A Level 2 signal says the book looks lopsided or a wall just vanished. It does not say whether price is likely to move your way in the next moment. BookPilot's Micro-Forecast adds that second opinion: a probability that the next 2-tick move of the mid price is up, used as a gate before any order is sent.
What the forecast estimates
The forecast produces one number, P up: the estimated probability that the mid price's next 2-tick move is upward. P up of 0.62 means the model leans towards an up move; 0.38 leans down; 0.50 means it has no opinion.
It is a logistic model, the same family of model used for many yes-or-no estimates. It combines four inputs from the book and the tape:
- Queue imbalance at the touch: how the size at the best bid compares with the size at the best ask.
- Microprice offset: where a size-weighted price sits between the bid and the ask.
- Recent flow: the buy/sell balance of the last Flow Window prints, 50 by default.
- The Pressure Break ratio from the scanner.
How it adapts to your instrument
The forecast starts with built-in weights. After about 1,600 labelled samples on your instrument it fits its own weights, then refits every 200 new samples. The fitted model is saved per instrument when the strategy stops, in Documents\NinjaTrader 8\VantedgeAlpha\forecast_<instrument>.txt.
Two practical consequences:
- Judge the filter only after a few sessions of data on that instrument. Early on it is running on the starting weights.
- Delete the file if you want it to start from scratch, for example after changing Flow Window, which changes one of its inputs.
The confirmation gate
With Forecast Confirm on (the default), a signal from either scanner must clear two checks before it reaches the Rules of Engagement:
| Setting | Default | What it requires |
|---|---|---|
| Confirm Threshold | 0.55 | The forecast must give the signal's direction at least this probability |
| Confirm Min Vectors | 0 (off) | How many of the 4 trend vectors must point the signal's way |
The trend vectors are the slope of the mid price over the last 5, 20, 40 and 150 price changes. Requiring two or more of them keeps BookPilot trading with the short-term trend and blocks counter-trend signals.
For a long signal at the default threshold, P up must be 0.55 or more. For a short, P up must be 0.45 or less, because the probability of a down move is 1 minus P up.
Turning the dial
- Raising Confirm Threshold demands stronger agreement and cuts the trade count. The guide notes that above about 0.7 trades become rare.
- Lowering it towards 0.5 makes the filter almost neutral. Most signals pass.
- Turning Forecast Confirm off sends every signal that passes the Rules of Engagement to the market. Pulled Wall trades can then fade the book pressure, because nothing checks them against it.
The guide's suggested range for Confirm Threshold is 0.52 to 0.65. Its Conservative profile uses 0.60 and its Active profile 0.52.
The forecast's other jobs
The same P up feeds two more parts of BookPilot:
- Microstructure Bias. At or above Bias Threshold (0.55) the bias is Long, at or below 0.45 it is Short, otherwise None. In With Bias mode, signals against the bias are refused.
- Early exits. With Risk Tolerance set to Low or Medium, BookPilot can exit when the forecast for your position falls below the Risk Tolerance Threshold (0.40 by default). The default Risk Tolerance is High, which leaves exits to the bracket.
Diagnosing it from the panel
When a signal is refused, the panel's Blocked line says why. If you mostly see Forecast Confirm or against the bias, the filters are strict for current conditions: lower Confirm Threshold slightly, or raise Bias Threshold so the bias is None more often. Make one change at a time and replay the same Market Replay days afterwards. Every setting mentioned here, with its range, is in the BookPilot guide.
Keep expectations sensible. The forecast is a short-horizon estimate built from a few book features. It shifts the odds you are working with; it does not know where price is going.
A probability filter does not remove the risk of trading futures. Test changes to it in Market Replay and SIM before trading an evaluation, funded or live account.