无法估计 ARIMA 模型 R 预测包中的错误 - auto.arima

Question

I am using auto.arima from forecast package for multiple datasets in HPC server. The train data always has 24 data points with month-wise data for the past 2 years. The data also contains an external regressor. It has been working fine since long, but I encountered an error suddenly for one dataset alone.The error message says-

<simpleError in search.arima(x, d, D, max.p, max.q, max.P, max.Q, max.order,     stationary, ic, trace, approximation, xreg = xreg, offset = offset,     allowdrift = allowdrift, allowmean = allowmean, parallel = parallel,     num.cores = num.cores): No ARIMA model able to be estimated>

The code that I used is -

auto_arima <- ts(data$y, start=min(data$DATE), frequency=12)
auto_arima <- auto.arima(auto_arima, xreg = data$x1, stepwise=FALSE, approximation=FALSE, seasonal=TRUE)

The dataset that was producing the error is

structure(list(DATE = structure(c(17804, 17835, 17865, 17896, 
17927, 17955, 17986, 18016, 18047, 18077, 18108, 18139, 18169, 
18200, 18230, 18261, 18292, 18321, 18352, 18382, 18413, 18443, 
18474, 18505), class = "Date"), x1 = c(21, 22, 21, 22, 22, 19, 
22, 21, 22, 21, 22, 22, 21, 22, 21, 22, 22, 19, 22, 21, 22, 21, 
22, 22), y = c(17532306871, 41826190703, 7748403270, 14959015398, 
15241717931, 18655759009, 15393151016, 15251081090, 14516497716, 
13974303432, 11254893416, 17410813458, 14446411746, 12876827127, 
17609512917, 12179664532, 17473491954, 18447419416, 28262667160, 
36623989143, 28711285833, 28711285833, 28711285833, 28711285833
)), row.names = c(NA, 24L), class = "data.frame")

When I run this code in my windows machine with Rversion 4.0.2 installed, it is working fine (ARIMA model (0,1,0) is being built) but when running the same code with the same data in HPC server, I am facing the error described above. I am facing the same error even when I try to execute the same code with Rversion 3.6.1 installed from RStudio in another windows machine.

Is this because of R version difference or am I missing something? Please help me with this.

Answer 1

The R version probably makes no difference here. The package version is what matters more. Here you are using the forecast package, and the auto.arima algorithm has had a few improvements and bug fixes over the years. The current CRAN version of forecast is v8.12 and it returns the ARIMA(0,1,0) model.