R中的时间序列预测误差

Question

I'm attempting to use the forecast function from the forecast package v4.06 on a time series object I create. Depending on the size of the vector I pass into the time series function to generate a time series object, I will get an exception thrown from the subsequent call to forecast.

For example, creating a time series object with a vector of length 6 like so :

tsObj <- ts(tsVector[1:6], frequency=12, start=c(2013,4))

and then calling the forecast function on that time series object:

pred <- forecast(tsObj, 1)

generates the following Exception:

Error in lsfit(1:maxn, y.sa[1:maxn]) : 0 responses, but only 2 variables
In addition: Warning message:
In lsfit(1:maxn, y.sa[1:maxn]) : 6 missing values deleted

In fact, vector lengths of 4 thru 11 generates this error, but lengths of 1,2,3, or >=12 succeed in generating an appropriate prediction. Any insights into the exception would be immensely appreciated. Thanks for the time!

================================================================================

I believe I found the source for lsfit and found the place where this exception is generated:

## check for compatible lengths
nrx <- NROW(x)
ncx <- NCOL(x)
nry <- NROW(y)
ncy <- NCOL(y)
nwts <- length(wt)
if(nry != nrx) stop(paste("X matrix has", nrx, "responses, Y",
   "has", nry, "responses."))
if(nry < ncx) stop(paste(nry, "responses, but only", ncx,     "variables"))

It looks like a mismatch in number of rows between the x and y vectors. However, I can't find where lsfit is called within the forecast package to figure out what "y" is in this case. Thanks for any/all help!

Answer 1

The issue is that you're setting a frequency of 12, but this isn't possible to estimate with less than the same number of tsVector inputs. So the minimum frequency you can set is the number of objects in your vector.

Both of these code blocks work for me:

tsVector <- c(1,2,3,4,5, 6)
tsObj <- ts(tsVector, frequency = 5, start = c(2013, 4))
pred <- forecast(tsObj, 1)

and

tsVector <- c(1,2,3,4,5, 6, 7, 8, 9 , 10, 11, 12)
tsObj <- ts(tsVector, frequency = 12, start = c(2013, 4))
pred <- forecast(tsObj, 1)

Hope this helps!