(EViews10): ARIMA Models (Diagnostics)
ฝัง
- เผยแพร่เมื่อ 18 ก.ย. 2024
- Follow these simple guides once you have selected the “most” appropriate model. (1) Check the residuals correlogram to see if there’s any information yet to be captured in this model; (2) a flat correlogram is most ideal; (3) if a lag is significant, re-estimate the model; (4) check the residuals correlogram again; (5) but, avoid over-fitting a model; (6) perform Ljung-Box test for squared residuals (autocorrelation test).
Always remember that the fundamental idea of the B-J methodology is that of parsimony (meagreness or stinginess); because parsimonious models produce better results than over-parameterised models. Too many variables in a model consumes degrees of freedom more so when those variables contribute little to the significance of the dependent variable; may result in negative adjusted R-squared. A flat correlogram of the residuals is most ideal and avoid “over-fitting” an ARIMA model to a data series. The forecast is based on the final ARIMA model and there cannot be an exact or perfect ARIMA model because it is more “of an art than of science”. How can the “most” appropriate model be estimated and selected from the tentative models identified? First, estimate all the tentative models and select the most appropriate using these criteria. Choose the model having (1) most significant coefficients (2) least volatility (3) highest adjusted R-squared (4) lowest AIC/SIC. Since, ARMA/ARIMA is a method among several used in forecasting variables, the tools required for identification are: correlogram, autocorrelation function and partial autocorrelation function. The partial autocorrelation (PAC) measures correlation between (time series) observations that are k time periods apart after controlling for correlations at intermediate lags (i.e., lags less than k). In other words, partial autocorrelation is the correlation between Yt and Yt−k after removing the effect of the intermediate Y’s (measures the marginal impact). To identify the appropriate ARMA/ARIMA model, I have outlines 5 procedures: (1) plot the series to visualise if stationary or not; (2) from the correlogram, calculate the ACF and PACF of the raw data. Check whether the series is stationary or not. If the series is stationary go to step 4, if not go to step 3; (3) take the first differences of the raw data and calculate the ACF and PACF from the correlogram; (4) visualise the graphs of the ACF and PACF and determine which models would be good starting points; and (5) estimate those models. Using EViews10, this video shows you how to perfrom diagnostics using the correlogram of the residuals. Here is the link to the Gujarati and Porter Ex21-1.wf1 dataset (EViews file) used for this tutorial cruncheconometr... datasets-2/. Endeavour to have a Google account for easy accessibility.
Follow up with soft-notes and updates from CrunchEconometrix:
Website: cruncheconometr...
Blog: cruncheconomet...
Forum: cruncheconometr...
Facebook: / cruncheconometrix
TH-cam Custom URL: / cruncheconometrix
Stata Videos Playlist: • (Stata13):Estimate and...
EViews Videos Playlist: • (EViews10):Interpret V...
TH-cam recently changed the way my content will be monetised. My channel now needs 1,000 subscribers. So it would be amazing if you show your support by both watching my videos and subscribing to my channel if you haven’t done so already. Monetising my videos allows me to invest back into the channel with some new equipment so this small gesture from you will be extremely huge for me. Many thanks for your support….CrunchEconometrix loves to teach, support my Channel with your subscription and sharing my videos with your cohorts.
hello i need some help. I have the following requirement: Estimate a univariate time series model (white noise, autoregressive, moving
average or ARMA) that you think best describes the time series you have
chosen and interpret your results.
I tried using KO (coca cola) returns but i'm not understanding which is the best prcoess#
Excellent videos. To the point. Very helpful specially from an application perspective. Have watched a lot of your content. You are doing brilliant work. Please keep this going!! Best wishes!! :)
Hi Prateek, I am humbled and encouraged by your positive feedback. I hope to do more to help as many that are willing to learn..may I know from where (location) you are reaching me?
Hi thank you so much these are incredible, no other videos like these available. Brilliant!!!
Thanks, Lisa for the positive feedback and remarks on my videos. Deeply appreciated! May I know from where (location) you are reaching me?
very informative and clear concept video maam. thank you so much for sharing.
You are welcome, Ali! 💞
Excellent videos. Big thanks from the UK, subscribed, liked and shared your videos to my friends.
U're da best, Sean!😄 I'm humbled by your comment and subscription. May God bless you...and thanks for sharing too🙏
The vedio is wonderrulf thnks for sharing.
Thanks for your encouraging feedback, deeply appreciated 🥰🙏
Is my pleasure!
thank you for saving my life, i love you.
Hahahaha Firas, compliments well taken! 💕 😊 May I know from where (location) you are reaching me?
@@CrunchEconometrix I'm from Tunisia(North Africa), I'm a Risk Management Student
@@firasmbarki4793 Awesome Firas! 💕 Kindly spread the word about my videos to your friends and academic community in Tunisia 🇹🇳! 💕 😊
Maam, u are the best and i love you fir for saving my life
Hahahaha Esse, glad you find this helpful!
love your videos, please keep sharing!!
Thanks, Camilla for the encouraging feedback. Deeply appreciated!!! May I know from where (location) you are reaching me?
Thank you so much, this videos help me perfectly! Way better explanation than my professor´s
Hi Alexandra, I am humbled by your positive feedback and remarks. I will continue to do my best to simplify the technicalities involved in econometric analysis. I hope that you will subscribe (it's FREE) and encourage your friends and colleagues to do same. Please may I know from where (location) you are reaching me?
Dear Madam,
Thanks for the clear presentation. Anyway, I may have a question.
Your estimation was done in two parts; first: ARMA (1, 1), ARMA (1, 1), ARMA (1, 8), ARMA (8, 1) and ARMA (8, 8). We found that ARMA (1, 8) was the best based on the criteria (AI, SBC, …). Then, the second stage was the estimation of AR(1) AR(12) MA(8) and AR(1) MA(8) MA(12). Getting the result, we found also that AR(1) MA(8) MA(12) was selected for forecasting. So, my question is why we should estimate into two steps like that? Isn’t it a wrong path if we estimate all the possible equation at once: ARMA (1, 1), ARMA (1, 1), ARMA (1, 8), ARMA (8, 1), ARMA (8, 8), ARMA (1, 12), ARMA (12, 1), ARMA (12, 8), ARMA (8, 11), and ARMA (12, 12); then, we will choose the best based on these set criteria? Because we know from the correlogram that lag 12 is significant. If we would do that; could we get the same result? I wanna try but I do not find the link of the downloadable data.
Thanks for your coming response and also for your good explanation.
Hi Lala, no harm in exploring your procedure. After all, there is no end to learning and there are several ways to get a bird. Also, the link to the dataset is shown in the video and available in the video description. May I know from where (location) you are reaching me?
@@CrunchEconometrix Lala from Madagascar; let me try to find the data and launch my e views. I ll be back in the discussion for the result ... Many thanks again for ur kindness
excellent videos --- I mean excellent
Thanks for the positive feedback, Yannick!!!
Superbly done. Plz discuss seasonal data or SARIMA model. plz discuss (3,1,1)(3,1,1) SARIMA model with Model equation.
At the moment, I have no idea about SARIMA. But I have noted the topic. Thanks.
Are the t-statistics for ARMA coefficients adjusted for the bias of the estimator? there is no really a good way to estimate ARMA coefficients bias free like in a regular regression setting. if not, they are wrong even in large samples as the bias is very persistent and consistency kicks in in the limit.
A very clear explanation to be understood.
Should we do heteroskedasticity test for the chosen model before go to forecast step?
Thank you.
Yes, you can.
Thank you for the extremely beneficial videos! I would like to ask, if one finds that there are some lags lying outside the bounds when he tests for autocorrelation, what could be the reason for this and what are the possible ways to solve this?
Thanks in advance!
Hi Musah, I explained this and also showed what to do either in this clip or in the Stata version. So, you may need to watch both and adapt the information to your study. Thanks.
@@CrunchEconometrix , thank you ma'am
thanks professor
why you didnt choose ARIMA(12 1 12)
I explained why in the video and the preceding one. Kindly watch again. Thanks.
You are welcome, Sir.
Thanks for the video...But I would like to ask...if re-estimation is done to include an additional AR or MA term...how would the new ARIMA expression look like?
I showed what to do. You may want to check out other online resources for more information.
It is good and simple to understand Thanks, Is it possible to forecast the best ARIMA model even the residual is not normal?
Hi Gebeyehu, thanks for the positive feedback. Normality of the residual must be established.
Hello Mrs. Thanks a lot for your videos. They are really helpful. I'd like to ask a quick question. When you adjust the model arima (1,1,8) with ma(12), how should we then write the new adjusted model ? Arima (1,1,8,12) ? I guess not, I'd like to understand this part. Thank you.
Hi Dimitri, if I were to do this, I will only make a note in my work on the adjusted model....and not necessarily specify another having initially done so. Thanks.
@@CrunchEconometrix Thank you very much.
How do you Write down the adjusted Model? I have the same question
Hi Seidias, there's no need to write down the adjusted model. Explaining what you did should suffice.
Dear Professor,
On determining the optimal model and including number of lags in the ARIMA (P, d, q) what is the decision when the Sigma2 and Adjusted R2 are not moving at the same direction. Consider the results below;
Model 1:
Significant Coefficient : 3
Sigma2 (volatility): 879.2982
Adjusted R2: 0.277531
AIC: 9.769049
SBIC: 9.910767
Model 2:
Significant Coefficient : 3
Sigma2 (volatility): 855.7793
Adjusted R2: 0.296856
AIC: 9.754889
SBIC: 9.896608
Which one is a better combination, i mean, which one i should use?
Thanks
Shaf, please "adapt" explanations. Not in my place to decide for you. I gave clear explanations on the criteria...please use them. You may need to watch the clip again. Thanks.
The results are not aligned with you said in the video. The Siqma2 and Adjusted R2 are moving in different direction.
@@MShaf-ic9td Use AIC Criteria and follow my explanation.
Thank you so much for the explanation and sorry for asking too many question.
Hello sir. Thank you for this helpful video. I have a question. With the ARIMA method, how can we obtain the results of the volatility of that variable for each period of a variable, for example in an annual data set, for each year. For example, 1990 --- 1991 ---- 1992.
Thank you.
Thanks Berk for the positive feedback...appreciated! I have no idea about the approach you are suggesting. You may need to check other online resources. Regards.
Thanks you for your video, I have learnt a lot...
i just have 1 question, i am comparing 2 models, one of them has high adjusted R2, low AIC & SBIC, but the sigmasq is also higher... which model should I choose madam?
Bishal, use the guides I gave to picking the ideal model and take a decision. Thanks.
@@CrunchEconometrix alright ma'am, thanks for your response
Hi! Thanks for your videos, I found them really helpful. May I just ask a technical question, please? If I am using EViews8, do I have to write: ar(1) ar(2) ar(3) .... ar(8) ma(1) ma(2) ...ma(8) and so on when estimating ARIMA(8,1,8), for instance? Or can I only include the necessary lag order without writing them from the first one to the necessary one? Thank you in advance
Hi Diana, I showed what to in ARIMA (Estimations). Kindly watch and adapt. Thanks.
hello Madam. MY name is Abu Bakar. thank you so much for this humanitarian gesture. i really appreciate your effort. my question is: I am working on panel data. but couldn't run ARIMA with this panel data in eviews. can you kindly help. Thank you
Hi Abubacarr, I never used ARIMA on panel data. You may need to check other online resources. Thanks.
thank you very much this videos. may us you, how can choice the best models all of alternative?
Hi Tuki, thanks for the positive feedback. Deeply appreciated! But it appears you skipped the prerequisite videos on "model estimation and selection". Kindly watch them for more information. Please may I know from where (location) you are reaching me?
if i added the variable but the correlogram of the residuals is not flat then what should i do?
Then redo the test with the instruction I gave in the video.
Hi thanks for the vidio but i have question. I see you remodelling from (1,1,8) in prev video and include the 12 because correlogram shows a not-flattened pattern, but why in the estimate in eviews u include ma (8) and ma (12) simultaneously? Can I just write the formula "c ar(1) ma(12)"? or still need to write "c ar(1) ma(8) ma(12)"?
Hi Pradipta, you may decide to follow my approach or do what you feel is right. You may also check other online resources for more information. Regards.
Hey so how to decide if you should log transform raw data or not?
Hi Shriya, as detailed in econometric textbooks, articles and online resources there are several reasons for log transformation - to control for heteroscedasticity, outliers etc. Kindly read up, thanks.
If in ljung box test all the p value is insignificant, but some of the lag spike are outside the band , is it acceptable?
Jeet, this video is detailed and well explained. You may want to watch it again and apply to your results. Thanks.
If my probability of all lags in ljung box test is less than 0.05, what should i do to my data or model?
Hi Chan, there is no direct answer to this. You can combine several measures: increase lags, change the series etc.
thanx for the lecture I have one question that to check serial correlation in ARIMA model we need to check correlogram Q statistics or correlogram residual squared
Hi Ramandeep, it's the residuals as explained in the video. May I know from where (location) you are reaching me?
What if null hypothesis of no autocorrelation in residuals is not rejected???? Plz i need guidance
You re-estimate the model by augmenting the lags.
Great video...can you please tell me how did you obtain column "dppires" (residuals)... Which command did you use to transform dppi to dppires?
Which part? In this video?
Hello Madam. First off I would like to say thank you so much for making this video as it helps me for my thesis. A question though; my q statistics checks out fine but I'm having some trouble for my residual squared correlogram. All the probability is valued below 5% and lags 1 through 10 isn't flat. Any solutions for this madam?
Hi, lags 1 to 10 are outside their CIs may imply that you did not use the appropriate ARIMA model from onset. Those lags still contain some information left uncaptured.
how do you perform correlogram for this residual
Kindly watch the entire videos on ARIMA to fully understand the process.
What should i do mam if autocorrelation values are still less then 5 %
Re-estimate the model at higher-order and/or change the ARIMA structure.
Dear Mrs.
Please add auto-substitute for English video. Thank you so much!
Hi Phurong, I wish I could but that will be too much for me to handle. Please bear with me. May I know from where (location) you are reaching me?
Subtítulos por favor!
Subtitles Please!!! :´)
Ronaldo I did not subtitle my TH-cam videos. My apologies.
Hi very informative madam can you please tell me how can i run or use AR(1)AR(12)MA(8),we use ARIMA (12,1,8)?how can i write two AR terms together .i use the R software .very greatful.
Hi Neelma, thanks for the positive feedback. Deeply appreciated! But I don't know how to use the R software, unfortunately.
I dont ask how to use in R but my qus is how can we write it ARIMA(.,.,.) i.e u use two ar terms 1 and 12
@@neelma2199 Why not follow my example? It is well explained.
Dear mam,
My AR terms are : 96, 192 and my MA terms are 84, 90, 96, 100 and 192
My best fit ARIMA model is (96,1,96) and I deliberately left the combinations with 192. When I ran the correlelogram of Q statistics for (96,1,96), I got significant values in 1, 4 for both ACF and PACF which is not at all there in my original AR and MA terms (AR terms are : 96, 192 and my MA terms are 84, 90, 96, 100 and 192
).
Now what should I do for Diagnostics mam?
I should go for AR(96),MA(96),MA(192)
and AR(96),AR(192),MA(96) ????
or I should proceed with the following combinations:
AR(96),MA(96),MA(1)
AR(96),AR(1),MA(1)
AR(96),MA(96)
, MA(4)
AR(96),AR(4),MA(96)
Please do help me mam.
.
Hi Arrthi, so what are the various outcomes of the different ARIMA models you estimated?
what should we do if there are still probability values more than 5% in auto correlation test ?
Thank you.
Okki, I explained that values above 5% indicate no autocorrelation.
@@CrunchEconometrix Thank you
good evening sir, i am camara from malaysia. i actually have two problem to run the arima model: 1) i transformed my annually data to quaterlly but i still have problem to run it. 2) i used also my annually data for arima modeling but if i want to forecast, the eviews tells me to not put new year. To sum up, i followed all your step in order to run it but there is no way, please help me....
Thank you in advance
Having transformed the data, then quarterly data and not the annual should be used for forecasting.
Sir, what if during check the autocorrelation, there is only one probability not greater than 0.05, other than that are greater than 0.05. Helps me
Hi Shahrin, your query is a bit unclear.
@@CrunchEconometrix how can i snap a pic to show you😭
@@CrunchEconometrix do u have instagram of facebook
I encourage queries posted on TH-cam, thanks.
No need to snap a pix. You can state what the query is and I'll do my best to guide you.
why sigmasQ is not showing in my result please help me
Did you follow my procedure?