Dynamic Prediction Length for Time Series with Sequence to Sequence Networks

TL;DR

This paper introduces a sequence-to-sequence model that predicts variable-length outputs for time series, balancing accuracy and prediction length through a novel loss function, and demonstrates its effectiveness on securities price prediction.

Contribution

The paper presents a new method allowing sequence-to-sequence models to predict variable lengths with a custom loss function and threshold-based evaluation, addressing a key limitation of fixed output lengths.

Findings

Longer predictions for stable securities

Balances prediction accuracy and length naturally

Effective on securities price forecasting

Abstract

Recurrent neural networks and sequence to sequence models require a predetermined length for prediction output length. Our model addresses this by allowing the network to predict a variable length output in inference. A new loss function with a tailored gradient computation is developed that trades off prediction accuracy and output length. The model utilizes a function to determine whether a particular output at a time should be evaluated or not given a predetermined threshold. We evaluate the model on the problem of predicting the prices of securities. We find that the model makes longer predictions for more stable securities and it naturally balances prediction accuracy and length.