Researcher has collected data of Stock Index daily closing price, Bond yield daily closing rate, daily Brent oil closing price and Currency exchange rate daily closing price.
Researcher is looking to find out impact of oil price, Bond yield and exchange rate fluctuations on stock market returns.
Therefore, Kindly suggest in what form all four variable are required to transform to assess the impact using VAR and VECM Model in Eviews.
What do you mean by "form". However I think the stationarity and cointegration properties of the variables would help to determine whether to use VAR or VECM.
Since you forecasting, I rather prefer you leave the variables in levels. However in specifying your model, you can do a double-log, log-linear or lin-log or you could use them without log. However be mindful of the integration order of your variables. If you really do understand the impact of shocks of your IVs on the DV very well you can decide to use the SVAR or SVECM model and forecast from there.
I would think that the effect of a shock on any of these variables is multiplicative. For example if your equity index is at 100 a certain shock might increase it by say 5 points or 5% to 105. If the index was at 200 would you expect an identical shock to increase the index to 205 or 210. If you conclude that 205 is the correct answer then treat the variables in levels. If you conclude that 210 is correct you should model the variables in logs. For my part I would think that the log transformation is correct.
Note that your bond yield is a return. Your other variable are price indices. You need to convert your bond yield to a bond price to make it compatible with the other data.
As some theory predicts that the log of asset prices is non-stationary you need to use a VECM to analyse these series. The length of time covered by the series is important. You should consider how long it takes for a disequilibrium to be dissipated and need a multiple of that period in a VECM analysis. The number of observations is not as important as the length of the series. You may also have some missing variables in your analysis (prices output etc.) that might be available monthly. Consider a monthly rather that a daily series if you need these extra variables
Dear John C Frain thanks for your reply. I pursued with VECM analysis and later used VDC and impulse response to finalize the same. Now I am facing heteroskedasticity in VECM model which have two co-integration equation and variables also failed normality test. Kindly provide the remedy to remove heteroskedasticity and make variables normal.
Variable used: logreturn, logbondyield, logexchnagerate and logcrudeprice.
There is no easy answer to your question. The first thing that I would look would be the possibility of some breaks in the data or other innovations. You might try to identify these from economic history or changes in the method of computation of the indices. your non-normality and heteroskedasticity may be caused by such events. It may be easier to manage these events in monthly rather than daily data. These problems are discussed at great length in Juselius (2006), The Cointegrated VAR model, Oxford. This covers the VECM model in much greater detail and depth than could be covered in a forum such as this.
Dear Gopal Saraswat
I expected you first developing a multiple regression model and you must show that The Derbun Watson factar less than R-Square in this case we expect the suprior regression then we must do the unit root test if we fine there are no stationary by level of all variables and they become stationary in the same difference then we can go on by applying VECM if we find first coeficient c1 minus and significance then we can continue with our analytic if not (ci not sig. Or not minus) we change to VAR model.
With Best Regard
- Badges
- Science topic
- Similar topics
- Psychology
- Mental Processes
- Learning