Lorentzian Classifier Indicator for NinjaTrader 8

Description

This classification model is built on the Lorentzian distance measure to find similarities between features extracted from a block of price bars. The Lorentzian distance is a special type of distance measure able to classify features in a multi-dimensional space — a novel way to implement Nearest-Neighbour algorithms. By using it, the indicator can find similar blocks of price action in historical training data and infer the next bar's outcome from a vote among the most similar historical features.

Here are the steps the indicator executes to predict the next price action outcome and attribute a score to that prediction:

• Consider a period of bars to analyze — a Prediction Period parameter sets how many of the most recent bars are used, and this sliding window progresses as new bars close.
• Extract features from those bars — not based on any indicator values, but on context-based Time, Price, and Volume values.
• Use the Lorentzian distance algorithm to find the most similar features among the training data loaded when the indicator is initialized.
• Take a number of the most similar samples — the count can be set manually via Voting Count, or determined automatically (via a clustering algorithm) when Voting Count is 0.
• Let those samples vote on the outcome — each votes whether the next candle will be bullish or bearish — and average the result into a predicted price movement, in points, for the next N candles.

Examples

Example 1: MES on the 1-Minute Timeframe

The indicator's simplest usage scenario, running on MES (Micro ES, the S&P 500) on the 1-minute timeframe, with a threshold value of 5 filtering out any signal scoring below 5 — leaving only high-probability signals, drawn as green arrows for bullish and red arrows for bearish. This example uses the RTH session (09:30-16:30 EST); time of day has a big impact on signal quality since the classification algorithm is context-based on price, time, and volume, and MES/US futures volume is much higher during RTH than during ETH (extended trading hours).

Example 2: MES on the 1-Minute Timeframe

With the same parameters, more signal arrows appear — some of good quality, some irrelevant (e.g. a green bullish arrow drawn during a bearish downtrend). Low-quality signals can usually be dismissed by checking the very next bar after the signal, while some signals do successfully predict a genuine change of market regime and trend.

Example 3: Composite Signal — a Range Based on Two Consecutive Signals

The indicator can produce a more complex composite signal from a sequence of two consecutive arrow signals close to each other and forming an increasing (bullish) or decreasing (bearish) price action — defining a price range that can be traded as a breakout entry setup, both on the bullish and bearish side.

Parameters

We always try to keep the parameters to a minimum.

Download & Installation Instructions

Installation Steps

1

Download the per-instrument training data files (one archive per instrument family, e.g. MES/ES, MNQ/NQ, MYM/YM, M2K/RTY, MCL/CL, MGC/GC) into a folder of your choice. Do not extract these files; this folder is what you'll set in the indicator's Data Folder parameter.

Sign in to download this file.

2

Import the downloaded .zip file into NinjaTrader using the Import NinjaScript menu item.

3

Open a new chart window.

4

Right-click the chart and choose Indicators... to add the indicator to the chart, pasting your license (found on the billing page) as you do.

5

After adding the indicator to the chart, it runs on the chart's historical data first, then continues to run on live market data.

Frequently Asked Questions

General

Release Notes