Search results

Machine learning (ML), and deep learning in particular, is gaining traction across a myriad of real-life applications. Portfolio management is no exception. This paper provides a systematic literature review of deep learning applications for portfolio management. The findings are likely to be valuable for industry practitioners and researchers alike, experimenting with novel portfolio management approaches and furthering investment management practice.

This review follows the guidance and methodology of Linnenluecke et al. (2020), Massaro et al. (2016) and Fisch and Block (2018) to first identify relevant literature based on an appropriately developed search phrase, filter the resultant set of publications and present descriptive and analytical findings of the research itself and its metadata.

The authors find a strong dominance of reinforcement learning algorithms applied to the field, given their through-time portfolio management capabilities. Other well-known deep learning models, such as convolutional neural network (CNN) and recurrent neural network (RNN) and its derivatives, have shown to be well-suited for time-series forecasting. Most recently, the number of papers published in the field has been increasing, potentially driven by computational advances, hardware accessibility and data availability. The review shows several promising applications and identifies future research opportunities, including better balance on the risk-reward spectrum, novel ways to reduce data dimensionality and pre-process the inputs, stronger focus on direct weights generation, novel deep learning architectures and consistent data choices.

Several systematic reviews have been conducted with a broader focus of ML applications in finance. However, to the best of the authors’ knowledge, this is the first review to focus on deep learning architectures and their applications in the investment portfolio management problem. The review also presents a novel universal taxonomy of models used.

Deep learning (DL) is a new and relatively unexplored field that finds immense applications in many industries, especially ones that must make detailed observations, inferences and predictions based on extensive and scattered datasets. The purpose of this paper is to answer the following questions: (1) To what extent has DL penetrated the research being done in finance? (2) What areas of financial research have applications of DL, and what quality of work has been done in the niches? (3) What areas still need to be explored and have scope for future research?

This paper employs bibliometric analysis, a potent yet simple methodology with numerous applications in literature reviews. This paper focuses on citation analysis, author impacts, relevant and vital journals, co-citation analysis, bibliometric coupling and co-occurrence analysis. The authors collected 693 articles published in 2000–2022 from journals indexed in the Scopus database. Multiple software (VOSviewer, RStudio (biblioshiny) and Excel) were employed to analyze the data.

The findings reveal significant and renowned authors' impact in the field. The analysis indicated that the application of DL in finance has been on an upward track since 2017. The authors find four broad research areas (neural networks and stock market simulations; portfolio optimization and risk management; time series analysis and forecasting; high-frequency trading) with different degrees of intertwining and emerging research topics with the application of DL in finance. This article contributes to the literature by providing a systematic overview of the DL developments, trajectories, objectives and potential future research topics in finance.

The findings of this paper act as a guide for literature review for anyone interested in doing research in the intersection of finance and DL. The article also explores multiple areas of research that have yet to be studied to a great extent and have abundant scope.

Very few studies have explored the applications of machine learning (ML), namely, DL in finance, which is a much more specialized subset of ML. The authors look at the problem from the aspect of different techniques in DL that have been used in finance. This is the first qualitative (content analysis) and quantitative (bibliometric analysis) assessment of current research on DL in finance.

The digital media recording and broadcasting classroom using Internet real-time intelligent image positioning and opinion monitoring in communication teaching is researched and analyzed.

First, spatial grid positioning and monitoring and image intelligent recognition technologies were used to extract and analyze teaching images by mastering Internet of Things (IoT) technology and establishing an intelligent image positioning and opinion monitoring digital media recording and broadcasting system framework. Next, a positioning node algorithm was utilized to measure the image distance, and then a moving node location model under the IoT was established. In addition, a radial basis function (RBF) neural network was used to realize the signal transmission function. The experimental data of the adopted RBF based on the optimization of the adaptive cuckoo search (ACS-RBF) neural network, particle swarm algorithm neural network, and method of least squares optimization were compared and analyzed. In addition, a more efficient RBF neural network was adopted. Finally, the digital media recording and broadcasting classroom scheme of real-time intelligent image positioning and opinion monitoring was designed. In addition, the application environment of digital media actual teacher teaching was detected, and recording and broadcasting pictures were analyzed and researched.

The actual value, predicted value, and the number of predicted samples of the ACS-RBF model were all better than those of the two other neural networks. According to the analysis and comparison of the sampling optimization Monte Carlo localization (SOMCL), Monte Carlo, and genetic algorithm optimization-based Monte Carlo positioning algorithms, the SOMCL algorithm showed better robustness, and its positioning efficiency was superior to that of the two other algorithms. In addition, the SOMCL algorithm greatly reduced the positioning and monitoring energy consumption.

The application of real-time intelligent image positioning and monitoring technology in actual communication teaching was realized in the study.

The choice of a sales channel for fresh vegetables is an important decision a farmer can make. Typically, the farmers rely on their personal experience in directing the produce to a sales channel. This study examines how sales forecasting of fresh vegetables along multiple channels enables marginal and small-scale farmers to maximize their revenue by proportionately allocating the produce considering their short shelf life.

Machine learning models, namely long short-term memory (LSTM), convolution neural network (CNN) and traditional methods such as autoregressive integrated moving average (ARIMA) and weighted moving average (WMA) are developed and tested for demand forecasting of vegetables through three different channels, namely direct (Jaivasree), regulated (World market) and cooperative (Horticorp).

The results show that machine learning methods (LSTM/CNN) provide better forecasts for regulated (World market) and cooperative (Horticorp) channels, while traditional moving average yields a better result for direct (Jaivasree) channel where the sales volume is less as compared to the remaining two channels.

The price of vegetables is not considered as the government sets the base price for the vegetables.

The existing literature lacks models and approaches to predict the sales of fresh vegetables for marginal and small-scale farmers of developing economies like India. In this research, the authors forecast the sales of commonly used fresh vegetables for small-scale farmers of Kerala in India based on a set of 130 weekly time series data obtained from the Kerala Horticorp.

Higher energy conversion efficiency of internal combustion engine can be achieved with optimal control of unsteady in-cylinder flow fields inside a direct-injection (DI) engine. However, it remains a daunting task to predict the nonlinear and transient in-cylinder flow motion because they are highly complex which change both in space and time. Recently, machine learning methods have demonstrated great promises to infer relatively simple temporal flow field development. This paper aims to feature a physics-guided machine learning approach to realize high accuracy and generalization prediction for complex swirl-induced flow field motions.

To achieve high-fidelity time-series prediction of unsteady engine flow fields, this work features an automated machine learning framework with the following objectives: (1) The spatiotemporal physical constraint of the flow field structure is transferred to machine learning structure. (2) The ML inputs and targets are efficiently designed that ensure high model convergence with limited sets of experiments. (3) The prediction results are optimized by ensemble learning mechanism within the automated machine learning framework.

The proposed data-driven framework is proven effective in different time periods and different extent of unsteadiness of the flow dynamics, and the predicted flow fields are highly similar to the target field under various complex flow patterns. Among the described framework designs, the utilization of spatial flow field structure is the featured improvement to the time-series flow field prediction process.

The proposed flow field prediction framework could be generalized to different crank angle periods, cycles and swirl ratio conditions, which could greatly promote real-time flow control and reduce experiments on in-cylinder flow field measurement and diagnostics.

Because of the key role of joint torque in industrial robots (IRs) motion performance control and energy consumption calculation and efficiency optimization, the purpose of this paper is to propose a deep learning torque prediction method based on long short-term memory (LSTM) recurrent neural networks optimized by particle swarm optimization (PSO), which can accurately predict the the joint torque.

The proposed model optimized the LSTM with PSO algorithm to accurately predict the IRs joint torque. The authors design an excitation trajectory for ABB 1600–10/145 experimental robot and collect its relative dynamic data. The LSTM model was trained with the experimental data, and PSO was used to find optimal number of LSTM nodes and learning rate, then a torque prediction model is established based on PSO-LSTM deep learning method. The novel model is used to predict the robot’s six joint torque and the root mean error squares of the predicted data together with least squares (LS) method were comparably studied.

The predicted joint torque value by PSO-LSTM deep learning approach is highly overlapped with those from real experiment robot, and the error is quite small. The average square error between the predicted joint torque data and experiment data is 2.31 N.m smaller than that with the LS method. The accuracy of the novel PSO-LSTM learning method for joint torque prediction of IR is proved.

PSO and LSTM model are deeply integrated for the first time to predict the joint torque of IR and the prediction accuracy is verified.

This paper assesses the co-movement between Pax gold and six fiat currencies. It also investigates the optimal time-varying hedge ratios in order to examine the properties of Pax gold as a diversifier and hedge asset.

This paper examines the volatility spillover between Pax gold and fiat currencies using the framework of wavelet analysis, BEKK-GARCH models and Range DCC-GARCH. Moreover, this paper proposes to use the covariance and variance structure obtained from the new range DCC-GARCH framework to estimate the time-varying optimal hedge ratios, the optimal weighs and the hedging effectiveness.

Wavelet coherence method reveals that, at low frequency, large zone of co-movements appears for the pairs Pax gold/EUR, Pax gold/JPY and Pax gold/RUB. Further, the BEKK results show unidirectional (bidirectional) transmission effects between Pax gold and EUR, GBP, JPY and CNY (INR, RUB) fiat currencies. Moreover, the Range DCC results show that the Pax gold and the fiat currency returns are weakly correlated with low coefficients close to zero. Thus, Pax gold seems to serve as a safe haven asset against the systematic risk of fiat currency markets. In addition, the results of optimal weights show that rational investor should invest more in Pax gold and less in fiat currencies. Concerning the hedge ratios results, the findings reveal that the INR (JPY) fiat currency appears to be the most expensive (cheapest) hedge for the Pax-gold market. However, the JPY’s fiat currency appears to be the cheapest one. As for hedging effectiveness results, the authors found that hedging strategies including fiat currencies–Pax gold pairs are most likely to sharply decrease the portfolio’s risk.

A comprehensive understanding of the relationship between Pax Gold and fiat currencies is crucial for refining portfolio strategies involving cryptocurrencies. This research underscores the significance of grasping volatility transmissions between these currencies, providing valuable insights to guide investors in their decision-making processes. Moreover, it encourages further exploration into the interdependencies of digital currencies. Additionally, this study sheds light on effective contagion risk management, particularly during crises such as Covid-19 and the Russia–Ukraine conflict. It underscores the role of Pax Gold as a safe-haven asset and offers practical guidance for adjusting portfolios across various economic conditions. Ultimately, this research advances our comprehension of Pax Gold’s risk-return profile, positioning it as a potential hedge during periods of uncertainty, thereby contributing to the evolving literature on cryptocurrencies.

This study’s primary value lies in its pioneering empirical examination of the time-varying correlations and scale dependence between Pax Gold and fiat currencies. It goes beyond by determining optimal time-varying hedge ratios through the innovative Range-DCC-GARCH model, originally introduced by Molnár (2016) and distinguished by its incorporation of both low and high prices. Significantly, this analysis unfolds within the unique context of the Covid-19 pandemic and the Russian–Ukrainian conflict, marking a novel contribution to the field.

This research presents machine learning models for predicting international tourist arrivals in Indonesia during the COVID-19 pandemic using multisource Internet data.

To develop the prediction models, this research utilizes multisource Internet data from TripAdvisor travel forum and Google Trends. Temporal factors, posts and comments, search queries index and previous tourist arrivals records are set as predictors. Four sets of predictors and three distinct data compositions were utilized for training the machine learning models, namely artificial neural networks (ANNs), support vector regression (SVR) and random forest (RF). To evaluate the models, this research uses three accuracy metrics, namely root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE).

Prediction models trained using multisource Internet data predictors have better accuracy than those trained using single-source Internet data or other predictors. In addition, using more training sets that cover the phenomenon of interest, such as COVID-19, will enhance the prediction model's learning process and accuracy. The experiments show that the RF models have better prediction accuracy than the ANN and SVR models.

First, this study pioneers the practice of a multisource Internet data approach in predicting tourist arrivals amid the unprecedented COVID-19 pandemic. Second, the use of multisource Internet data to improve prediction performance is validated with real empirical data. Finally, this is one of the few papers to provide perspectives on the current dynamics of Indonesia's tourism demand.

Driven by the anticipated global stagflation, this straightforward yet novel study examines the cost of inflation as a macroeconomic factor by investigating its influence on stock market growth. Thus, this paper aims to examine the impact of inflation on the probability of initial public offering (IPO) withdrawal decision.

The paper employs a large dataset that covers the period January 1995–December 2019 and comprises 33,536 successful or withdrawn IPOs from 22 nations with various legal and cultural systems. This study applies a probit model utilizing version 15 of Stata statistical software.

This study finds that inflation is substantially and positively correlated with the likelihood of IPO withdrawal. Results of this study show that the IPO withdrawal decision increases up to 90% when the inflation rate climbs by 10%. Multiple robustness tests provide consistent findings.

This study's implications are important for researchers, investment banks, underwriters, issuers, regulators and stock exchanges. When processing IPO proposals, investment banks, underwriters and issuers must consider inflation projections to avoid negative effects, as demonstrated by the findings. In addition, regulators and stock exchanges must be aware of the detrimental impact of inflation on competitiveness in attracting new listings.

To the best of the authors’ knowledge, this study is the first to present convincing evidence of a major relationship between IPO withdrawal decision and inflation.

This paper aims to introduce a new perspective on long-term stock return predictability by focusing on the relative (individual and hybrid) informative power of a wide range of accounting (firm-related), technical and macroeconomic factors while considering the past performance of the stocks using machine learning algorithms.

The sample includes a panel data set of 94 non-financial firms listed in Egyptian Exchange 100 index from 2014: Q1 to 2019: Q4. Relativity has been investigated by comparing relevant factors’ individual and combined informative power and differentiating between losers and winners based on historical stock returns. To predict the quarterly stock returns, Gaussian process regression (GPR) has been used. The robustness of the results is examined through the out-of-sample test. This study also uses linear regression (LR) as a benchmark model.

The past performance and the presence of other predictors influence the informative power of relevant factors and hence their predictive ability. The out-of-sample results show a trade-off between GPR and LR with proven superiority to GPR in limited experiments. The individual informative power outperforms the hybrid power, in which macroeconomic indicators outperform the remaining sets of indicators for losers, while winners show mixed results in terms of various performance evaluation metrics. Prediction accuracy is generally higher for losers than for winners.

This study provides interesting insight into the dynamic nature of the predictor variables in terms of stock return predictability. Hence, this study also deepens the understanding of asset pricing in a way that directly contributes to practitioners’ portfolio diversification strategies.

In concern of the chaos of factors in the literature and its accompanying misleading conclusions, this study takes another look at the approach that studies stock return predictability. To the best of the authors’ knowledge, this is the first study in the Egyptian context that re-examines the predictive power of the previously discovered factors from a different perspective that highlights their relative nature.