Thetaf creates extreme unrealistic high forecasts

Over the time I've seen some rare cases, where the thetaf forecast creates extreme unrealistic big forecasts. Since these cases are rare and some days ago I found one of them, I thought it would be good to show it to you.

Here is a MWE:

raw <- c(
  203.18450, 0.00050, 0.00050, 0.00050, 0.00050, 0.00050, 0.00050,
  0.00050, 0.00050, 0.00050, 0.00050, 0.00050, 196.42155, 0.00050,
  0.00050, 0.00050, 0.00050, 0.00050, 0.00050, 0.00050, 0.00050,
  0.00050, 674.73050, 130.85868, 269.89313, 0.00050, 0.00050, 0.00050,
  0.00050, 0.00050, 0.00050, 81.39669, 74.31876, 0.00050, 603.28991,
  0.00050, 0.00050, 0.00050, 0.00050, 0.00050, 0.00050, 210.23732,
  0.00000, 105.11868, 301.64441, 0.00050, 513.91272, 210.23732, 0.00050,
  0.00050, 0.00050, 121.71787, 0.00050, 0.00050, 0.00050, 0.00050,
  0.00050, 0.00050, 0.00050, 0.00050, 220.24812, 0.00050, 0.00050,
  0.00050, 0.00050, 243.43418, 0.00050, 0.00050, 0.00050,
  420.47686, 0.00050
)

bad <- forecast::thetaf(ts(raw, frequency = 12), 24)

# max(raw)      ->     674.7305
# max(bad$mean) -> 8678388

This is the result on my win 10 x86_64 laptop:

      Point Forecast    Lo 80    Hi 80      Lo 95    Hi 95
Dec 6  1090444.82330 -3201579  5382469 -5473639.8  7654529
Jan 7  6877335.56059  2585311 11169360   313250.9 13441420
Feb 7       20.12222 -4292004  4292044 -6564064.6  6564105
Mar 7       20.32050 -4292004  4292045 -6564064.4  6564105
Apr 7   591235.07224 -3700789  4883259 -5972849.7  7155320
May 7       13.25812 -4292011  4292038 -6564071.5  6564098
Jun 7   891127.10654 -3400897  5183151 -5672957.7  7455212
Jul 7       24.12397 -4292000  4292049 -6564060.7  6564109
Aug 7   793328.32517 -3498696  5085353 -5770756.6  7357413
Sep 7  1393072.46158 -2898952  5685097 -5171012.4  7957157
Oct 7       25.28653 -4291999  4292050 -6564059.7  6564110
Nov 7  7799462.42959  3507438 12091487  1235377.5 14363547
Dec 7  1227483.62699 -3064541  5519508 -5336601.4  7791569
Jan 8  7732669.08532  3440645 12024694  1168584.0 14296754
Feb 8       22.59914 -4292002  4292047 -6564062.5  6564108
Mar 8       22.79644 -4292002  4292047 -6564062.3  6564108
Apr 8   662549.57376 -3629475  4954574 -5901535.6  7226635
May 8       14.84139 -4292010  4292039 -6564070.3  6564100
Jun 8   996496.18522 -3295528  5288521 -5567589.0  7560581
Jul 8       26.94862 -4291998  4292052 -6564058.3  6564112
Aug 8   885320.53560 -3406704  5177345 -5678764.7  7449406
Sep 8  1553063.37312 -2738961  5845088 -5011021.9  8117149
Oct 8       28.16310 -4291997  4292053 -6564057.2  6564113
Nov 8  8678387.58569  4386363 12970412  2114302.2 15242473

Do you have any idea where this comes from and what one can do about it?

robjhyndman / forecast

Thetaf creates extreme unrealistic high forecasts #867