You had an oversight on the differenced KPSS test. The result of differenced inflation stationarity test using KPSS shows that inflation is stationary at first difference since the p-value is 0.1 which is 10%. Your videos are quite informative and very useful. Thanks for all you have been doing.
@@sergiolescano The null hypothesis of KPSS is that the series is stationary, which is the opposite of the null hypothesis of ADF which is that the series is non-stationary. So for KPSS a p-value > 0.05 means we fail to reject the null hypothesis and conclude that the series is stationary. He either forgot that KPSS is the opposite of ADF or he misread the p-value (or misspoke entirely).
Thanks for the nice explanation, i get confused sometimes when reporting unit root tests we take absolute value or consider the sign of value for example test that compares calculated t statistics with tabulated value example t calt= -2.56 and t tabulated= -4.63 In this do we reject null or except . Thanks
Hi! Thanks for these videos, which are very helpful. One question: on the last test KPSS for dinf, we see a high p-value supporting the null hypothesis that the series is stationary: it is then agreeing with the rest of the tests to my understanding so why do you say the test indicates that dinf is non-stationary?
He made a mistake, either he forgot KPSS has the opposite hypothesis test to the ADF, or he misread the p-value. You are correct in saying that a p-value of 0.1, at the 0.05 significant level, means we fail to reject the null hypothesis and therefore conclude that the data is stationary.
Hii can anyone of you help me why p value shows NA and dicky fuller static 'NaN' when I run adf test, I'm new to R and I've to use ARIMA model to forecast.
Mmm one question, does the number of lags to be chosen in the ADF test depend on the data frequency? For example if I have monthly data could I use 12 lags or something like that?
There is no specific rule about the number of lags your should choose, but there are recommended bounds on the number of lags. If you are using data that you believe has significant seasonality then you should use no more lags than the number of lags in your seasonal-period. In your case, since your data is monthly and I assume your seasonal-period is a year, you should include no more than 12 lags in your ADF. From this point there is no concrete rule about how many lags to include in the test. There are multiple competing rules: Schwert (1989), for example, recommends a max. ceiling of 12 · (n /100 )^(1/4) for number of lags, where n denotes your data sample size. Whereas the rule used by the adf(.) function in R by default (if you don't specify the number of lags) is the largest integer less than 4 · (n/100 )^(2/9). Combine this information with the seasonality rule and you might settle on a number. With time series modelling in general, eyeballing the graph of your data/differenced-data is a good way to judge whether or not you're on the right track with stationarity - if it looks quite stationary and you have a decent sample size of data points, then it probably is stationary (i.e. the data was generated by a stationary process).
what if the results of the three test are contradictory??? both adf and kpss shows non stationarity at level however, the pp test shows stationarity. i tried differencing my data and then the adf and pp shows stationarity but my kpss shows non stationarity. it's very confusing
Probably you are making a mistake somewhere in your interpretation. The ADF and PP goes in the same direction in their hypothesis but the KPSS is a reverse of this. ADF and PP are unit root tests with the aim of testing for non-statationarity of data while KPSS's null hypothesis is that series is stationary.