Suurtel keelemudelitel põhinevate investeerimisportellide tulemuslikkuse hindamine ChatGPT näitel
Laen...
Kuupäev
Autorid
Ajakirja pealkiri
Ajakirja ISSN
Köite pealkiri
Kirjastaja
Tartu Ülikool
Abstrakt
Viimastel aastatel on suurte keelemudelite kiire areng toonud märkimisväärseid muutusi finantssektorisse, pakkudes investoritele uuenduslikke võimalusi portfellide koostamiseks. Käesolevas magistritöös uuriti, kuivõrd
suudab tehisintellektil põhinev keelemudel ChatGPT
koostada investeerimisportfelle vastavalt tootluse, riski ja ajahorisondi kriteeriumitele. Teoreetilises osas anti ülevaade tehisintellekti arengust, suurte keelemudelite toimimisest ja investeerimisportfellide hindamise metoodikatest. Empiirilises osas analüüsiti nelja eri eesmärgiga ChatGPT poolt genereeritud portfelli, kasutades ajalooliste hinnaliikumiste andmeid ja finantsnäitajaid (sh tulumäär, Sharpe'i suhe, beetakordaja, maksimaalne langus).
Tulemused näitasid, et kuigi ChatGPT suudab koostada loogilisi ja riskitasemele vastavaid portfelle, jäi nende tulemuslikkus konkreetsete investeerimiseesmärkide saavutamisel piiratuks. Töö järeldab, et suured keelemudelid võivad olla väärtuslikud tööriistad
investeerimisel, ent ei saa veel asendada
spetsialisti põhjalikku analüüsi.
Edasised uuringud võiksid hõlmata pikemaid perioode,
erinevaid varaklasse ning mitme keelemudeli võrdlust muutuvas turukeskkonnas.
In recent years, the rapid development of large language models has introduced notable changes to the financial sector, providing investors with innovative opportunities for portfolio construction. This master’s thesis examines the extent to which ChatGPT, an artificial intelligence–based language model, is capable of generating investment portfolios that align with predetermined criteria concerning return, risk, and investment horizon. The theoretical section outlines the evolution of artificial intelligence, the operating principles of large language models, and methodologies for assessing portfolio performance. The empirical analysis evaluates four portfolios generated by ChatGPT, each targeting a distinct investment objective. Historical price movement data and financial indicators (including return, Sharpe ratio, beta coefficient, and maximum drawdown) were used as the basis of this analysis. Findings indicate that although ChatGPT can produce structurally coherent portfolios corresponding to specified risk profiles, its effectiveness in achieving precise investment goals remains limited. The study concludes that while large language models may serve as valuable tools for supporting investment decisions, they cannot yet replace the comprehensive analysis conducted by human experts. Future research could benefit from exploring longer time horizons, a broader array of asset classes, and comparative assessments of multiple language models under changing market conditions.
In recent years, the rapid development of large language models has introduced notable changes to the financial sector, providing investors with innovative opportunities for portfolio construction. This master’s thesis examines the extent to which ChatGPT, an artificial intelligence–based language model, is capable of generating investment portfolios that align with predetermined criteria concerning return, risk, and investment horizon. The theoretical section outlines the evolution of artificial intelligence, the operating principles of large language models, and methodologies for assessing portfolio performance. The empirical analysis evaluates four portfolios generated by ChatGPT, each targeting a distinct investment objective. Historical price movement data and financial indicators (including return, Sharpe ratio, beta coefficient, and maximum drawdown) were used as the basis of this analysis. Findings indicate that although ChatGPT can produce structurally coherent portfolios corresponding to specified risk profiles, its effectiveness in achieving precise investment goals remains limited. The study concludes that while large language models may serve as valuable tools for supporting investment decisions, they cannot yet replace the comprehensive analysis conducted by human experts. Future research could benefit from exploring longer time horizons, a broader array of asset classes, and comparative assessments of multiple language models under changing market conditions.