0
$\begingroup$

How can we detect trend shifts in time series? I know of Ets in R which tries to make trigonometric analysis of seasonal data. However I have not found yet a way of finding periodic trend shifts in time series. I feel the current methods cannot predict trend shifts. They can however guess the extrapolation of the current trend which is more of a localised prediction for trend. Pardon me if my question is not correct. Thanks in advance.

$\endgroup$
  • $\begingroup$ No extrapolation method can $predict$ trend shifts. Only domain knowledge and judgement can predict future trend shift. And you are incorrect in saying that ets models use trigonometric analysis of seasonal data. $\endgroup$ – forecaster Jul 16 '15 at 19:54
0
$\begingroup$

Software can/does automatically detect changes in trend be the trend a constant in a non-stationary ( differenced ) ARIMA formulation or a Transfer Function model with a trend that is handled by input series reflecting the counting numbers e.g 1,2,3,,,, and 0,0,0,0,0,1.2.3.4 . See ARIMA model selection and stochastic vs deterministic trend/seasonality in time series forecasting. Anachronistic methods like fitting time and times squared and time cubed are not generally advisable unless there is a strong prior with respect to the subject e.g. growth models .

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.