Paper Type
Complete
Abstract
The rapid expansion of artificial intelligence has generated unprecedented investment and expectations of transformative productivity gains, yet concerns about the IT productivity paradox remain unresolved. This study investigates whether innovation inflection points can signal business-cycle transitions before they become apparent in aggregate statistics. Building upon Real Business Cycle theory and IS literature on the productivity paradox, we analyze 7,044 daily observations of Microsoft Corporation (1998–2026), integrating 51 technical indicators with 56 LSEG MarketPsych sentiment variables. We employ Random Forest, CNN, LSTM, and Decision Tree classifiers to detect four business cycle phases. Results show that all models achieve comparable F1 scores (0.671–0.677), with Random Forest yielding the best performance. Feature analysis reveals that technical indicators dominate predictive importance (68.5%), while sentiment variables provide a complementary 28.7%. These findings bridge aggregate business-cycle analysis and micro-level digital evidence to provide insights into technology-driven economic turning points.
Paper Number
1193
Recommended Citation
Xu, Yilin and Subramanian, Hemang, "Detecting Innovation Inflection Points in Firm-Level Market Data" (2026). AMCIS 2026 Proceedings. 10.
https://aisel.aisnet.org/amcis2026/sig_si/sig_si/10
Detecting Innovation Inflection Points in Firm-Level Market Data
The rapid expansion of artificial intelligence has generated unprecedented investment and expectations of transformative productivity gains, yet concerns about the IT productivity paradox remain unresolved. This study investigates whether innovation inflection points can signal business-cycle transitions before they become apparent in aggregate statistics. Building upon Real Business Cycle theory and IS literature on the productivity paradox, we analyze 7,044 daily observations of Microsoft Corporation (1998–2026), integrating 51 technical indicators with 56 LSEG MarketPsych sentiment variables. We employ Random Forest, CNN, LSTM, and Decision Tree classifiers to detect four business cycle phases. Results show that all models achieve comparable F1 scores (0.671–0.677), with Random Forest yielding the best performance. Feature analysis reveals that technical indicators dominate predictive importance (68.5%), while sentiment variables provide a complementary 28.7%. These findings bridge aggregate business-cycle analysis and micro-level digital evidence to provide insights into technology-driven economic turning points.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.