simple forecast with AR model

This function performs a forecast, h-periods-ahead, supposing the process follows an AR process. The best number p of lags is detrmined by the AIC principle, with a simplified formula. Once the best number of lags is determined, the algorithm performs a forecast, choosing an iterative o direct method.
The iterative method performs a first forecast for the next period, then uses this forecast as the last observation of the time series, and perform again a forecast using this last informtion. Simple OLS is used to find parameters of the forecast.
The direct method performs an OLS regression of the variable into its h-th lags, thus it does not uses "new" information, but the variable is regressed directly from its past values.