Instrument
The regime
detector
It reads seven inputs and prints a plain description of the conditions in the window that just closed. That is the whole job. It exists so that "my strategy stopped working" can be replaced with "the conditions this strategy was written for stopped being present".
There is no score. There is nothing to act on here.
Inputs
- Reads
- Close-to-close dispersion over a rolling window
- Measured as
- Standard deviation of log changes, 30 day window, annualised
- Why it is in there
- The same rule behaves differently in a quiet market and a loud one. This is the input that tells you which one you are in.
- Reads
- Dispersion of the volatility series itself
- Measured as
- Standard deviation of the 30 day realised series over 90 days
- Why it is in there
- Distinguishes a market that is loud and stable from one that is changing character week to week.
- Reads
- How much of the move is directional rather than round trip
- Measured as
- Net change divided by the sum of absolute daily changes, 60 day window
- Why it is in there
- A number near zero says price went a long way and ended up nowhere. That is a description, not a forecast.
- Reads
- How similarly the tracked assets are moving
- Measured as
- Mean pairwise correlation of daily changes, 30 day window
- Why it is in there
- When everything moves together, holding several things is one position wearing several hats.
- Reads
- Order book depth within a fixed distance of mid
- Measured as
- Median resting size inside 50bps, sampled hourly, daily median
- Why it is in there
- Thin books change what a given position size actually costs to enter and leave.
- Reads
- Distance from the running high of the tracked set
- Measured as
- Percentage below the trailing 365 day maximum
- Why it is in there
- Purely positional context. It says where the market is relative to its own recent history and nothing about where it goes next.
- Reads
- Coverage of the inputs above for the current window
- Measured as
- Share of expected observations present
- Why it is in there
- If the inputs are patchy the readout says so instead of quietly printing a stale label.
Example outputs
These are worked examples of the shape of a readout. They are not live values and they are not a claim about any market, past or present.
| Realised vol | Low band |
|---|---|
| Vol of vol | Low band |
| Trend persistence | High |
| Dispersion | Moderate |
| Liquidity | Normal |
| Completeness | 100% |
Conditions are calm and moves have been carrying rather than round tripping. This is a description of the window that just happened.
| Realised vol | High band |
|---|---|
| Vol of vol | High band |
| Trend persistence | Low |
| Dispersion | Compressed |
| Liquidity | Thin |
| Completeness | 100% |
Large moves, little net travel, everything moving together, thinner books. A window in which most rule sets built on calm data are outside the conditions they were written for.
| Realised vol | Rising from low |
|---|---|
| Vol of vol | Elevated |
| Trend persistence | Falling |
| Dispersion | Widening |
| Liquidity | Normal |
| Completeness | 98% |
The inputs disagree with each other. The honest label for a window like this is that it does not have a settled label yet.
| Realised vol | — |
|---|---|
| Vol of vol | — |
| Trend persistence | — |
| Dispersion | — |
| Liquidity | — |
| Completeness | 61% |
Below the coverage threshold, the detector prints nothing rather than guessing. A blank readout is a valid readout.
What it is not
- It does not produce a score, a rating, a percentage confidence or a dial
- It does not say buy, sell, enter, exit, hold, top or bottom
- It does not forecast anything, over any horizon
- It does not know what you own and has no opinion on it
- It is not evidence that any strategy works, including the one it sits next to
Method
The reference implementation, the parameter file and the replay harness are part of the Advanced tier. The description on this page is complete enough to rebuild it yourself, which is the point of publishing it. What the tiers contain.
Education about trading process. Not investment advice. Nothing here is a recommendation to buy or sell anything. Most retail traders lose money.