Automatic pattern recognition of Head-And-Shoulder

Implementation for Head-And-Shoulder (Lo et al., 2000, Journal of Finance) with simulated data

This prototype shows a reduced approach, how to implement an automated pattern recognition algorithm for the Head-And-Shoulders pattern (Lo et al., 2000, Journal of Finance) in MATLAB.
Lo et al. (2000) use a moving window consisting of 38 price observations. With every step, the oldest price will be deleted and a new price appended to the price queue. A kernel regression smooths the raw prices (observations 4.-38.) in order to filter noise. The pattern recognition algorithm identifies a specific visual pattern, which is described by the position of the last five local extrema (called E1, ..., E5) in the smoothed prices. For example, the arrangement E1 < E3 > E5 and (E2, E4) < (E1, E3, E5) constitutes a Head-And-Shoulder TOP pattern, if the 4. price is a local extrema. The remaining newer prices (3. – 1.) protect again look-ahead bias. A TOP pattern implies a decreasing price and BOTTOM pattern vice versa. A timer generates 80 random prices. The command-line and a figure present the output.