Capturing the Order Imbalance with Hidden Markov Model: A Case of SET50 and KOSPI50

Based on the empirical evidence of the recent strand of the literature, Market Efficiency creation process is not instantaneous, but it is rather attained over short-horizon of time. In the low liquid market, the price movement of financial assets can be predicted by order imbalance indicators. In contrast, in a more liquid market, the predictability of return can substantially decrease. In this study, we implement one of the well-known machine learning models for capturing the pattern recognition known as the hidden Markov model. We document the role of order imbalance in forecasting the price movement of selected stocks in markets with different levels of liquidity which are the stock exchange of Thailand and Korea exchange. As the consequence, we can create an algorithmic trading strategy based on the states of risky assets captured by the models. The main finding is consistent with the previous literature that both the predictability of the models and the profitability of the strategy diminish as the frequency decreases and market liquidity increases. Remarkably, our model in the market with lower liquidity is able to generate signal that achieves average hit ratio of 0.83 in predicting the risky assets as the positive price movement at frequency of 5 min.