Forecasting Danish stock prices in 2022 under macroeconomic distress using macro indicators - a deep learning approach
Authors
Pedersen, Melani Lærke ; Lyngø, Mikkel Bak
Term
4. term
Education
Publication year
2023
Submitted on
2023-10-23
Pages
78
Abstract
The stock market is hard to predict because prices move quickly and for many different reasons. Many theories try to forecast stock prices, but none works in every situation. This thesis tests LSTM models, a type of neural network for sequence data, to predict stock prices. It examines whether adding macroeconomic factors (broader economic conditions) to the model can improve accuracy. The analysis focuses on a period of macroeconomic distress in autumn 2022. The study tunes and compares a plain LSTM and an LSTM-X (an LSTM with external inputs) to better cope with volatility in the data. The findings show that neither model achieved its full potential in this period, but the work provides useful insights. The thesis outlines adjustments for future forecasting models to simplify handling volatility and improve accuracy.
Aktiemarkedet er svært at forudsige, fordi priserne kan svinge hurtigt og af mange forskellige grunde. Der findes mange teorier for at forudsige aktiekurser, men ingen virker i alle situationer. Denne afhandling afprøver LSTM-modeller, en type neuralt netværk til tidsserier, til at forudsige aktiekurser. Den undersøger, om det kan øge træfsikkerheden at tilføje makroøkonomiske faktorer (bredere økonomiske forhold) til modellen. Analysen fokuserer på en periode med makroøkonomisk uro i efteråret 2022. Studiet tuner og sammenligner en standard LSTM og en LSTM-X (LSTM med eksterne input) for bedre at håndtere volatilitet i data. Resultaterne viser, at ingen af modellerne levede op til deres potentiale i denne periode, men arbejdet giver værdifulde indsigter. Afhandlingen peger på justeringer til fremtidige prognosemodeller, som kan gøre det enklere at håndtere volatilitet og forbedre nøjagtigheden.
[This apstract has been rewritten with the help of AI based on the project's original abstract]