Skip to content
All library documents

Using an SVM with MACD and Parabolic SAR for Forex Signals

Article QuantInsti blog
Read the original

Summary

The article outlines a supervised learning workflow for classifying EUR/USD direction. It introduces features, feature selection, and support vector machines, then describes a model using hourly EUR/USD data dating back to 2010, with MACD and Parabolic SAR as inputs. Indicator values are lagged to reduce look-ahead bias, and the data is split into training and test sets before fitting the model and making predictions.

The reported model accuracy is 53%. The author then interprets the plotted predictions as threshold rules based on the price–SAR difference and MACD histogram. Short and long rules are reported at 54% and 50% accuracy, respectively. These figures are limited evidence: the document gives no fuller out-of-sample evaluation, transaction costs, risk-adjusted results, or evidence of profitability. It presents the rules as preliminary and says further backtesting is needed to assess viability.

Key ideas

  • The example uses MACD and Parabolic SAR to build features for an hourly EUR/USD classification model.
  • Lagging indicator values is described as a way to reduce look-ahead bias.
  • The SVM model reports 53% accuracy, while the derived short and long rules report 54% and 50% accuracy.
  • The reported accuracy figures do not establish profitability; the proposed rules require further backtesting.

Tags

ForexMachine learningTechnical indicatorsBacktesting

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