Testing Trend Stationarity in Time Series

  • Context: Graduate 
  • Thread starter Thread starter womata
  • Start date Start date
Click For Summary

Discussion Overview

The discussion revolves around the concept of trend stationarity in time series analysis, particularly in the context of cointegration methods applied to energy demand data. Participants explore the implications of ADF test results on the stationarity of the series and the validity of proceeding without detrending.

Discussion Character

  • Exploratory
  • Technical explanation
  • Debate/contested

Main Points Raised

  • One participant questions the trend-stationarity of their time series after conducting the ADF test, which indicates non-stationarity in level and stationarity in first difference.
  • Another participant notes that while ADF tests check for trend-stationarity, there are instances where visual inspection may suggest trend-stationarity that the tests fail to detect.
  • A participant mentions that finance time series are typically I(1) and co-integrated of order zero, suggesting a common scenario in financial data.
  • One participant seeks clarification on whether it is valid to proceed without detrending given their ADF test results for energy demand data.
  • Another participant argues that distinguishing between trend-stationary and difference-stationary series can be challenging, emphasizing the importance of understanding the context of the time series, such as seasonal demand variations in energy consumption.

Areas of Agreement / Disagreement

Participants express differing views on the implications of ADF test results for trend stationarity, with no consensus reached on whether the series can be treated as trend-stationary or if detrending is necessary.

Contextual Notes

Participants highlight the limitations of statistical tests in detecting trend stationarity, suggesting that contextual understanding of the time series may influence the interpretation of results.

womata
Messages
6
Reaction score
0
Hello,

I have two sets of time series that I found to be I(1), so I went ahead with using cointegration methods to find a relation between the two variables.

Now I'm questioning if the series is trend-stationary, which would mean I'd need a deterministic time trend in my cointegration. I have done the ADF test on the series and found that even when including a time trend there, I still find that the series is non-stationary in level and stationary in first difference.

Does this mean my series is not trend-stationary and that my initial approach is still valid? If what I did is wrong, how does one test for trend-stationarity?

Thank you.
 
Physics news on Phys.org
ADF and the large family of unit root tests check exactly for that, yet there are cases where trend-stationarity is obvious in a plot and yet the tests do not detect it.

Finance time series are typically I(1) and co-integrated of order zero. So if your time series have anything to do with finance that's the most likely scenario.
 
It is a series for energy demand. If the ADF test says it is non-stationary in level even if I include a trend, and that my series is I(1), is it valid to proceed without detrending since all the statistical tests don't show a time trend?
 
womata said:
It is a series for energy demand. If the ADF test says it is non-stationary in level even if I include a trend, and that my series is I(1), is it valid to proceed without detrending since all the statistical tests don't show a time trend?

Sometimes it is not easy to distinguish a trend stationary series from a difference stationary one, that is why it is always a good idea to think about what kind of time series you are dealing with, for example, in countries with cold winters there will be a higher demand in winter than summer since everyone will use energy to warm their houses, so you know that you have a trend here and you can safely ignore whatever the test say, that is, the higher demand in winter is not due to a random process.

Similarly in the stock market it's difficult to justify a trend and, unless it is a very special time series, you are better off assuming the existence of unit roots.
 
Thank you.
 
You're welcome :smile:
 

Similar threads

Undergrad Stationary time series with seasonality...
  • · Replies 8 ·
Replies
8
Views
3K
Undergrad Time series and why would we remove seasonality and trend?
  • · Replies 10 ·
Replies
10
Views
3K
Undergrad Time series analysis and data transformation
  • · Replies 8 ·
Replies
8
Views
2K
Graduate Which Dickey-Fuller test should I apply to this time series?
  • · Replies 3 ·
Replies
3
Views
2K
Graduate Reasonable length of forecast horizon in a time series
  • · Replies 2 ·
Replies
2
Views
2K
Graduate Correlation between parameters in a likelihood fit
  • · Replies 3 ·
Replies
3
Views
2K
Undergrad How do you implement the Dickey-Fuller test?
  • · Replies 4 ·
Replies
4
Views
2K
Undergrad Time Series Models vs Models Used for Cross-sectional Data
  • · Replies 14 ·
Replies
14
Views
3K
Graduate Help understanding ARIMA Model and ACF
  • · Replies 1 ·
Replies
1
Views
2K
Graduate Stationarity of Time Series: Tests
  • · Replies 1 ·
Replies
1
Views
11K