Anpassning av ARIMA-modeller till försäljningsdata
(2010) STAK11 20101Department of Statistics
- Abstract (Swedish)
- The purpose of this thesis is to display our results received from the assignment GfK addressed to us. GfK is an independent part who aid companies when they need statistical assistance. In order to forecast sales for one of GfK’s customers we have build models with time series analysis. We received 26 products with monthly data from 2005 to 2008. In this thesis we have narrowed it down to six of these products, which we find representative in the range of methods we used and the problems we encountered.
We have been using the program package of Minitab when processing the time series and building the models. The method used is for the most part the Box-Jenkins method, but with some adjustments. First step in the process was to make our... (More) - The purpose of this thesis is to display our results received from the assignment GfK addressed to us. GfK is an independent part who aid companies when they need statistical assistance. In order to forecast sales for one of GfK’s customers we have build models with time series analysis. We received 26 products with monthly data from 2005 to 2008. In this thesis we have narrowed it down to six of these products, which we find representative in the range of methods we used and the problems we encountered.
We have been using the program package of Minitab when processing the time series and building the models. The method used is for the most part the Box-Jenkins method, but with some adjustments. First step in the process was to make our time series stationary and even though the method alters from product to product, there are some general theories which have been presented in chapter 2. In order to make adjustments for inconsistent variance we have been using Box-Cox transformation or the log-transformation and to process trend we have used detrending and differentiating. Furthermore in cases where outliers are present we have used dummy variables and when building models ARIMA has been our instrument of choice. With these models we finally let Minitab forecast sales for the next coming year.
In our thesis, for the products we have analysed, we find that sales are best explained with a seasonal model. Out of the six products we got two SAR(1)12, three SMA(1)12 and one combined MA(1)*SMA(1)12. The fit of our models has in most cases been good since the forecasts seem to fit well with the original time series. On the contrary we have encountered some problems during the process which has been discussed in chapter 4. First and foremost it would have been favourable if we would have had a larger amount of observations to work with, but we have also discussed alternative methods that could have improved our thesis. All in all we are satisfied with the models built from the applied theory. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/1612874
- author
- Andersson, Freddy LU and Fältman, Ellinor LU
- supervisor
- organization
- course
- STAK11 20101
- year
- 2010
- type
- M2 - Bachelor Degree
- subject
- keywords
- time series, trend, transformering, avtrendning, säsongsmodell, prognostisering, tidsserie, ARIMA, sales, differencing, detrending, försäljning, differentiering, transformation, forecasting, Seasonal models
- language
- Swedish
- id
- 1612874
- date added to LUP
- 2010-06-15 08:51:15
- date last changed
- 2010-06-15 08:51:15
@misc{1612874,
abstract = {{The purpose of this thesis is to display our results received from the assignment GfK addressed to us. GfK is an independent part who aid companies when they need statistical assistance. In order to forecast sales for one of GfK’s customers we have build models with time series analysis. We received 26 products with monthly data from 2005 to 2008. In this thesis we have narrowed it down to six of these products, which we find representative in the range of methods we used and the problems we encountered.
We have been using the program package of Minitab when processing the time series and building the models. The method used is for the most part the Box-Jenkins method, but with some adjustments. First step in the process was to make our time series stationary and even though the method alters from product to product, there are some general theories which have been presented in chapter 2. In order to make adjustments for inconsistent variance we have been using Box-Cox transformation or the log-transformation and to process trend we have used detrending and differentiating. Furthermore in cases where outliers are present we have used dummy variables and when building models ARIMA has been our instrument of choice. With these models we finally let Minitab forecast sales for the next coming year.
In our thesis, for the products we have analysed, we find that sales are best explained with a seasonal model. Out of the six products we got two SAR(1)12, three SMA(1)12 and one combined MA(1)*SMA(1)12. The fit of our models has in most cases been good since the forecasts seem to fit well with the original time series. On the contrary we have encountered some problems during the process which has been discussed in chapter 4. First and foremost it would have been favourable if we would have had a larger amount of observations to work with, but we have also discussed alternative methods that could have improved our thesis. All in all we are satisfied with the models built from the applied theory.}},
author = {{Andersson, Freddy and Fältman, Ellinor}},
language = {{swe}},
note = {{Student Paper}},
title = {{Anpassning av ARIMA-modeller till försäljningsdata}},
year = {{2010}},
}