HMM-Stock-Forecast

Stock forecasting based on (gaussian) Hidden Markov Model

Model

Model implementation is based on these two articles:

• [1] Nguyen N. Hidden Markov Model for Stock Trading. International Journal of Financial Studies. 2018; https://doi.org/10.3390/ijfs6020036.
• [2] M. R. Hassan and B. Nath, Stock market forecasting using hidden Markov model: a new approach 5th International Conference on Intelligent Systems Design and Applications (ISDA’05), 2005, pp. 192-196, doi: 10.1109/ISDA.2005.85.

Arguments

Application accept these arguments:

• -t stock ticker (e.g. AAPL)
• -f path to csv soubor (4 columns: Open, Low, High, Close)
• -s data start date
• -e data end date
• -w window size (see [1] for more information about window)
• -m type of hmm implementation:
  • HMM - our implementation (default)
  • pomegranate - https://github.com/jmschrei/pomegranate

Run

• In first step program find number of optimal hidden states (n) based on:

  • AIC (Akaike information criterion)
  • BIC (Bayesian information criterion)
  • HQC (Hannan-Quinn information criterion)
  • CAIC (Bozdogan Consisten Akaike information criterion)

• Then prediction is estimated for w (window) business days. Everytime historical data 0-x are used to predict closing price at day x+1

Results

• Ticker: SPY

• Start date: 2019-01-01

• End date: 2022-01-01

• Window size: 100

• MAPE: 0.861 %

• r2: 82 %

Dependencies

• numpy
• matplotlib
• pandas
• sklearn