[R-SIG-Finance] Appropriate model to data?

[Andreas Klein]

Dear R Users.

In the attachment I have a *.txt file with two datasets:
- present_data (#246)
- future_data (#39)

I tried to model the present_data to simulate the future_data (scenario-simulation with Monte Carlo based on 100.000 cases), so the forecast will be the 99% and the 99.9% quantile of all the simulations based on the model of the present_data. My aim is to reach the peak at the end of the future_data.

The different unit root tests give hints for stationary, but some also showed a unit root. I worked with both.

- I fitted an AR(1) model.

- I fitted an ARIMA(1,1,1) model, but I couldn't verify the model and the coefficients with bootstrap. The model was unstable while resampling with replacement, which is an indicator for overdifferencing of the series in my oppinion, so the series has no unit root.

- I fitted a threshold autoregressive model with 2 regimes and a threshold variable of 0.02259983 and the order of the lower and upper regime of 1.

- I tried to model a GARCH(1,1) model, but I got partially insignificant coefficients.

In all four cases I was not able to reach the peak of the future_data with the above described simulation type.

The innovations were always non-normal.
So as innovation distribution for simulation I used a not justified normal distribution, a more justified skewed-t-distribution and the empirical distribution obtained with bootstrapping.



I would be glad if anyone could have a look at the data and give me some hints what models I can try next, which will give me more succes in my attempt to find an appropriate model for the given present_data in reaching my aim!



Thanks in advance.



Regards,
Andreas.


      
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: data.txt
URL: <https://stat.ethz.ch/pipermail/r-sig-finance/attachments/20090224/a92b3b98/attachment.txt>