A Markovian Approach to Financial Market Forecasting

Abstract: This thesis aims to investigate the feasibility of using a Markovian approach toforecast short-term stock market movements. To assist traders in making soundtrading decisions, this study proposes a Markovian model using a selection ofthe latest closing prices. Assuming that each time step in the one-minute timeframe of the stock market is stochastically independent, the model eliminates theimpact of fundamental analysis and creates a feasible Markov model. The modeltreats the stock price’s movement as entirely randomly generated, which allowsfor a more simplified model that can be implemented with ease. The modelis intended to serve as a starting ground for more advanced technical tradingstrategies and act as useful guidance for a short-term trader when combinedwith other resources. The creation of the model involves Laplace smoothing toensure there are no zero-probabilities and calculating the steady-state probabilityvector of the smoothed matrix to determine the predicted direction of the nexttime step. The model will reset daily, reducing the impact of fundamental factorsoccurring outside trading hours and reducing the risk of carrying over bias fromprevious trading day. Any open positions will hence be closed at the end of theday. The study’s purpose is to research and test if a simple forecasting modelbased on Markov chains can serve as a useful tool for forecasting stock prices atshort time intervals. The result of the study shows that a Markov-based tradingstrategy is more profitable than a simple buy-and-hold strategy and that theprediction accuracy of the Markov model is relatively high.