My data consists of quantity of vehicles sold and the date on which it was sold. It shows an increase at the month ends, ie; on 30th and 31st of the month ends. Data is of 6 month date wise. Since i have only variable "quantity sold" and i want to predict the future sales, i thought of going with univariate time series analysis. What else can i use for my prediction? how to model my above case where the sales tend to increase at the month ends? should i use the multiplicative model or additive time series model?
$\begingroup$
$\endgroup$
1
-
$\begingroup$ post your data and I will try to take a look at it to suggest an approach. Specify the first date. $\endgroup$– IrishStatCommented Sep 16, 2016 at 20:52
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
3
- You can consider an explanatory model in which you could look at factors such as Holidays, Promotions, etc and how they affect the seasonal part of the time-series data. This could help you with the prediction of the seasonal portion.
- To check if your time series is additive or multiplicative, A visual examination can be used for preliminary analysis. In an increasing trend, If the size of fluctuations in the data increases with increase in the level of the series, then higher probability of a multiplicative model and if the size of the fluctuations remains constant irrespective of the level of the series, then higher probability of additive model.
To statistically confirm, you can refer to the following post to check if your time series is a better fit for an additive or multiplicative model.
https://www.r-bloggers.com/is-my-time-series-additive-or-multiplicative/
The above link will help you understand whether multiplicative or additive is a better fit for your time-series. In brief, It compares the auto correlation (acf) between the remainder components in the time series for both additive and multiplicative. The decomposition which gives a lower acf is a better fit
-
$\begingroup$ +1 but please include some context for links, preferably quoting (but crediting) the most critical parts (if sufficiently brief). $\endgroup$– Glen_bCommented May 25, 2017 at 7:38
-
$\begingroup$ I will keep that in mind. The link will help you understand whether multiplicative or additive is a better fit for your time-series. In brief, It compares the auto correlation (acf) between the remainder components in the time series for both additive and multiplicative. The decomposition which gives a lower acf is a better fit. $\endgroup$ Commented May 25, 2017 at 8:55
-
$\begingroup$ Could you edit that into the answer ? Just click the "edit" link under your present answer. $\endgroup$– Glen_bCommented May 25, 2017 at 9:32