Generalized Distribution Prediction for Asset Returns

TL;DR

This work tackles distributional forecasting of asset returns across diverse asset classes by introducing a two-stage LSTM framework (qLSTM) that first estimates asset-specific quantiles of volatility-normalized returns and then scales them with market data, followed by a Quantile-Smooth Density Estimation to obtain full distributions. A Hybrid variant (qHybrid) blends Gaussian quantiles with learned quantiles to improve calibration across distribution shapes. The approach uses asset-neutral features and opensource data to enable cross-asset generalization, showing strong tail-risk performance on real data and robust calibration on synthetic data across multiple metrics such as Wasserstein distance and CRPS. Overall, the method advances practical tail-risk assessment and derivative pricing with reproducible, multi-asset distribution forecasts.

Abstract

We present a novel approach for predicting the distribution of asset returns using a quantile-based method with Long Short-Term Memory (LSTM) networks. Our model is designed in two stages: the first focuses on predicting the quantiles of normalized asset returns using asset-specific features, while the second stage incorporates market data to adjust these predictions for broader economic conditions. This results in a generalized model that can be applied across various asset classes, including commodities, cryptocurrencies, as well as synthetic datasets. The predicted quantiles are then converted into full probability distributions through kernel density estimation, allowing for more precise return distribution predictions and inferencing. The LSTM model significantly outperforms a linear quantile regression baseline by 98% and a dense neural network model by over 50%, showcasing its ability to capture complex patterns in financial return distributions across both synthetic and real-world data. By using exclusively asset-class-neutral features, our model achieves robust, generalizable results.