Perceptrons for Classification and Algorithmic Trading

Summary

This guide introduces the perceptron, a basic neural network model for binary classification. It outlines activation functions and learning, then demonstrates how weights and a bias can be updated from classification errors. Examples use iris flower measurements to show a simple linearly separable task and a harder classification problem, with changes to the learning rate and error tracking across training epochs.

The trading example applies perceptrons to EUR/USD using volatility and recent return features. It defines targets from subsequent price movement, trains on rolling historical windows, and uses the resulting long and short signals to enter positions. The guide illustrates a workflow rather than establishing a profitable strategy: it provides no performance results, transaction cost analysis, or detailed risk controls. A perceptron learns only a linear decision boundary, and the demonstrated targets, features, and training setup would need careful validation to address overfitting and changing market conditions.

Key ideas

• A perceptron predicts a class using a weighted sum of inputs, a bias, and an activation rule.
• Training adjusts weights and bias in response to classification errors.
• The iris examples illustrate both linearly separable and more challenging binary classification tasks.
• The trading example uses volatility and recent returns to predict directional EUR/USD outcomes.
• Rolling training and explicit out-of-sample evaluation are needed before treating signals as evidence of an edge.

Tags

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