Dear Cecilla!
On 22 July 2013 14:08, Cecilia Carmo <cecilia.carmo at ua.pt>
wrote:> I have the following model:
> Cost of debt = intercept + information quality + control variable1
> + control variable2 + ? + error term
>
> I want to perform 2sls because I think I could have two situations:
> First: maybe information quality is correlated with the error term
> (because of omitted variables)
>
> Second: maybe information quality depends on the cost of debt like this:
> information quality = intercept + Cost of debt + control variable1 +
> control variable2 + ? + error term
>
> I have some variables (instruments) for information quality.
> Now I need to know how to use systemfit with this information
> in each of those situations.
The use of systemfit depends on the econometric model that you want to
estimate. In the first case, the following specification might be
suitable:
result1 <- systemfit( Cost_of_debt ~ intercept + information_quality +
control_variable1 + control_variable2, inst = ~ control_variable1 +
control_variable2 + instrument_variable1 + instrument_variable2,
method = "2SLS" )
Please note that you have to assume that the instrumental variables
(instrument_variable1 and instrument_variable2) do not influence the
cost of debt.
In the second case, you can only estimate the model if you have some
exclusion restrictions, i.e. at least one exogenous variable that does
not influence the cost of debt (e.g. instrument_variable1 and
instrument_variable2) and at least one exogenous variable that does
not influence the information quality (e.g. control_variable2). In
this case, the following specification might be suitable:
sys <- list( Cost_of_debt ~ intercept + information_quality +
control_variable1 + control_variable2, information_quality ~ intercept
+ Cost_of_debt + control_variable1 + instrument_variable1 +
instrument_variable2 )
result2 <- systemfit( sys, inst = ~ control_variable1 +
control_variable2 + instrument_variable1 + instrument_variable2,
method = "2SLS" )
You could also use the 3SLS method in this case.
Please do not forget to cite the systemfit package in your publications. Thanks!
Best regards,
Arne
--
Arne Henningsen
http://www.arne-henningsen.name