HP Filter array size problems in Matlab 2023a

I am running the HP filter using the function built into Matlab for a vector with 100.000 rows.

Until last week, I was using Matlab 2021a on a computer with 8GB of RAM and the HP filtering worked flawlessly.

Since today, I'm running the same code using Matlab 2023a on a computer with 32GB of RAM and the HP filter runs into the error:

"Error using \

Requested 100002x100000 (74.5GB) array exceeds maximum array size preference (31.7GB). This might cause MATLAB to become unresponsive."

How is this possible and any idea what I can do about it?

[Mario Malic]

First thing you check is to compare the function changes between those versions. https://uk.mathworks.com/help/doc-archives.html

Answer 1

[Star Strider]

It would help to have the data and the code.

There are likely version differences, specifically that the arguments to the mldivide, \ functions in R2023a are a row vector and a column vector, where in R2021a, they were both (most likely) column vectors. (Without seeing the code, I cannot determine the reason for this change.) Probably the best approach is to manually transpose one of the vectors so that they are both the appropriate dimensions.

[Moritz Scheidenberger]

I see, I guess you are right.

This line of code replicates the issue.

[~,ufiltered] = hpfilter(ones(100000,1));

It's a shame because even when using a shorter series, the HP filter has become much more computationally expensive now.

Before, it was a matter of a second for a series with 100.000 periods.

[Star Strider]

Thank you!

The documentation for Y states ‘Each column of Y corresponds to a variable, and each row corresponds to an observation.’ so it wants a column vector (or data organised as column vectors), not a row vector.

According to Version History there were changes in R2022a and R2023a that may have affected the way the function operates. (I do not have the Econometrics Toolbox, so I have no experience with the hpfilter function.)

I thought you were referring to a highpass filter, such as would be produced by the Signal Processing Toolbox.

If my Answer helped you solve your problem, please Accept it!

.

[Moritz Scheidenberger]

I see, I copied the wrong line of code.

The problem is the same for a column vector.

[~,ufiltered] = hpfilter(ones(1,100000));

Nevertheless, I guess your answer solved it since it must come down to a version change. Disapponting to see such a big hit to performance of the hpfilter function.

Thank you for the quick responses.

[Star Strider]

As always, my pleasure!