Trading through Earnings Seasons using Self-Supervised Contrastive Representation Learning

TL;DR

This research introduces the Contrastive Earnings Transformer (CET) model, a self-supervised learning approach rooted in Contrastive Predictive Coding (CPC), aiming to optimise the utilisation of earnings data, and conducts a comparative study of CET against benchmark models across diverse sectors.

Abstract

Earnings release is a key economic event in the financial markets and crucial for predicting stock movements. Earnings data gives a glimpse into how a company is doing financially and can hint at where its stock might go next. However, the irregularity of its release cycle makes it a challenge to incorporate this data in a medium-frequency algorithmic trading model and the usefulness of this data fades fast after it is released, making it tough for models to stay accurate over time. Addressing this challenge, we introduce the Contrastive Earnings Transformer (CET) model, a self-supervised learning approach rooted in Contrastive Predictive Coding (CPC), aiming to optimise the utilisation of earnings data. To ascertain its effectiveness, we conduct a comparative study of CET against benchmark models across diverse sectors. Our research delves deep into the intricacies of stock data, evaluating how various models, and notably CET, handle the rapidly changing relevance of earnings data over time and over different sectors. The research outcomes shed light on CET's distinct advantage in extrapolating the inherent value of earnings data over time. Its foundation on CPC allows for a nuanced understanding, facilitating consistent stock predictions even as the earnings data ages. This finding about CET presents a fresh approach to better use earnings data in algorithmic trading for predicting stock price trends.

Figures (5)

• Figure 1: Representation learning of the combined price, volume, and earnings data via CPC. Price and volume data, as well as earnings data, are separately encoded and combined through element-wise addition to form the context vector $C_{t}$ whose mutual information with future price and volume data representations, $z_{t+k}$, are optimised.
• Figure 2: Structure of Transformer Encoder which is employed to produce the embedding representation of input time series $z_{t} \in \mathbb{R}_{d}, t=0, \dots, \omega$ with dimension $d$, which is the result of previously feeding minutely-frequency price and volume data through a non-linear network. Operations of the Transformer is described in section \ref{['sec:encoding_of_data']}
• Figure 3: Chart representation of data in table \ref{['tab:results_section_5.3']} portraying varying model performances over the 4 days since the release of earnings data. The y-axis is prediction success rate. For each model, the four bars going left to right represent day 2 -- day 5.
• Figure 4: Ablation study results portraying the impacts by a changing latent step size on the InfoNCE loss and similarity score between predicted context vector and the corresponding positive sample vector
• Figure 5: Ablation study results portraying the impacts by a changing latent step size on the InfoNCE loss and similarity score between predicted context vector and the corresponding positive sample vector