M/Ph dZddlZddlmZddlmZGddZedkrdej gdZ e e Z ej e Zej d e d zd ej e zz d eje zzZd eje d zzZej e ed dzd edd zzd eje zzZed ddedd zzZeje Ze dddfekeZeje dkZeje dkZeejeej gdzZeejeej gdzZeeeejeed Zeed deedd zz ed Zeed deedd zz dzeej e Z!eZ!eee!e Z"e"#\Z$Z%ee$de%j&dSdS)a'Paneldata model with fixed effect (constants) and AR(1) errors checking fast evaluation of groupar1filter quickly written to try out grouparfilter without python loops maybe the example has MA(1) not AR(1) errors, I'm not sure and changed this. results look good, I'm also differencing the dummy variable (constants) ??? e.g. nobs = 35 true 0.6, 10, 20, 30 (alpha, mean_0, mean_1, mean_2) estimate 0.369453125 [ 10.14646929 19.87135086 30.12706505] Currently minimizes ssr but could switch to minimize llf, i.e. conditional MLE. This should correspond to iterative FGLS, where data are AR(1) transformed similar to GLSAR ? Result statistic from GLS return by OLS on transformed data should be asymptotically correct (check) Could be extended to AR(p) errors, but then requires panel with larger T N)optimize)OLSc(eZdZddZdZdZdZdS)PanelAR1Nc|jd}||_|||_tj|dk|_|j|_dS)Nr)shapeendogexognpdiff groups_start groups_valid)selfr r groupsnobss g/var/www/html/test/jupyter/venv/lib/python3.11/site-packages/statsmodels/sandbox/regression/ar_panel.py__init__zPanelAR1.__init__sJ{1~  DIWV__a/!..cH|dd||ddzz |jS)N)r)rxyalphas r ar1filterzPanelAR1.ar1filter)s)122CRC($*;<rJs*.333333!!!!!!!!F zRX((())F 3v;;D BIdOOE 29QtAv  YRYt__!4 4s29??4;P;P7P PD biood1f%% %C 29T??SW $s3ss8| 3c")//$:O:O6O OD qrr7SSbS\ !D")F##K!!!D&>[088==KBGFOOQ&LBGFOOQ&L vrvk828LLL#9#9:::A  XRXlll%;%;<<