Timeline for Daily forecasting using ARIMA in R
Current License: CC BY-SA 4.0
22 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 22, 2018 at 13:34 | history | edited | Ferdi | CC BY-SA 4.0 |
deleted 12 characters in body
|
| Nov 22, 2014 at 17:43 | comment | added | Tom Reilly | Yes. you can try it both ways and see which way you like better. | |
| Nov 21, 2014 at 18:15 | comment | added | jjreddick | Tom, When you recommended "a regression model with NO ARIMA" do you mean using a simple multi-variate linear regression model by creating those dummy variables for week/month? | |
| Jul 26, 2013 at 15:01 | comment | added | Tom Reilly | Pankaj, You need to have a column for each holiday. Otherwise you are assuming that the impact is the same. Please adjust the file and name the holidays at the top of the column. Chl---I am in the US and pankaj is in india...we are responding infrequenly...pankaj...contact me at [email protected] | |
| Jul 26, 2013 at 12:41 | comment | added | pankaj jha | dropbox.com/s/yq1u1qzhreedlwy/totdata1.xlsx.... Done Tom. I have replaced the missing value with the average of that month. Please let me know if you like the missing value in any other way . | |
| Jul 26, 2013 at 12:27 | comment | added | chl | Please, consider creating a chat room to continue this discussion. | |
| Jul 26, 2013 at 11:53 | comment | added | pankaj jha | the sales figure for these 11 days is missing . Before fitting a model I replaced it with the average of March months. I removed it from the data provided to you so that you could suggest somthing else as treatment. I will do the same in the data kept at dropbox. For holiday I have created a field in column AH. | |
| Jul 26, 2013 at 11:45 | comment | added | Tom Reilly | Pankaj, Your data set has missing observations. If there was zero volume you still need to consider it. A time series by definition is observations over an equally spaced period of time. For example, you have no March 1, 2012 or March 2,2012. There are 11 days you are missing. Also, you have no holiday variables. That is fine if you think that there is no impact, but highly doubt that they are not important. Report to dropbox when you are ready. | |
| Jul 26, 2013 at 4:50 | comment | added | pankaj jha | dropbox.com/s/yq1u1qzhreedlwy/totdata1.xlsx .... Please check this link | |
| Jul 25, 2013 at 20:44 | comment | added | Tom Reilly | Pankaj, It looks like the file wasn't uploaded properly. Please check. | |
| Jul 25, 2013 at 6:15 | comment | added | pankaj jha | dropbox.com/s/n94e0it09l4eslu/totdata1.csv.... the variables marked in red are target and marked in grey are used in model. I am not using variable AJ to AL. Lag depends a lot on different variables usually a weeks lag is good as suggested by my clients. As of now I have not used any lag. Also I need weekly prediction. Thanks for the help Tom. | |
| Jul 24, 2013 at 14:50 | comment | added | Tom Reilly | This is good, but can you label what variables you are actually using in the model. Highlight them in red? Also highlight what the Y variable is. I like that you are trying to use a causal to explain the history like "numofuniqcustproductlevel". You will have to provide a forecast of this as required by regression. It also seems like you might be bringing in too many of these (ie dist_invnogrplevel). Columns AJ to AL make no sense to use in a regression. Maybe you are not using these? Again, I need to see what you actually want to use. Do any have a known lead or lag relationship? | |
| Jul 24, 2013 at 12:12 | comment | added | pankaj jha | dropbox.com/s/n94e0it09l4eslu/totdata1.csv ... have put the daily data here . The data belongs of a wholesale shop for delhi , india region – pankaj jha 10 mins ago edit | |
| Jul 21, 2013 at 19:04 | comment | added | Tom Reilly | Open an account with dropbox.com and post the data there and then the link here. Post any output that you have in Excel or text file....actuals, your causals fit and your model. | |
| Jul 21, 2013 at 9:41 | comment | added | pankaj jha | Thanks Tom ,I am doing the outliers treatment and also using a latent variable to tag those events. I am new to stackflow. Really don't know how to put post data here | |
| Jul 19, 2013 at 20:59 | comment | added | Tom Reilly | Pankaj, You can use a trend variable (ie 1,2,3,4,etc) as a regressor. The trend variable is a surrogate, but it is trying to handle something that you are not in terms of explanatory power. For example, you can include a causal variable that can help like "# of stores" or "# of sales reps" and future values of stores which can explain that "growth" provide "guidance" to the forecast. You likely are not handling other issues that I have mentioned already (ie outliers, lead/lags before holidays). I see you did not post your data. You can scale the data to anonyomize!! Still the same model! | |
| Jul 19, 2013 at 13:03 | comment | added | pankaj jha | Thanks Tom , I am trying the regression approach as it will help the client to visualize the seasonality in a better way(by coefficient of different months and days of week as well). But one small question how to handle yearly growth . As the business is new I can see there is decent growth in volume of business from last year. I have included the monthly dummy variables, days of week , holidays and even daily promotional offers. But still results is not coming in the desirable variance. Any help would be greatly appreciated | |
| Jul 18, 2013 at 11:53 | comment | added | Tom Reilly | You can also post your model and results to dropbox.com | |
| Jul 18, 2013 at 11:39 | comment | added | Tom Reilly | We agree that it is difficult with such a short dataset, but these are the cards you were dealt. ARIMA lag 7 could be used, but if you consider forecasting accuracy from many origins you might see disaster around holidays. Post your data and state the starting date and country where the data resides and we can take a look. | |
| Jul 17, 2013 at 19:25 | comment | added | pankaj jha | Thanks Tom for replying. I understand it is difficult to handle seasonality with one year data. I am trying the regression approach suggested by you and will share the result. Also I was trying to fit arima simply. I was informed by my clients that they are getting good result from moving average only . So I thought ARIMA will for sure improve on that. Any Comment on this approach | |
| Jul 17, 2013 at 17:37 | history | edited | Gala | CC BY-SA 3.0 |
added 3 characters in body
|
| Jul 17, 2013 at 17:30 | history | answered | Tom Reilly | CC BY-SA 3.0 |