Public articles linked to the same research event.
arXiv DIVINE introduces a cross-market stock pretraining framework that reconstructs 77 targets derived from 16 standard technical indicators out of raw OHLCV history; after joint pretraining on six equity market datasets and transferring only the learned encoder to downstream stock ranking, it achieves the strongest average portfolio performance across all six markets, outperforming pretraining baselines and matching or exceeding substantially larger financial foundation models with a lightweight 0.05M-parameter encoder, while analyses indicate complementary gains from indicator diversity and market diversity in transfer.
DIVINE introduces a cross-market stock pretraining framework that reconstructs 77 targets derived from 16 standard technical indicators out of raw OHLCV history; after joint pretraining on six equity market datasets and transferring only the learned encoder to downstream stock ranking, it achieves the strongest average portfolio performance across all six markets, outperforming pretraining baselines and matching or exceeding substantially larger financial foundation models with a lightweight 0.05M-parameter encoder, while analyses indicate complementary gains from indicator diversity and market diversity in transfer.
DIVINE introduces a cross-market stock pretraining framework that reconstructs 77 targets derived from 16 standard technical indicators out of raw OHLCV history; after joint pretraining on six equity market datasets and transferring only the learned encoder to downstream stock ranking, it achieves the strongest average portfolio performance across all six markets, outperforming pretraining baselines and matching or exceeding substantially larger financial foundation models with a lightweight 0.05M-parameter encoder, while analyses indicate complementary gains from indicator diversity and market diversity in transfer.
DIVINE introduces a cross-market stock pretraining framework that reconstructs 77 targets derived from 16 standard technical indicators out of raw OHLCV history; after joint pretraining on six equity market datasets and transferring only the learned encoder to downstream stock ranking, it achieves the strongest average portfolio performance across all six markets, outperforming pretraining baselines and matching or exceeding substantially larger financial foundation models with a lightweight 0.05M-parameter encoder, while analyses indicate complementary gains from indicator diversity and market diversity in transfer.