R: why AICs equal to inf in arima models with non-zero mean in forecast v6.1 while it has a value in forecsat v5.8

Question

I had installed the package forecast v5.8 and after a while I updated it to forecast v6.1. I applied auto.arima() function on my hourly data. But I have inf in arima models with non-zero mean in forecast v6.1 while it has a value in forecsat v5.8.

why did that occur? I will put just some models of result

forecast v5.8

>auto.arima(ep_m,approximation = F,stepwise = F,trace = T)
  ARIMA(0,0,0)(0,0,1)[24] with non-zero mean : 3693.726
  ARIMA(0,0,2)(2,0,0)[24] with non-zero mean : 1712.235
  ARIMA(0,0,0)(0,0,2)[24] with non-zero mean : 3292.21
  ARIMA(1,0,0)(1,0,0)[24] with non-zero mean : 1450.052

  Series: ep_m 
  ARIMA(2,0,2)(1,0,0)[24] with non-zero mean 

  Coefficients:
           ar1      ar2      ma1     ma2    sar1  intercept
        1.8472  -0.9088  -0.7574  0.0879  0.5968    29.5869
  s.e.  0.0219   0.0217   0.0479  0.0391  0.0473     0.2868

  sigma^2 estimated as 0.3774:  log likelihood=-700.47
  AIC=1414.94   AICc=1415.1   BIC=1447.23

forecast v6.1

 >auto.arima(ep_m,approximation = F,stepwise = F,trace = T)
  ARIMA(0,0,0)(0,0,1)[24] with non-zero mean : inf
  ARIMA(0,0,2)(2,0,0)[24] with non-zero mean : inf
  ARIMA(0,0,0)(0,0,2)[24] with zero mean     : Inf
  ARIMA(1,0,0)(1,0,0)[24] with non-zero mean : Inf

  Series: ep_m 
  ARIMA(1,0,2)(2,0,0)[24] with non-zero mean 

  Coefficients:
           ar1      ma1     ma2    sar1    sar2  intercept
        0.9058  -0.0197  0.0924  0.4766  0.4571    29.2759
  s.e.  0.0178   0.0422  0.0396  0.0341  0.0343     2.2677

  sigma^2 estimated as 0.3087:  log likelihood=-642.71
  AIC=1299.41   AICc=1299.57   BIC=1331.7

that's giving me a different best ARIMA.

Answer 1

See the Changelog for changes in successive versions of the forecast package.

The tests for near-unit-roots were made stricter in v6.0, which meant that an Inf AIC was then returned for some models which previously were considered ok.