Short-Term Stock Price-Trend Prediction Using Meta-Learning

TL;DR

This paper introduces a meta-learning framework with convolutional neural networks for short-term stock trend prediction, effectively handling limited data and improving prediction accuracy and profitability on the S&P500.

Contribution

It proposes a novel meta-learning approach with slope-detection labeling for stock trend prediction using CNNs, addressing data scarcity issues.

Findings

Meta-learning significantly improves prediction accuracy.

The framework enhances profitability in stock trend forecasting.

Effective on S&P500 data with limited labeled samples.

Abstract

Although conventional machine learning algorithms have been widely adopted for stock-price predictions in recent years, the massive volume of specific labeled data required are not always available. In contrast, meta-learning technology uses relatively small amounts of training data, called fast learners. Such methods are beneficial under conditions of limited data availability, which often obtain for trend prediction based on time-series data limited by sparse information. In this study, we consider short-term stock price prediction using a meta-learning framework with several convolutional neural networks, including the temporal convolution network, fully convolutional network, and residual neural network. We propose a sliding time horizon to label stocks according to their predicted price trends, referred to as called slope-detection labeling, using prediction labels including "rise plus," "rise," "fall," and "fall plus". The effectiveness of the proposed meta-learning framework was evaluated by application to the S&P500. The experimental results show that the inclusion of the proposed meta-learning framework significantly improved both regular and balanced prediction accuracy and profitability.