Introduction
The Particle Filter method for prediction is a sequential Monte Carlo technique that uses a set of random samples, or "particles," to represent the probability distribution of a dynamic system's state, updating these particles based on observed data to improve predictions over time.
To Buy or rent PARTy, use this link.
Complete user manual and a set file are included in the zip file attached at the end of this blog post.
- PARTy is an expert advisor that uses the Particle Filter algorithm to predict the trend and make the decision of buy or sell.
- The Particle Filter algorithm is a Monte Carlo approximation of the probability density function (PDF) of the future symbol price. The advantage is that it does not make any assumptions on the shape of the PDF, which makes this algorithm a more robust predictor than others that assume normality, or other type of probability function.
- PARTy creates a predictive model based on up to 200 past candles (user input), and updates its prediction with the arrival of each new candle.
- PARTy is multi-timeframe. It makes a prediction in a higher time frame (HTF) depending on the user input, while trades are made in a lower time frame (LTF).
- If the HTF prediction is above the current price with a good confidence level (user input), a BUY order is sent.
- If the HTF prediction is below the current price with a good confidence level (user input), a SELL order is sent.
-
Think of it like a ship's captain navigating the ocean:
-
The HTF Particle Filter is the large-scale nautical chart. It tells you the direction of the main ocean current (the underlying trend). It ignores the small, choppy waves (the LTF price changes). Based on this chart, the captain decides the overall course: "We are heading northeast (the trend)."
-
The LTF Chart is the view from the ship's bridge. The crew uses this view to navigate around smaller waves or other boats, perhaps making slight adjustments (buy or sell entries). But they do all of this while still maintaining the general "northeast" course set by the captain (the trend).
-
- PARTy trades ANY Forex symbol, ANY metal symbol. Use major ones: combinations of USD, GBP, JPY, EUR, AUD, CAD, XAU.
-
User may run any number of symbols in parallel by running an instance of this robot on each symbol.
-
PARTy does not use grid, martingale, or hedging strategies. Don’t be fooled by other EAs that show exponential growth in back testing.
-
Account type: Netting or Hedging. It always use Netting rules.
-
Initial Balance, lot size and leverage are subject to your broker's limits.
-
User level required: from beginner to expert traders. It requires user to know the MetaTrader 5
Optimization tools as we recommend to optimize parameters before running live. -
Users can optimize variables to get strategies with high annual return, low risk, high win rate, high expected payoff, high probability of success, and more, based on the training data.
-
All time frames are available. Preferred time frames are below one day (D1).
-
All trades are opened with a Stop Loss automatically. Never lose more of what you risk in a given trade.
-
User may select which days of the week to trade.
-
User may select a time window for trades (start/end hours).
-
User may select a maximum number of bars to keep the position open.
-
User may select a specific time of the day to close the position.
-
There are multiple methods for opening Take Profit (TP) calculation.
-
There are multiple methods for the opening and trailing Stop Loss (SL) calculation.
-
There are multiple methods to handle large changes in price (positive or negative).
Methodology
The following image describes the iterative nature of the Particle Filter method. Particles model the price and speed of price, and each particle has a weight. At the beginning, historic data is used and particles are randomly spread for a large amounts of prices. A prediction is made and compared with measurements (new bars). From the difference between the predictions and the measurements, weights are updated, and a resampling of particles is performed. As more new bars arrive, new predictions are proposed, and the process continue. Predictions are made in a higher time frame to smooth them.
