About This Project

Bubble Trouble is an independent research project that applies a systematic, rule-based framework to monitor the AI infrastructure investment cycle for bubble dynamics. It was built by an independent researcher with an interest in capital cycles, market structure, and the recurring patterns of technology infrastructure booms and busts.

The project is not affiliated with any financial institution, hedge fund, brokerage, or investment advisory firm. It generates no revenue, accepts no advertising, and has no commercial relationship with any data provider.

This project attempts to be transparent about its limitations. Here is what you should know:

The sample size is tiny

The framework's backtest covers three events (2000, 2008, 2022). This is an n=3 sample, which is statistically insufficient for any meaningful confidence about future performance. The backtest provides directional intuition about whether the signal types are measuring the right things, but it cannot prove the system works. We acknowledge this openly.

Known limitations are documented

Every signal has known weaknesses. Private credit markets (Signal 2) are opaque and difficult to monitor in real time. Margin debt data (Signal 6) has a 6-week reporting lag, which means the system is always looking at stale data for this signal. The capex-revenue divergence (Signal 1) only updates quarterly. These limitations are not hidden — they are described in the methodology and backtest pages.

The system is designed for early decline, not the exact peak

The peak detection signals are confirmatory — they tell you the unwind has begun, not that it will begin. By the time these signals fire and meet persistence requirements, some decline will have already occurred. The system is designed to capture 70–80% of the move, not 100%. Trying to exit at the exact peak is a fool's errand; the goal is to exit while there's still significant downside remaining.

Private credit data is particularly challenging

Signal 2 (Private Credit Stress Basket) relies on manually assembled public information about inherently private markets. BDC discount data is publicly available, but redemption gate incidents, bank financing pullback, and PIK loan stress require careful reading of financial news, regulatory filings, and industry reports. This signal is the most subjective in the system.

The system can be wrong

It may produce false positives (alerting when there is no bubble) or false negatives (failing to alert when a bubble bursts through an unmonitored channel). When the system is wrong, we will document it transparently in the history log.

"Better to be approximately right than precisely wrong."

The system deliberately uses a small number of well-understood signals rather than a large number of noisy ones. It prefers transparent, explainable rules over opaque statistical models. It includes multiple safeguards against premature escalation (persistence requirements, escalation blockers, maximum speed limits) because false positives are costly — they cause you to miss upside in a market that may continue to rise.

The framework acknowledges that bubbles are easier to identify in hindsight than in real time, and that timing the exit is the hardest problem in investing. It does not claim to solve this problem — it claims to provide a structured, systematic way to think about it.