Quasi-Maximum Likelihood Estimation Method

[Thomas Meurs]

Dear all,

As a starting point for my own thesis I'm trying to replicate a paper regarding capital structure adjustment speed.(http://www.researchgate.net/publicat...justment_Speed)
In short it is about how macroeconomic conditions affect the speed with witch a firm moves to a certain (optimal) ratio between debt and equity.

One of the first steps is to obtain this optimal leverage ratio, which is described by the following formula:
D_i,t* = γMacro_t-1 + βX_i,t-1

Where D* is the optimal leverage and is a function of prior period macroeconomic variables and firm characteristic variables.

As Papke & Wooldridge (1996) show there are some issues with linear regressions and fractional data and therefore they 'use the quasi-maximum likelihood estimation method (QMLE) to estimate the fitted value of the equation above as the proxy for target leverage.'
This is basically all the info they provide on how they have done it. As I am not the greatest statistician en most certainly not great with Stata I was hoping you could help me how to use QMLE to obtain the optimal leverage?

I have data regarding 'all' firms during the period 1977 - 2006 and includes the items such as short/long term debt and total assets (for the ratio I suppose) and quite some data that function as the explanatory variables (macroeconomic and firm characteristic) such as GDP growth, term spread, market-to-book etc.

Kind regards,
Thomas

[Joao Santos Silva]

Hi Thomas,

If you have Stata 14, have a look at -fracreg-; otherwise you will need -glm-.

All the best,

Joao

[Thomas Meurs]

Allright, I have Stata 13 so I tried this:

//QMLE
glm booklevr mtb tang ebitr dep lnta rnd rdd se, fam(bin) link(logit) scale(x2) 
estimates store C43
predict gp43 if e(sample)
predict yp43 if e(sample), xb

A code I got from one of the authors of this paper:
http://we.vizja.pl/en/issues/volume/6/issue/3#art250
Which is a replication of the paper mentioned earlier and required for my research.

Only problem is Stata is running for like 4 hours straight and is coming up with iteration 750 at the moment and does not seem like it is stopping shortly.

Any ideas why it is so running so long? Dataset too big? I have around 130.000 firm year observations.

[Richard Williams]

You are welcome to try the user-written fracglm. For info see

http://www3.nd.edu/~rwilliam/stats3/...onseModels.pdf

fracreg does most of the same things that fracglm does so I may never bother to formally release it, but it does have the advantage of only requiring Stata 11.2. I know it replicates various things Wooldridge has done but I am not sure about the specific model you are dealing with.

[Joao Santos Silva]

Dear Thomas,

The problem may be that you have perfect predictors and in that case the estimator will not exist. To check for this you can try to use the user-written command -flex- (to install type: ssc install flex). This estimates a slight generalization of the Papke & Wooldridge model and checks for perfect predictors.

All the best,

Joao