Panel A. Monthly Data Observations |
|---|
PMLFt | IMA,tMRIt | (1-IMA,t)MRIt | IMA,t+1MRIt+1 | (1-IMA,t+1)MRIt+1 |
0.136*** | | | | |
0.105*** | -0.178*** | -0.174** | | |
0.130*** | -1.071*** | -1.147*** | 0.981*** | 1.063*** |
Panel B. Quarterly Data Observations |
PMLFt | IMA,tMRIt | (1-IMA,t)MRIt | IMA,t+1MRIt+1 | (1-IMA,t+1)MRIt+1 |
0.292*** | | | | |
0.259*** | -0.007 | -0.008 | | |
0.396*** | -0.059*** | -0.073*** | 0.077*** | 0.106*** |
- Note. Our benchmark model is PMGFt+1=C+PMLtF+εt+1, where PMGF and PMLF are filtered observations. Filtered observations are used to alleviate the contamination of autocorrelations in PMG and PML.Regression with mean reversion indicator controlled is presented as follows,
-
\( {\displaystyle \begin{array}{l}{{PMG^F}_t}_{+1}=\kern0.5em \mathrm{C}\kern0.5em +\kern0.5em \upalpha {PML_t}^F+\kern0.5em {\upbeta}_1{\mathrm{I}}_{\mathrm{MA},\mathrm{t}}{\mathrm{MRI}}_{\mathrm{t}}+{\upbeta}_2\left(1\hbox{-} {\mathrm{I}}_{\mathrm{MA},\mathrm{t}}{\mathrm{MRI}}_{\mathrm{t}}\right)+{\upvarepsilon}_{\mathrm{t}+1,}\\ {}{{PMG_t}^F}_{+1}=C+\upalpha {PML_t}^F+{\upbeta}_1{\mathrm{I}}_{\mathrm{MA},\mathrm{t}}{\mathrm{MRI}}_{\mathrm{t}}+{\upbeta}_2\left(1\hbox{-} {\mathrm{I}}_{\mathrm{MA},\mathrm{t}}{\mathrm{MRI}}_{\mathrm{t}}\right)+{\upbeta}_3{\mathrm{I}}_{\mathrm{MA},\mathrm{t}+1}{\mathrm{MRI}}_{\mathrm{t}+1}+{\upbeta}_4\left(1\hbox{-} {\mathrm{I}}_{\mathrm{MA},\mathrm{t}+1}{\mathrm{MRI}}_{\mathrm{t}+1}\right)+{\upvarepsilon}_{\mathrm{t}+1.},\end{array}} \)
- This regression follows Huang et al. (2015) who use the following state-dependent regression to predict stock returns, rt+1
- rt + 1 = C + β1IMA, tMRIt + β2(1 ‐ IMA, tMRIt) + εt + 1
- The constant C is not reported in the table for space-saving. ***, **, * mean respectively significance at the level of 1%, 5% and 10%
