Does anyone know if it is possible to use regressors with the TBATS model in the R forecast package?
From the help file it looks like you can add additional parameters for ARMA and since ARIMA takes regressors (i.e xreg = X ) I was wondering if I could pass regressors to TBATS if I use ARMA (using.arma.erros=TRUE) in my TBATS model.
Im still a pretty new R use so I would really appreciate any guidance.
No, that is not possible. The xreg argument will be used when selecting the order of the ARMA model for the errors, but then it will be ignored when doing the estimation and forecasting.
I think TBATS does not accept regressors but only one regressor and that is the time which may be days, hours, years etc
Related
Suppose I fit AR(p) model using R arima function from stats package. I fit it using a sample x_1,...,x_n. In theory, when predicting x_{n+1} using this model, it needs an access x_n,...x_{n-p}.
How does the model know which observation I want to predict? What if I wanted to actually predict x_n based on x_{n-1},...,x_{n-p-1} and how my code would differ in this case? Can I make in-sample forecasts, similar to Python's functionality?
If my questions imply that I think about forecasting in a wrong way, please kindly correct my understanding of the subject.
I have a mydata.ts which is around 200 rows. I used stationary tests, took differences and examined ACF and PACF. So I decided to try ARIMA(1,1,1)(0,1,1) for instance.
Which R function should I use to find fitted values and forecasts? Arima, arima or auto.arima?
And can I trust the MAPE, MAD and other error results on summary(model)? Because I read an answer and it was saying the results are not the real but approximated or something.
auto.arima will find the whole model specification that is the 'best' based on AIC, BIC.
IF you know the order (1,1,1) or (0,1,1) then use Arima from forecast package(same as arima, but little more general)
Arima(your_data, order=c(1,1,1)) will give the basic answer.
Seee the documentation for forecast.
then actual out-of-sample forecast can be done with the forecast function.
I am trying to fit a transfer function model using R in order to apply the fitted model to a validation set of data, because SPSS doesn't allow me to (or I don't know how to) compute point forecasts just like the function Arima() from forecast package does. It does let me apply the model, but it does not use the dependet variable's lagged values, that's why I am trying R.
Anyone know how I could get those type of "updated" or validation forecasts using the arimax() function? I am not looking for the following type of predictions:
predict(vixari011, n.ahead=12)
But rather these:
Arima(test$VIX, model = vixari)
From what I have been reading there is no prediction function for the arimax() function, any ideas about how I could forecast to evaluate point-by-point performance? I can just think of computing manually using a spreadsheet...
I had the same problem. I know this post is old but this can help someone.
I used this it worked just fine
forecast(fitted(arimax_ts_model), h=11)
I'm trying to find whether there is any package in R that deals with multiple seasonality for ARIMA models and, if not, if there is any way of going through it.
I have an hourly series and would like to test seasonality for lags=c(24,7*24,30*24)
Many thanks in advance.
You can use TBATS model, and her's an example
Does anyone here know how I can specify additional external variables to an ARIMA model ?
In my case I am trying to make a volatility model and I would like to add the squared returns to model an ARCH.
The reason I am not using GARCH models, is that I am only interested in the volatility forecasts and the GARCH models present their errors on their returns which is not the subject of my study.
I would like to add an external variable and see the R^2 and p-values to see if the coefficient is statistically significant.
I know that this is a very old question but for people like me who were wondering this you need to use cbind with xreg.
For Example:
Arima(X,order=c(3,1,3),xreg = cbind(ts1,ts2,ts3))
Each external time series should be the same length as the original.