1) The document summarizes 6 research papers on trading strategies and algorithms.
2) The papers propose and test various automated trading strategies using techniques like expert advisors, deep reinforcement learning, technical analysis rules and machine learning models.
3) The strategies are backtested on major currency pairs and show profits outperforming simple buy and hold strategies in most cases.
1) The document summarizes 6 research papers on trading strategies and algorithms.
2) The papers propose and test various automated trading strategies using techniques like expert advisors, deep reinforcement learning, technical analysis rules and machine learning models.
3) The strategies are backtested on major currency pairs and show profits outperforming simple buy and hold strategies in most cases.
1) The document summarizes 6 research papers on trading strategies and algorithms.
2) The papers propose and test various automated trading strategies using techniques like expert advisors, deep reinforcement learning, technical analysis rules and machine learning models.
3) The strategies are backtested on major currency pairs and show profits outperforming simple buy and hold strategies in most cases.
Finamatrix Journal Automated Gold Trading with MT4 Domain/Subdomain and Contribution Domain : Trading Strategy Subdomain : Expert Advisor Contribution : Proposed a framework to improve the performance and accuracy of a strategy by using EA and back test on MT4 Material & Methods Expert Advisor is an automated trading bot. An algorithm is designed and it backtested for several months. Experiment & Result Obtained The algorithm settings include the following but no limited to risk per trade, maximum total risk, maximum account risk, stop loss pips, take profit pips, start hour, end hour etc. A low spread level of less than 10 pips will improve test results while a spread level of more than 30 pips can reduce test results. A successful expert advisor with short term intraday trades should be able to be successful during period of lower spreads The test result shows the profitability of this strategy, nearly 4 times profits in 9 months with 479 trades. The result also displays a limited winning percentage of 45% in total trades Conclusion The tests have provided evidence that investors are able to use micro accounts to create high returns on gold trading with similar EAs. The low percentage of profit trades does not translate to low returns. Nearly or less than 50% of profit trades is enough, provided that the average profit trade is greater than the average loss trade, as well as most of the profit trades achieve more absolute profit than most of the loss trades. Paper # 2 Journal Name & Paper Title Physica A Short-Term Predictions in Forex Trading Domain/Subdomain and Contribution Domain : Forex Trading Subdomain : Kinetic Equation Contribution : Trading Algorithm on major 3 pairs Material & Methods Forex Currency Pairs EURUSD, EURCHF, USDCHF Apply a recent model in kinetic theory that is used to model trubelance. As far as this is the first application of this kinetic approach to a financial system. Experiment & Result Obtained In experiment the kinetic equation used the trading live data pairs like USDCHF, EURCHF and after successful analysis the model makes the prediction for the buy. In result model shows much accuracy. Conclusion There are more currencies than this model can cope with. Any set of three currencies may be used, and it is possibl to develop a methodology that uses many sets of three currencies to cover more ground. There will be redefined models for the short term evolution of the stock market. Paper # 3 Journal Name & Paper Title ABAC Journal Vol. 39 Robust Forex Trading With Deep Q Network Domain/Subdomain and Contribution Domain : Forex Trading Subdomain : Deep Learning, Machine Learning & Q Learning Contribution : Developing a trading system which have adaptive capabilities and synchronicity with the market and compare with traditional trading strategies. Material & Methods Forex Trading 6 currency pairs which were EURUSD, GBPUSD, USDCAD, USDJPY, USDCHF and AUDUSD. The appropriate method to study how machine learns to trade is to translate financial trading problems to a reinforcement learning problem and then train the computer through a Deep Q Learning algorithm. Experiment & Result Obtained In experiment it maps reinforcement learning to financial trading. It performs 4 functions : Set of States, Set of Actions, Reward Function and Experience Tuple. After performing 2 experiments on 2 currencies with total historical data of 15 years. The assumptions are
· The initial capital of 100,000USD
· No transaction cost
· The position sizing is 1% for each trade
· One position can be opened at a time
· We enter using the close price of that dayThe AI agent’ s
performance is significantly superior to the buy and hold performance Conclusion This study makes several contributions to academics, such as the application of artificial intelligence in algorithmic trading systems development. It is a desirable method to replace the human decisionmaking system because the computer can read hidden profitable price patterns better than a human and a computer can execute the trade swiftly and accurately, compared to a human who tends to perform a suboptimal decision-making process when they trade. The AI is the best candidate to replace humans in this situation. Paper # 4 Journal Name & Paper Title Managerial Finance Investigating the Profitability of Technical Analysis System on FX Domain/Subdomain and Contribution Domain : Forex Trading Subdomain : Technical Trading Analysis and Rules Contribution : Profit Testing System Material & Methods Data series for this paper consists of daily spot exchange rates for USD/DM and USD/BP. Short-term, monthly Eurodollars, Euromark, and Eurosterling interest rates are also used. The technical analysis software used is “Windows on WallStreet (version 2.1.2)”. Data analysis is carried in two steps. First, the profitability of specific technical systems is calculated for the whole period.Then, these profits are compared to “buy and hold” trading strategy. Experiment & Result Obtained In considering the USD/DM exchange rate over the whole period, all technical rules used were profitable. More specifically, all variants of the MACD system were profitable while, from the momentum rules only, one had losses. For the USD/BP rate, over the whole period and for fifty technical rules, 43 were profitable and 7 unprofitable. This time series shows big losses for the second sub-period, a fact that leads us to consider more carefully the persistent profitability of technical analysis Conclusion Empirical results in this study do not lead to a clear-cut answer regarding the profitability of using technical analysis in foreign exchange markets. Firstly, technical rules used were profitable over the whole period of the investigation. Secondly, by comparing the profitability of these rules against a simple “buy and hold” strategy, the former still lead to profits. Paper # 5 Journal Name & Paper Title Knowledge And Information System Transferring Trading Strategy Knowledge to DL Models Domain/Subdomain and Contribution Domain : Trading Strategies Subdomain : Deep Learning Contribution : Transfer the knowledge of real algorithmic trading strategies to Neural Network models for generating signals. Material & Methods Information from the financial market may consist of qualitative information, sentiment of the news articles and attributes of an asset price movement. The Quantitative methods are used as the source of learned trading signals, but the presented methods can be applied on any source of signals, as long as the information that was taken into consideration to produce them is contained within the provided input to the model. Experiment & Result Obtained The collected data is implemented on the model by using SMA (Simple Moving Average) long and short. Whenever short SMA cross the long SMA it generates the buy signal. The CNN and LSTM model is tested on both training and test data while both performance are same on training data but LSTM performs better than CNN on test data. Conclusion The knowledge from production strategies that emit trading signals is successfully transferred to an LSTM model. Although in this instance the employed strategies can be algorithmically derived from the OHLC price time series, the presented model can be applied in cases where that is not true, such as strategies directly generated by human traders. This can be useful for companies to ensure that a successful trader’s behavior can be simulated by such model to continue the same trading activity if they become unavailable. The proposed model is compared to another deep learning model, namely a convolutional neural network (CNN) which it surpasses at all performance metrics. Paper # 6 Journal Name & Paper Title ETASR(Engineering,Technology and Applied Science Research) Modeling And Trading the EURUSD Exchange Rate Using Machine Learning Techniques Domain/Subdomain and Contribution Domain : Forex Trading Subdomain : EURUSD Subdomain : Machine Learning Contribute : Producing Trading Strategy using KNN Material & Methods The European Central Bank (ECB) publishes a daily fixing for selected EUR exchange rates.The reference exchange rates are published both by electronic market information providers and on the ECB’s website shortly after the concentration procedure has been completed. The Naive Strategy, MACD Strategy and Machine Learning Model(KNN) Experiment & Result Obtained The currency pair EURUSD data is implemented on all models. But the KNN shows the most accurate result as compare to traditional trading strategies. Conclusion Implementing a variety of machine learning techniques in the problem of modeling and trading with the EURUSD exchange rate. From all the applied machine learning techniques, Random Forests has not been applied in this problem again while being one of the most accurate classifiers. The machine learning techniques were benchmarked with two traditional trading strategies Naïve strategy and MACD strategy. Paper # 7 Journal Name & Paper Title Expert System with Applications A Forex Trading Expert System Based on a new approach to the rule base evidential reasoning Domain/Subdomain and Contribution Domain : Foreign Exchange Market Subdomain : Technical Analysis System Contribution : Propose a forex trading expert system based on some new technical analysis indicators and a new approach to the rule-base evidential reasoning. Material & Methods A proposed system will be implemented on the data collected from the Foreign Exchange Market. Experiment & Result Obtained The proposed Forex trading system was implemented with the use of a Forex trading platform MetaTrader 4, on the base of data from Investor online FX.Generally, the developed trading system include such parameters as Stop Loss, Take profit and special parameters that enhance the quality of decision. It is well known that even very successful trading systems providing excellent results at the stage of optimization, usually produce good or satisfactory results only during some limited (testing) time after optimization period. The developed Forex trading system based on the proposed new approach to the rule-base evidential reasoning may be successfully used in practice for different currency pairs and time frames. Conclusion A new approach to the rule-base evidential reasoning (RBER) is proposed.Since the combination rules play a key role in RBER, a comparative analysis of them is provided. A real Forex trading system was developed. It is shown that this system provides good results and may be successfully used in practice for different currency pairs and time frames. Paper # 8 Journal Name & Paper Title IEEE Transactions On Neural Networks Computational Learning Techniques For Intraday FX Domain/Subdomain and Contribution Domain : Forex Trading Subdomain : AI Contribution : Proposed a trading algorithm Material & Methods Genetic Algorithm is the application of artificial intelligence (AI)techniques to technical trading and finance has experienced significant growth. Reinforcement learning has so far found only a few financial applications. Consider trading rules defined in terms of eight popular technical indicators used by intraday FX traders. The indicators using are the price channel breakout, adaptive moving average, relative strength index, stochastics, moving average convergence/divergence, moving average crossover, momentum oscillator, and commodity channel index. Experiment & Result Obtained The current RL implementation requires about eight minutes CPU time on a 650 MHz Athlon per single training optimization . The GA is implemented in the interpreted language Scheme, but evaluation is parallelised over multiple similar CPUs. It also takes about eight minutes CPU time per optimization on a single machine. The Markov chain and heuristic approaches execute in four seconds and approximately four minutes, respectively Conclusion In this paper developed three trading strategies basedon computational learning techniques and one simple heuristic based on trading thresholds over a fixed horizon. The strategies based on the genetic (programming) algorithm (GA) and reinforcement (Q-)learning train at 15-min intervals on the buy sell signals from eight popular technical trading indicators some of which require a number of previous observations and current positions over a one year period of GBPUSD FX data Paper # 9 Journal Name & Paper Title Journal Of Computers Forex Trading Robot With Technical And Fundamental Analysis Domain/Subdomain and Contribution Domain : Forex Trading Subdomain : Trading Bot Contribution : This paper build forex robot that analyze forex market based on fundamental analysis. Fundamental news periodically extracted from a website that provide forex news calendar www.forexfactory.com. Material & Methods The data will be collect from the forexfactory through the trading bots and then they execute trade according it. There are 2 automated trading bots; Technical Robot and Fundamental Robot. Technical Robot works based on the crossing of moving average. Robot Rules for buy position 1. Wait for the currency to trade above both the 50 SMA and 100 SMA. 2. Exit trade when the price breaks below the 50 SMA by 10 pips. Robot Rules for sell positon 1. Once the price has broken below the closest SMA by 10 pips or more. 2. Exit trade when the price breaks above the 50 SMA by 10 pips. On other side Fundamental Robot the algorithm of the trading robot works based only on fundamental analysis. This robot analyze news from a news website (www. forexfactory.com). This robot use the news to gauge market sentiment that will move the market price. News influences people’s bullish or bearish sentiment on a particular market, which in turn creates volatility as those people buy and sell. Experiment & Result Obtained Some experiments were conducted to compare the performance attained by the fundamental robot and the technical robot. To obtained this purpose, three of the most dynamic currency pairs in forex trading were used EURUSD, USDJPY, and GBPUSD. The testing duration for all of these experiments is 30 days. The performance charts of technical robot form rising linear curve, while the performance charts of fundamental robot form jagged saw curve. This means that the performance of technical robot is more likely to be stable in profiting compared with fundamental robot. Conclusion Experiments show that the performance of technical robot is more robust and stable compare with fundamental robot. The reason is because there are a lot of factors that affect the news, like market’s sentiments, multiple news in one time, countinuos effect of the previous news. This makes fundamental robot can’t depend only on the forecast, previous, and actual value from the news website, but also other factor such as the affect of the previous subsequence news. On the same timeframe, the technical robot result the more profitable balance than fundamental robot. In this re-search, the technical robot makes trading decision based on the statistics of the latest market movement. Paper # 10 Journal Name & Paper Title Journal Of Risk And Financial Management Take Profit And Stop Loss Trading Strategies Comparison in Combination with an MACD Trading System Domain/Subdomain and Contribution Domain : Forex Trading Subdomain : Automated Trading System Contribution : Implement a combination of an adaptive MACD Expert Advisor that uses back-tested optimized parameters per asset with price levels defined by the ATR indicator, used to set limits for Stop Loss. In this research, the researcher tested and compared six different TP and SL strategies used in combination with an algorithmic. Material & Methods Trading Systems, Automated Trading, Private Investors, Take profit and Stop loss. Trading System will automatic execute trade with mentioned tp and sl. It's an intelligent system it create using 6 different strategies. The system knows when enter into a trade or when exit from a trade. The system detects which technique to incorporate with an existing MACD strategy and which to avoid. Experiment & Result Obtained Conducted our experiments on the different Take Profit – Stop Loss strategies for nine assets from the Forex, Metals, Commodities, Energy, and Cryptocurrencies categories: AUDUSD, EURGBP, EURUSD, GBPUSD, USDCHF, USDJPY, XAUUSD, OIL, and BTCUSD. Firstly compared the results for the simple MACD Expert Advisor with the default parameters and without holding positions over weekends, the simple MACD Expert Advisor with our selected parameters and without holding positions over weekends, the simple MACD Expert Advisor with our selected parameters with holding open positions over the weekends It can be concluded that holding open positions over the weekends helps cut one’s losses, but it also takes away a proportion of the profits for all the examined assets. Also, using the default parameters almost always results in losses. Conclusion In this research, it is examined that various Take Profit and Stop Loss strategies added to a simple MACD automated trading system used in trading 10 assets from the Forex, Metals, Energy, and Cryptocurrencies categories. In order to make the MACD parameters less important in our research, we chose parameters based on the characteristics of their neighborhoods of ±2 and used them for all our experiments.