I need to model an ARIMA with a time-series data. But my data is the statistics of land area, and it's annual data, so I have 64 points between 1950~2014. Because it increased by a stable rate, So I think maybe it can used to build ARIMA. Since I know the time series analysis may need as many observation as possible, so I'm little concerned about my data. But in other hand, the land area wouldn't change dramatically like interest rate, price. So I think may be I have enough observations for ARIMA modelling. So can anyone give some advice?
$\begingroup$
$\endgroup$
2
-
$\begingroup$ You should cross-reference relevant questions you asked previously. See e.g. stats.stackexchange.com/questions/126577/… Here the question is seriously misleading in detail as you don't have 64 data points at all. You have data every 5 years and are interpolating to years. (1950-2014 would give 65, but that's pedantic.) $\endgroup$– Nick CoxCommented Dec 5, 2014 at 10:35
-
$\begingroup$ Yeah, it's true. But I have decided to do the data processing to generate another 51 true points and I have given up the interpolating methods at all. $\endgroup$– Pig_called_luluCommented Dec 5, 2014 at 13:54
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
4
64 observations are quite sufficient for ARIMA, since you have yearly observations (so you won't do seasonal differencing). You could also look at other standard forecasting methods, like Exponential Smoothing.
answered Dec 4, 2014 at 11:38
-
1$\begingroup$ Unless there is some reason you have to use ARIMA (e.g. for a class) I'd follow Kolassa's suggestion and try a Holt's exponential smoothing (since you say the series has a trend). Note a Holt's is equivalent to one type of ARIMA model, but is simpler and is often used with shorter time series. $\endgroup$ Commented Dec 5, 2014 at 2:57
-
1$\begingroup$ This is good advice given the (initial) wording of the question, but the question is disingenuous. See stats.stackexchange.com/questions/126577/… $\endgroup$– Nick CoxCommented Dec 5, 2014 at 10:35
-
$\begingroup$ +1 to @NickCox' comment. Fitting an ARIMA model to heavily interpolated data makes indeed very little sense. $\endgroup$ Commented Dec 5, 2014 at 11:38
-
$\begingroup$ Yeah,I'm sorry~ When I found that my sample maybe too small for a time-series modelling, my first idea is whether can I get a bigger sample by interpolating, thus it will be easier. But after checked answers in Cross Validated and asked someone else for help, I know that it is unreasonable. So at last, I decided to generate true values by spatial analysis. With 5 days spent and 6 PCs' help, now I finally have 65 true values. And I prepare to use some time-series modeling to fit the values~Thanks for your advise, I will read some materials about exponential smoothing. $\endgroup$ Commented Dec 11, 2014 at 3:28