Learning Illiquid Asset Prices
Researchers present a new semiparametric model using neural networks to price illiquid assets like mortgage-backed securities (MBS), which rarely trade and are hard to value. Analyzing over 45 million MBS trades, the model achieved an R2 of 88 percent, far surpassing traditional methods. It adapts to market changes and reveals that delayed disclosure of trade details increases uncertainty, a key regulatory insight. The model’s complexity may limit use by smaller institutions.
What it examines
This paper develops a new method to estimate prices for illiquid assets, like mortgage-backed securities, using a semiparametric model and machine learning. The approach uses asset and market features to provide accurate point and interval price estimates, helping with portfolio valuation and regulatory compliance.
What it concludes
The method delivers highly accurate price predictions and reliable uncertainty intervals, outperforming existing techniques. It is useful for valuing illiquid assets in finance, risk management, and regulation. The approach is flexible, scalable, and can be applied to other markets where prices are rarely observed.
Evidence objects
Researchers unveil a neural network-based model for pricing illiquid assets like mortgage-backed securities, using real-time market and asset-level data to deliver accurate prices and robust uncertainty measures, surpassing traditional methods.
key_findings bullet 1 · key_findings · validation V0
Analyzing over 45 million MBS trades, the model achieved an impressive $R^2$ of 88%, adapting to market shifts within a single day and revealing which bond and market features most influence pricing.
key_findings bullet 2 · key_findings · validation V0
The study finds delayed disclosure of key trade information (CUSIPs) increases pricing uncertainty, raising regulatory concerns; however, the models reliance on large, detailed datasets may limit accessibility for smaller institutions.
key_findings bullet 3 · key_findings · validation V0
This paper pioneers a semiparametric conditional factor model for quantile estimation in illiquid mortgage-backed securities (MBS), integrating neural networks for flexible factor loadings and cross-conformal prediction for calibrated intervals. Its assumption-light, dynamically adaptive approach delivers rigorous uncertainty quantification, outperforming existing methods and offering actionable insights, marking a significant advance in quantitative finance.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
prices. Empirical results for mortgage-backed securities demonstrate the method's effectiveness and represent the first treatment of illiquid MBS pricing in
Source row: 1184 · abstract type: snippet