Time series

I have time series 374 439 441 396 372 422 428 391 375 450 541 552 648 730 759 739 826 1156 1321 1364 1345 1348 1215 1260 1216 1564 1868 1776 1845 2155 2150 2047 2169 3484 3529 1368 965 1038 1404 1678 1652 2068 2055 1944 2216 2417 2055

I don't exactly know it is stationary or nonstationary time series because different tests in matlab give me different results. It's necessary for me in order to understand, should I integrate time series in order to make it stationary or not.

I'll be grateful if you could help me.

Answer 1

[tom reilly]

Natasha,

Not every nonstationary time series should be handled with integration. For example, this example would be better suited to use a DETERMINSTIC level shift variable. See more on this by searching reviewing Nathan Balke's paper on level shifts. http://econpapers.repec.org/article/besjnlbes/v_3a11_3ay_3a1993_3ai_3a1_3ap_3a81-92.htm

It would make a lot of sense to have 6 dummy variables. Two one-time pulses, one level shift and three changes in seasonality (ie seasonal pulse) where a seasonal dummy would start at 24, 25 and 26.

Y(T) = 597.89 +[X1(T)][(+ 1240.7 )] :LEVEL SHIFT 20 +[X2(T)][(+ 1690.4 )] :PULSE 35 +[X3(T)][(+ 1645.4 )] :PULSE 34 +[X4(T)][(- 748.05 )] :SEASONAL PULSE 25 +[X5(T)][(- 537.55 )] :SEASONAL PULSE 26 +[X6(T)][(- 524.55 )] :SEASONAL PULSE 24 + + [A(T)]