Predicting Trend in the Next-Day Market by Hierarchical Hidden Markov Model

contribution Suggested a new structure of HMM, Trend HMMM, and improved stock forecasting accuracy

strength Analyzed multiple components of time series by employing 3 layer hmm: trend + level + variation -> decoupling trend, variation

Can consider different types of variables: Two indexes, mean deviation rate and index variation

Can make use of the newly available data at ach morning: different percent of btw today’s open and yesterday’s close

weakness Low flexiblityband reactivity due to the intertia of trend especially for the beginnig of the trend

Does the model structure account for any interaction between each layer?

Would be better if the trend had been categorized into more specified groups instead of just up stable and down


data

S&P CNX Nifty index within the period January 1st 2008 – June 30rd 2010

model

decision