So far I know that regression model or equation is not used to test or investigate causal relationships.

A regression model can be used to investigate causal relationship.  If you obtain a model with input variables that explain changes in output variables, you can use that model to prediction.  A causal relationship occurs when noise is zero (or enough small).  The real problem is find the input variables: for example news can affect currencies market (mercado de divisas)., but what type of news?.

In addition to what Gabriel said, you need to control for confounding variables in order to rule out alternative explanations. This is especially a problem if you do not have experimental data.

There is a nice explanation of Prediction vs. Causation in Regression Analysis by Paul Allison published at : http://www.statisticalhorizons.com/prediction-vs-causation-in-regression-analysis.

Regards, Karin

Yes regression model can be used to investigate the cause and effect relation between variables

Regression analysis alone may not be sufficient to establish causality in proper scientific sense. This is because there could be many nonsense regressions, especially in time series. As an example, suppose you record the time of returning from office of two persons and express it as number of minutes past 12-00 pm. These two persons (say A and B) work in two different offices and are unrelated to each other. One (A) returns everyday around 5-00 pm and another (B) around 6-00 pm. Suppose you observe them for 100 days, getting a pair of 100 observations. You will get very good result if you either (i) regress A on B or, (ii) regress B on A. You may even interpret B causes A (B returning on an average 23 hours before A everyday) or A causes B (A returning one hour before B on an average every day), depending upon your specification. However, in reality, neither variables CAUSE the other and each is determined by other variables not known to you. That is why in case of regression it is extremely important to have the right kind of theory. If theory suggests that one variable (say, X) is a cause and the other (say, Y) is an effect and you get good results regressing Y on X, then only you may interpret that X causes Y. In Econometrics, when one uses the term causality in the context of regression, that is often in predictive sense rather than pure scientific sense. For example, in case of Granger causality, if you regress Y on its own past and the past of X and find that some of the coefficients of X are statistically significant, you may interpret that X Granger causes Y.

Try to look at the methodology.

These studies gives some insight on how to implement GC test in regression models.

Thesis The interdependence between stock markets of BRIC and develo...

Article Investigation of the consequences of ignoring daily data non...

Usually, the decision to use a regression model represents an implicit belief that (at least some) of the right-hand-side (RHS) variables "cause" the left-hand-side (LHS) variable. The Eviews Granger Causality tests are in fact carried out within a regression model where the RHS contains lagged values of the LHS variable and of at least one other variable. The causality test investigates whether one variable's lags can be dropped from the regression model without significantly reducing its ability to explain the LHS variable.

It can be used some how but reliability has issues  specially with time series dat , it is never a good option. spurious results may distort actual relationships

It depends on your theory and the type of regression analysis being used. Regression can be used to determine a causal relationship between X and Y in a controlled environment. However, to determine the certainty of the cause you may need to pay attention to the mechanism (the process through which the cause occurs). This means a different type of data and analysis may be required (usually qualitative data). Or a different kind of modelling may be required.

All the conventional tests for causality have some weak points, first of all correlation doesn't show us causation. The regression results also doesn't show us causation even we have economic theory, Suppose we have valid theory and one has estimated a regression model to find caual ordering, this will also lead to spurious causal results because may be x and y both varaibles is cause by third variable z which is not included in the model, it means that we have to use a model in which all the causal paths should be included in the model and also capture the effect of third variable or confounding variable. The existing tests haven't the ability to capture it and I only suggest one model that is developed by Peter, Pearl, Clark called PC causality algorithm which has the ability to find the true causal relationship.