AI Fear & Greed Index

Methodology

The score blends five signals about the AI and robotics stocks Supercycle tracks, each worth an equal 20 percent:

• Momentum. How far the AI composite sits above or below its own 125-day average.
• Breadth. The share of tracked stocks trading above their own 50-day average, a read on participation rather than a few large names carrying the tape.
• 52-week position. Whether more tracked stocks sit near their 52-week highs or near their 52-week lows.
• Volatility. How much the composite has been moving, annualized over the trailing 30 days. Calm markets read as greed, turbulent ones as fear.
• Exposure tilt. Whether money is favoring pure AI plays over peripheral names over the trailing 20 days, or the reverse. A layer with no priced Pure Play members is scored on the remaining four signals instead, with their weights renormalized to fill the composite, rather than folding in a fabricated flat tilt. Any gauge below built on fewer than five signals says so.

Each signal is not scored on its raw value. It is percentile ranked against roughly three years of its own history first, so a reading of 70 means "higher than about 70 percent of the last three years," not an arbitrary point on a fixed scale. The five percentile scores are then averaged at equal weight into the 0-100 index. This history currently begins Sep 14, 2023, but the very first few weeks of rows have too few prior observations to rank against and are dropped rather than scored on thin air, so the earliest possible reading today is Oct 11, 2023. Both dates advance as the rolling three-year window moves forward with each nightly row, rather than staying fixed.

Buildout momentum sits outside the score on purpose. The panel above reports firm power contracts and construction confirmations, which are facts about what is physically being built, not a sentiment signal. It doesn't have the years of history a percentile rank needs, and folding an unscored, un-rankable number into a weighted score would manufacture precision that isn't there. The two are shown side by side instead: sentiment can read fearful while the physical buildout keeps moving, and that gap is the point, not a contradiction to resolve.

Limitations:

• Survivorship bias. History is computed over today's universe of tracked stocks. Companies that were delisted, acquired, or removed from tracking are absent from past readings, not just today's.
• Look-ahead bias in the exposure tilt. Today's AI Exposure tier assignments are applied back over historical prices, so a stock's current classification, not the classification it would have had at the time, drives its past tilt contribution.
• Universe growth. The constituent count is recorded with every row. It started at 449 stocks at the current beginning of this history (Sep 14, 2023) and stands at 501 today, so earlier and later readings are drawn from differently sized universes.
• A young percentile window. Three years of history is a short base for a sector this new. As more years accumulate the percentile ranks will settle; today's tails are thinner than they will be later.
• Cap weighting uses current share counts. Historical share counts aren't stored, so today's share count is applied against historical closing prices wherever the index needs a market cap weight.

This is an editorial sentiment reading, not a financial metric or investment advice. See the AI Exposure Score methodology on the disclaimer page for how the exposure tiers behind the tilt signal are assigned.