How to return forecasts from a TSLM model with Fourier Seasonality in R?

I have the below time series for weekly fish caught in a specific location (period=52 for weekly data) and only 55 datapoints

Time Series:
Start = c(1, 1)
End = c(2, 3)
Frequency = 52
 [1]  773 1239  842  567  686  930 1165  952 1277  820  364  343  342  444
[15]  432  503  463  372  372  367  423  378  423  459  350  399  433  439
[29]  382  331  326  345  497  579  381  306  423  403  549  412  354  471
[43]  435  420  410  455  534 1064  816  485  744 2260 1542 1988 1233

I have been following Rob J Hyndman's method for using Fourier transformation for the seasonality https://robjhyndman.com/hyndsight/tslm-decomposition/ aspect of the TSLM code, and grid searched the correct K between 1:26 and the minimum AIC is at 26.

So, essentially, I have the below code for the TSLM:

decompose_df <- tslm(fish_ts ~ trend + fourier(fish_ts,26))

and then trying to use the forecast function to get forecasts for the next 10 periods:

fish_fcst <- forecast(decompose_df, newdata=data.frame(fourier(fish_ts,26,10)))

But I get this warning: Warning: 'newdata' had 10 rows but variables have 55 rows and the forecast returns 55 values that are slightly different than the input ts object, but seem to follow a similar trend, instead of 10 forecasted periods. Can anyone tell me how to get forecasts or what I am doing wrong?

editing with the dput, sorry!

structure(c(773, 1239, 842, 567, 686, 930, 1165, 952, 1277, 820, 
364, 343, 342, 444, 432, 503, 463, 372, 372, 367, 423, 378, 423, 
459, 350, 399, 433, 439, 382, 331, 326, 345, 497, 579, 381, 306, 
423, 403, 549, 412, 354, 471, 435, 420, 410, 455, 534, 1064, 
816, 485, 744, 2260, 1542, 1988, 1233), .Tsp = c(1, 2.03846153846154, 
52), class = "ts")

Thanks in advance!

Solution

Your code works for me using v8.15 of the forecast package. So perhaps you are using an old version of package -- there were some issues with matching regression variable names a few versions ago.

In any case, the model makes no sense. You have 55 observations, yet your model has 53 degrees of freedom. Perhaps you are misunderstanding the AIC values. They are on a scale from -∞ to ∞, and you want the one closest to -∞, not the one closest to zero. I would expect a value of K less than 5 with so few observations.