论文实证参考

Methods

Baseline

Moderating effects 调节效应（M变量不同条件下，XY关系是否发生改变）

ref2:

• firm size

• ownership structure: SOEs versus non-SOEs

• corporate competitiveness

• geographical location

• Subsample
  We use a pooled sample median as the demarcation point rather than the yearly/monthly/etc median.

We use a pooled sample median as the demarcation point rather than the yearly median, because time trends may bias the classification of strong and weak EG. For example, IOC has been increasing steadily over time: the mean IOC is 18% in 1992, 24% in 1999, and 30% in 2006. Thus, classifying 2006 firm observations into high and low IOC by the 2006 median may wrongly classify a firm-year into low IOC when it has a high IOC relative to the entire sample.
For ICR, we use the median of all firm-year observations rather than the median industry to obtain more balanced subsamples between high and low ICR. Because there are more firms in low concentration industries, using the median industry would yield very unbalanced subsamples.
Ref: Kim, E.H., Lu, Y., 2011. CEO ownership, external governance, and risk-taking. Journal of Financial Economics 102, 272–292. https://doi.org/10.1016/j.jfineco.2011.07.002

Mediating effect 中介效应

Robustness test

ref1

IV estimations

DID

other

ref1

Alternative dependent variable

Alternative independent variable

Control new variable

Consider cross-sectional dependence

reestimates with Driscoll-Kraay standard errors (Hoechle, 2007)

Sys-GMM

To further mitigate the endogeneity issues arising from reverse causality, we employ the method of SYS-GMM (Dong et al., 2024).

ref:
[1] Ding, T., Li, H., Liu, L., Feng, K., 2024. An inquiry into the nexus between artificial intelligence and energy poverty in the light of global evidence. Energy Economics 136, 107748. https://doi.org/10.1016/j.eneco.2024.107748 【code & data】
[2] https://onlinelibrary.wiley.com/doi/10.1111/acfi.12674 【U型曲线】