Lazy Trading Part 7: Developing self-adapting Trading System

Refresher of Big Strategy, Decision Support System...

Types of algorithmic systems. In this course we will look into Model based as well as Hybrid system!

Comparison of proposed method of Pattern Recognition vs 'conventional' algorithmic trading that implies Optimization

Theory about how to get and transform data to train models, verify them and perform predictions

Work on going

Getting the code

How to Clone code for R package lazytrade:

- get link from GitHub.com

A. Using GitHub Desktop (see Course 1)

- use link in menu File [Clone, from URL]

B. Using R-Studio

- use link in R Studio [create new project from version control...]

In this video we are using R Studio to get the code from GitHub.

In this video we demonstrate how to create R Studio project inside existing folder!

GitHub desktop can act like a hub of all [Version Control] projects that you might have

but how to add the projects that were created using other applications?

Before you can use modified R package it is necessary to 'build' or 'compile' package:

• use Description file to define a new version

• run code: devtools::document() to create documentation for the package

• menu Build -> More -> Clean and Rebuild

• optional: menu Build -> More -> Build Source Package [this will build package archive that someone else could use to install this package]

Function helps to 'compress' code into one functional block. We show how to understand this code by looking into function source code stored in the folder 'lazytrade/R'

This lecture features that R package allows automated test and checks:

• Write tests

• Define examples

• Use build menu to run tests + checks

• Check Readme.rmd document within lazytrade package for examples and instructions

There are two goals of this section:

1. Make sure latest current data is kept updated

2. Deploy a mechanism to aggregate latest data and keep it for later step (modelling)

Explain use of DataWriter _ v 6.01 to write data to files

purpose: this data will be used for 2 purposes:

1. use latest data for predictions

2. bring data for aggregation

• deploy robot Data Writer v 6.01 in H1 chart

• UseBarsCollect2200

• retrieve data timeframe 1 Hour

• Pair Selection 'All 28 pairs'

• check that the files are generated e.g. AI_RSIADX<SYMBOL>60.csv

Note: data will kept updated at every new bar (every hour)

In short, data writer robot is executing the function collectandwrite() that:

1. executed on the beginning of a new bar

2. function is executed for an array of symbols

3. we can specify our own symbols (e.g. Futures, Commodities, etc) using a comma separated string e.g. : "#DJ30_U20","#EUR50_U20","FRA40_U20","#GER30_U20","#JP225_U20","#UK100_U20","#US100_U20"

4. for every asset we have a combination of indicators on specific shifts (we use RSI and ADX)

5. we also save close price of the latest close bar as well as the price 34 bars ago (this is to be able to calculate price difference for predictions as well as for strategy test)

Modify user path and run this script line by line [code is in the repository R_selflearning/PROD1]

purpose: This script is needed to collect historical data separately for every asset. This data will be used to train deep learning regression model

execute code line by line using R package source code, find these elements:

• function 'fail safe' elements

• read data from csv file

• calculate 'LABEL' column

• 'trim' data amount

• write data into 'rds' format

deploy windows task scheduler:

• run this task 3 times x week

• recur task every 1 week

• at 21:50

• Building a new model

• Testing the model

• Rebuilding the model

• Analyse logs from the model performance

in this lecture we will be using function to make model

see script: 10_ForceModelUpdateM60.R

purpose: create deep learning models for the first time

Part1: Define variables of the function

• read purpose of these variables

• select and define them into Global Environment

Part2: Study Examples

• Best model using RMSE as performance

• Best model using 'Objective Test'

• Model with Fixed or Random Neural Network structure

Part3: Random Neural Network Structure

• Reproducibility of Neural Network modeling is usually bad

• We write model into sub-folders

• Select best model from sub-folder...

How to create logs:

• log date and time

• log time it takes to build model

• log amount of data

Deploy script: 06_TestBuildTestModelM60.R

purpose:

• perform strategy test on latest data

• rebuild model again

• repeat...

• measure the time it takes to build model

deep dive into the function aml_test_model - prepare

Creating variables to execute code within the function

Strategy test part

Recording logs from the strategy test

run script analyze_results1.R

purpose:

analyse logs of strategy tests on different order hold conditions

Example on how to find the value of the first quantile from the data column

goals of this section:

• deploy script to generate predictions

• understand how functions work

Deploy script 07_ScoreDataM60.R

purpose:

• apply model on new data

• write predicted price changes to the files

:: Use this code in 'production'

"C:\Program Files\R\R-4.0.3\bin\Rscript.exe" "C:\Users\<... user path .. >\R_selflearning\_PROD1\07_ScoreDataM60.R"

Function documentation and reproducible examples directly in R:

execute in R console

?lazytrade::aml_score_data

Key concepts of the Decision Support System Robot:

• Works from files (Entries, Filters)

• Contains ‘filter’ logic to enter

• Only executes trades, manages Stop Loss, Take Profit, etc

• Logging of data (results, store/recover data)

Introducing more robust robot with corrected and tested behaviour

How does it work - functional description using a block diagram

To test this robot it is possible to:

• use demo account;

• use M1 timeframe

• change content of files inside the terminal's sandbox

• check out function GetTradeFlagCondition() for exact entry conditions

• files examples for EURUSD charts are attached

Main logic of this robot is stored inside the function

Explanation how to log information from MQL4 variables to the flat files

Visual understanding why robot Falcon F2 uses arrays to keep information about opened positions

Brief introduction of the trading robot FALCON A

Repository with the code of the Trading Robot FALCON_A

Trying to reverse-engineer the trading robot

Visual understanding of knn algorithm

Review how to practically implement kNN in the DSS [Decision Support System]

This lecture is to review how new trading strategy with kNN integrated algorithm was implemented inside of the FALCON trading robot template

Added another trading robot adapted to work with Decision Support System

A little summary of the course... hopefully it was fun and interesting!