在R中使用ARIMA和xreg进行预测

Question

I'm having trouble with my first forecasting implementation in R. What I'd like to achieve is to predict the variable Y with 2 exogenous variables X1 and X2. The 3 datasets are each represented as a single column with 12 rows.

From another Stackpost I followed a similar approach:

DataSample <- data.frame(Y=Y[,1],Month=rep(1:12,1),
                     X1=X1[,1],X2=X2[,1])

predictor_matrix <- cbind(Month=model.matrix(~as.factor(DataSample$Month)), 
                          X1=DataSample$X1,
                          X2=DataSample$X2)
# Remove intercept
predictor_matrix <- predictor_matrix[,-1]

# Rename columns
colnames(predictor_matrix) <- c("January","February","March","April","May","June","July","August","September","October","November","X1","X2")

# Variable to be modeled
var <- ts(DataSample$Y, frequency=12)

#Find ARIMA
modArima <- auto.arima(var, xreg = predictor_matrix)

At this line I get the following error:

Error in optim(init[mask], armaCSS, method = optim.method, hessian = FALSE, : non-finite value supplied by optim

I presume that my predictor_matrix is not in the correct format but I can't find the error.

Any help would be appreciated,

Answer 1

You have indicated "datasets are ... 12 rows". Your predictor matrix has 13 columns (11 months [of dummy variables?] and 2 other variables). Therefore, you necessarily have a linear dependence among the columns and the optimization procedure fails.

You need (ideally much) more data to support the number of predictor variables and/or a sparser set of predictors.