Greetings to all gretl users and fans,
There is a new package GlobalFactors 0.1 now available.
The package deals with PC estimation of multilevel factor models, in
which the factors are not only pervasive (i.e. common to all groups) but
also semi-pervasive (i.e. common to a subset of groups only). Multilevel
factor models make it possible to distinguish between factors and
explore their hierarchical influences.
The main ideas behind the package (so far) are: (a) to consistently
estimate the number of global factors (b) Conditional on that,
consistently estimate the number(s) of local factors. Conditional on
these, consistently estimate global factors and loadings and local
factors and loadings (all these are based on Principal Components
estimates so there are no significant computational "complexities").
(Of course) there is a catch. Special care should be taken to initially
"identify" the global factors through a pairwise canonical correlation
procedure. Identification here, means to avoid choosing a pair of
countries - or more generally of groups - that strongly correlate not
due to global factors but due to regional or local or within group
However, the package is on its way for extension to include two ultra
(1) Chen (2022). Circularly Projected Common Factors for Grouped Data.
develops a consistent
selection criteria for identifying the number of the global factors
(2) Lin & Sin (2022). Generalised Canonical Correlation Estimation of
the Multilevel Factor Model, https://ssrn.com/abstract=4295429
deals (a) with consistent estimation of the number of global factors and
(b) consistent estimation of global and local factors (c) also providing
inference theory. LS(2022) do not require either the orthogonality
between the global and local factors or the selection of any tuning
parameters. As LS (2022) state "this makes the GCC criterion more
general than existing studies".
Hopefully, the fully extended/updated package will be presented at the
coming 8th Gretl Conference Gdańsk, Poland.
Please send me an email for any bugs, naïve code writing etc. for this
package version and/or ideas for extending the package.
(well hopefully there are no errors...)
Thank you for your time.