Napovedovanje trenda cen delnic na podlagi sporočil za javnost

Lukšič, Nal (author), Hovelja, Tomaž (mentor)

V magistrskem delu se ukvarjamo s problematiko napovedovanja trenda cen delnic podjetij iz farmacevtskega sektorja. Cilj raziskave je izdelati kakovosten napovedni model, ki bi glede na vhodne spremenljivke z visoko natančnostjo napovedal, ali bo cena padla ali narasla. S tem si želimo doseči višje donose investicije od že obstoječih modelov. Glavni parameter, na katerem temelji napovedni model so objave podjetij na straneh sporočil za javnost. Te smo uporabili zaradi velikega vpliva na nihanje cen. Poleg objav dogodkov v delu uporabimo še momentum cene pred in po objavi dogodka ter trend iskanosti podjetja v spletnem iskalniku Google. S pomočjo vhodnih parametrov nato na osnovi naključnih gozdov zgradimo napovedni model. Končni rezultati kažejo, da predlagani sistem dosega do 85,2% točnost napovedovanja primerov, ko cena zraste. V primeru investicije 1000 evrov bi s pomočjo napovednega modela v roku 34 mesecev ustvarili 490-odstotni donos, oziroma 4928 evrov dobička.

• Language: Slovenian
• Keywords: odločitveno drevo, naključni gozd, napovedovanje cen delnic, strojno učenje, klasifikacija
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2022
• PID: 20.500.12556/RUL-134409
• COBISS.SI-ID: 96305667
• Publication date in RUL: 13.01.2022

Secondary language

• Language: English
• Title: Forecasting stock price trends based on press releases

Abstract

The research looks into the problem of pharma stocks price trend prediction. The aim of this research is to create a prediction model that would classify stock trends with high precision. That would allow the user to achieve higher returns on the initial investment than the existing models deliver. The main parameter on which the predictive model is based is news posted on press release pages of pharma companies. These have been used due to the large impact they have on price fluctuations. In addition, price momentum before and after an event has been used, as well as the search trends in Google's search engine. The final results show that the proposed system achieves an 85.2% accuracy. In the case of an investment of EUR 1,000, the forecast model would generate a 490\%\ return or EUR 4,928 of profit within 34 months.

• Keywords: decision tree, random forest, stock price prediction, machine learning, classification