Using Kohonen Maps to Map Price Inputs to a Trading Signal

Summary

The article explains self-organizing, or Kohonen, maps and applies them to a trading model that compresses several price features into a one-dimensional output. It distinguishes historical input data from the future value being estimated, then describes an MQL5 design in which neurons associate feed data with a single output value. The example uses changes in highs and lows as inputs and the latest bar’s open-to-close change as the output; it emphasizes that traders should choose features to match their own market view.

The author reports training on EURJPY data from October 2018 through June 2021, followed by forward testing from the training end date. Two risk settings are compared, with reported Sharpe ratios of 0.43 and 0.85; the text says the more conservative case produced the lower ratio. The article provides no detailed report figures in the supplied text, and the short training period limits what can be concluded. It calls for further risk and position sizing work and testing on broker real-tick data over longer periods before deployment.

Key ideas

• Kohonen maps organize high-dimensional observations into a lower-dimensional representation while retaining relationships among observations.
• The example maps several historical price features to a single output dimension.
• Input and output features should reflect the trader’s own market assumptions and can be customized.
• The EURJPY example was trained on data from October 2018 to June 2021 and then forward tested.
• The reported comparison favors the more aggressive setting by Sharpe ratio, but further testing and risk tuning are needed.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.