Dear present or future experts,
Table 3.6 in [1] contains minimum data requirements for different
forecasting techniques, including six those designed for seasonal time
series. I think some of you may also be of a mind to conclude that the
table is rather weird. It's not clear from the text if the table is
based on any consistent approach to systematize the information across
many methods. For example, it is not clear why SMA can be used with
minimum 30 observations and SES with only 2.
In particular, it is not really clear if there is a solid
theoretical/practical research behind their recommendation to use the
following six methods with the following minimum amount of data:
1) seasonal exponential smoothing 2 x s
2) adaptive filtering 5 x s
3) classical decomposition 5 x s
4) Census X-12 6 x s
5) Box-Jenkins 3 x s
6) Time series multiple regression 6 x s
s - length of seasonality
The authors write that the table is a summary (generalized
information) that can be used as a starting point when deciding which
forecasting technique is to be used with such and such amount of data.
However, they do not refer to particular works of their own or others
to help an interested reader analyze the table content deeper.
Further, within the same book they write (chapter 9) that ARIMA models
require 6-10 years of data as minimum, depending on the size of s.
Source [2] reports that for ARIMA models different econometric
software require information no less than for five to six full
seasonal cycles. Finally, I read it recently from a very reliable
source of statistical knowledge that it is sometimes possible to fit
some seasonal ARIMA models to very short data sets (only one year!).
These three do not seem to help explain the line: 5) Box-Jenkins 3 x
s. Consequently, I doubt if I can use the above table as a good
reference in my research. Am I wrong?
My ability to use a wider range of sources is limited at the moment. I
hope a brief discussion of the information given in the table on
minimum data requirements for the above forecasting techniques will
slake my curiosity. Thanks. Any ideas appreciated. Any relevant
references welcome. Any sharing of personal experience may help. Any
further doubts regarding their table may help end in our certainties.
References:
[1] Business Forecasting by Hanke, Reitsch and Wichern, 7th ed., p.73
[2] Statistical Forecasting Methods by Dubrova, p.161 (in Russian)
---
Best wishes,
Andrey Kostenko
---
Quote of the day:
If a man will begin with certainties, he shall end in doubts, but if
he will content to begin with doubts, he shall end in certainties.
Francis Bacon
|