NEWS

hdflex 0.3.0 (2024-10-10)

Enhanced parallelization using RcppThreads for the stsc() function.
Improved (computational) performance
Added S3 class method for stsc and dsc objects: summary.stsc_obj and summary.dsc_obj for generating plots showing the evolution of the tuning parameter, as well as standard accuracy metrics such as Mean-Squared-Error, Continuous-Ranked-Probability-Score, and Predictive-Log-Likelihood-Score.
Introduction of the new argument bias for stsc() and tvc, allowing users to decide whether bias correction should be applied to the F-Signals in the TVC-models.
Addition of the new argument incl for stsc() and dsc, enabling users to specify whether certain signals are required to be included in the subsets.
Improved internal structure and performance for dsc().
Renamed the argument burn_in_tvc to burn_in and sample_length to init.
Consolidated the arguments risk_aversion, min_weight, and max_weight into portfolio_arguments.

Fixed a bug in the computation of the time-varying coefficients in the first step of the stsc() method.
The forgetting factor delta in the second step of the stsc() method now already applies to the most recent predictive likelihood score in t-1, as stated in Equation (13) in Adaemmer et al. (2023). Previously, the score in t-1 was given a weight of 1.0
Added new argument to stsc() to decide whether the subset combinations in the second step of the method should be combined with equal weights (as proposed in Adaemmer et al. (2023)) or with weights derived from the predictive log-likelihood scores.

Added the function stsc() to directly apply the STSC-algorithm from Adaemmer, Lehmann and Schuessler (2023). This function is faster and more memory efficient than subsequently applying tvc() and dsc() as it is now completely written in Rcpp.
Fixed the package overview help file.
Updated documentation
Updated example