OpenSourceAP.DownloadR 0.2.0
- Data are now read from the OpenAP Hugging Face dataset (https://huggingface.co/datasets/OpenSourceAP/CrossSection)
instead of Google Drive. Files are parquet, read with nanoparquet.
Google Drive scraping is removed, and with it the rvest, stringr,
data.table, lubridate, withr and magrittr dependencies.
- Release ids are unchanged;
"2023" and
"2022" are the 2023.08 and 2022.03 releases.
2026_10 (Data Release 2026.10, data through December
2025) is added and is the default.- New releases:
2020_05, 2020_07,
2021_03, 2021_04 and 2022_04.
Each release’s datasets come from its own file listing, so older
releases offer only what they contain (e.g. no daily portfolios before
2021.04).
- New portfolio datasets:
placebo_op,
vw_force, ff93, holdper1,
holdper3, holdper6, holdper12 and
the daily portfolios (daily_op_ew,
daily_op_vw, daily_deciles_ew,
daily_deciles_vw, daily_quintiles_ew,
daily_quintiles_vw), where the release has them.
dl_signal() downloads only the requested columns of the
release’s wide firm characteristics file and errors on unavailable
signals.- Signing is consistent:
signed = FALSE returns all
signals, including Price, Size and STreversal, unsigned, and
signed = TRUE returns them all signed.
dl_all_signals() now defaults to
signed = TRUE, which matches what it returned before; its
signed = TRUE no longer flips signs twice.
- CRSP signals come from the CRSP CIZ monthly file
(
crsp.msf_v2), since the legacy file is no longer
updated.
dl_port() warns about unavailable predictors, and
portfolio port is character ("01", …,
"LS"), as in the release files.- Removed internal methods
get_individual_signal_url()
and apply_sign_logic().
individual_signal_id_map is kept as an empty data frame.
New fields release_id and hf_revision.
OpenSourceAP.DownloadR 0.1.0