Considering the main issues of regression models in econometric analysis,
in practice, the researcher is often confronted with a situation in which received the regression is a "bad", that is, t statistics, most estimates are small, reflecting the insignificance of relevant independent variables. At the same time, the F statistic can be quite large, which indicates the importance of the overall regression. A possible reason for this phenomenon is known as multicolinearity and occurs when there is a high correlation between the factors.
One of the conditions of classical regression model is the assumption of linear independence of the explanatory variables. When this condition, that is, when one of the variables is a linear combination of the others, is called a complete collinearity. In this situation, you cannot use Ordinary least squares (MQO). In practice, the complete collinearity occurs very rarely. More often faced with a situation that among the factors that there is a high correlation. Then indicate the presence of multicolinearity. In this case, the least squares estimate (method of least squares) formally exists, but has a "bad" properties.
Multicolinearity can occur due to various reasons. For example, several of the independent variables may have a tendency of ordinary time, which undergo small oscillations. In particular, this can happen when the value of an independent variable values are dated to the other.
only some of the most characteristic signs of multicolinearity.
1. a small change in the source data (such as adding new observations) leads to a significant change in estimates of the model.
2. estimates have large standard errors, low value, while the overall model is significant (high coefficient of determination R2 and F-statistics).
3. estimates of the coefficients are incorrect from the viewpoint of the theory of signs, or overly large.
What do you do when all the signs that multicolinearity? The answer to this question is no, and econometristas have different opinions about it. When confronted with the multicolinearity problem can be a natural desire to drop the "extras" independent variables, that can serve as your cause. However, be aware that it can cause new problems. First, it is not always clear which variables are redundant in this sense. Multicolinearity just means a relationship about linear factors, but not always allocate the "extra" variables. Secondly, in many cases, the removal of all independent variables can have a significant impact on the essence of the model. Finally, the rejection of the so-called essential variables, i.e. independent variables that really affect the dependent variable studied, results in a change of model. In practice, usually when a multicolinearity clean factor at least important to the analysis and then repeat the calculations
