An Investigation of the Neural Network Predictability of Asset Prices in Environments with Trading Frictions
The document presents comprehensive neural network models for asset return prediction and portfolio optimization, assessing trading frictions' impact.
What it examines
This study uses neural network models to predict asset returns and optimize portfolio construction by incorporating realistic trading frictions, risk factors, and different rebalancing horizons. It aims to capture tradable predictive signals and overcome limitations of traditional models by accounting for turnover, liquidity, and market impact.
What it concludes
The research shows that optimized neural network portfolios can produce positive, risk-adjusted returns with small capital, though performance declines with larger investments and longer horizons. Applications include enhanced asset pricing, portfolio management, and risk assessment, with future studies recommended to further refine trading friction models.
Evidence objects
Traditional equalweighted and valueweighted portfolio strategies collapse when realistic trading costs are applied, turning profitable trades into losses; however, neural network predictions yield excess returns, especially in small-scale capital scenarios.
key_findings bullet 1 · key_findings · validation V0
The study introduces a robust portfolio optimization framework that dynamically adjusts positions, integrating effective spread and market impact models while controlling for microcap biases, significantly enhancing trading cost modeling, remarkably.
key_findings bullet 2 · key_findings · validation V0
Extensive out-of-sample tests across varied rebalancing frequencies and datasets underscore the rigorous approach; though effective, scalability challenges for large portfolios persist, warranting further investigation into broader market applications in detail.
key_findings bullet 3 · key_findings · validation V0
The paper introduces a novel neural network approach integrating realistic trading frictions and market microstructure into portfolio optimization and market prediction. Its originality lies in innovative modeling of trading frictions. This compelling methodology mitigates microcap bias, enriches practical machine learning finance applications, and extends traditional equal- and value-weighted investment strategies.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
- … trading friction model that is rooted in the fundamentals of market microstructure. This … combining the spread cost of a trade (Novy-Marx and Velikov 2016) with the trade’s …
Source row: 193 · abstract type: snippet