246 Second, we use ordered logit regressions with Rating Discrepancy as the dr aetpi ne gn doef ndti fvf ea rr ei anbt l Ce Rt oA si n(vi .ees. ,t ri gaat itne gi fdtihs cerreepias nac yd)i .s Ta go roe be smeer nv et br aettiwn ge edni stchree pc ar endc yi t, wB ye eolni ml y iinnactliundgeatlrlasni nc hgel es-trhaat et dh at vr aenrcehc ee si v(e2d9a2t lter aa sntcthwe os )c, rwe de i ot br at at iinngos uart i s s u a n c e . multiplerated subsample that consists of 1,865 tranches. Our key independent variable is the categorical variable Risk Retention Methods. Consistent with our third model, we control for several security-design characteristics (No. of Tranches, Log Tranche Value, Log Transaction Value and Subordination), Frequent Originator, Credit Rating, Year and Security Type. The specification of our fourth model is:
RkJQdWJsaXNoZXIy MTk4NDMw