Geometric Structured Trend Tunneling: A Hybrid VARIMA-SVR Model for Synthetic Stock Time Series Generation

Abstract

This study presents a novel hybrid framework, Geometric Structured Trend Tunneling (GSTT), for generating synthetic multivariate time series data, specifically applied to stock price data of Medco Energi Internasional (MEDC), a major player in Indonesia’s energy sector. The proposed model integrates the statistical power of Vector Autoregressive Integrated Moving Average (VARIMA) with the nonlinear pattern-capturing capability of Support Vector Regression (SVR), enabling high-fidelity reconstruction of temporal structures and feature dependencies in financial datasets. The dataset used spans over two decades (2003–2024) and includes core trading indicators such as Open, High, Low, and Close prices. Experimental results demonstrate that GSTT achieves excellent performance across multiple evaluation metrics, including MAE, RMSE, R², and KS tests, while preserving inter-feature correlations and distributional fidelity. Visual comparisons and descriptive statistics further confirm the model’s ability to replicate realistic market behavior. Unlike deep generative models such as GANs or VAEs, GSTT offers a more interpretable, stable, and computationally efficient alternative for financial data augmentation, simulation, and robust AI training. This work contributes a scalable solution for addressing data scarcity in financial modeling, with potential applications in backtesting, risk analysis, and algorithmic trading simulations.