Ultra-short term trading using a neural-network based ensemble of financial technical indicators in a closed world market

Abstract

The proposed paper presents the analysis, design, implementation and evaluation of an ultra-short-term frequency trading system for the foreign exchange market (FOREX), which features all stages of the trading process (Pre-Trade Analysis, Trend Forecasting, Trade Execution). The system uses artificial intelligence techniques in an environment that is constantly changing according to the decisions of the participating trading simulators. Our goal is to simulate the judgment and decision-making of the human expert (technical analyst or broker) in a closed world trading system that constantly adjusts exchange rates. We examine the system in terms of its contribution to how exchange rates are affected in an environment which solely consists of traders, without any exogenous factors, and show that our system can outperform all conventional technical indicators, thereby indicating that it could also play the role of facilitating the stabilization of a market by squeezing out non-intelligent traders. We designed and implemented a self-adjusting trading environment whose exchange rate price is only affected by trading within the environment (closed world assumption). We based our work on a modified series of technical indicator simulators, which are fed to an artificial neural network architecture, to eventually generate the trend forecasting signal and a series of customizable ultra-short term automated trading machines, which perform real-time virtual transactions based on the generated forecasting signals. A comparative analysis of the results is carried out to confirm that the proposed architecture outperforms traders based only on the conventional technical indicators while we also document the behavior of the system towards facilitating the attainment of an equilibrium.

Keywords

Foreign exchange technical analysis neural networks trend forecasting self-adjusting environment

1. Introduction

The largest share of foreign exchange market profits, especially foreign exchange (FOREX) market profits [1], derives from extended margin-based leveraging [2]. Leverages as high as 1–200 (i.e. someone with an initial capital of 1000 € is allowed to risk a capital of 200.000 €) are a source of high risk for trading (even within a few minutes). Therefore, it has been argued that forecasting models as well as the accompanying systems of algorithmic trading [3] should be based on short time periods.

Economists have been trying for several decades to build models for successfully forecasting trends. These efforts gave birth to the field termed as technical analysis. Despite a large number of long and arduous attempts no indicator or model of a sufficiently generic nature has been developed so far that can forecast with great success the trend of the financial markets. The main reason for this is that technical analysis does not take into account the most recent changes of fundamentals, which have not been yet recorded, nor the effect of breaking news on the psychology of investors. Moreover, examining technical analysis indicators and their usage in trading [4] requires so much time that the appreciation of short term (sometimes, within seconds) changes in exchange rates is rendered practically impossible. In markets of considerable depth and volume, such as the FOREX one, exploiting micro-changes within a minimum time frame is of paramount importance and can be achieved with ultra-short term trading [5].

In our previews work [6] we have designed and built a tick-to-tick ultra-short term trading system, which includes all stages of the trading process, namely, pre-trade analysis, production of the transaction mark (trend prediction) and execution of transactions, for ultra-short term transactions in the FOREX market. The system is fully customizable and has been built using a lean object-oriented approach. This has allowed us to simultaneously test a large number of automated trading machines, which, in turn, drew predictive data from a series of technical indicator simulators, as well as from a neural network system attempting to online learn how to correctly forecast a new technical indicator. The dataset consisted of over 10 million data points and we have fed it through a set of 32 automated trading machines, each one featuring a different combination of forecasting signal source and trading parameters. Based on the results of our extensive experimentation, we have concluded that, for a large collection of appropriately set parameters, the implemented ultra-short term trading system performs fully autonomously at a (simulated) profit when tested with real data [6].

The aim of this paper is to examine whether our system can perform satisfactorily in a self-adjusting trading environment whose exchange rate is only affected by transactions within the environment, thus allowing us to investigate whether this environment can be driven to equilibrium (The exchange rate changes little or not at all).

The rest of this paper is structured in four subsequent sections. We first briefly review related work on predicting exchange rates using computational intelligence and on data generation. Then, we describe the system architecture and we proceed with presenting and analyzing the experimental results on the two main questions, that of relative performance and that of the attainment of an equilibrium. We conclude in the last section, where we also set out a couple of future work.

2. A brief background on predicting exchange rates using computational intelligence

As earlier mentioned, FOREX traders do use technical analysis tools [7] to predict exchange rates but higher profits are normally achieved only by automated systems [3], which trade huge sums of money based on forecasting models. However, automated systems also tend to follow the “avalanche” model [5] by training each other, and thus to re-enforce ascending or descending trends. In shallow markets, such models may lead to significant distortions. A typical example is the stock market crash of Monday October 19, 1987 (Black Monday), with the resulting sell-off in the S&P 500 and the Dow Jones generating a price fall in excess of 20%, where a particular automated trading system exacerbated the sell-off by attempting to hedge a portfolio of stocks against market risk by short-selling stock index futures. As it automatically began to sell stocks, as stop-loss targets were hit, it triggered a domino effect of other programs following suit and, with falling prices triggering further stop-loss orders, a vicious cycle ensued.

Conventional wisdom in trading is being captured by technical analysis, with methods having resulted both in successes and failures. Failures are usually due to undetected changes in fundamental values and market psychology and forecasting inaccuracies tend to increase with shorter-term forecasting [9].

Forecasting methods are broadly divided into two categories, traditional and non-traditional. Traditional ones are usually based on static algorithms, which are not altered and are not influenced by input data [11]. Basically, these are econometric models which help us interpret the results. Furthermore, they allow hypothesis control, which is a standard quality assurance procedure in technical analysis [11].

Non-traditional methods include all methods which are based on data and auto-correct themselves [11]. Such methods are based on fuzzy logic [12], on Artificial Neural Networks (ANN) [8], neuro-fuzzy architecture (hybrid systems) [13] and genetic algorithms [14]. Non-traditional methods can be quite competitive to econometric methods, due to the generalized operations they perform [15]. While they can function as general models, they do not guarantee satisfactory results; nevertheless, they are usually better than conventional models where data is associated with linear relationships [15] especially when modelling the market response.

Machine-learning based methods have been extensively used for predicting trading patterns and are now considered strong enough to deal with FOREX forecasting based on past trading data [8].

Still, the hidden layers of ANN systems represent an internal representation of relations between variables and, as a result, they do not satisfy certain pre-requisites required by palindromic models, such as variability between epochs, smoothness of background noise etc. Neural networks can also perform quite well in cases of sparse data, as opposed to regression models where serious problems arise [16]. Moreover, neural networks are suited to complex phenomena for which satisfactory performance measures do exist but there is limited knowledge of the relationships between these phenomena. They are also relatively successful for forecasting and prognosis [16]. Additionally, genetic algorithms have been also used to learn trading rules and, then, combined with an echo-state network to predict the market trend [14], with results demonstrating better results both on bull and bear markets compared to the usual buying and holding strategy of retail traders.

We now briefly review some key contributions to the field of predicting exchange rates, using ML techniques.

Yong et al. [17] examined the closing price as well as the various closing price technical indicators to determine their impact on the FOREX trend provided by n ANN model.

Abraham et al. [18] attempted to compare the performance of a Takagi Sugeno Type neuro-fuzzy system to that of a neural network to predict the average monthly exchange rate values. The Australian dollar was used as the basis and its exchange rates with the US dollar, the Singapore dollar, the New Zealand dollar, the yen and the pound sterling. It was reported that the proposed models were able to predict average exchange rates but only when dealing with a time horizon of at least one month.

Wang et al. [19] have shown that a neural network with three layers combined with an Auto regressive integrated moving average mode outperforms the global modeling techniques in terms of profit returns.

Vanstone and Finnie [20] presented a methodology for the design of robotic trading systems using artificial neural networks. They outlined the key steps (selecting inputs and outputs, partitioning available data, determine architecture, setting threshold and stop trading signals, real world constraints and benchmarking) for building a neural network to be used in stock trading.

A hybrid model was developed by Ni and Yin [21], consisting of a mix of different neural network models (mostly unsupervised learning) and simulators of technical indicators to predict exchange rates. The model uses some of the most popular technical indicators such as the moving average [7], convergence/deviation, and the relative resistance index (RSI) [7]. A genetic algorithm is used to “mix” the forecasting of the generators and the technical neural networks, producing the overall system forecasting.

The combination of neural networks with technical analysis, and, in particular, the relative resistance index (RSI), has been used to improve the trading systems [22] of the IBEX-35 stock index.

Further experiments with two other indicators, the weighted moving average and the momentum [7], have shown that, when fed into a technical neural network, they produce a better forecast compared to the original (i.e. the forecast produced by the indicator before being introduced into the artificial neural network) [23].

Khirbat et al. [24] studied the forecasting of stock market prices by feeding the time series of stock prices to a multi-layer back propagation neural network which also attempts to deal with other non-deterministic input, such as the Earning per Share value and a public confidence index.

Sermpins et al. [25] used a hybrid Neural Network Optimization and Adaptive Radial Basis Function (ARBF-PSO) to implement a leverage negotiation strategy based on the Glosten, Jagannathan and Runkle (GJR) variability forecasts. By comparing the results of the ARBF-PSO with those of three different architectural neural networks, a Neighbor Neighborhood (k-NN) algorithm, a Moving Average Model (MA) and a Moving Average Convergence/Deviation Median (MACD) for the EUR/JPY, EUR/GBP andEUR/JPY exchange rates, for the period January 1999 – March 2011 (daily closing prices from the ECB), the ARBF-PSO has been shown to be superior to other models in terms of statistical accuracy and transaction efficiency.

A demonstration of how to efficiently approach the problem of automated trading with a large portfolio strategy that continuously consumes streams of data across multiple diverse markets appears in [10], where a simple scalable trading model that learns to generate profit from multiple inter market price predictions and markets’ correlation structure is presented.

In our previous work, we designed and implemented an ANN, which tries to predict market signals in the money market [8] based on an input which consists of a series of econometric models (technical indicators). Essentially, the ANN “corrected” the econometric model, combining the advantages of technical analysis and ANN in causal modeling and case control. It is important to note that the ANN wasn’t trained with a set of data and then tested with another data set but, rather, it was trained in real time and produces its price forecast in real-time, being constantly trained throughout its life (at the resolution of one price per minute).

At the next stage we presented [26] the analysis, design, implementation and evaluation of a neural network based system for forecasting in the foreign exchange (FOREX) market. We aimed to simulate the judgment and decision making of the human expert (technical analyst or broker) with a system that responds in a timely manner to changes in market conditions, thus facilitating the optimization of ultra-short term transactions. We designed and implemented a series of technical indicator simulators, which are fed to a novel neural network architecture to generate the trend forecasting signal.

Although the aforementioned works do highlight the advantage of these novel architectures, compared to various previous models and to classic technical indices, they have not been applied to automatic trading systems. Thus, it is unclear whether their implementation can be translated to real market profits.

To solve the problem, we also designed and implemented a series of customizable ultra-short term automated trading machines [6], which receive as inputs the generated forecasting signals and perform real-time virtual transactions. A comparative analysis of the results of both automated trading machines and each machine is carried out for a comprehensive variety of trend forecasting sources.

In a self-adjusting trading environment, the field of data generation plays an important role, so we now briefly review some key contributions to this field.

The process of data generation plays a significant role in various areas of computer science. An appropriate data generator is suitable and necessary for almost every type of testing (including automated): the regression tests, null value tests, coverage, security and performance test. With the rise of data science, the data generation is as well used in machine learning, data mining, data visualization, financial industry etc. Important aspect of the generated data is that the data needs to be realistic but not real, which embrace the confidentiality and privacy. Popić et al. [27] they gave a short survey on the different types of generators from the architecture point of view and their intended usage, as well as they listed their pros and cons. Finally, they gave an overview of the used data generation algorithms and the best practices in different areas.

Liu et al. [28] they proposed a data generator to produce synthetic data set which can be as big as the experiments demand that the size of attributes, rules, and samples of the synthetic data sets can be easily changed to meet the demands of evaluation on different learning algorithms. In the generator, related attributes are created at first. And then, rules are created based on the attributes. Samples are produced following the rules. Three decision tree algorithms are evaluated used synthetic data sets produced by the proposed data generator.

Ramos and Rego [29] they propose a methodology which uses the original data, as a template, to generate candidate datasets, to finally accept only those datasets which resemble the template, based upon parameterized features.

Kondratyev and Schwarz [30] used a special type of generative neural networks – a Restricted Boltzmann Machine (RBM) – to build a powerful generator of synthetic market data that could replicate the probability distribution of the original market data.

Raberto and Cincotti [31] presented an artificial dual auction financial market populated by heterogeneous agents who exchange a risky asset for cash. Agents issued random orders subject to financial constraints. It appeared that the waiting times of orders are exponentially distributed, and even the waiting times of trades are also exponentially distributed.

Chen et al. [32] proposed an artificial financial market, under the market matching hypothesis, using artificial intelligence algorithms. This market included three types of agents with different investments and risk preferences, representing the heterogeneity of traders. Genetic network programming was combined with a SARSA( $\lambda$ )( $\lambda$ ) (state-action-reward-state-action) algorithm to design the market to reflect the adaptation of technical agents. A simplified expression of the Chinese stock exchange auction mechanism is used as the pricing mechanism.

Yagi et al. [33] used agent-based simulations to compare the major liquidity indicators in an artificial market where a high-frequency trader participated was compared to one where no high-frequency traders participated. The results showed that all liquidity indicators in the market where a high-frequency trader participated improved more than those in the market where no high-frequency traders participated.

All research, as summarized above, investigated the behavior of an artificial financial market. It is notable that a comparison of the behavior of autonomous agents participating in the artificial financial market with the behavior of the same autonomous agents when participating in a closer-to-the-real-world scenario (where, for example, market prices are drawn from real, global financial market sources) has been lacking.

In this paper we design and implement an exchange rate price generator, whose prices will be derived based on the buy or sell orders given by the various participating traders in our closed trading system (this is why we refer to our system as “closed”: its transactions are modify a simulated exchange rate which, in turn, generates or applies to new transactions). Our goal is to investigate the behavior of our architecture [6] in such a closed trading system (self-adjusting trading environment). To do that, we compare the performance of simulated autonomous traders, which use technical indicators as prediction signals, to traders which rely on our architecture to receive prediction signals.

3. A detailed system description

In this section we describe how we analyzed, designed and implemented an algorithmic ultra-short term trading system, comprising all stages of the (price) negotiation process and Transaction Execution as described in [6] which exists and operates into a closed auto-adjusting trading world. This system includes the data generator which feed the auto-adjusting closed trading world and the fundamental stages of Pretrade Analysis, Transaction Signal Production (Trend Forecasting), Transaction Execution. An overview of the system architecture is shown in Fig. 1, which we will also use as a reference for describing various system components.

First, we designed and implemented the closed trading world where all traders live and operate. The orders (buy and sell) from these traders determine the price of the exchange rate.

At the analysis stage the data used in the subsequent steps are selected.

Then, at the trend forecasting stage, a series of technical indicator simulators [6] were designed and implemented to generate the trend forecasting signal. These signals are fed as inputs to the artificial neural network system (Fig.1), which also yields a trend forecasting signal [6] with appropriate design changes in parameterization to be consistent with the attempted ultra-short term trading.

At the transaction stage [6], a series of customizable high-frequency automated trading machines were designed and implemented, which receive as inputs the forecasting signals generated in the previous stage and perform real-time virtual transactions. Next, a comparative analysis of the results of both automated trading machines and each machine is made for the different sources of trend forecasting (the different technical indices and the ANN).

3.1 “Closed” trading world and experimental data source

We chose to experiment with the EUR/USD rate so as to render our results comparable to previous work.

Figure 1.

An overview of the system – architecture.

Figure 2.

A snapshot of the simulation.

The first stage of our work is the designing and implementation of the closed trading world in which the traders participating in it will live. In this world, the exchange rate price generator will generate prices based on the buy and sell orders that the closed world will receive from the traders participating in it.

The “closed” world of eur/usd consists of$10,000,000,000 in value. Our basic assumption is that every dollar trading order in the world affects the world proportionally: (trading order amountin dollars)/10,000,000,000,000. In the “closed” world there are 32 traders with $5,600,000 in total capital. The remaining capital of $9,994,400,000 consists of the value of the world’s fundamentals. The fundamentals during the experiment are considered stable. The above simplified hypothesis serves the purpose of this paper which does not aim to simulate the real world currency market.

In our previews work [6] we have experimented with the tick-to-tick EUR/USD exchange rate of July, August and September 2020. The data initially consists of over 10 million values, which decrease after pre-processing in order to squeeze out flat areas (where the exchange rate does not change at all). In order to make the experiments comparable (with our previews work [6]) we used the same time data (day, time) for which our “closed” world algorithm generates its own price data every time. Each dollar sell order increases the eur/usd exchange rate by 0.1 billionths. Similarly, each buy order to buy a dollar decreases the eur/usd exchange rate by 0,1 billionths.

3.2 Prediction stage

We chose some technical indicators for our experiments [7], which are consistent with the short-term forecast (Fig. 1), namely the arithmetic moving averages (MA) of 600, 900 and 1800, the RSI-600 oscillator, the CCI-600 oscillator, the Williams-600 oscillator and the price oscillator (MA-600, MA-900, MA-1800) based on our previous work [6].

These technical indicators are computed and produce a forecast as described in Annex 1.

The system accepts exchange rate, time and date as inputs (Fig. 1) and, based on the predicted trend signal and the configurations of its auto-trading agents, simulates ultra-short term trading and produces performance logs for simulating profit or loss. We are using the purchase price and the selling price as input, to the accuracy of five decimal places, and we also use a tick-to-tick frequency [35].

For the first 20,000 values the system works exactly as in our previous work [6] with real eur/usd exchange rate data, as extracted after processing by truefx. This is done in order to train both the technical indicators and the neural network system with real data before they themselves start to influence the course of the “closed” world. That is, we examine what would happen if suddenly the eur/usd exchange rate world suddenly started to be influenced solely by the traders participating in our system.

3.3 Transaction stage

This part includes a series of high frequency automated trading mechanisms which perform virtual transactions with real data. It uses as input the forecasting from the previous level, the price of the exchange rate and it manages a number of simulated transaction machines, each with a different configuration (to be presented below), producing for each of them a detailed log with all (simulated) performed transactions and obtained financial results.

All input data (date, time, purchase price and sales price, as well as the trend forecasting of the technical indicators simulators and the technical neural network system) are fed to a series of high frequency automated trading machines.

Automated trading machines only take into account the trend forecasting signals ( $-$ 1, $+$ 1, $-$ 1.5, $+$ 1.5, $-$ 2, $+$ 2) to determine their strategy. Each trading machine is described using a set of parameters, as described in our previews work [6].

Automated trading machines display in real time the following data for each transaction.

1.
The opening time of the transaction position with the accuracy of a second.
2.
The opening price of the transaction position.
3.
The Type of position (short or long) [35, 36].
4.
The intensity of the trend forecasting signal ( $-$ 1, $+$ 1, $-$ 1.5, $+$ 1.5, $-$ 2, $+$ 2) based on the opening of the position.
5.
The time of closing the position with the accuracy of a second.
6.
The closing price of the transaction position.
7.
The profit or loss of the transaction.

It is worth noting that the opening of a long position [35, 36] takes place at the buying price and its closing at the selling price. Correspondingly, the opening of a short position [35, 36] takes place in the sale price and its closing at the buying price. In other words, the result of the ultra-short term simulated algorithmic transactions include the spreads [1] between the buying and selling prices of the exchange rate.

Table 1
ANN parameterization

A/A Parameter Value

1 Number of ANN Epochs 10

2 Number of ANN Hidden Neurons 14

3 Learning rate of synapses between Hidden layer Neurons and Input Layer Neurons (LR-Inputs) 0.001

4 Learning rate of synapses between Hidden layer Neurons and Output Layer Neurons (LR-Output) 0.001

5 Number of Hidden Layers 1

6 Number of Output Neurons 1

7 Number of Input Neurons 7

8 Period (in a number of values) of auxiliary MA 10

9 “Actual Trend” Algorithm Selection Parameter (1: Based of table 3.4 or 2: Based of table 3.3) 1

Table 2
Parameterization of back-propagation ANN’s

A/A Parameter ANN-1 ANN-2 ANN-3

1 M(x) (In a number of prices approx. 1price $=$ 1sec) 30 60 90

2 Trend Value $\pm$ 2 1,0090 1,0090 1,0090

3 Trend Value $\pm$ 1.5 1,0060 1,0060 1,0060

4 Trend Value $\pm$ 1 1,0030 1,0030 1,0030

5 Trend Value $\pm$ 0.5 1,0015 1,0015 1,0015

6 $a(x)$ 0.50 0.25 0.25

Table 3
Parameterization of technical indicators simulators

A/A Parameter Value

1 Period (in a number of values) Oscillator RSI 600

2 Period (in a number of values) Oscillator Williams 600

3 Period (in a number of values) Oscillator CCI 600

4 Period (in a number of values) Short-Term MA 600

5 Period (in a number of values) Mid-Term MA 900

6 Period (in a number of values) Long-Term MA 1800

7 Auxiliary Moving Averages of Price Oscillator (600, 900, 1800)

3.4 System development and use

A/A	Parameter	Value
1	Number of ANN Epochs	10
2	Number of ANN Hidden Neurons	14
3	Learning rate of synapses between Hidden layer Neurons and Input Layer Neurons (LR-Inputs)	0.001
4	Learning rate of synapses between Hidden layer Neurons and Output Layer Neurons (LR-Output)	0.001
5	Number of Hidden Layers	1
6	Number of Output Neurons	1
7	Number of Input Neurons	7
8	Period (in a number of values) of auxiliary MA	10
9	“Actual Trend” Algorithm Selection Parameter (1: Based of table 3.4 or 2: Based of table 3.3)	1

A/A	Parameter	ANN-1	ANN-2	ANN-3
1	M(x) (In a number of prices approx. 1price $=$ 1sec)	30	60	90
2	Trend Value $\pm$ 2	1,0090	1,0090	1,0090
3	Trend Value $\pm$ 1.5	1,0060	1,0060	1,0060
4	Trend Value $\pm$ 1	1,0030	1,0030	1,0030
5	Trend Value $\pm$ 0.5	1,0015	1,0015	1,0015
6	$a(x)$	0.50	0.25	0.25

A/A	Parameter	Value
1	Period (in a number of values) Oscillator RSI	600
2	Period (in a number of values) Oscillator Williams	600
3	Period (in a number of values) Oscillator CCI	600
4	Period (in a number of values) Short-Term MA	600
5	Period (in a number of values) Mid-Term MA	900
6	Period (in a number of values) Long-Term MA	1800
7	Auxiliary Moving Averages of Price Oscillator	(600, 900, 1800)

The system has been developed in Java using the Apache NetBeans IDE 11.2. The application is fully parameterizable by means of an appropriately labeled Type parameter file (.fxipf standing for FX Intelligence Parameter File), whose contents are shown in Annex 1.

Figure 2 shows a desktop snapshot when executing the application. We view the central simulation screen {1}, the activated automated trading machines {2}, and the summary of aggregated results of automated trading machines {3}.

On the displays of automated trading machines, we see in real time the transactions and all their data and elements.

At the end of the simulation, all transaction records are exported and stored in the data folder.

4. Experimentation and results

The values of the parameters chosen for all neural networks (back-propagation and feed-forward series) are the same as in our previous work, so that the results are comparable (Table 1).

As in our previous work, the series of back-propagation and feed-forward ANNs consist of three ANNs each. We have also six parameters of each back-propagation ANN as shown in Table 2.

The parameter values for the technical analysis simulators are also similar (Table 3).

There are thirty-two (32) parameterized automated trading machines in the experimentation, evenly allocated to the ‘0’ class (opposite signals close each other’s position) and ‘1’ class (opposite signals do not close each other’s position).

In class “1” machines, the maximum waiting time should be limited, otherwise we are at risk of staying with several positions open at the end of the session. Half of the machines of each class (0 or 1) are of a higher sensitivity (value 1) and the other half of a lower sensitivity (value 10), giving rise to four combinations, overall, which are codified as machines of type I through IV (numbering in 1atin).

Based on the above, we group the machines into four general configurations (general machines or machine types) that we will now call general machines (or just machine Types). Each technical indicator simulator (there are seven altogether) and the forecasting machine of the ANN feeds a separate forecasting signal to a machine of each Type. Therefore, we activate 4 $\times$ (7 $+$ 1) $=$ 32 automated trading machines to simulate.

All values of the parameters of the thirty-two automated trading machines to be used in the context of experimentation are shown in Table 4.

Table 4
Parameterization of automated transaction machines

N/A		General type I	General type II	General type III	General type IV

1	Machine Sensitivity (1000 to 0).	1	1	10	10
2	Machine Class (0 or 1).	0	1	0	1
3	Take Profit Factor for trend signals of very high intensity $\pm$ 2.	1.00090	1.00090	1.00090	1.00090
4	Take Profit Factor for trend signals of high intensity $\pm$ 1.5.	1.00060	1.00060	1.00060	1.00060
5	Take Profit Factor for trend signals of normal intensity $\pm$ 1.	1.00030	1.00030	1.00030	1.00030
6	Stop Loss factor.	1.0012	1.0012	1.0012	1.0012
7	Revision Time of Take Profit factor.	60	60	60	60
8	Maximum Waiting Time.	120	120	120	120
9	Revised Take Profit Factor.	1	1	1	1
10	Initial Capital ($)	50000	50000	300000	300000
11	Capital per Transaction ($)	10000	10000	10000	10000
12	Maximum Leverage Ratio	1:1	1:1	1:1	1:1
Forecasting Trend Sources		Price	Price	Price	Price
ANN $\to$ 0, PriceOsc(600,,900,1800) $\to$ 2, CCI-600 $\to$ 2, Williams-600 $\to$ 3, RSI-600 $\to$ 4 MA-600 $\to$ 5, MA-900 $\to$ 6, MA-1800 $\to$ 7,		0 to 7	0 to 7	0 to 7	0 to 7
Total Machines:		8	8	8	8

Table 5

Performance benchmarks in current “closed world” vs our previous work with real data [6]

	Current “closed world”				Previous work with real data [1]
General machine types with ANN	Type I	Type II	Type III	Type IV	Type I	Type II	Type III	Type IV
% Days with profit per machine	100%	100%	100%	100%	56%	57%	63%	63%
Average net profit per transaction	$2,607	$2,528	$2,926	$2,846	$0,193	$0,181	$0,173	$0,160

Figure 3.

Total experimental results per machine and forecasting trend source.

Figure 4.

Total monthly experimental results per transaction per machine and forecasting trend source.

Figure 5.

Days with profit per machine.

Figure 6.

Total profit per forecasting source in all machines (Total pips vs Average profit per transaction).

Figure 7.

Total profit per machine type (total pips vs average profit per transaction).

Figure 8.

Percentage gain in investment capital.

Figure 9.

The comparator sequences for each general machine type in relation to the source of the forecasting trend signal.

In our experimentation we have no leverage.

4.1 Presentation and analysis of results

As we describe in Section 3.1, the change in exchange rate prices depends entirely on the movements of the 32 traders participating in our “closed world”. Each trader has a certain amount of initial capital that he can allocate to his trades (Table 4). When a trader’s remaining capital is not sufficient for one trade ( $<$ $10,000), that trader withdraws. The more traders withdraw, the fewer transactions that take place in our “closed world” market, the smaller the exchange rate fluctuations. If a tipping point is passed, then so few trades take place that the market in our “closed world” becomes essentially flat.

The tipping point occurred on the simulated 11th day of transactions on 13 July 2020, when 13 of the 28 traders (excluding the ANN traders) had withdrawn. The smaller the fluctuations in the exchange rate, the more difficult it is for the remaining traders to place buy or sell orders, with the result that even smaller fluctuations feedback on themselves. In other words, even traders who had retained capital to trade, eventually did not do so because they were not receiving a forecast) from the technical indicators that guided them) which would trigger the placement of an order.

We notice that ANN machines were the only ones with a profit at the end of 13 July 2020. All machines fed with the technical indicators forecast had losses. Thirteen of them lost almost all of their initial capital.

The best performance in absolute profit ($ 139,551) is achieved by the Type I machine which uses the ANN as a source of forecasting, followed by the Type II machine, again with the ANN as a forecasting source($ 139,086) (Fig. 3). Among the top four we observe the Type III and Type IV machines with the ANN as forecasting source. It is noteworthy that the first places in terms of performance are occupied by the ANN. All the others forecasting sources fare quite badly. It is worth recalling that ANNs accept as inputs the forecasts of the technical indicators that feed their machines. Although these predictions lead their machines to significant losses, when fed into the ANN they lead to significant gains. That is, the ANN architecture essentially corrects the erroneous predictions of the technical analysis.

In terms of average earnings per transaction, the best performance is achieved by the Type I machine with ANN ($ 2,926 per transaction), followed very closed by the Type II machine with ANN ($ 2,846 per transaction) (Fig. 4).

The last entries to the top four are the Type III machine with ANN, the Type IV machine with ANN. We notice that all machines with any other technical indicator have losses.

Type I, II, III and IV machine with ANN forecasting are profitable all days within the total period of experimentation (11/11) (Fig. 5).

Figure 6 shows the total profit per forecasting source for all machines, in total pips and at average profit per transaction. ANN turns out to be the most efficient with $ 310,594. All technical indicators underperform and lose the entire investment capital in most cases. In terms of average profits, the ANN (2,849 $/transaction) outperform the profit of all others in absolute figures.

Figure 7 shows the total profit per machine Type and the average profit per transaction for the ANN forecasting source.

Figure 8 shows the comparative monthly profit rates of the four general types of Automatic Trading Machines based on the ANN forecast.

Figure 9 shows the cumulative profit for each general machine type and the forecasting signal of ANN.

The ANN in the general type IV machine behaves identically to the type III machine. Similarly, the ANN in the general type I machine behaves identically to the type II machine.

We note that the slopes of the lines decrease after July 2. A further obvious decrease in the slope of the lines is observed on the 8th and the 10th of July. This is because, after July 2, the traders who have as trend forecast the various technical indicators begin withdrawing and as a result the volatility in the exchange rate is decreasing and therefore the trend forecast signals are weaker. After 10 July the exchange rate changes are small and the trades that take place are similar, while on 13 July the exchange rate seems to be hovering within such narrow limits that it is no longer possible to generate further buy or sell signals.

In our previous work [6] with the same architecture and ANN system configuration, the performance of ANN-based traders is inferior to the performance of ANN-based traders in our current work. Table 5 shows the performance benchmarks between them.

The above observation is very important. In a closed world where prices are only affected by the technical analysis we use as input to our ANN system, our architecture performs extremely well both relative to the original technical analysis and to our architecture itself when applied to the real market. This is a clear indication that if more sophisticated forecasting indicators are used as inputs to our architecture which are weighted towards the trend of the real currency market, the forecasting of our architecture will outperform.

By way of introduction, we remind you that MAs are trend indicators, i.e. they are indicators that follow the following general rule: “If the trend is bullish/bearish for the time period X0 to X, it will be bullish/ bearish for the time period X $+$ 1”. In contrast, oscillators detect the ‘fatigue’ of the existing trend by giving signals against it, considering the market overbought or oversold. If we focus on ANN-based machines, we see that they make profits both when MAs based machines make profits and when Oscillators based machines make profits and when both lose. That is, our architecture picks the right side, every time, even when all other existing traders lose. That is, it uses the technical indicators as both indicators and anti-indicators with great success.

Annex 2 features the detailed results of the experimentation per day, machine and source of forecasting.

5. Conclusions

We simulated the judgment and decision-making of the human expert (technical analyst or broker) in a closed trading system (trading world) that adjusts exchange rates and examined the strengths of the system in terms of the weight of its contribution to pricing of exchange rates in an environment that is not affected by other factors than the traders participating in it. Specifically, we examined whether the applied architecture [6] can outperform technical indicators in a closed trading system (unaffected by other exogenous factors) and whether this leads to an ability to adjust from towards a volatile exchange rate towards an equilibrium exchange rate.

The change in exchange rate prices depends entirely on the movements of the traders participating in our “closed world”. Each trader has a certain amount of initial capital that it can allocate to its transactions. When a trader’s remaining capital is not sufficient for one transaction, that trader withdraws. The more traders withdraw, the fewer transactions take place in our “closed world” market, and the smaller the exchange rate fluctuates. If a tipping point is passed (most traders have withdrawn), then so few transactions take place that the market in our “closed world” becomes essentially flat.

The best performance in absolute profit is achieved by automated traders which uses the ANN as a source of forecasting, All the others forecasting sources seem to be quite weak. It is worth recalling that ANNs accept as inputs the forecasts of the technical indicators, each of which feeds a corresponding trading machine as well. Although these predictions lead the corresponding machines to significant losses, when collectively fed into the ANN they lead to significant gains. That is, the ANN architecture essentially corrects the wrong predictions of the technical analysis.

ANN architecture based traders make profits both when MA-based machines make profits, when Oscillator-based machines make profits and when both lose. That is, our architecture picks the right winning side, every time, even when all other existing traders lose. This is an indication that if more sophisticated forecasting indicators are used as inputs to our architecture, which may manage to closely follow the trend of the real currency market, the performance of our architecture could further improve.

It is evident that by opening up our implementation and our research in such a way, one expects that we can pursue research at a variety of levels: we can investigate how one produces industrial-quality trend prediction, while also researching how one can utilize such trend predictions, either for profit-making enterprises or as a tool for tilting a market towards an equilibrium for hedging purposes.

Future work should focus on examining alternative ANN training techniques as well as investigating the exploitation of conventional fundamentals, break events and news which affect the FOREX market and might be announced at any time point during a 24-hours day. Such data can be harnessed from the internet in real time. Also, different neural networks models (such as ARIMA and LSTM [34]), with various input data set combinations (technical indicators predictions) will be considered and compared.

Footnotes

Annex 1. System parameters

Conditions	Trend Forecasting Signal
MA_M(t) $<$ MA_10(t) && MA_M(t $-$ 1) $\geqslant$ MA_10(t $-$ 1)	$+$ 1
MA_M(t) $>$ MA_10(t) && MA_M(t $-$ 1) $\leqslant$ MA_10(t $-$ 1)	$-$ 1
MA_M(t) $<$ MA_10(t)	$+$ 0,5
MA_M(t) $>$ MA_10(t)	$-$ 0,5
Other Cases	0
Simulators of Moving Averages
MA_M(t): Moiving Average of M values, MA_10: Moving Average of 10- values
Oscillators Simulators
Conditions of CCI	Trend Forecasting Signal
CCI(t) $<$ $-$ 150 && CCI(t) $<$ CCI(t $-$ 1) && CCI(t $-$ 1) $<$ CCI(t $-$ 2) && CCI(t $-$ 2) $<$ CCI(t $-$ 3)	$+$ 2
CCI(t) $>$ 150 && CCI(t) $>$ CCI(t $-$ 1) && CCI(t $-$ 1) $>$ CCI(t $-$ 2) && CCI(t $-$ 2) $>$ CCI(t $-$ 3)	$-$ 2
CCI(t) $<$ $-$ 150	$+$ 1,5
CCI(t) $>$ 150	$-$ 1,5
CCI(t) $<$ $-$ 100	$+$ 1
CCI(t) $>$ 100	$-$ 1
CCI(t) $<$ CCI(t $-$ 1) && CCI(t $-$ 1) $<$ CCI(t $-$ 2) && CCI(t) $<$ 0	$+$ 0,5
CCI(t) $>$ CCI(t $-$ 1) && CCI(t $-$ 1) $>$ CCI(t $-$ 2) && CCI(t) $>$ 0	$-$ 0,5
Other Cases	0
Conditions of Williams	Trend Forecasting Signal
WILL(t) $<$ $-$ 99 && WILL(t) $<$ WILL(t $-$ 1) && WILL(t $-$ 1) $<$ WILL(t $-$ 2) && WILL(t $-$ 2) $<$ WILL(t $-$ 3)	$+$ 2
WILL(t) $>$ $-$ 1 && WILL(t) $>$ WILL(t $-$ 1) && WILL(t $-$ 1) $>$ WILL(t $-$ 2) && WILL(t $-$ 2) $>$ WILL(t $-$ 3)	$-$ 2
WILL(t) $<$ $-$ 99	$+$ 1,5
WILL(t) $>$ $-$ 1	$-$ 1,5
WILL(t) $<$ $-$ 98	$+$ 1
WILL(t) $>$ $-$ 2	$-$ 1
WILL(t) $<$ $-$ 80	$+$ 0,5
WILL(t) $>$ $-$ 20	$-$ 0,5
Other Cases	0
Conditions of RSI	Trend Forecasting Signal
RSI(t) $<$ 5 && RSI(t) $<$ RSI(t $-$ 1) && RSI(t $-$ 1) $<$ RSI(t $-$ 2) && RSI(t $-$ 2) $<$ RSI(t $-$ 3)	$+$ 2
RSI(t) $>$ 90 && RSI(t) $>$ RSI(t $-$ 1) && RSI(t $-$ 1) $>$ RSI(t $-$ 2) && RSI(t $-$ 2) $>$ RSI(t $-$ 3)	$-$ 2
RSI(t) $<$ 5	$+$ 1,5
RSI(t) $>$ 90	$-$ 1,5
RSI(t) $<$ 15	$+$ 1
RSI(t) $>$ 85	$-$ 1
RSI(t) $<$ 30	$+$ 0,5
RSI(t) $>$ 70	$-$ 0,5
Other Cases	0
Conditions of Price Oscillator	Trend Forecasting Signal
PROSC(t) $<$ $-$ 12	$+$ 2
PROSC(t) $>$ 12	$-$ 2
PROSC(t) $<$ $-$ 9	$+$ 1,5
PROSC(t) $>$ 9	$-$ 1,5
PROSC(t) $<$ 6	$+$ 1
PROSC(t) $>$ $-$ 6	$-$ 1
PROSC(t) $<$ 0	$+$ 0,5
PROSC(t) $>$ 0	$-$ 0,5
Other Cases	0

Line	Parameter	Value
Content of the system’s configuration file
1 ${}^{\text{st}}$	Number of ANN epochs	10
2 ${}^{\text{nd}}$	Number of ANN Hidden Neurons	14
3 ${}^{\text{rd}}$	Learning rates between synapses of Neurons of Hidden Layer with Input Neurons (LR-Inputs)	0.001
4 ${}^{\text{th}}$	Learning rates between synapses of Neurons of Hidden Layer with Output Neurons (LR-Output)	0.001
5 ${}^{\text{th}}$	Number of ANN Hidden layers	1
6 ${}^{\text{th}}$	Number of Exit Neurons	1
7 ${}^{\text{th}}$	Number of Import Neurons	7
8 ${}^{\text{th}}$	Period (in a number of values) of Oscillator RSI	600
9 ${}^{\text{th}}$	Period (in a number of values) of Oscillator Williams	600
10 ${}^{\text{th}}$	Period (in a number of values) of Oscillator CCI	600
11 ${}^{\text{th}}$	Period (in a number of values) of Short-Term MA	600
12 ${}^{\text{th}}$	Period (in a number of values) of Mid-Term MA	900
13 ${}^{\text{th}}$	Period (in a number of values) of Long-Term MA	1800
14 ${}^{\text{th}}$	Period (in a number of values) of auxiliary MA (used instead of a pair of instantaneous values so that there is no possible momentary deviation of values due to tick to tick data)	10
15 ${}^{\text{th}}$	Number of ANN pairs	3
16 ${}^{\text{th}}$	M(x) (In a number of prices approx. 1price =1sec)	30
17 ${}^{\text{th}}$	Trend Value $\pm$ 2	1.00090
18 ${}^{\text{th}}$	Trend Value $\pm$ 1,5	1.00060
19 ${}^{\text{th}}$	Trend Value $\pm$ 1	1.00030
20 ${}^{\text{th}}$	Trend Value $\pm$ 0,5	1.00015
21 ${}^{\text{th}}$	a(x)	0,5
22 ${}^{\text{th}}$	Number of Automated Trading Machines.	32
23 ${}^{\text{th}}$	Machine Sensitivity (1000 to 0). Paragraph 3.4.1.	1
24 ${}^{\text{th}}$	Machine Class. Paragraph 3.4.1.	0
25 ${}^{\text{th}}$	Take Profit Factor for trend signals of very high intensity $\pm$ 2. Paragraph. 3.4.1.	1.00090
26 ${}^{\text{th}}$	Take Profit Factor for trend signals of high intensity $\pm$ 1.5. Paragraph. 3.4.1.	1.00060
27 ${}^{\text{th}}$	Take Profit Factor for trend signals of normal intensity $\pm$ 1. Paragraph. 3.4.1.	1.00030
28 ${}^{\text{th}}$	Stop Loss Factor. Paragraph. 3.4.1.	1.0012
29 ${}^{\text{th}}$	Revision Time of take profit factors. Paragraph. 3.4.1.	60
30 ${}^{\text{th}}$	Maximum Waiting Time. Paragraph. 3.4.1.	120
31 ${}^{\text{th}}$	Revised Take Profit Factor. Paragraph. 3.4.1.	1
32 ${}^{\text{th}}$	Source of Trend Forecasting Signals (0 – 7). Paragraph. 3.4.1.	0
33 ${}^{\text{th}}$	Total Investment Capital	$10000
34 ${}^{\text{th}}$	Capital per Transaction	$10000

Lines 16 through 21 are repeated as many times as the number of ANN pairs as set on the 15 ${}^{\text{th}}$ line. Lines 23 through 34 are repeated as many times as the number of machines as set on the 22 ${}^{\text{th}}$ line. So each machine has its own exclusive parameterization.

Annex 2. Analytical results of experimentation per day,machine and source of forecast

	Date
	1–Jul	2–Jul	3–Jul	5–Jul	6–Jul	7–Jul	8–Jul	9–Jul	10–Jul	11–Jul	13–Jul	Total
Type III (ANN)	856	2232	600	29	726	716	629	173	147	5	9	6122	Count of Trades
Type III (ANN)	2255	7916	906	42	1251	1393	1146	362	624	22	40	15957	Profit of Trades
Type I (ANN)	6720	18319	4438	204	5507	5468	4731	1185	1047	26	50	47695	Count of Trades
Type I (ANN)	19390	67564	8457	422	11995	12995	10458	3430	4513	120	207	139551	Profit of Trades
Type IV (ANN)	893	2250	654	30	768	750	649	175	147	5	9	6330	Count of Trades
Type IV (ANN)	2220	7950	885	47	1277	1429	1143	363	624	22	40	16000	Profit of Trades
Type II (ANN)	6969	18459	4738	206	5678	5685	4827	1189	1047	26	50	48874	Count of Trades
Type II (ANN)	18974	67804	8327	431	11980	12979	10320	3431	4513	120	207	139086	Profit of Trades
Type III (PROSC)	1265	2698	402	21	513	547	421	3082	0	0	0	8949	Count of Trades
Type III (PROSC)	$-$ 4899	$-$ 22858	1598	75	1884	1985	1475	$-$ 9688	0	0	0	$-$ 30428	Profit of Trades
Type I (PROSC)	12554	26946	3897	204	4964	5270	4080	30394	0	0	0	88309	Count of Trades
Type I (PROSC)	$-$ 46192	$-$ 223482	16795	781	19988	20790	15978	$-$ 94834	0	0	0	$-$ 290176	Profit of Trades
Type IV (PROSC)	1265	2698	402	21	513	547	421	3082	0	0	0	8949	Count of Trades
Type IV (PROSC)	$-$ 4899	$-$ 22858	1598	75	1884	1985	1475	$-$ 9688	0	0	0	$-$ 30428	Profit of Trades
Type II (PROSC)	12554	26946	3897	204	4964	5270	4080	30394	0	0	0	88309	Count of Trades
Type II (PROSC)	$-$ 46192	$-$ 223482	16795	781	19988	20790	15978	$-$ 94834	0	0	0	$-$ 290176	Profit of Trades
Type III (CCI)	928	2846	913	32	1061	1104	1194	3350	1488	104	1603	14623	Count of Trades
Type III (CCI)	$-$ 5885	$-$ 21401	3067	87	3532	3711	4158	$-$ 8694	6520	437	6316	$-$ 8152	Profit of Trades
Type I (CCI)	9187	28047	7971	301	9328	9625	9414	32131	6231	430	6617	119282	Count of Trades
Type I (CCI)	$-$ 58813	$-$ 220351	28391	819	32485	33559	32594	$-$ 88662	27875	1800	19244	$-$ 191059	Profit of Trades
Type IV (CCI)	932	2888	1100	42	1302	1326	1406	3373	1522	105	2358	16354	Count of Trades
Type IV (CCI)	$-$ 5913	$-$ 21333	3315	101	3821	3937	4408	$-$ 8657	6502	436	5652	$-$ 7731	Profit of Trades
Type II (CCI)	9207	28307	9192	360	10836	11039	10570	32267	6329	431	8502	127040	Count of Trades
Type II (CCI)	$-$ 59009	$-$ 220144	29550	915	33967	35192	34543	$-$ 88361	27961	1798	15846	$-$ 187742	Profit of Trades
Type III (WILL)	967	2702	453	18	516	511	456	3112	225	8	256	9224	Count of Trades
Type III (WILL)	$-$ 6354	$-$ 22958	975	7	1039	881	887	$-$ 9637	987	40	641	$-$ 33492	Profit of Trades
Type I (WILL)	9534	25735	0	0	0	0	0	0	0	0	0	35269	Count of Trades
Type I (WILL)	$-$ 61266	$-$ 229343	0	0	0	0	0	0	0	0	0	$-$ 290609	Profit of Trades
Type IV (WILL)	967	2711	488	22	566	540	495	3116	238	9	267	9419	Count of Trades
Type IV (WILL)	$-$ 6354	$-$ 22952	1024	4	1101	917	952	$-$ 9641	1012	45	652	$-$ 33240	Profit of Trades
Type II (WILL)	9535	25734	0	0	0	0	0	0	0	0	0	35269	Count of Trades
Type II (WILL)	$-$ 61276	$-$ 229343	0	0	0	0	0	0	0	0	0	$-$ 290619	Profit of Trades
Type III (RSI)	934	2593	24	1	25	25	29	3034	226	265	297	7453	Count of Trades
Type III (RSI)	$-$ 5706	$-$ 23330	34	2	34	50	45	$-$ 9507	515	560	634	$-$ 36669	Profit of Trades
Type I (RSI)	9219	25916	145	1	149	158	202	877	0	0	0	36667	Count of Trades
Type I (RSI)	$-$ 55082	$-$ 228908	271	2	278	420	322	$-$ 7990	0	0	0	$-$ 290687	Profit of Trades
Type IV (RSI)	934	2593	24	1	25	25	29	3034	327	361	417	7770	Count of Trades
Type IV (RSI)	$-$ 5706	$-$ 23330	34	2	34	50	45	$-$ 9507	423	525	544	$-$ 36886	Profit of Trades

	Date
	1–Jul	2–Jul	3–Jul	5–Jul	6–Jul	7–Jul	8–Jul	9–Jul	10–Jul	11–Jul	13–Jul	Total
Type II (RSI)	9219	25916	145	1	149	158	202	877	36667	0	0	73334	Count of Trades
Type II (RSI)	$-$ 55082	$-$ 228908	271	2	278	420	322	$-$ 7990	0	0	0	$-$ 290687	Profit of Trades
Type III (MA600)	1496	3223	2094	97	2612	2714	2630	3603	687	0	0	19156	Count of Trades
Type III (MA600)	$-$ 353	6468	$-$ 9056	$-$ 432	$-$ 10894	$-$ 11389	$-$ 10510	$-$ 1367	$-$ 2475	0	0	$-$ 40008	Profit of Trades
Type I (MA600)	14762	31676	18993	880	23696	24658	18898	0	0	0	0	133563	Count of Trades
Type I (MA600)	$-$ 2404	71171	$-$ 80162	$-$ 3554	$-$ 96153	$-$ 100601	$-$ 78450	0	0	0	0	$-$ 290153	Profit of Trades
Type IV (MA600)	1696	3611	3318	145	4079	4247	4249	4050	3815	244	4813	34267	Count of Trades
Type IV (MA600)	$-$ 492	6994	$-$ 7553	$-$ 350	$-$ 8729	$-$ 8915	$-$ 7903	$-$ 664	$-$ 3390	$-$ 186	$-$ 2445	$-$ 33633	Profit of Trades
Type II (MA600)	16603	34923	29457	1327	36275	37741	37377	39127	33159	2131	37593	305713	Count of Trades
Type II (MA600)	$-$ 3435	75103	$-$ 67003	$-$ 2948	$-$ 77393	$-$ 78597	$-$ 62567	27406	10578	873	10349	$-$ 167634	Profit of Trades
Type III (MA900)	1481	3195	2006	94	2482	2592	2526	3617	450	0	0	18443	Count of Trades
Type III (MA900)	$-$ 1070	6446	$-$ 8991	$-$ 419	$-$ 11005	$-$ 11507	$-$ 10471	$-$ 1317	$-$ 1666	0	0	$-$ 40000	Profit of Trades
Type I (MA900)	14452	31268	17835	834	22188	23044	16463	0	0	0	0	126084	Count of Trades
Type I (MA900)	$-$ 9337	71013	$-$ 78373	$-$ 3483	$-$ 95798	$-$ 100652	$-$ 73623	0	0	0	0	$-$ 290253	Profit of Trades
Type IV (MA900)	1691	3605	3308	147	4067	4236	4243	4055	3831	245	4824	34252	Count of Trades
Type IV (MA900)	$-$ 1005	7070	$-$ 7379	$-$ 316	$-$ 8505	$-$ 8845	$-$ 7865	$-$ 763	$-$ 3130	$-$ 186	$-$ 2531	$-$ 33455	Profit of Trades
Type II (MA900)	16603	34923	29457	1327	36275	37741	37377	39127	33159	2131	37594	305714	Count of Trades
Type II (MA900)	$-$ 9383	75159	$-$ 67484	$-$ 2789	$-$ 77253	$-$ 78663	$-$ 62761	27675	10824	852	10476	$-$ 173347	Profit of Trades
Type III (MA1800)	1567	3299	2344	110	2958	3053	2866	3597	1920	116	0	21830	Count of Trades
Type III (MA1800)	$-$ 1959	6901	$-$ 8260	$-$ 281	$-$ 9179	$-$ 9564	$-$ 8899	$-$ 1515	$-$ 6836	$-$ 408	0	$-$ 40000	Profit of Trades
Type I (MA1800)	15493	32388	21553	1027	27163	28005	26013	35241	16587	1069	4247	208786	Count of Trades
Type I (MA1800)	$-$ 18371	74356	$-$ 73709	$-$ 2445	$-$ 81549	$-$ 85397	$-$ 73905	17072	$-$ 35899	$-$ 2213	$-$ 7940	$-$ 290000	Profit of Trades
Type IV (MA1800)	1691	3605	3308	147	4067	4236	4243	4055	3831	245	4824	34252	Count of Trades
Type IV (MA1800)	$-$ 1882	7244	$-$ 6860	$-$ 291	$-$ 7790	$-$ 8203	$-$ 7196	$-$ 711	$-$ 3171	$-$ 190	$-$ 2463	$-$ 31513	Profit of Trades
Type II (MA1800)	16603	34923	29457	1327	36275	37741	37377	39127	33159	2131	37594	305714	Count of Trades
Type II (MA1800)	$-$ 17571	77006	$-$ 62740	$-$ 2567	$-$ 70758	$-$ 72954	$-$ 57420	25775	10967	872	10194	$-$ 159196	Profit of Trades

References

Copeland

. Exchange rates & international finance. Pennsylvania USA: Trans-Atlantic Publications, 6th edition, 2014.

Margin. Wikipedia: Margin (finance). 2022. http//en.wikipedia.org/wiki/Margin_%28finance%29.

Hanif

Elliott

. Algorithmic, electronic, and automated trading. The Journal of Trading. 2012; 7(4): 78-86.

Trader. Wikipedia: Trader. 2022. http://en.wikipedia.org/wiki/Trader.

Kumiega

Vliet

BEV

. Automated finance: The assumptions and behavioral aspects of algorithmic trading. Journal of Behavioral Finance. 2012; 13: 51-55.

Zafeiriou

Kalles

. Ultra-short-term trading system using a neural network-based ensemble of financial technical indicators. Neural Comput & Applic. 2021; 1-26. doi: 10.1007/s00521-021-05945-4.

Taylor

Allen

. The use of technical analysis in the foreign exchange market. Journal of International Money and Finance. 1992; 11: 304-314.

Zafeiriou

Kalles

. Short-term trend forecasting of foreign exchange rates with a neural-network based ensemble of financial technical indicators. International Journal on Artificial Intelligence Tools. 2013; 22(3): 1350016. doi: 10.1142/S0218213013500164.

Hirshleifer

Teoh

. Herd behaviour and cascading in capital markets: A review and synthesis. European Financial Management. 2003; 25-66.

10.

Ruta

. Automated trading with machine learning on big data. Anchorage(AK): IEEE International Congress on Big Data. 2014; 824-830.

11.

Bunn

. Non-traditional methods of forecasting. European Journal of Operational Research. 1992; 92: 528-536.

12.

Aladaga

Yolcu

Egrioglu

Dalar

. A new time invariant fuzzy time series forecasting method based on particle swarm optimization. Applied Soft Computing. 2012; 12(10): 3291-3299.

13.

Zhang

Wan

. Statistical fuzzy interval neural networks for currency exchange rate time series forecasting. Applied Soft Computing. 2007; 7: 1149-1156.

14.

Septiawan

Hayati

Kusuma

. Forecasting of currency exchange rate in forex trading system using genetic algorithm. International Interdisciplinary Conference on Science Technology Machineering Management Pharmacy and Humanities Held. 2017.

15.

Venugopal

Baets

. NeuralNetworks and statistical techniques in marketing reaserch: An conceptual comparison. Marketing Intelligence and planning. 1994; 12: 30-38.

16.

Chavarnakul

Enke

. Intelligent technical analysis based equivolume charting for stock trading using neural networks. Expert Systems with Applications. 2008; 34: 1004-1017.

17.

Yong

David

Lee

. Technical indicators for forex forecasting: A preliminary study. Advances in Swarm and Computational Intelligence. 2015; 87-97.

18.

Abraham

Chowdhury

Petrovic-Lazarevic

. Australian forex market analysis using connectionist models. Management Journal of Management Theory and Practice. 2003; 29: 18-22.

19.

Wang

Tang

Chai

. Exchange rate forecasting model analysis based on improved artificial neural network algorithm. IEEE International Conference on Communication Systems and Network Technologies. 2016.

20.

Vanstone

Finnie

. An empirical methodology for developing stock market trading systems using artificial neural networks. Expert Systems with Applications. 2009; 36: 6668-6680.

21.

Yin

, Exchange rate forecasting using hybrid neural networks and trading indicators. Neurocomputing. 2009; 72: 2815-2823.

22.

Rodríguez-González

García-Crespo

Colomo-Palacios

Iglesias

Gómez-Berbís

. CAST: Using neural networks to improve trading systems based on technical analysis by means of the RSI financial indicator. Expert Systems with Applications. 2011.

23.

Vargas

Anjos

Bichara

Evsukoff

. Deep leaming for stock market prediction using technical indicators and financial news articles. International Joint Conference on Neural Networks (IJCNN). 2018.

24.

Khirbat

. Optimal neural network architecture for stock market forecasting. IEEE International Conference on Communication Systems and Network Technologies. 2013.

25.

Sermpins

Theofilatos

Karanthanpoulos

Georgopoulos

Dunis

. Forecasting foreign exchange rates with adaptive neural networks using radial basis functions and particle swarm optimization. European Journal of Operational Research. 2013; 225: 528-540.

26.

Zafeiriou

Kalles

. Intraday Ultra-Short-Term Forecasting of Foreign Exchange Rates using an Ensemble of Neural Networks based on Conventional Technical Indicators. 11th Hellenic Conference on Artificial Intelligence. 2020; 224-231. doi: 10.1145/3411408.3411418.

27.

Popić

Pavković

Velikić

Teslić

. Data generators: a short survey of techniques and use cases with focus on testing. IEEE 9th International Conference on Consumer Electronics (ICCE-Berlin). 2019; 189-194. doi: 10.1109/ICCE-Berlin47944.2019.8966202.

28.

Liu

Fang

Tang

Chan

PPK

. Synthetic data generator for classification rules learning.7th International Conference on Cloud Computing and Big Data (CCBD). 2016; 357-361. doi: 10.1109/CCBD.2016.076.

29.

Ramos

Rego

. Feature-based generators for time series data. Proceedings of the Winter Simulation Conference. 2005. doi: 10.1109/WSC.2005.1574558.

30.

Kondratyev

Schwarz

. The market generator. SSRN. 2019; doi: 10.2139/ssrn.3384948.

31.

Raberto

Cincotti

. Modeling and simulation of a double auction artificial financial market. Physica A: Statistical Mechanics and its Applications. 2005; 34-35. doi: 10.1016/j.physa.2005.02.061.

32.

Chen

. Agent-based artificial financial market with evolutionary algorithm. Economic Research-Ekonomska Istraživanja. 2022; 1-21. doi: 10.1080/1331677X.2021.2021098.

33.

Yagi

Masuda

Mizuta

. Analysis of the impact of high-frequency trading on artificial market liquidity. IEEE Transactions on Computational Social Systems. 2020; 7(6): 1324-1334. doi: 10.1109/TCSS.2020.3019352.

34.

Sima

Namin

. Forecasting Economics and Financial Time Series: ARIMA vs. LSTM. arXiv.org. 2018. doi: 10.48550/arXiv.1803.06386.

35.

Tick. Investopedia: Tick. 2022. http://www.investopedia.com/terms/t/tick.asp.

36.

Long/Short Equity. Investopedia: Long/Short Equity. 2022. http://www.investopedia.com/terms/l/long-shortequity.asp.

Ultra-short term trading using a neural-network based ensemble of financial technical indicators in a closed world market

Abstract

Keywords

1. Introduction

2. A brief background on predicting exchange rates using computational intelligence

3. A detailed system description

3.1 “Closed” trading world and experimental data source

3.3 Transaction stage

4. Experimentation and results

Table 4 Parameterization of automated transaction machines

5. Conclusions

Footnotes

Annex 1. System parameters

Annex 2. Analytical results of experimentation per day,machine and source of forecast

References

Table 4
Parameterization of automated transaction machines