1) run a pooled OLS estimation of the coefficients in order to get a numerical value for the
residuals;
2) use the OLS residuals to make a biased but consistent estimate of the two variance
23
components 3
where:
N [Tav.
it
2
εj
T
ave
_ УУ е,., and еi =
N у T it i '
1
T- Уе
Ti
it
3)
run a control random effect estimation for computing the true estimated model; this task is
carried out separately for each package and provides a tru. valu. for the dependent variable and for the
structural coefficients;
4) generate two random samples drawn from two normal distributions with zero mean and the
standard deviations computed in step 2;
5) add the two simulated residuals to the true value of the dependent variable computed in
step 3; in this way it is simulated a new draw for the dependent variable;
6) perform a random effect estimation using this new dataset;
7) compute the average and the standard deviation of the replications.
In what follows the previous seven-step algorithm will be referred to as MCA1 (Monte Carlo A1).
A first interesting result from the application of MCA1 is the CPU time required for the completion
of the Monte Carlo experiment. This is shown in the following graph where we compare execution time
versus number of replications for the Monte Carlo exercises. TSP is the fastest software.
23 See Wallace and Hussain (1969) for the balanced panel case.
13
More intriguing information
1. Towards a framework for critical citizenship education2. The Cost of Food Safety Technologies in the Meat and Poultry Industries.
3. The name is absent
4. Unemployment in an Interdependent World
5. The name is absent
6. Rent Dissipation in Chartered Recreational Fishing: Inside the Black Box
7. A Principal Components Approach to Cross-Section Dependence in Panels
8. Research Design, as Independent of Methods
9. Micro-strategies of Contextualization Cross-national Transfer of Socially Responsible Investment
10. The name is absent