The Wind-Change Index: Reading Uncertainty as a Number
Volatility answers how hard prices shake, not whether structure is being replaced. The Wind-Change Index tries to measure the latter: entropy, dispersion of behavior and network reconfiguration, with the failure conditions stated up front.
Investors share one habit for dealing with uncertainty: turn it into a number. Volatility, VaR, maximum drawdown, credit spreads — all attempts to describe risk through the distribution of prices. They are useful, and they share a blind spot. They speak the language of second moments. They describe the amplitude of shaking, not the change of structure.
A structural switch usually shows up first as "all correlations suddenly fail": two assets that hedged each other fall together, two unrelated sectors rise together. Volatility may have moved only a little, but half your usable models are gone. That is wind-change.
1. Why volatility is not enough
Volatility has three structural defects.
It is symmetric. A rally and a crash both count as volatility, yet structural change is deeply asymmetric: re-rating upward and collapse downward run on different mechanisms.
It is a function of price. Volatility sees the price series, not who is trading, under what rules, and where constraints are tightening. When the composition of participants changes, volatility usually lags.
It does not separate "noisy" from "changing". A high-volatility range market and a market switching tracks can produce the same reading.
2. Three components
The index is built from three independently computable components. The test for including them: each can be calculated from public or semi-public data, each can be questioned on its own, and each can fail on its own.
Entropy H. Discretise the market state into observable buckets — direction of flows, order of sector rotation, shape of option skew. Entropy measures how disordered the distribution of those signals is. Rising entropy means the dominant path is being diluted.
Dispersion of behavior D. Approximated by divergence in positioning, opinion or trade structure. When participants move from consensus to split judgment, the future develops several possible attractors. Dispersion is direct evidence of multiple solutions.
Network reconfiguration R. Build a network of assets or sectors and measure how fast edge weights change per unit time. A structural switch does not typically look like a denser or sparser network; it looks like connections being rapidly reassembled.
The current composition is quantile-normalised then equally weighted into a 0–100 scale. Equal weighting is deliberate: I do not want to fake precision with a more elaborate weighting scheme before it has been tested.
3. How to read it, and how not to
Above 70: a structural switch. Old models lose explanatory power quickly. Rely less on point estimates; move to scenarios and trigger signals.
Below 30: calm. Not the same as safe. Low wind-change plus high leverage is one of the more dangerous combinations, because calm itself encourages leverage.
30–70: transition. Most of the time is spent here. The value is in the rate of change rather than the level: a rapid rise is more worth noticing than a high plateau.
It should not be used as a timing signal, a position sizing rule, or a single basis for a decision. It describes the nature of the environment, not the return of an asset.
4. Three ways it necessarily fails
An honest indicator has to say when it does not work. I currently see three cases.
- A single-point policy intervention. When structure changes in one day by administrative decision, none of the three components can react in time, and the index lags badly.
- Liquidity drought. When trading dries up, estimates of dispersion and edge weights both become unreliable; the reading can be too high or too low.
- Collinear components. In extreme conditions all three may be driven by the same factor, and the composite then carries less information than any single component.
This is why I insist on publishing the method before building a product. An indicator that does not state its failure conditions becomes rhetoric the moment it enters a decision room.
5. Next
The v1.0 whitepaper is in progress: full definitions, parameter sensitivity, and segmented reviews across different market regimes. After that I plan to publish the reading quarterly together with a change log — what changed, why, and whether old readings were recomputed.
What I want most is not users but opponents. If you can show that a component necessarily fails under a particular regime, or offer a simpler substitute, the index has finally started working.