Pandelis Mitsis

Questions and Answers (40) See all

• Answer added in Econometrics
  50 Johansen Cointegration to test the long run relationship between exchange rate and macroeconomic fundamentals.
  By Sujan Ghosh · Bangladesh Institute of Bank Management
  Pandelis Mitsis · University of Cyprus

  Radia, You didn't reply to my question, but here it goes: 1) Already answered. 2) If you have ONLY ONE integrated variable and the rest of them are ... [more]

  Radia, You didn't reply to my question, but here it goes: 1) Already answered. 2) If you have ONLY ONE integrated variable and the rest of them are stationary then econometric theory says that there may NOT be co-integration (long-run relationship). From what you say you already found no-cointegration in your tests. Have you included all variables in the VAR/VECM models? (you mentioned some additional explanatory variables). If you did that was correct, since the model should include all, so there is not really anything else you can do to test for a long run relationship. 3) Thus IF your dependent variable is the ONLY integrated variable then you must put in in first differences and use a VAR for a short-run relationship and explain why the tests suggest there is NO long-run relationship. 4) An ARDL model is sensible if your VAR is unnecessary. This is related with the endogeneity assumption and the Granger causality tests (that I already mentioned twice, above). I.e. when the relationship between some variables is only one-way, not two-way, then is not necessary to use a system of equations (not e.g. a VAR), to model them. A single equation model (e.g. ARDL). is enough. From the information you gave me there is nothing more I can say. I wish you all the luck.
  Following
• Answer added in Econometrics
  50 Johansen Cointegration to test the long run relationship between exchange rate and macroeconomic fundamentals.
  By Sujan Ghosh · Bangladesh Institute of Bank Management
  Pandelis Mitsis · University of Cyprus

  It is not usual to use different models for the integrated variables and for the stationary ones. Is the crude oil the only I(1) variable in your mode... [more]

  It is not usual to use different models for the integrated variables and for the stationary ones. Is the crude oil the only I(1) variable in your model? How many variables you have in total? (stationery or not)
  Following
• Answer added in Econometrics
  50 Johansen Cointegration to test the long run relationship between exchange rate and macroeconomic fundamentals.
  By Sujan Ghosh · Bangladesh Institute of Bank Management
  Pandelis Mitsis · University of Cyprus

  Indeed. The related coefficients of the unrestricted VAR model would give you any evidence for short-run relationship between your variables. Just mak... [more]

  Indeed. The related coefficients of the unrestricted VAR model would give you any evidence for short-run relationship between your variables. Just make sure that your VAR is correctly specified (no auto-correlation, ho heteroskedasticity, etc). And you may complete your analysis with Granger causality tests, Variance decomposition Analysis and Impulse/Response.
  Following
• Answer added in Econometrics
  50 Johansen Cointegration to test the long run relationship between exchange rate and macroeconomic fundamentals.
  By Sujan Ghosh · Bangladesh Institute of Bank Management
  Pandelis Mitsis · University of Cyprus

  Radia, then you use a well-specified VAR model (see my first post). Attention: Since you have not co integration, ALL variables should be stationary... [more]

  Radia, then you use a well-specified VAR model (see my first post). Attention: Since you have not co integration, ALL variables should be stationary. That means that the non-stationary variables should be made stationary by differencing them.
  Following
• Answer added in Applied Econometrics
  11 Can I make a regression model with the whole population?
  By Lucia Mauro · Universidad Nacional de Mar del Plata
  Pandelis Mitsis · University of Cyprus

  Dear Muhammad, I agree with your emphasis that any results from the particular sample should be used, since they actually cover 83% of the population... [more]

  Dear Muhammad, I agree with your emphasis that any results from the particular sample should be used, since they actually cover 83% of the population, but I disagree that the statistical significance is of no relevance. You see the entire population is only a draw (as series of observations) from a statistical distribution and in that manner the statistical significance says to as how meaningful are our results. To put it differently, OLS gives results (i.e. alpha and beta coefficients), for any pair of variables, whether related or not. Professor Spanos of the Virginia Tec made several contributions in that area, with the most known example when he recorded the number of socks an elder relative of his knitted for a period of time and run regressions with national economic variables of a country and found not only estimates, but also statistical significance!!! So, I conclude that yes, Lucia must use the sample she has, but the statistical significance (and model mis-specification tests), should not be compromised. As, I mentioned, she can use the results she has (e.g. the beta) even if it is not statistically different from zero, mentioning that maybe the standard error was over-estimated due to the small sample-size.
  Following