No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Portfolio management based on a reinforcement learning framework
Abstract
Portfolio management is crucial for investors. We propose a dynamic portfolio management framework based on reinforcement learning using the proximal policy optimization algorithm. The two‐part framework includes a feature extraction network and a full connected network. First, the majority of the previous research on portfolio management based on reinforcement learning has been dedicated to discrete action spaces. We propose a potential solution to the problem of a continuous action space with a constraint (i.e., the sum of the portfolio weights is equal to 1). Second, we explore different feature extraction networks (i.e., convolutional neural network [CNN], long short‐term memory [LSTM] network, and convolutional LSTM network) combined with our system, and we conduct extensive experiments on the six kinds of assets, including 16 features. The empirical results show that the CNN performs best in the test set. Last, we discuss the effect of the trading frequency on our trading system and find that the monthly trading frequency has a higher Sharpe ratio in the test set than other trading frequencies.
Since the techniques of Reinforcement learning (RL) can actually produce dynamic decisions under uncertainty in the financial portfolio optimization, therefore, it is a critical area of research. A review in recent advancements in the application of RL for portfolio management has been done in this paper, with an emphasis on its possibility of enhancing the associated risk management as well as optimizing returns in complex financial markets. First, we outline the basic principles of RL and discuss its wide range of applications in portfolio optimization. Thereafter, the paper moves on to major challenges in this field, that is, big data issues, non-stationary environments, and computational complexities. Last, we will be showing future directions for the research which may include the integration of metalearning, multi-agent systems, and real-time adaptability for further enhancing the performance of RL-based portfolio optimization systems.
In this paper we describe the integration of machine learning (ML) techniques into the framework of Markowitz's portfolio selection and show how they can help advance the robust mathematical strategies necessary for modern financial markets. By combining traditional econometrics with cutting-edge ML methodologies, we show how to enhance portfolio management processes including alpha generation, risk management, and optimization of risk metrics like conditional value at risk. ML's capacity to handle vast and complex datasets allows for more dynamic and informed decision-making in portfolio construction. Moreover, we discuss the practical applications of these techniques in real-world portfolio management, highlighting both the potential enhancements and the challenges faced by portfolio managers in implementing ML strategies.
Stock market prediction has been identified as a very important practical problem in the economic field. However, the timely prediction of the market is generally regarded as one of the most challenging problems due to the stock market’s characteristics of noise and volatility. To address these challenges, we propose a deep learning-based stock market prediction model that considers investors’ emotional tendency. First, we propose to involve investors’ sentiment for stock prediction, which can effectively improve the model prediction accuracy. Second, the stock pricing sequence is a complex time sequence with different scales of fluctuations, making the accurate prediction very challenging. We propose to gradually decompose the complex sequence of stock price by adopting empirical modal decomposition (EMD), which yields better prediction accuracy. Third, we adopt LSTM due to its advantages of analyzing relationships among time-series data through its memory function. We further revised it by adopting attention mechanism to focus more on the more critical information. Experiment results show that the revised LSTM model can not only improve prediction accuracy, but also reduce time delay. It is confirmed that investors’ emotional tendency is effective to improve the predicted results; the introduction of EMD can improve the predictability of inventory sequences; and the attention mechanism can help LSTM to efficiently extract specific information and current mission objectives from the information ocean.
Models of stock price prediction have customarily utilized technical indicators alone to produce trading signals. In this paper, we construct trading techniques by applying machine-learning methods to technical analysis indicators and stock market returns data. The resulting prediction models can be utilized as an artificial trader used to trade on any given stock trade. Here the issue of stock trading decision prediction is enunciated as a classification problem with two class values representing the buy and sell signals. The stacking technique utilized in this paper is to assist trader with applying the proposed algorithms in their trading using random forest which was staked with different algorithms which incorporates Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM) and Neural Network (NN). The experimental results indicated that Top Layer of Random Forest (TRF) produced the best performance among all the algorithms compared. This is an indication that it is a promising strategy for forecasting Nigerian stock returns.
The non-linearity and high change rates of stock market index prices make prediction a challenging problem for traders and data scientists. Data modeling and machine learning have been extensively utilized for proposing solutions to this difficult problem. In recent years, deep learning has proved itself in solving such complex problems. In this paper, we tackle the problem offorecasting the Turkish Stock Market BIST 30 index movements and prices. We propose a deep learning model fed with technical indicators and oscillators calculated from historical index price data. Experiments conducted by applying our model on a dataset gathered for a period of 27 months on www.investing.com demonstrate that our solution outperforms other similar proposals and attains good accuracy, achieving 0.0332, 0.109, 0.09, 0.1069 and 0.2581 as mean squared error in predictingBIST 30 index prices for the next five trading days. Based on these results, we argue that using deep neural networks is advisable for stock market index prediction.
The study of the impact of investor sentiment on stock returns has gained increasing momentum in the past few years. It has been widely accepted that public mood is correlated with financial markets. However, only a few studies discussed how the public mood would affect one of the fundamental problems of computational finance: portfolio management. In this study, we use public financial sentiment and historical prices collected from the New York Stock Exchange (NYSE) to train multiple machine learning models for automatic wealth allocation across a set of assets. Unlike previous studies which set as target variable the asset prices in the portfolio, the variable to predict here is represented by the best asset allocation strategy ex post. Experiments performed on five portfolios show that long short-term memory networks are superior to multi-layer perceptron and random forests producing, in the period under analysis, an average increase in the revenue across the portfolios ranging between 5% (without financial mood) and 19% (with financial mood) compared to the equal-weighted portfolio. Results show that our all-in-one and end-to-end approach for automatic portfolio selection outperforms the equal-weighted portfolio. Moreover, when using long short-term memory networks, the employment of sentiment data in addition to lagged data leads to greater returns for all the five portfolios under evaluation. Finally, we find that among the employed machine learning algorithms, long short-term memory networks are better suited for learning the impact of public mood on financial time series.
Today’s investors’ dilemma is choosing the right stock for investment at right time. There are many technical analysis tools which help choose investors pick the right stock, of which RSI is one of the tools in understand whether stocks are overpriced or under priced. Despite its popularity and powerfulness, RSI has been very rarely used by Indian investors. One of the important reasons for it is lack of knowledge regarding how to use it. So, it is essential to show, how RSI can be used effectively to select shares and hence construct portfolio. Also, it is essential to check the effectiveness and validity of RSI in the context of Indian stock market. EPS and P/E ratio tend to reflect the profitability of a stock. It is also important to find which of the above better reflects the profitability of the organization so as to make better decisions regarding investment. In this case 20 stocks are chosen from NSE, 10 of which are based on highest P/E ratio and remaining 10 are based on highest EPS. Here, the portfolio consists of stocks both for short term investment and long term investment. For short term investment, we find 14 day RSI for all 20 scripts and for long term investment 56 day RSI is being found out for all the 20 scripts. RSI values are calculated for the time period 2011 January to 2013 December. Ten scripts that are chosen based on RSI from the 20 scripts are included in the portfolio. In order to find the validity of RSI in Indian stock markets, we evaluate the performance of short term investments by computing 14 day RSI for the chosen shot term investment stocks at a future point of time (March 2014), and the performance is evaluated by comparing it with the initial 14 day RSI. In this case most of the results proved positive, thus showing that RSI is valid in Indian stock markets. Also out of the 10 selected scripts six are the ones which have highest P/E ratio, this indicates that P/E ratio is a better indicator of profitability when compared to EPS.
Financial portfolio management is the process of constant redistribution of a fund into different financial products. This paper presents a financial-model-free Reinforcement Learning framework to provide a deep machine learning solution to the portfolio management problem. The framework consists of the Ensemble of Identical Independent Evaluators (EIIE) topology, a Portfolio-Vector Memory (PVM), an Online Stochastic Batch Learning (OSBL) scheme, and a fully exploiting and explicit reward function. This framework is realized in three instants in this work with a Convolutional Neural Network (CNN), a basic Recurrent Neural Network (RNN), and a Long Short-Term Memory (LSTM). They are, along with a number of recently reviewed or published portfolio-selection strategies, examined in three back-test experiments with a trading period of 30 minutes in a cryptocurrency market. Cryptocurrencies are electronic and decentralized alternatives to government-issued money, with Bitcoin as the best-known example of a cryptocurrency. All three instances of the framework monopolize the top three positions in all experiments, outdistancing other compared trading algorithms. Although with a high commission rate of 0.25% in the backtests, the framework is able to achieve at least 4-fold returns in 30 days.
Portfolio management is the decision-making process of allocating an amount of fund into different financial investment products. Cryptocurrencies are electronic and decentralized alternatives to government-issued money, with Bitcoin as the best-known example of a cryptocurrency. This paper presents a model-less convolutional neural network with historic prices of a set of financial assets as its input, outputting portfolio weights of the set. The network is trained with 0.7 years' price data from a cryptocurrency exchange. The training is done in a reinforcement manner, maximizing the accumulative return, which is regarded as the reward function of the network. Backtest trading experiments with trading period of 30 minutes is conducted in the same market, achieving 10-fold returns in 1.8 months' periods. Some recently published portfolio selection strategies are also used to perform the same back-tests, whose results are compared with the neural network. The network is not limited to cryptocurrency, but can be applied to any other financial markets.
Some studies published recently (Dejan Eric, 2009; R. Rosillo, 2013; Terence Tai-Leung Chong, 2008; Ülkü and Prodan, 2013) uncover that moving average convergence divergence (MACD) trading rules have predictive ability in many countries. The MACD trading strategies applied by these papers to execute the trading signals are various. This study analyzes the performance of a MACD trading strategy (MACD-4 in the current study), which is applied popularly by practitioners, but was not tested by prior academicians. Furthermore, the author compares the performance of each of the strategies on a group of markets to identify the best one. Before considering the costs, the author finds that the MACD-4 trading strategy has predictive ability. The best performance is MACD strategy applied by Terence Tai-Leung Chong (2008). This strategy is also the most effective one if it is applied in a high trading cost environmentm because the numbers of trades created are the lowest. Especially, the strategy applied by R. Rosillo (2013) is unpredictable in the selected samples.
Predicting financial market changes is an important issue in time series analysis, receiving an increasing attention in last two decades. The combined prediction model, based on artificial neural networks (ANNs) with principal component analysis (PCA) for financial time series forecasting is presented in this work. In the modeling step, technical analysis has been conducted to select technical indicators. Then PCA approach was applied to extract the principal components from the variables for the training step. Finally, the ANN-based model called NARX was used to train the data and perform the time series forecast. TAL1T stock of Nasdaq OMX Baltic stock exchange was used as a case study. The mean square error (MSE) measure was used to evaluate the performances of proposed model. The experimental results lead to the conclusion that the proposed model can be successfully used as an alternative method to standard statistical techniques for financial time series forecasting.
Using daily data for the Swiss franc/US dollar exchange rate, this paper studies the trading profitability of the technical indicator Relative Strength Index (RSI). The authors find that for the past decade or so, using the standard configuration of RSI < = 30 and RSI > = 70 as buy or sell threshold, RSI offers no trading profit, but a small loss instead. However, when the buy/sell threshold parameters are altered, to deviate from the combination most commonly used, using RSI as the trading signal still yields profits. The authors also provide an explanation of this phenomenon. One implication of our findings is that consistent profit opportunities should no longer exist in what is already commonly and widely known, but taking a path less travelled could still lead to profit opportunities not yet discovered and utilized.
Trend-following strategies take long positions in assets with positive past returns and short positions in assets with negative past returns. They are typically constructed using futures contracts across all asset classes, with weights that are inversely proportional to volatility, and have historically exhibited great diversification features especially during dramatic market downturns. However, following an impressive performance in 2008, the trend-following strategy has failed to generate strong returns in the post-crisis period, 2009-2013. This period has been characterised by a large degree of co-movement even across asset classes, with the investable universe being roughly split into the so-called Risk-On and Risk-Off subclasses. We examine whether the inverse-volatility weighting scheme, which effectively ignores pairwise correlations, can turn out to be suboptimal in an environment of increasing correlations. By extending the conventionally long-only risk-parity (equal risk contribution) allocation, we construct a long-short trend-following strategy that makes use of risk-parity principles. Not only do we significantly enhance the performance of the strategy, but we also show that this enhancement is mainly driven by the performance of the more sophisticated weighting scheme in extreme average correlation regimes.
The popular Q-learning algorithm is known to overestimate action values under
certain conditions. It was not previously known whether, in practice, such
overestimations are common, whether this harms performance, and whether they
can generally be prevented. In this paper, we answer all these questions
affirmatively. In particular, we first show that the recent DQN algorithm,
which combines Q-learning with a deep neural network, suffers from substantial
overestimations in some games in the Atari 2600 domain. We then show that the
idea behind the Double Q-learning algorithm, which was introduced in a tabular
setting, can be generalized to work with large-scale function approximation. We
propose a specific adaptation to the DQN algorithm and show that the resulting
algorithm not only reduces the observed overestimations, as hypothesized, but
that this also leads to much better performance on several games.
Financial market prediction and trading presents a challenging task that attracts great interest from researchers and investors because success may result in substantial rewards. This paper describes the application of a hierarchical coevolutionary fuzzy system called HiCEFS for predicting financial time series. A novel financial trading system using HiCEFS as a predictive model and employing a prudent trading strategy based on the price percentage oscillator (PPO) is proposed. In order to construct an accurate predictive model, a form of generic membership function named irregular shaped membership function (ISMF) is employed and a hierarchical coevolutionary genetic algorithm (HCGA) is adopted to automatically derive the ISMFs for each input feature in HiCEFS. With the accurate prediction from HiCEFS and the prudent trading strategy, the proposed system outperforms the simple buy-and-hold strategy, the trading system without prediction and the trading system with other predictive models (EFuNN, DENFIS and RSPOP) on real-world financial data.
Q-learning (Watkins, 1989) is a simple way for agents to learn how to act optimally in controlled Markovian domains. It amounts to an incremental method for dynamic programming which imposes limited computational demands. It works by successively improving its evaluations of the quality of particular actions at particular states.
This paper presents and proves in detail a convergence theorem forQ-learning based on that outlined in Watkins (1989). We show thatQ-learning converges to the optimum action-values with probability 1 so long as all actions are repeatedly sampled in all states and the action-values are represented discretely. We also sketch extensions to the cases of non-discounted, but absorbing, Markov environments, and where manyQ values can be changed each iteration, rather than just one.
Learning to store information over extended time intervals by recurrent backpropagation takes a very long time, mostly because of insufficient, decaying error backflow. We briefly review Hochreiter's (1991) analysis of this problem, then address it by introducing a novel, efficient, gradient based method called long short-term memory (LSTM). Truncating the gradient where this does not do harm, LSTM can learn to bridge minimal time lags in excess of 1000 discrete-time steps by enforcing constant error flow through constant error carousels within special units. Multiplicative gate units learn to open and close access to the constant error flow. LSTM is local in space and time; its computational complexity per time step and weight is O. 1. Our experiments with artificial data involve local, distributed, real-valued, and noisy pattern representations. In comparisons with real-time recurrent learning, back propagation through time, recurrent cascade correlation, Elman nets, and neural sequence chunking, LSTM leads to many more successful runs, and learns much faster. LSTM also solves complex, artificial long-time-lag tasks that have never been solved by previous recurrent network algorithms.
This article studies momentum in stock returns, focusing on the role of industry, size, and book-to-market (B/M) factors.
Size and B/M portfolios exhibit momentum as strong as that in individual stocks and industries. The size and B/M portfolios
are well diversified, so momentum cannot be attributed to firm- or industry-specific returns. Further, industry, size, and
B/M portfolios are negatively autocorrelated and cross-serially correlated over intermediate horizons. The evidence suggests
that stocks covary “too strongly” with each other. I argue that excess covariance, not underreaction, explains momentum in
the portfolios.
Forecasting of wheat production is of great importance for farmers and agriculture policy makers to improve production planning decisions. Numerous studies proved that traditional econometric techniques face significant challenges in out of sample predictability tests due to model uncertainty and parameter instability. Recent studies introduce several machine learning algorithms to improve time series prediction accuracy. The purpose of this study is to develop a precise wheat production model using artificial neural networks (ANN). A total of 71 years' wheat production data from 1948 to 2018 is divided into training data and test data. The model is trained by using 53 years' data and forecasts the future wheat production for the remaining 14 years. There are 16 indicators used as input variables for wheat production and top ten most important variables highlighted. The findings show, that the model captures much of the trend, and some of the undulations of the original series. The results reveal that the most important features in wheat production includes production prevailing trends, momentum and volatility.
Recurrent reinforcement learning (RRL) techniques have been used to optimize asset trading systems and have achieved outstanding results. However, the majority of the previous work has been dedicated to systems with discrete action spaces. To address the challenge of continuous action and multi-dimensional state spaces, we propose the so called Stacked Deep Dynamic Recurrent Reinforcement Learning (SDDRRL) architecture to construct a real-time optimal portfolio. The algorithm captures the up-to-date market conditions and rebalances the portfolio accordingly. Under this general vision, Sharpe ratio, which is one of the most widely accepted measures of risk-adjusted returns, has been used as a performance metric. Additionally, the performance of most machine learning algorithms highly depends on their hyperparameter settings. Therefore, we equipped SDDRRL with the ability to find the best possible architecture topology using an automated Gaussian Process (GP) with Expected Improvement (EI) as an acquisition function. This allows us to select the best architectures that maximizes the total return while respecting the cardinality constraints. Finally, our system was trained and tested in an online manner for 20 successive rounds with data for ten selected stocks from different sectors of the S&P 500 from January 1st, 2013 to July 31st, 2017. The experiments reveal that the proposed SDDRRL achieves superior performance compared to three benchmarks: the rolling horizon Mean-Variance Optimization (MVO) model, the rolling horizon risk parity model, and the uniform buy-and-hold (UBAH) index.
Forecasting cryptocurrency prices is crucial for investors. In this paper, we adopt a novel Gradient Boosting Decision Tree (GBDT) algorithm, Light Gradient Boosting Machine (LightGBM), to forecast the price trend (falling, or not falling) of cryptocurrency market. In order to utilize market information, we combine the daily data of 42 kinds of primary cryptocurrencies with key economic indicators. Results show that the robustness of the LightGBM model is better than the other methods, and the comprehensive strength of the cryptocurrencies impacts the forecasting performance. This can effectively guide investors in constructing an appropriate cryptocurrency portfolio and mitigate risks.
Deep reinforcement learning for multi-agent cooperation and competition has been a hot topic recently. This paper focuses on cooperative multi-agent problem based on actor-critic methods under local observations settings. Multi agent deep deterministic policy gradient obtained state of art results for some multi-agent games, whereas, it cannot scale well with growing amount of agents. In order to boost scalability, we propose a parameter sharing deterministic policy gradient method with three variants based on neural networks, including actor-critic sharing, actor sharing and actor sharing with partially shared critic. Benchmarks from rllab show that the proposed method has advantages in learning speed and memory efficiency, well scales with growing amount of agents, and moreover, it can make full use of reward sharing and exchangeability if possible.
We propose a new family of policy gradient methods for reinforcement learning, which alternate between sampling data through interaction with the environment, and optimizing a "surrogate" objective function using stochastic gradient ascent. Whereas standard policy gradient methods perform one gradient update per data sample, we propose a novel objective function that enables multiple epochs of minibatch updates. The new methods, which we call proximal policy optimization (PPO), have some of the benefits of trust region policy optimization (TRPO), but they are much simpler to implement, more general, and have better sample complexity (empirically). Our experiments test PPO on a collection of benchmark tasks, including simulated robotic locomotion and Atari game playing, and we show that PPO outperforms other online policy gradient methods, and overall strikes a favorable balance between sample complexity, simplicity, and wall-time.
Dynamic control theory has long been used in solving optimal asset allocation problems, and a number of trading decision systems based on reinforcement learning methods have been applied in asset allocation and portfolio rebalancing. In this paper, we extend the existing work in recurrent reinforcement learning (RRL) and build an optimal variable weight portfolio allocation under a coherent downside risk measure, the expected maximum drawdown, E(MDD). In particular, we propose a recurrent reinforcement learning method, with a coherent risk adjusted performance objective function, the Calmar ratio, to obtain both buy and sell signals and asset allocation weights. Using a portfolio consisting of the most frequently traded exchange-traded funds, we show that the expected maximum drawdown risk based objective function yields superior return performance compared to previously proposed RRL objective functions (i.e. the Sharpe ratio and the Sterling ratio), and that variable weight RRL long/short portfolios outperform equal weight RRL long/short portfolios under different transaction cost scenarios. We further propose an adaptive E(MDD) risk based RRL portfolio rebalancing decision system with a transaction cost and market condition stop-loss retraining mechanism, and we show that the proposed portfolio trading system responds to transaction cost effects better and outperforms hedge fund benchmarks consistently.
The portfolio optimization model has limited impact in practice because of estimation issues when applied to real data. To address this, we adapt two machine learning methods, regularization and cross-validation, for portfolio optimization. First, we introduce performance-based regularization (PBR), where the idea is to constrain the sample variances of the estimated portfolio risk and return, which steers the solution toward one associated with less estimation error in the performance. We consider PBR for both mean-variance and mean-conditional value-at-risk (CVaR) problems. For the mean-variance problem, PBR introduces a quartic polynomial constraint, for which we make two convex approximations: one based on rank-1 approximation and another based on a convex quadratic approximation. The rank-1 approximation PBR adds a bias to the optimal allocation, and the convex quadratic approximation PBR shrinks the sample covariance matrix. For the mean-CVaR problem, the PBR model is a combinatorial optimization problem, but we prove its convex relaxation, a quadratically constrained quadratic program, is essentially tight. We show that the PBR models can be cast as robust optimization problems with novel uncertainty sets and establish asymptotic optimality of both sample average approximation (SAA) and PBR solutions and the corresponding efficient frontiers. To calibrate the right-hand sides of the PBR constraints, we develop new, performance-based k-fold cross-validation algorithms. Using these algorithms, we carry out an extensive empirical investigation of PBR against SAA, as well as L1 and L2 regularizations and the equally weighted portfolio. We find that PBR dominates all other benchmarks for two out of three Fama–French data sets.
This paper was accepted by Yinyu Ye, optimization.
By discussing the board game (Go) and the recent successful computer system (AlphaGo), associate EIC Jim X. Chen presents an overview of the evolution of computing and hardware efficiency. The emphasis is on sharing information about the evolution of computing, which has a significant impact on artificial intelligence.
For a long time moving averages has been used for a financial data smoothing. It is one of the first indicators in technical analysis trading. Many traders debated that one moving average is better than other. As a result a lot of moving averages have been created. In this empirical study we overview 19 most popular moving averages, create a taxonomy and compare them using two most important factors - smoothness and lag. Smoothness indicates how much an indicator change (angle) and lag indicates how much moving average is lagging behind the current price. The aim is to have values as smooth as possible to avoid erroneous trades and with minimal lag - to increase trend detection speed. This large-scale empirical study performed on 1850 real-world time series including stocks, ETF, Forex and futures daily data demonstrate that the best smoothness/lag ratio is achieved by the Exponential Hull Moving Average (with price correction) and Triple Exponential Moving Average (without correction).
Several different decisions, including asset allocation, security selection and market timing, affect the return to a pension fund (or any investor). The impact of each type of decision can be measured by comparing a portfolio’s actual return with the return on a hypothetical portfolio that does not reflect a particular decision that went into the real portfolio. The critical element in the comparison is defining the naive alternative to the decision. When asset allocation is the decision being evaluated, the naive alternative is not obvious. If treasury bills are the appropriate naive alternative, then asset allocation is, as commonly thought, the single decision with the greatest impact on a typical pension fund’s return. But if a diversified mix (such as the average asset mix across large pension funds) is the alternative, then the impact of departing from this naive allocation may be no greater than the impact of other decisions, including security selection.
In order to delineate investment responsibility and measure performance contribution, pension plan sponsors and investment managers need a clear and relevant method of attributing returns to those activities that compose the investment management process—investment policy, market timing, and security selection. The authors provide a simple framework based on a passive, benchmark portfolio representing the plan's long-term asset classes, weighted by their long-term allocations. Returns on this "investment policy" portfolio are compared with the actual returns resulting from the combination of investment policy plus market timing (over- or underweighting within an asset class). Data from 91 large U.S. pension plans over the 1974-83 period indicate that investment policy dominates investment strategy (market timing and security selection), explaining on average 95.6 percent of the variation in total plan return. The actual mean average total return on the portfolio over the period was 9.01 percent, versus 10...
We quantify the effects of closing price manipulation on trading characteristics and stock price accuracy using a unique sample of prosecuted manipulation cases. Based on these findings we construct an index of the probability and intensity of closing price manipulation. As well as having regulatory applications, this index can be used to study manipulation in the large number of markets and time periods in which prosecution data are not readily available.
This paper analyzes the portfolio problem of an investor who can invest in bonds, stock, and cash when there is time variation in expected returns on the asset classes. The time variation is assumed to be driven by three state variables, the short-term interest rate, the rate on long-term bonds, and the dividend yield on a stock portfolio, which are all assumed to follow a joint Markov process. The process is estimated from empirical data and the investor's optimal control problem is solved numerically for the resulting parameter values. The optimal portfolio proportions of an investor with a long horizon are compared with those of an investor with a short horizon such as is typically assumed in ‘tactical asset allocation’ models: they are found to be significantly different. Out of sample simulation results provide encouraging evidence that the predictability of asset returns is sufficient for strategies that take it into account to yield significant improvements in portfolio returns.
This study presents a hybrid AI (artificial intelligence) approach to the implementation of trading strategies in the S&P 500 stock index futures market. The hybrid AI approach integrates the rule-based systems technique and the neural networks technique to accurately predict the direction of daily price changes in S&P 500 stock index futures. By highlighting the advantages and overcoming the limitations of both the neural networks technique and rule-based systems technique, the hybrid approach can facilitate the development of more reliable intelligent systems to model expert thinking and to support the decision-making processes. Our methodology differs from other studies in two respects. First, the rule-based systems approach is applied to provide neural networks with training examples. Second, we employ Reasoning Neural Networks (RN) instead of Back Propagation Networks. Empirical results demonstrate that RN outperforms the other two ANN models (Back Propagation Networks and Perceptron). Based upon this hybrid AI approach, the integrated futures trading system (IFTS) is established and employed to trade the S&P 500 stock index futures contracts. Empirical results also confirm that IFTS outperformed the passive buy-and-hold investment strategy during the 6-year testing period from 1988 to 1993.
Portfolio optimization problem has been studied extensively. In this paper, we look at this problem from a different perspective. Several researchers argue that the USA equity market is efficient. Some of the studies show that the stock market is not efficient around the earning season. Based on these findings, we formulate the problem as a classification problem by using state of the art machine learning techniques such as minimax probability machine (MPM) and support vector machines (SVM). The MPM method finds a bound on the misclassification probabilities. On the other hand, SVM finds a hyperplane that maximizes the distance between two classes. Both methods prove similar results for short-term portfolio management.
The performance of ROC on the BSE. Theoretical and applied
- Anghel D. G.
Attention mechanism in the prediction of stock price movement by using LSTM evidence from: The Hong Kong stock market
- Chen S.
Stock trend prediction with technical indicators using SVM. Independent Work Report Stanford Univ
- X Di
- Campbell J. Y.
Algorithm trading using q‐learning and recurrent reinforcement learning
- Du X.