I have only 13 year observations in my linear time series. i can not apply for ARDL that requires at least 30 observations. Some advice me to use boostratpping ARDL. but the booststrapping ARDL is to cater and estimate confidence intervals for long-run multipliers which are non-linear functions of the ARDL coefficients. It is not to cater for small samples!
Any help?
Non-parametric stat tests.
Or extreme groups-design.
Hope this helps,
Matt
It is clear that you need to consider another technique as no time series analysis will stabilize with that few observations. You might consider a regression analysis. It will not offer the precision of a time series, but it can be done with fewer observations and it is more precise than nonparametrics.
BOOTSTRAPPING is normally used in larger sample size where series of subsets are taken from the larger sample to use as an estimate for the whole sample. In this case, you start with 13 elements, bootstrapping is not appropriate. In order to bootstrap, you might need to generate a set of simulated data and then apply bootstrap.
MONTE CARLO SIMULATION may be a better choice. Since you have a small sample, try to use the range (max - min) as the boundaries of your observation set and then subject to Monte Carlo simulation by: (i) determine the N repetition in Monte Carlo, (ii) use excel to generate simulated data equal to that N repetitions, and (iii) now use the newly generated data set (simulated data) to run your ARDL. You should not gave any problem with sample size now since your N may be as large as 20,000 simulated elements from which you can also tell excel to randomly select 30, 40, or 50 subsets.
CAVEAT: simulation is simulation, not real data. Here you start with 13 elements, by using Monte Carlo tool, you may generate over thousands of simulated random elements within the range. Whatever structure produce is artificially produce by the original 13 elements. See experiment of unit circle to general pi (3.14).
LINK:
https://en.wikipedia.org/wiki/Monte_Carlo_method
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series