Make Trading Decisions Research Articles

EarnHFT: Efficient Hierarchical Reinforcement Learning for High Frequency Trading

High-frequency trading (HFT) is using computer algorithms to make trading decisions in short time scales (e.g., second-level), which is widely used in the Cryptocurrency (Crypto) market, (e.g., Bitcoin). Reinforcement learning (RL) in financial research has shown stellar performance on many quantitative trading tasks. However, most methods focus on low-frequency trading, e.g., day-level, which cannot be directly applied to HFT because of two challenges. First, RL for HFT involves dealing with extremely long trajectories (e.g., 2.4 million steps per month), which is hard to optimize and evaluate. Second, the dramatic price fluctuations and market trend changes of Crypto make existing algorithms fail to maintain satisfactory performances. To tackle these challenges, we propose an Efficient hieArchical Reinforcement learNing method for High Frequency Trading (EarnHFT), a novel three-stage hierarchical RL framework for HFT. In stage I, we compute a Q-teacher, i.e., the optimal action value based on dynamic programming, for enhancing the performance and training efficiency of second level RL agents. In stage II, we construct a pool of diverse RL agents for different market trends, distinguished by return rates, where hundreds of RL agents are trained with different preferences of return rates and only a tiny fraction of them will be selected into the pool based on their profitability. In stage III, we train a minute-level router which dynamically picks a second-level agent from the pool to achieve stable performance across different markets. Through extensive experiments in various market trends on Crypto markets in a high-fidelity simulation trading environment, we demonstrate that EarnHFT significantly outperforms 6 state-of-art baselines in 6 popular financial criteria, exceeding the runner-up by 30% in profitability.

Examining the Moderating Role of the Automated Trading System on the Financial Performance of the Ghana Stock Exchange

The study investigates the role of the automated trading system on stock market capitalization, liquidity, efficiency, returns and volatility on the Ghana stock exchange. The project solely focus on secondary data from the internet databases of the Ghana Stock Exchange (GSE), and Bank of Ghana (BOG). The data is evaluated using a multivariate regression approach and includes monthly time series data from 2000 to 2020 as well as the volume of stock traded before and after the exchange is automated. The study finds that market performance has an impact on the automated trading system, and it also finds that the amount of stock traded is positively impacted by the automated trading system, exchange rates, and interest rates. Once more, the study draws the conclusion which asserts the economic indicator variable of the nation are marked via erratic and significant fluctuations in conversion rates. According to the report, the exchange’s transaction guidelines need to alter in order to provide shareholders the knowledge and confidence to make trading decisions on their own, without the assistance of a licensed stockbroker.

Carbon emission price forecasting in China using a novel secondary decomposition hybrid model of CEEMD-SE-VMD-LSTM

Effective forecasting of carbon prices helps investors to judge carbon market conditions and promotes the environment and economic sustainability. The contribution of this paper is constructing a novel secondary decomposition hybrid carbon price forecasting model, namely CEEMD-SE-VMD-LSTM. It is noteworthy that the sample entropy is introduced to identify the highly complex signals rather than empirically determined in previous studies. Specifically, the complementary ensemble empirical mode decomposition (CEEMD) model is used to decompose the original price signals. The sample entropy (SE) and variational mode decomposition (VMD) are conducted to recognize and secondary decompose the highly complex components, while the long short-term memory (LSTM) model is employed to forecast the carbon price by summing up the predicted intrinsic mode function (IMF) components. The conclusion shows the proposed model has the smallest forecasting errors with the values of RMSE, MAE and MAPE are 0.2640, 0.1984 and 0.0044, respectively, the secondary decomposition models are better than other primary decomposition models and the forecasting performances of LSTM-type models are better than those of other GRU-type models. Further evidence convinces us that short-term forecasting accuracy is superior to long-term forecasting. Those conclusions and model innovation can provide a valuable reference for investors to make trading decisions.

Modeling the Subscription Ratio and IPO Returns Non-Linear Relationship: Evidence from the Tunisian Stock Market

To capture the herd behavior of IPO investors, the relationship binding IPO underpricing and the subscription ratio is often assumed to represent a straight line in the relevant literature, due mainly to recurrent general linear modeling in the data analysis process. In Tunisia, however, and following the subscription result declaration, all IPOs are usually met with noticeable success owing to over subscription issues, resulting in persistent herd behavior. In this context, the present paper is deigned to highlight how Tunisian investors tend to mimic each other on setting out to make trading decisions. More particularly, we attempt to specify the non-linear relationship binding the subscription ratio and IPO returns within a quantile regression, highlighting the investors’ noticeable sensitivity to high subscription, thereby, easily inducing the spread of herd behavior and increasing IPO returns.

Automated cryptocurrency trading approach using ensemble deep reinforcement learning: Learn to understand candlesticks

Despite their high risk, cryptocurrencies have gained popularity as viable trading options. Cryptocurrencies are digital assets that experience significant fluctuations in a market operating 24 h a day. Recently, considerable attention has been paid to developing trading bots using machine-learning-based artificial intelligence. Previous studies have employed machine learning techniques to predict financial market trends or make trading decisions, primarily using numerical data extracted from candlesticks. However, these data often overlook the temporal and spatial information of candlesticks, leading to a limited understanding of their significance. In this study, we utilize multi-resolution candlestick images containing temporal and spatial information. Our rationale for using visual information from candlestick charts is to replicate the decision-making processes of human trading experts. To achieve this, we employ deep reinforcement learning algorithms to generate trading signals based on a state vector that includes embedded candlestick-chart images. The trading signal is generated using a multi-agent weighted voting ensemble approach. We test the proposed approach on two BTC/USDT datasets under both bullish and bearish market scenarios. Additionally, we use an attention-based technique to identify significant areas in the candlestick images targeted by the proposed approach. Our findings demonstrate that models using candlestick images 'as-is', outperform those using raw numeric data and other baseline models.

Research on Optimal Investment Strategy Combination Based on ARIMA Model and mean-variance analysis -- Taking Gold and Bitcoin assets as examples

Gold and Bitcoin are popular trading products in today's trading market. In order to build a trading portfolio that maximizes returns, the prices of two trading products need to be predicted first. This article utilizes ARIMA to deal with the non-stationarity and predict the future prices of gold and bitcoin. In this article, the choice of parameters is ARIMA (4, 1, 4) for both bitcoin and gold. To find the best timing to sell and buy the two assets, the article first rate them with well-designed rating system by three important factors: Changes in value, Moving averages, and Bias. Then based on these factors, the model further linearly composes the indicator for risk and trend. By utilizing the information, the model gets with the main factor to make trading decisions.

A hybrid multicategory framework for generating automated trading systems

AbstractThis work proposes a framework for producing automatic trading systems that suggest making trading decisions in financial markets. The trading systems thus generated utilize multicategory classifiers. A collection of important technical indicators are considered as input features for the underlying multicategory classifiers. The proposed trading systems also have an option to use random forest (RF) algorithms for feature selection. The trading range breakout strategy is used to train the proposed trading systems thus generating “BUY/SELL/WAIT” trading signals on daily open prices. The performances of the proposed trading systems are evaluated over five future indices. Empirical findings suggest that the day trading systems based on the proposed multicategory SVM classifiers along with the RF technique outperform the day trading systems built using the multicategory classifiers taken from the literature. The trading range breakout strategy‐based systems are found to be superior to the traditional BUY‐HOLD strategy and that of RF‐PSVM (BUY/SELL) based strategy.

Combining machine learning and effective feature selection for real-time stock trading in variable time-frames

The unpredictability and volatility of the stock market render it challenging to make a substantial profit using any generalised scheme. Many previous studies tried different techniques to build a machine learning model, which can make a significant profit in the US stock market by performing live trading. However, very few studies have focused on the importance of finding the best features for a particular period for trading. Our top approach used the performance to narrow down the features from a total of 148 to about 30. Furthermore, the top 25 features were dynamically selected before each time training our machine learning model. It uses ensemble learning with four classifiers: Gaussian naive Bayes, decision tree, logistic regression with L1 regularisation and stochastic gradient descent, to decide whether to go long or short on a particular stock. Our best model performed daily trade between July 2011 and January 2019, generating 54.35% profit. Finally, our work showcased that mixtures of weighted classifiers perform better than any individual predictor about making trading decisions in the stock market.

Joint Decision-Making Model of Power Manufacturer Participation in Green Certificate Market and Medium - and Long-Term Power Market

To effectively guide power manufacturers in the medium and long-term market and green certificate market trading behavior. This paper firstly constructs the model of power manufacturers' participation in the green certificate market and the model of participating in the medium and long-term market of power, and establishes the coupling constraint of power manufacturers' participation in the power market and the green certificate market. Secondly, considering the speculative behavior of power manufacturers and using conditional value-at-risk to depict the profit risk of electricity generation due to the uncertainty of market price, the optimal comprehensive benefit considering the annual expected profit and risk of power manufacturers is taken as the decision-making objective, and the joint decision-making model of power manufacturers participating in the green certificate market and the medium and long term power market is established. It is helpful for power manufacturers to make trading decisions in medium and longterm market and green certificate market.

Deep Reinforcement Learning-Based Trading Strategy for Load Aggregators on Price-Responsive Demand.

With the development of the Internet of things and smart grid technologies, modern electricity markets seamlessly connect demand response to the spot market through price-responsive loads, in which the trading strategy of load aggregators plays a crucial role in profit capture. In this study, we propose a deep reinforcement learning-based strategy for purchasing and selling electricity based on real-time electricity prices and real-time demand data in the spot market, which maximizes the revenue of load aggregators. The deep deterministic policy gradient (DDPG) is applied through a bidirectional long- and short-term memory (BiLSTM) network to extract the market state features that are used to make trading decisions. The effectiveness of the method is validated using datasets from the New England electricity market and Australian electricity market by introducing a bidirectional LSTM structure into the actor-critic network structure to learn hidden states in partially observable Markov states through memory inference. Comparative experiments of the method show that the method can provide greater yield results.

Quantitative Trading through Random Perturbation Q-Network with Nonlinear Transaction Costs

In recent years, reinforcement learning (RL) has seen increasing applications in the financial industry, especially in quantitative trading and portfolio optimization when the focus is on the long-term reward rather than short-term profit. Sequential decision making and Markov decision processes are rather suited for this type of application. Through trial and error based on historical data, an agent can learn the characteristics of the market and evolve an algorithm to maximize the cumulative returns. In this work, we propose a novel RL trading algorithm utilizing random perturbation of the Q-network and account for the more realistic nonlinear transaction costs. In summary, we first design a new near-quadratic transaction cost function considering the slippage. Next, we develop a convolutional deep Q-learning network (CDQN) with multiple price input based on this cost functions. We further propose a random perturbation (rp) method to modify the learning network to solve the instability issue intrinsic to the deep Q-learning network. Finally, we use this newly developed CDQN-rp algorithm to make trading decisions based on the daily stock prices of Apple (AAPL), Meta (FB), and Bitcoin (BTC) and demonstrate its strengths over other quantitative trading methods.

GPM: A graph convolutional network based reinforcement learning framework for portfolio management

Portfolio management is a decision-making process of periodically reallocating a certain amount of funds into a portfolio of assets, with the objective of maximizing the profits constrained to a given risk level. Due to its nature of learning from dynamic interactions and planning for long-run performance, reinforcement learning (RL) recently has received much attention in portfolio management. However, most of existing RL-based PM approaches in general only consider price changes of portfolio assets and the implicit correlations between price changes, while ignoring the rich relations between companies in the market, such as two assets in the same sector or two companies have a supplier-customer relation, which are very important for trading decision making. To address these limitations, in this paper, we propose GPM, a novel graph convolutional network-based reinforcement learning framework for portfolio management, which first employs Relational Graph Convolutional Network (R-GCN) to extract asset relational features, and then combines relational features with multi-scale temporal features to make trading decisions for better performance. We also introduce softmax with temperature to increase portfolio diversity, which leads to further increase in profits. Experimental results on two real-world datasets: NASDAQ and NYSE, validate the effectiveness of GPM over state-of-the-art PM methods. Further, more experiments are conducted to evaluate GPM with different graph neural networks and different number of network layers, to explore proper settings for different markets.

Social media and money: How our tweets affect the financial markets

Social media, such as Twitter, has replaced traditional media as a dominant information source (Gan et al., 2020). With social media being a fast and more efficient information dissemination channel, investors can quickly access, update and exchange stock-related information before making trading decisions. In this study, I examine the mechanism by which social media sentiment affects stock prices. I first collect a sample of 2.2 million tweets and quantify each tweet using Natural Language Processing (NLP) techniques to infer whether the tweet has a positive or negative tone. Next, I employ a modified vector autoregression (VAR) model to assess the impact of tweets on stock returns at the minute frequency. I find that Twitter sentiment affects stock prices through its impact on trades. Trades have stronger price impacts with an increase in tweet volume or investor sentiment. In addition, both bullish and bearish tweets amplify the impact of trades on stock returns. Twitter sentiment causes a permanent price movement until a new equilibrium is reached, showing Twitter sentiment contains substantial information to drive stock prices. These findings suggest that trades in the stock market become more informative as tweet activity increases, as well as when tweets contain higher sentiment. To the best of my knowledge, this is the first study looking at the exact mechanism of how social media affects the stock market and investors' response behaviour.

LGBT policy, investor trading behavior, and return comovement

Investors are attentive to lesbian, gay, bisexual, and transgender (LGBT) topic and a firm's adoption of LGBT-supportive policy. Using a sample of new LGBT adopters from KLD database, we show that mutual funds with a strong (weak) preference for LGBT stocks increase (decrease) their holdings in new LGBT adopters and receive more (less) capital flows when investor sentiment toward LGBT is high. We also find significant evidence that LGBT-induced trading activities lead to comovements in stock returns and share turnover. Specifically, LGBT adopters experience an increase (decrease) in return comovement with a portfolio of existing (non-) LGBT stocks. Our additional analyses based on an alternative sample from the Human Rights Campaign yield consistent results and suggest that investors consider not only the presence or lack of LGBT-supportive policy in a firm but also its LGBT performance when they make trading decisions. This research makes an important contribution to our limited understanding of the LGBT policy effect on investors, a key group of the firm's stakeholders.

Investors pay attention to multi-level financial news ——"A pure waste of time" or "It’s worth it"

In recent years,the efficiency of news communication has been greatly improved with the advance of Internet technology, and whether investors should pay attention to multiple levels of news when making trading decisions is worth pondering. However, Chinese scholars have paid little attention to the news hierarchy, and do not bring different levels of news into the prediction system. Therefore, this paper studies the stock price trend prediction from this perspective. This paper selects the medical industry as the representative, and adds its sub industry news and industry news to form a multi-level news system on the basis of individual stock news. In order to make better use of and extract the information contained in different levels of news, the multi-kernel learning model is incorporated on the basis of using SVM. The results suggest that the use of multi-level news can improve the accuracy of stock price trend prediction. Compared with the SVM model solely considering individual stock news, the prediction accuracy of multi-kernel learning model based on multi-level news is improved by 10%, which is of great significance for investment decision-making.

The Prediction Model and System of Stock Rise and Fall Based on BP Neural Network

The purpose of forecasting the rise and fall of the stock market is for investors to analyze the future stock price trends, trends and other information, and combine the characteristics of the stocks to choose appropriate methods to make trading decisions. The purpose of this paper to study the stock price prediction model is to improve the preventive measures against stock risks. This article mainly uses the experimental method and the comparative method to conduct an experimental research on the stock price fluctuation forecast model. Experimental results show that the accuracy of BP neural network algorithm in stock prediction can reach 91.5%. This means that the stock price prediction model based on BP neural network has certain feasibility, and its research value is also relatively large.

Combining machine learning and effective feature selection for real-time stock trading in variable time-frames

The unpredictability and volatility of the stock market render it challenging to make a substantial profit using any generalised scheme. Many previous studies tried different techniques to build a machine learning model, which can make a significant profit in the US stock market by performing live trading. However, very few studies have focused on the importance of finding the best features for a particular trading period. Our top approach used the performance to narrow down the features from a total of 148 to about 30. Furthermore, the top 25 features were dynamically selected before each time training our machine learning model. It uses ensemble learning with four classifiers: Gaussian Naive Bayes, Decision Tree, Logistic Regression with L1 regularization, and Stochastic Gradient Descent, to decide whether to go long or short on a particular stock. Our best model performed daily trade between July 2011 and January 2019, generating 54.35% profit. Finally, our work showcased that mixtures of weighted classifiers perform better than any individual predictor of making trading decisions in the stock market.

A Deep Reinforcement Learning-Based Decision Support System for Automated Stock Market Trading

Presently, the volatile and dynamic aspects of stock prices are significant research challenges for stock markets or any other financial sector to design accurate and profitable trading strategies in all market situations. To meet such challenges, the usage of computer-aided stock trading techniques has grown in prominence in recent decades owing to their ability to rapidly and accurately analyze stock market situations. In the recent past, deep reinforcement learning (DRL) methods and trading bots are commonly utilized for algorithmic trading. However, in the existing literature, the trading agents employ the historical and present trends of stock prices as an observing state to make trading decisions without taking into account the long-term market future pattern of stock prices. Therefore, in this study, we proposed a novel decision support system for automated stock trading based on deep reinforcement learning that observes both past and future trends of stock prices whether single and multi-step ahead as an observing state to make the optimal trading decisions of buying, selling, and holding the stocks. More specifically, at every time step, future trends are monitored concurrently using a forecasting network whose output is concatenated with past trends of stock prices. The concatenated vectors are subsequently supplied to the DRL agent as an observation state. In addition, the suggested forecasting network is built on a Gated Recurrent Unit (GRU). The GRU-based agent captures more informative and inherent aspects of time-series financial data. Furthermore, the suggested decision support system has been tested on several stock markets such as Tesla, IBM, Amazon, CSCO, and Chinese Stocks as well as equity markets i-e SSE Composite Index, NIFTY 50 Index, US Commodity Index Fund, and has achieved encouraging profit values while trading.

Decision-Making Techniques using LSTM on Antam Mining Shares before and during the COVID-19 Pandemic in Indonesia

Stocks, apart from having volatile and chaotic characteristics, also have various kinds of noise, non-linear and non-stationary movements, making them difficult to predict accurately. Therefore, the risk of investing in stocks depends on the skills of investors or traders in making judgments and decisions. This study aims to use Long Short-Term Memory (LSTM) as a decision-making technique with historical stock prices as the sole predictor, then implement it in conditions before and during the COVID-19 pandemic. The study results concluded that Long Short-Term Memory (LSTM) could be used as a decision-making technique in conditions before and during the COVID-19 pandemic with historical price inputs as the sole predictor. Based on the research that has been done, the following conclusions can be drawn: The LSTM model can predict stock prices well using historical stock prices as the sole predictor. The LSTM model can be used as a trading decision-making technique for day traders. The risk of stock prediction using the LSTM method in 2019 before the COVID pandemic was proven to be lower than in 2020 during the COVID pandemic. For further research, researchers can conduct more in-depth research on the risk criteria for making trading decisions as an essential reference that can be used to select the LSTM model.

LSTM-DDPG for Trading with Variable Positions.

In recent years, machine learning for trading has been widely studied. The direction and size of position should be determined in trading decisions based on market conditions. However, there is no research so far that considers variable position sizes in models developed for trading purposes. In this paper, we propose a deep reinforcement learning model named LSTM-DDPG to make trading decisions with variable positions. Specifically, we consider the trading process as a Partially Observable Markov Decision Process, in which the long short-term memory (LSTM) network is used to extract market state features and the deep deterministic policy gradient (DDPG) framework is used to make trading decisions concerning the direction and variable size of position. We test the LSTM-DDPG model on IF300 (index futures of China stock market) data and the results show that LSTM-DDPG with variable positions performs better in terms of return and risk than models with fixed or few-level positions. In addition, the investment potential of the model can be better tapped by the reward function of the differential Sharpe ratio than that of profit reward function.