ACD/README.md

# ACD

A live market activity gauge in native MQL5: how busy the quote stream is
right now, compared with what is normal for this time of day, updated on
every quote.

Companion code for the MQL5 article: https://www.mql5.com/en/articles/24742

## What it does

Tick volume answers "how many quotes arrived in this bar", which is a
different question from "is the market unusually busy right now". Most of its
variation is the clock: on EURUSD the quietest half hour of the day waits
roughly forty-seven times longer between quotes than the busiest one, every
day. It is also complete only when the bar closes, and two bars with the same
count look identical whether their quotes arrived evenly or in one burst.

The library works on the waits between consecutive quote changes instead.
`CAcdEvents` turns the tick stream into events, folding ticks that share a
millisecond and marking waits longer than five minutes as a closed market.
On a retail feed these are quote changes, not trades, and nothing here
pretends otherwise. `CAcdDiurnal` learns the time-of-day profile on the
training period only, files each wait under the time it started, interpolates
between half-hour bins, and scales the profile so an adjusted wait of 1 means
"exactly as long as usual for this hour".

What remains still clusters, and `CAcdModel` fits the autoregressive
conditional duration model of Engle and Russell (1998) to it by maximum
likelihood, with exponential or Weibull surprises and omega pinned by mean
targeting. Every fit is scored on a held-out period against a rhythm-only
baseline that gets its own best Weibull shape, so the recursion receives no
credit for dispersion it did not explain. On six symbols the clustering beat
that baseline out of sample on every one, and on EURUSD it removed 99.3% of
the Ljung-Box statistic of the adjusted waits.

`CAcdState` runs the fitted model live in constant time per quote, through
the same one-step update and the same stored profile as the fit, so a state
replayed over the fitting history reproduces the model's expectations
exactly. Its main reading is the activity ratio: 1 at the normal pace for the
hour, 2 when quotes are expected twice as often, 0.5 when half as often. A
silence ratio covers the gap between quotes, when the expectation cannot move.

`ACD_Evidence.mq5` asks whether any of this helps forecast the volatility of
the next few minutes, with a ladder of regressions scored on a later test
period, at the bar close and at a varying second inside each bar. Mid-bar,
every live measure beats tick volume from finished bars on all six symbols.
Most of that gain, however, is freshness rather than activity: live realised
volatility alone, measured up to the same second, beats the ACD gauge alone
everywhere. Activity still adds something prices do not. With live
volatility already in the forecast, adding the gauge improved it on all six
symbols at both horizons, by a small margin that is largest on the indices.
Whether the gauge or plain rhythm-adjusted rolling counts add more depends on
the market and the horizon: the gauge at five minutes and on the indices,
the counts at one minute on currencies and gold.

## Layout

```
Include/ACD/AcdTypes.mqh              parameters, fit report, buffer map, the one-step update and log-density
Include/ACD/AcdEvents.mqh             tick stream to events: same-millisecond folding, session gaps, daily loading
Include/ACD/AcdDiurnal.mqh            time-of-day profile with interpolation and unit-mean scaling
Include/ACD/AcdModel.mqh              maximum-likelihood fit, rhythm-only baseline, residual diagnostics
Include/ACD/AcdState.mqh              live constant-time state: activity ratio, surprise, expected wait, silence
Scripts/ACD/ACD_Calibrate.mq5         fits one symbol: feed check, rhythm, both fits, residuals, live parameters
Scripts/ACD/ACD_Evidence.mq5          multi-symbol volatility-forecast ladder, at the bar close and mid-bar
Indicators/ACD/ACD_Activity.mq5       live activity gauge with a tick volume contrast and buffers for EAs
```

Run `ACD_Calibrate.mq5` first, on the symbol you care about with thirty days
of quotes. It stops if the feed looks timer-driven, since a clustering model
fitted to a clock describes the broker's server rather than the market. Check
that the clustering gain on the test period is positive, then paste the
printed alpha and beta into `ACD_Activity.mq5`. The indicator learns its own
rhythm from the trading days before the plot. Recalibrate about once a month
and whenever you change symbol or broker.

An Expert Advisor should read buffer 3, the plain activity ratio, through
`iCustom`. Buffer 4 holds the expected wait in milliseconds for the next
quote on the forming bar; divide the time since the last tick by it for a
silence ratio of your own, since timer events do not reach an indicator
created through `iCustom`.

## Disclaimer

Educational code. The gauge measures market conditions and says nothing about
direction, and nothing here demonstrates a trading edge. The evidence covers
thirty days of quotes from one broker, the gains from activity over live
volatility are small, and the ranking between the gauge and simpler counts
shifted with horizon and asset class within that window. Test on your own
account and broker feed before relying on any of it.