change time makes factor's ic drops a lot

[ukifune99]

❓ Questions and Help

I change the train/valid/test time of conf_baseline/conf_combined_factors_sota_model/conf_combined_factors.yaml to
train: [2017-01-01, 2021-12-31]
valid: [2022-01-01, 2023-12-31]
test: [2024-01-01, 2025-11-30]
in /qlib/experiment/factor_template and /qlib/experiment/model_template

then I run rdagent fin_factor, the ic of each factor is lower than 0.01, which is much lower than the result before I change the time

[SunsetWolf]

Hi, @ukifune99 , Hi, the problem you mention is as expected.

By shifting the train/valid/test windows to 2017–2025, the experiment is effectively evaluated in a very different market regime compared to the original configuration (2008–2016).

In recent years, especially post-2020, single-factor ICs below 0.01 are common and normal.