`describe` for time series with basic checks ?

[josef...@gmail.com]

just an idea for some semi-automatic tools for specification search,
similar to auto_xxx features for forecasting:


econometrics textbooks or lecture notes often start with the
recommendation that we need to do some basic checks first, e.g. check
for stationarity.
If the model is misspecified, then users often run into convergence
problems in the estimation, e.g. ARMA imposes stationarity
restrictions by default.


I'm thinking of some basic statistics for a rough initial
specification search, e.g.
- trend t-value for a trend regression
- seasonality peaks in spectral density or autocorrelation function
- non-linearity, e.g. variance stabilizing transforms as in Box-Cox PR
- unit roots (adfuller, ...)
- possible outliers ?

and more for the multivariate case
?

related aside: for automatic forecasting those checks are all put
together inside a specification search part, e.g. by Hyndman

Complementary to this, we still have post-estimation diagnostic tools,
hypothesis test and plots for various specification assumptions.


suggestion, volunteers?

Josef