which each variable is in turn explained by its own lagged values, plus past values of the

remaining n-1 variables. It combines the traditional VAR approach, which treats all the

variables in the system as endogenous, with the panel-data approach, which allows for

unobserved individual heterogeneity (Love and Zicchino 2004).

Several design issues about PVAR model are worthy noting. First of all, a critical

issue to resolve when using PVAR model is the number of lags of the dependent and

independent variables used in the equations. There is always a dilemma, if there are too

few lags, the model will be inconsistent, but if there are too many lags, the model will be

inefficient (Burkhart and Lewis-Beck 1994, Soysa 2003). For a PVAR model, it is even

more difficult to decide. Here I will use a tradition method to decide the number of lag

terms by comparing results using different lag terms. If the results using one lag term,

two lag terms, three lag terms, and four lag terms are consistent, then I will prefer using

one or two lag terms. So I specify a first-order VAR model with i lag terms as follows:

Democracy

t

= a

1

+ b

1

FDI from democratic countries

t-i

+ b

2

Democracy

t-i

FDI from democratic countries

t

= a

2

+b

3

FDI from democratic countries

t-i

+ Democracy

t-i

(i =1, 2, 3, 4, 5….n)

The second issue is control variables. In my analysis I will not include any control

variables in the PVAR model. The reason is that by controlling the lagged dependent

variable, I capture the effects of variables contributing to the dependent variable. That is,

the values of previous democracy levels contain all the information about the previous

conditions promoting democratic features, including the economic and social situations.

In this case, it is difficult for spurious effects to be reported. Furthermore, the inclusion of

the lagged dependent variable might “soak up the variations in the dependent variable

13