Multi-Input LSTM Networks for Volatility Forecasting in Precious Metals A Retail Hedging Framework for Small-Scale Bullion Enterprises

Read full text Summarise with VIRAT
Authors: Ayush Jain
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Small and Medium Enterprises (SMEs) in the bullion sector face a Systemic Information Gap, where family-run shops rely on intuition while institutional traders utilize high-frequency algorithms. This disparity leaves local jewelers vulnerable to Inventory Value Decay (IVD)—a quantifiable loss in "paper value" caused by market price retractions during the typical 90-180 day holding period. This research introduces "BullionGuard," a retail hedging framework designed to democratize financial engineering by utilizing a Multi-Input Long Short-Term Memory (LSTM) network. Unlike traditional univariate models, this architecture leverages a branched Functional API to simultaneously process sequential silver price data and non-sequential macro-economic "shocks," such as the US Dollar Index (DXY) and 10-Year Treasury Yields. Key Methodology and Results: Engineering for the Edge: The system was optimized for consumer-grade hardware, reducing the memory footprint from over 4GB to less than 800MB. Computational Efficiency: Full model re-training is completed in under 10 minutes on a standard i5 processor, providing a "Zero-Lag" decision engine. Predictive Superiority: The Multi-Input LSTM achieved a 69% directional accuracy and a Mean Absolute Error (MAE) of $0.62, significantly outperforming the Moving Average baseline ($1.42 MAE). Operational Impact: A 12-month simulation against a family retail ledger demonstrated that the model's "Accumulation Signals" reduced IVD by 14.2% and expanded net margins by approximately $1.15 per ounce. By replacing reactive buying with a Predictive Hedging Signal, this framework proves that deep learning can serve as a protective shield for traditional SMEs, transforming inventory management from a game of chance into a precise, accessible science.

Fields of Study & Subjects

Stock Market Forecasting Methods Forecasting Techniques and Applications Financial Risk and Volatility Modeling

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
22 March 2026
Language
EN