ gamma0 + gamma1 * accrual + gamma2 * cash
>abnor_rt.formula = abnor_rt ~ beta * (earnings - gamma0
- gamma1.star * accrual
- gamma2.star * cash)

I also have the datasets of earnings\ accrual\ cash\ abnor_rt. My aim is to fit the equations and compare their coefficients.

I am now using nlsystemfit function from r package systemfit. But I find it difficult to determine the instrument variables (is it ~accrual+ cash). Also, may I ask that how can I determine the start value? Will the start value influence the final results, for example, two sets of different start values which can both make convergence possible.