使用R更新ARIMA预测

Question

I would like to know how can I do a Updating ARIMA forecast on R. I want to forecast a monthly time series with a lead time equal to 1 month. for example, if my last observation is for february and I want forecast for March, April and May. after a first step, the value for March becomes available and I would like to use this new data with previous input data to forecast April and May. at the second step, I want to use my original input data and the estimated value of March and April to forecast May. etc. but my original input data is equal to 600 month and I want to forecast 300 months. Please, Could you help me?

ts.month <- ts(month, start=c(1970,10), frequency=12)
tsf.month <- diff(ts.month)
tss.month <- diff(tsf.month)


T <- length (tss.month)     # observed data from 1970 to 2000
index <- 1:(T-1)
res <- forecast(arima(tss.month[index], c(12, 0, 1)), h=1,
                level=c(80,95), fan=FALSE, xreg=NULL,  
                bootstrap=FALSE, npaths=5000)
ts.res <- ts(res$mean, start=c(1991,10), frequency=12)

I tried this following method for the first and second month, but of course I can't use it for 300 months, and the result is wrong maybe cause of the cbind.

T <- length (tss.month)
index <- 1:(T-1)
res1 <- forecast(arima(tss.month[index], c(12, 0, 1)), h=1,
            level=c(80,95), fan=FALSE, xreg=NULL,  
            bootstrap=FALSE, npaths=5000)

ts.res <- ts(res1$mean, start=c(1991,10), frequency=12)
ttt <- ts.intersect(tss.month, ts.res)
ts.tt <- ts(ttt, start=c(1960,12), frequency=12)


T <- length (ts.tt)
index <- 1:(T-1)
res2 <- forecast(arima(ts.tt [index], c(12, 0, 1)), h=1,
             level=c(80,95), fan=FALSE, xreg=NULL,  
             bootstrap=FALSE, npaths=5000)
ts.res2 <- ts(res2$mean, start=c(1991,11), frequency=12)
ttt2 <- rbind(tss.month, ts.res2)
ts.tt2 <- ts(ttt2, start=c(1960,12), frequency=12)

Answer 1

You can append the value through rbind and re-run the ARIMA and forecast. My example:

tail(data)
             total      dates
2015-06-02 37071.08 2015-06-02
2015-06-03 36722.21 2015-06-03
2015-06-04 28785.25 2015-06-04
2015-06-05 24800.79 2015-06-05
2015-06-06 13361.47 2015-06-06
2015-06-07 14050.93 2015-06-07
fit<-auto.arima(data$total,ic=c("aicc","aic", "bic"),allowdrift=TRUE)
forecast(fit,h=1)
Point Forecast    Lo 80    Hi 80    Lo 95    Hi 95
1883 28960.05 25876.42 32043.69 24244.04 33676.07

To add the new ACTUAL data point when you receive it you could:

toAdd<-data.frame(total=34455.21,dates="")
new.data<-rbind(data,toAdd)

This data set had strong seasonality (ARIMA(3,1,3).

fit<-auto.arima(new.data$total,ic=c("aicc","aic", "bic"),allowdrift=TRUE)
forecast(fit,h=1)
Point Forecast    Lo 80    Hi 80    Lo 95    Hi 95
1884       28684.03 25597.14 31770.92 23963.04 33405.02