HypothesisTests.jl
HypothesisTests.jl copied to clipboard
List of tests common to time series analysis
- [x] Augmented Dickey-Fuller (unit root)
- [ ] Phillips-Perron (unit root)
- [ ] KPSS (unit root)
- [ ] Shapiro-Wilks (normality)
- [x] Jaques-Berra (3rd and 4th moments for normality) (TimeModels has this implemented)
- [x] Ljung-Box (auto correlation)
- [x] Durbin-Watson (auto correlation)
- [x] Breusch-Godfrey (serial correlation)
I have an ADF test somewhere in my files, will post it as a gist, so you can just use it.
Ok, nice.
Have a look at this. Didn't properly backtest everything, but seems to work mostly fine.
I'm going to work on implementing these - has any work been done on this?
Not really.
@rawls238 cool, go for it!
I have started implementing autocorrelation tests in #87 and #89.
Ok, ADF test is here #92. Please feel free to comment.
@milktrader Should we move the Jarque-Berra test to HypothesisTests as well? It seems a bit lost in TimeModels if all other tests are here.
@BenjaminBorn yes, I think so. TimeModels has not been developed for some time now.
TimeModels has not been developed for some time now.
In particular, I don't think it even works on Julia versions later than 0.4. If it won't be maintained going forward, perhaps it should be deprecated and the relevant pieces moved to other packages as appropriate? @BenjaminBorn has already been doing excellent work implementing the tests in HypothesisTests; it would just be a matter of finding appropriate homes for the other things.
I agree @ararslan. I could move the package back to my repos and let it sit there until I have resources to develop it.
I have added the Jarque-Bera test to HypothesisTests (see #95).
Concerning TimeModels, it seems it consists of two main parts. ARIMA modelling based on a state-space approach and some GARCH functionally (transferred from GARCH.jl). Would it make sense to split it again? There are also ARIMA.jl and RARIMA.jl (an R wrapper) but they also seem to be dormant.
Yes, I think a big reason for dormant activity is because it's become too unwieldy.
Durbin-Watson test added in #102.
@pedrochaim has UnitRoots.jl with DFGLS & KPSS.
@tomaskrehlik you have some nice code for the Johansen test for cointegration. Would it be possible to port it here?
ARCHModels.jl has several time series tests.
@tomaskrehlik you have some nice code for the Johansen test for cointegration. Would it be possible to port it here?
I haven't really done any Julia in a while but I will put it on my agenda and see when it fits in. :)