Given that you data is time based and most likely has a seasonal component I would suggest that you use ARIMA. The basic test of significance for each parameter in ARIMA is T-Test. I have included a reference below that discusses the details of ARIMA. Depending on what software you use (SAS, STATA, Etc.) there should be a great number of operational examples online.

Chris

Box, G. E. P., Jenkins, G. M., and Reinsel, G. C. (1994), Time Series Analysis: Forecasting and Control, 3rd Edition, Englewood Cliffs, NJ: Prentice-Hall.

the VAR analysis is usually applied for monetary policy transmission purpose both in standard reduced-form and SVAR

Likewise, Yuthana, I would recommend you to use the SVAR. Thus, with impulse responses you can be able to see the immediate and the long-run impacts.

Dear Pahlaj,

It depends on various factors. Firstly, Sample size. since you have economic growth, usually it is on annual basis or it might be on quarterly basis so the data is not as high frequency as usually required. If data is annual, no seasonal components are to be calculated, if quarterly then seasonal adjustments are to be done (available in E-views)

Secondly, it depends on Stationarity of data sets. if all are integrated of same order, you will go for cointegration followed by Causality. otherwise you will go for ARDL.

Thirdly, it also depends on your research question, if you are interested in to figure out structural breaks, then SVAR or if you want to figure out impact of policy then Quasi experiment model.

Hope it will help.

I would use VECM if I do a single entity with long data and I would use PVAR if I do cross-sectional and time series analysis. For PVAR, check out my paper titled, The Impact of Subnational Fiscal Policies on Economic Growth: A Dynamic Analysis Approach here. Good luck!

Erick, how  do I identify long term effect from impulse response? So far, I can get only medium-term effect. Thanks!

I would recommend that you use ARIMA for the time series as mentioned in your question/

I would suggest the following steps. First, check the stationarity of the series. Secondly, check for cointegration using different methods if the series are I(1), . If cointegration is evident apply VECM. Check model validity, significance of the error correction term and coefficient estimates and also their signs. If everything is okay estimate variance decompositions, Granger causality and impulse response functions within the VECM framework. You can then assess the impact level of the monetary policy variables on growth. Note that choice of cointegrating regression method depends on sample size and the unit root characteristics of the series. If the series are not I(1), still you can apply ARDL method.

Hope this will help.

Best wishes. 

Yes, if the variables are I(1) and cointegrated, one can follow the path of VECM, Variance decompositions, Impulse response functions and Persistence profile with a view to discerning the Granger-causality. However, if the variables are not I(1), one can follow the path of ARDL for testing cointegration and if the variables are found cointegrated, then proceed with the Granger-causality analysis for finding out the impact of monetary policy on economic growth.

Thanks.

Best wishes,

Mansur Masih

i recommend computable general equilibrium model for this type of economic modeling.

Hi,

It depends on various factors and on what you seek to do in your research. If you are interested in interactions between economic growth and other indicators. I can recommend you the VAR analysis.

From the results of some Johansen cointegration tests, you can find out if you are obliged to use a VAR or VECM model in your research.

Also, causality issues can be treated in this kind of question but there exists some big confusions concerning the interpretation of the results of VAR analysis and causality analysis in empiric literature. These are two distinct concepts.

So, figuring out this kind of details can distinguish your paper from others.

Also, Bruneau and Jondeau paper from Central Bank of France can be a good reference in order to understand that these are two different concepts.

Good luck,

Regards,

Thank you all for your valuable time in providing me various useful suggestions. We plan to use two independent variables: 1) money supply M2, and 2) Interest rate, and we use GDP growth as dependent. Time series data set is available, in quarterly form, only for the period from 1993 to 2012. Cambodian economy is also affected by dollarization.

In this case, any specific suggestion would still be much appreciated. I am new in this field.  

Dear Pahlaj,

If economy is affected by dollarization, then you can incorporate "exchange rate" as predictor.

You should first test for stationarity of the variables.GDP growth rate is usually stationary but M2 and interest rate need not be.I have seen some papers in which interst is considered in log it is not correct .

The framework for examining the impact of M2 and interest rates on growth is actually the monetary policy transmission mechanism. This is actually a short run analysis that examines how a shock to the short-term interest rate is transmitted to intermediate variables such as M2, Exchange rates, inflation, employment and growth. the VAR or VEC Impulse response analysis is the usual model for this type of analysis. the analysis should begin by examining the time series properties of the data, such as unit roots, granger causality,  cointegration, etc.  Good luck

You can use VECM model after testing stationarity.

Hi Pahlaj,

I would start with the methodology described in Levine & Renelt (1992), mainly due to the following reasons:

a. Their analysis covers some very extensive sensitivity analysis tests.

b. Their analysis offers a division into time periods, which is heavily relied on in later researches.

c. Their methodology is straight-forward, yet is capable of producing robust results.

However, since your study deals with less studied data (the Cambodian economy), I would be cautious in applying methods used in studying US/EU markets. It's true that VECM models are your safest bet, but try applying some Sub-Model models (like SMAR), since sampled data may be subject to delayed effects caused by fixed exchange rates or inflation taxes. As a thumb rule: the greater the intervention of the central bank in the economy, the less chances are VECM model results be significant, since the influence of the independant variables does not neccesarily have a continous effect on GDP growth (meaning, exchange rate on t-1 will have no effect, while on t-2 will be significant). In this case, UVAR models are simply not robust enough to capture the effect of government policy on GDP growth... 

In my previous study, I did use VECM for each individual country. If I have to do it again, I would like to try switching regime in which the methodology can be changed depending on time period and whether the data has structural breaks. I don't know much about switching regime, but someone here might know. I would like to learn more about it.  

By implication,  you are trying to model the transmission mechanism (how changes  in nominal variables translate to real variables). Though I don't know much about the country  you are studying, prior experiences have shown that GDP growth rate and M2 are likely to be of different order of integration. I therefore suggest that the first step in your analyses is to test for stationarity and cointegration of  your variables likely using the ARDL bounds tests.  Depending on your findings from the test, you can now apply the  VECM. 

Again, if your analysis goes beyond academic exercise, there is need for you to know the target of the monetary policy rules in the country and put it into consideration in your model; know the state of openness of the economy and  the impact of dollarisation on it  for you to include exchange rate in your model.  Also, try to read the Taylor rule if you are not familiar with it.

@Pahlaj, In addition to conventional VAR and VECM models, you may try applying the 'Structural VAR' methodology for analyzing the dynamic response of GDP growth due to monetary policy shock. 'Structural VAR' methodology applies theory driven restrictions on the model variables and you can perform tests to assess the validity of these restrictions. In addition, instead of using 'money supply' as the policy variable, you should use the 'bank rate' as the monetary policy variable according to contemporary empirical research. There are some problems when you use 'money supply' as the policy variable. For the sake of brevity, I don't want to lengthen my discussion on this particular issue. Please check relevant literature.