How to detect and remove auto correlation (serial correlation)
Вставка
- Опубліковано 14 жов 2024
- hello guys.. in this video i have showed how to detect auto correlation and how to remove it... there are two methods of detecting serial correlation i have shown you both.....
You can also buy an amazing book of ""BASIC ECONOMICS"" BY ""THOMAS SOWELL"" from the link given below:
amzn.to/35EVJKl
Disclaimer: Affiliate link included, As an amazon Associate i earn from qualifying purchases..
However, when you run the model with its lagged value, only the lagged value is significant, how therefore would you interprete this?
SIr, including lagged values of the dependent variable in the model may also introduce other issues, such as multicollinearity or overfitting.
Sir what is the interpretation of the model if we take the lagged values and how to write it in any research paper
Can i know for the Durbin Watson range was it depend on the statistics? Like we need to look at our n and k to decide the range between DL and DU. For examples, if the n is 30 and the k is 5, the range is between 0.877 to 1.606?
same question here, how to know the range please?
In no autocorrelation durbin watson stat should be near to 2 irrespective of anything else. To be on safe side we say 1.7 to 2.2
so what will the new equation be after adding the Lag?
What is the removal method called ? What is the name of removal method?
Sir how to remove multicollinearity in the lag model if you have any video than plz share Link hare 🙏
Hello , what do you mean by first order and higher orders ? Are you talking about differences ??
First order means correlation with 1 period back observation. However higher order means 2 or more than 2 period back observation
@@econacademy16 thanks , I wish you help with a study i am doing . I have one dependent variable Y and 7 variables which i want to test their ability to predict y . but i am not sure which model is more suitable. All series are non stationary and not normal at level .I took the logs and first difference for all series and I ran VAR Eviews model . After that I tested the optimal lag and based on AIC it was 2 . Then i tested for Co-integration using 2 lags . There was two cointegrations defined . As far as my understanding if the variables are stationary at first difference and there is co-integeration , I shall go to VECM model with (p-1) . Once i run the VECM model , the residuals are not normal, serially correlated and has heteroskedasticity. What shall I do in this case ? What model is suitable ?
sir, i take Inflation as an dependent variable and GDP, GCF and GFCF as an independent variables but during bound test these all variables are insignificant,i also change the variables for significant results but not gain the desirable results plzz suggest me now what i do ?
Did you connect your model with previous literature?
Hi, although l used the one period lag of dependent variable as independent variable, there is still serial autocorrelation. What is your suggestion? Thanks
There is also a built-in method for removal of serial correlation... I will make a video of it very soon
I've the problem of autocorretion in my model. The value of Durbin watson test is 0.47.
I've added 2 lags of dependent varibale (GDP) but still the problem exist.
What should I do?
Add some relevant variables
Sir my Durbin Watson test value is now 2.13 , is it showing no autocorrelation now ? If yes then why my probability chi square is 0.0000 less than 5%??
Durbin Watson is used for 1st order autocorrelation while durbin h is used for higher order autocorrelation
Hi, what if when i go to residual diagnostic in eviews, i dont have 'Serial Correlation LM Test'. I dunno where is wrong. I only have Arellano-Bond Serial Correlation Test. Please help me
You can also use this test or try get eviews 10 from somewhere
@@econacademy16 Im using eviews11 now. I still can detect auto-correlation by using Durbin-Watson Stat right ? Please correct me if im wrong.
Yes
Hello. Is the rule of thumb the same for panel data?
the observation in panel data is relatively larger so the range between 1.5 to 2.5 is also sometimes considered as no autocorrelation. the more it is closer to 2 in absolute form the less there will be autocorrelation
@@econacademy16 thanks for the response.
Dear Sir, when I use 1 lag my result is there is not serial correlation. But if I use more that 1 lag my result shows serial correlation. Can you tell me what should I do?
So it is higher order autocorrelation. Plz watch videos on treatment of higher order auto corellation
Hello. Is it okay if the |t-statistic| < 1,96 after we have removed auto correlation ?
Check probability value of t-stats
If probability is less than 10 then it will be significant at 10% level of significance
@@econacademy16 thank you very much
Doing what you said but still I have serial correlation n hetroscadasticity
Very good