VGSF - WU Vienna - LC

Alberto Rossi, Robert H. Smith School of Business

Campus WU D3.0.225 11:00 - 12:30

Organizer VGSF

As part of the ser­ies of the "Fin­ance Re­search Sem­inar", VGSF wel­comes Al­berto Rossi from the Robert H. Smith School of Busi­ness to present his re­search pa­pers.
Per­sonal Webpage

Pa­pers

Pre­dict­ing Stock Mar­ket Re­turns with Ma­chine Learn­ing

We em­ploy a semi-­para­met­ric method known as Boos­ted Re­gres­sion Trees (BRT) to fore­cast stock re­turns and volat­il­ity at the monthly fre­quency. BRT is a stat­ist­ical method that gen­er­ates fore­casts on the basis of large sets of con­di­tion­ing in­form­a­tion without im­pos­ing strong para­met­ric as­sump­tions such as lin­ear­ity or mono­ton­icity. It ap­plies soft weight­ing func­tions to the pre­dictor vari­ables and per­forms a type of model aver­aging that in­creases the sta­bil­ity of the fore­casts and there­fore pro­tects it against over­fit­ting. Our res­ults in­dic­ate that ex­pand­ing the con­di­tion­ing in­form­a­tion set res­ults in greater out-of-sample pre­dict­ive ac­cur­acy com­pared to the stand­ard mod­els pro­posed in the lit­er­at­ure and that the fore­casts gen­er­ate prof­it­able port­fo­lio al­loc­a­tions even when mar­ket fric­tions are con­sidered. By work­ing dir­ectly with the mean-vari­ance in­vestor’s con­di­tional Euler equa­tion we also char­ac­ter­ize semi-­para­met­ric­ally the re­la­tion between the vari­ous co­v­ari­ates con­sti­tut­ing the con­di­tion­ing in­form­a­tion set and the in­vestor’s op­timal port­fo­lio weights. Our res­ults sug­gest that the re­la­tion between pre­dictor vari­ables and the op­timal port­fo­lio al­loc­a­tion to risky as­sets is highly non-­lin­ear.

Who Be­ne­fits from Robo-ad­vising? Evid­ence from Ma­chine Learn­ing

We study the ef­fects of the largest US robo-ad­viser, Van­guard Per­sonal Ad­visor Ser­vices (PAS), on in­vestor per­form­ance. Across all cli­ents, PAS re­duces in­vestors hold­ings in money mar­ket mu­tual funds and in­creases bond hold­ings. It re­duces the hold­ings of in­di­vidual stocks and US act­ive mu­tual funds, and moves in­vestors to­wards low-­cost in­dexed mu­tual funds. Fi­nally, it in­creases in­vestors’ in­ter­na­tional di­ver­si­fic­a­tion and in­vestors’ over­all risk-­ad­jus­ted per­form­ance. From sign-up, it takes ap­prox­im­ately six months for PAS to ad­just in­vestors’ port­fo­lios to the new al­loc­a­tions. We use a ma­chine learn­ing al­gorithm, known as Boos­ted Re­gres­sion Trees (BRT), to ex­plain the cross-sec­tional vari­ation in the ef­fects of PAS on in­vestors’ port­fo­lio al­loc­a­tion and per­form­ance. The in­vestors that be­ne­fit the most from robo-ad­vising are the cli­ents with little in­vest­ment ex­per­i­ence, as well as the ones that have high cash-hold­ings and high trad­ing volume pre-ad­op­tion. Cli­ents with little mu­tual fund hold­ings and cli­ents in­ves­ted in high-fee act­ive mu­tual funds also dis­play sig­ni­fic­ant per­form­ance gains.



Back to overview