Quantifying Speculative Intensity: The Impact of Bubble Size on Forward Drawdown Risk in AI stocks

Silu Yang

Co-Presenters: Individual Presentation

College: College of Business and Public Management

Major: BS.FINANCE

Faculty Research Mentor: Durmaz, Nazif  

Abstract:

This study aims to construct a bridge across this theoretical and applied divide by introducing and empirically testing a novel Bubble-Drawdown Dynamic Model. We move the inquiry beyond the now well-trodden question of "is there a bubble?" to address a more consequential and underexplored problem: for a curated sample of core AI stocks, does the measurable size or intensity of a statistically detected price bubble possess significant predictive power for the depth of price decline over subsequent short to medium-term horizons? This research question reframes bubbles not merely as binary events to be identified but as continuous processes whose characteristics may inform the scale of their eventual correction. To operationalize this, our methodological framework unfolds in three integrated stages. First, we extend the GSADF methodology to generate a proprietary, time-series BubbleSize metric. This metric quantifies, for each stock at each point in time, the degree to which its price dynamics exceed the threshold of explosive behavior, thus transforming a detection signal into a scalable, continuous variable that captures the "temperature" of the bubble. Second, we rigorously calculate both historical maximum drawdown and, more critically, a suite of forward-looking drawdown series. These forward-looking measures, defined as the peak-to-period-end loss over defined future windows (e.g., 30, 60, 90, 180 days), are designed to precisely represent the realized loss an investor would incur, aligning with the practical risk management perspective advocated in drawdown research. Third, we integrate these components within a comprehensive panel data econometric model. The core specification examines the predictive relationship: DD_i,t+h = α + β * BubbleSize_i,t + γ * Controls + ε_i,t, where a statistically and economically significant positive β would confirm our central hypothesis that larger bubbles precede deeper crashes. This approach allows us to move from the pattern-matching of Potrykus (2024) or the connectedness analysis of Eteman (2024) to a direct, quantifiable test of the bubble-crash risk nexus.