National Yang Ming Chiao Tung University (NYCU) brings together Information Management and Finance to form an interdisciplinary program that integrates information engineering, data analytics, and financial strategy. By embedding AI technologies into both research and education, the institute has built an intelligent financial research platform that enables large-scale data analysis, quantitative modeling, and real-world strategy development—empowering students and researchers to accelerate innovation and data-driven decision-making.
Moving a financial strategy from an initial model to practical validation requires repeated access to historical trading data, parameter adjustments and comparisons across different conditions. The laboratory also supports researchers and students analyzing individual quotes and order-book changes, creating concurrent demand for low-latency data access. When backtesting and shared-file access compete for storage resources, time spent waiting for data slows research validation.
This constraint matters when research addresses industry problems. Asset allocation studies must compare sentiment signals with strategy performance, while anti-money-laundering research must examine transaction patterns and risk features. The team therefore needed infrastructure capable of supporting large datasets and repeated testing so that its hypotheses could be evaluated in settings closer to practical use.
NYCU deployed a dual-tier QSAN storage architecture to separate performance-sensitive database workloads from shared research files. The all-flash tier, QSAN XF5 series, serves SQL and database servers accessing data for AI analysis, strategy backtesting and strategy execution research. A second tier, QSAN XN2 series unified storage, centralizes research datasets, analytical results and model files for shared access.
This arrangement gives backtesting reads and routine file access separate storage resources, reducing opportunities for contention between the workloads. The front-end platform replays tick-level historical data, including individual quotes and five-level order-book changes. Custom indicators, performance calculations, equity curves and profit-and-loss analysis let researchers observe market activity, test hypotheses and review results within one workflow.
The architecture links storage to the research process: the all-flash tier supports repeated backtesting, while shared storage retains datasets and models for continued refinement and practical validation. Approved Series Names and the deployed connectivity configuration remain pending confirmation.
Annual backtesting fell from 180 seconds to 12 seconds, a 15x speedup. For a team repeatedly adjusting parameters, shorter waits help researchers compare strategy settings and review results sooner. This brings backtest feedback into research and classroom model discussions more quickly, although the number of additional validation runs has not been measured.
The concurrent multi-strategy test fell from 320 seconds to 15 seconds, a speedup of about 21x. This processing-time improvement is useful for research workflows that compare multiple strategies or model settings. The laboratory’s industry research includes sentiment-driven asset allocation and anti-money-laundering transaction-pattern analysis and risk assessment, providing practical problems against which methods can be evaluated. Available sources do not quantify improvements in investment performance or risk detection for these projects.
We’re here to help you find the best, most cost‑effective solution for your business.
