Deep reinforcement learning stock market trading, utilizing a CNN with candlestick images.

Billions of dollars are traded automatically in the stock market every day, including algorithms that use neural networks, but there are still questions regarding how neural networks trade. The black box nature of a neural network gives pause to entrusting it with valuable trading funds. A more recent technique for the study of neural networks, feature map visualizations, yields insight into how a neural network generates an output. Utilizing a Convolutional Neural Network (CNN) with candlestick images as input and feature map visualizations gives a unique opportunity to determine what in the input images is causing the neural network to output a certain action. In this study, a CNN is utilized within a Double Deep Q-Network (DDQN) to outperform the S&P 500 Index returns, and also analyze how the system trades. The DDQN is trained and tested on the 30 largest stocks in the S&P 500. Following training the CNN is used to generate feature map visualizations to determine where the neural network is placing its attention on the candlestick images. Results show that the DDQN is able to yield higher returns than the S&P 500 Index between January 2, 2020 and June 30, 2020. Results also show that the CNN is able to switch its attention from all the candles in a candlestick image to the more recent candles in the image, based on an event such as the coronavirus stock market crash of 2020.

Projecting Quarantine Utilization During a Pandemic.

Objectives. To develop an approach to project quarantine needs during an outbreak, particularly for communally housed individuals who interact with outside individuals. Methods. We developed a method that uses basic surveillance data to do short-term projections of future quarantine needs. The development of this method was rigorous, but it is conceptually simple and easy to implement and allows one to anticipate potential superspreading events. We demonstrate how this method can be used with data from the fall 2020 semester of a large urban university in Boston, Massachusetts, that provided quarantine housing for students living on campus in response to the COVID-19 pandemic. Our approach accounted for potentially infectious interactions between individuals living in university housing and those who did not. Results. Our approach was able to accurately project 10-day-ahead quarantine utilization for on-campus students in a large urban university. Our projections were most accurate when we anticipated weekend superspreading events around holidays. Conclusions. We provide an easy-to-use software tool to project quarantine utilization for institutions that can account for mixing with outside populations. This software tool has potential application for universities, corrections facilities, and the military. (Am J Public Health. 2022;112(2):277-283. https://doi.org/10.2105/AJPH.2021.306573).

Travel time prediction of urban public transportation based on detection of single routes.

Improving travel time prediction for public transit effectively enhances service reliability, optimizes travel structure, and alleviates traffic problems. Its greater time-variance and uncertainty make predictions for short travel times (≤35min) more subject to be influenced by random factors. It requires higher precision and is more complicated than long-term predictions. Effectively extracting and mining real-time, accurate, reliable, and low-cost multi-source data such as GPS, AFC, and IC can provide data support for travel time prediction. Kalman filter model has high accuracy in one-step prediction and can be used to calculate a large amount of data. This paper adopts the Kalman filter as a travel time prediction model for a single bus based on single-line detection: including the travel time prediction model of route (RTM) and the stop dwell time prediction model (DTM); the evaluation criteria and indexes of the models are given. The error analysis of the prediction results is carried out based on AVL data by case study. Results show that under the precondition of multi-source data, the public transportation prediction model can meet the accuracy requirement for travel time prediction and the prediction effect of the whole route is superior to that of the route segment between stops.

Precision Reimbursement for Precision Medicine: Using Real-World Evidence to Evolve From Trial-and-Project to Track-and-Pay to Learn-and-Predict.

Basic scientists and drug developers are accelerating innovations toward the goal of precision medicine. Regulators create pathways for timely patient access to precision medicines, including individualized therapies. Healthcare payors acknowledge the need for change but downstream innovation for coverage and reimbursement is only haltingly occurring. Performance uncertainty, high price-tags, payment timing, and actuarial risk issues associated with precision medicines present novel financial challenges for payors. With traditional drug reimbursement frameworks, payment is based on an assumed randomized controlled trial (RCT) projection of real-world effectiveness, a "trial-and-project" strategy; the clinical benefit realized for patients is not usually ascertained ex post by collection of real-world data (RWD). To mitigate financial risks resulting from clinical performance uncertainty, manufacturers and payors devised "track-and-pay" frameworks (i.e., the tracking of a pre-agreed treatment outcome which is linked to financial consequences). Whereas some track-and-pay arrangements have been successful, inherent weaknesses include the potential for misalignment of incentives, the risk of channeling of patients, and a failure to use the RWD generated to enable continuous learning about treatments. "Precision reimbursement" (PR) intends to overcome inherent weaknesses of simple track-and-pay schemes. In combining the collection of RWD with advanced analytics (e.g., artificial intelligence and machine learning) to generate actionable real-world evidence, with prospective alignment of incentives across all stakeholders (including providers and patients), and with pre-agreed use and dissemination of information generated, PR becomes a "learn-and-predict" model of payment for performance. We here describe in detail the concept of PR and lay out the next steps to make it a reality.

Age-adjusted quick Sequential Organ Failure Assessment score for predicting mortality and disease severity in children with infection: a systematic review and meta-analysis.

We assessed the diagnostic accuracy of the age-adjusted quick Sequential Organ Failure Assessment score (qSOFA) for predicting mortality and disease severity in pediatric patients with suspected or confirmed infection. We conducted a systematic search of PubMed, EMBASE, the Cochrane Library, and Web of Science. Eleven studies with a total of 172,569 patients were included in the meta-analysis. The pooled sensitivity, specificity, and diagnostic odds ratio of the age-adjusted qSOFA for predicting mortality and disease severity were 0.69 (95% confidence interval [CI] 0.53-0.81), 0.71 (95% CI 0.36-0.91), and 6.57 (95% CI 4.46-9.67), respectively. The area under the summary receiver-operating characteristic curve was 0.733. The pooled sensitivity and specificity for predicting mortality were 0.73 (95% CI 0.66-0.79) and 0.63 (95% CI 0.21-0.92), respectively. The pooled sensitivity and specificity for predicting disease severity were 0.73 (95% CI 0.21-0.97) and 0.72 (95% CI 0.11-0.98), respectively. The performance of the age-adjusted qSOFA for predicting mortality and disease severity was better in emergency department patients than in intensive care unit patients. The age-adjusted qSOFA has moderate predictive power and can help in rapidly identifying at-risk children, but its utility may be limited by its insufficient sensitivity.

Active Aging through Volunteerism: A Longitudinal Assessment of Perceived Neighborhood Safety as a Predictor among Older Adults, 2008-2018.

Volunteering can play an important role in active aging. The resource theory of volunteering posits that volunteerism depends on human, social, and cultural capital. Benefits of volunteering have been documented at the micro-, meso-, and macrolevels, positively affecting individual older people as well as their local communities and society at large. Taking a process-oriented theoretical approach, this study focused on the mesolevel factor of the environment with the purpose of determining the relationship between perceived neighborhood safety and volunteerism over the course of a decade and the extent to which this relationship differs by gender and race. Longitudinal data from the Health and Retirement Study in the United States of America between 2008 and 2018 were used (N = 72,319 adults 60 years and older). Generalized estimating equations (GEE) with robust standard errors were employed while controlling for a number of covariates. A third of the sample volunteered in the past year (33%). The probability of volunteering among older adults who rated their perceived neighborhood safety as excellent was greater compared with those who rated their perceived neighborhood safety as fair/poor after controlling for all other model covariates (ME: 0.03, 95% CI: 0.02, 0.05). Among males rating their perceived neighborhood safety as excellent, the probability of volunteering was higher (ME: 0.04, 95% CI: 0.02, 0.07). Among females, the probability of volunteering was higher among those who perceived their neighborhood safety to be excellent (ME: 0.03, 95% CI: 0.01, 0.05) or very good (ME: 0.02, 95% CI: 0.00, 0.04). White respondents who rated their neighborhood safety as excellent (ME: 0.05, 95% CI: 0.03, 0.07) or very good (ME: 0.04, 95% CI: 0.02, 0.06) had a higher probability of volunteerism. Results were not significant among Black respondents and those who described their race as "other." This study's process-oriented theoretical approach indicates that initiatives aimed at improving neighborhood safety and older adults' perceptions of neighborhood safety could increase social capital and lead older adults to engage in more volunteering, providing benefits at micro-, meso-, and macrolevels-to older individuals, their local communities, and society at large.

Improving stock trading decisions based on pattern recognition using machine learning technology.

PRML, a novel candlestick pattern recognition model using machine learning methods, is proposed to improve stock trading decisions. Four popular machine learning methods and 11 different features types are applied to all possible combinations of daily patterns to start the pattern recognition schedule. Different time windows from one to ten days are used to detect the prediction effect at different periods. An investment strategy is constructed according to the identified candlestick patterns and suitable time window. We deploy PRML for the forecast of all Chinese market stocks from Jan 1, 2000 until Oct 30, 2020. Among them, the data from Jan 1, 2000 to Dec 31, 2014 is used as the training data set, and the data set from Jan 1, 2015 to Oct 30, 2020 is used to verify the forecasting effect. Empirical results show that the two-day candlestick patterns after filtering have the best prediction effect when forecasting one day ahead; these patterns obtain an average annual return, an annual Sharpe ratio, and an information ratio as high as 36.73%, 0.81, and 2.37, respectively. After screening, three-day candlestick patterns also present a beneficial effect when forecasting one day ahead in that these patterns show stable characteristics. Two other popular machine learning methods, multilayer perceptron network and long short-term memory neural networks, are applied to the pattern recognition framework to evaluate the dependency of the prediction model. A transaction cost of 0.2% is considered on the two-day patterns predicting one day ahead, thus confirming the profitability. Empirical results show that applying different machine learning methods to two-day and three-day patterns for one-day-ahead forecasts can be profitable.

Where Do We Go From Here? A Framework for Using Susceptible-Infectious-Recovered Models for Policy Making in Emerging Infectious Diseases.

OBJECTIVES: Throughout the coronavirus disease 2019 pandemic, susceptible-infectious-recovered (SIR) modeling has been the preeminent modeling method to inform policy making worldwide. Nevertheless, the usefulness of such models has been subject to controversy. An evolution in the epidemiological modeling field is urgently needed, beginning with an agreed-upon set of modeling standards for policy recommendations. The objective of this article is to propose a set of modeling standards to support policy decision making. METHODS: We identify and describe 5 broad standards: transparency, heterogeneity, calibration and validation, cost-benefit analysis, and model obsolescence and recalibration. We give methodological recommendations and provide examples in the literature that employ these standards well. We also develop and demonstrate a modeling practices checklist using existing coronavirus disease 2019 literature that can be employed by readers, authors, and reviewers to evaluate and compare policy modeling literature along our formulated standards. RESULTS: We graded 16 articles using our checklist. On average, the articles met 6.81 of our 19 categories (36.7%). No articles contained any cost-benefit analyses and few were adequately transparent. CONCLUSIONS: There is significant room for improvement in modeling pandemic policy. Issues often arise from a lack of transparency, poor modeling assumptions, lack of a system-wide perspective in modeling, and lack of flexibility in the academic system to rapidly iterate modeling as new information becomes available. In anticipation of future challenges, we encourage the modeling community at large to contribute toward the refinement and consensus of a shared set of standards for infectious disease policy modeling.

Impact of assumptions on future costs, disutility and mortality in cost-effectiveness analysis; a model exploration.

INTRODUCTION: In cost-effectiveness analyses, the future costs, disutility and mortality from alternative causes of morbidity are often not completely taken into account. We explored the impact of different assumed values for each of these factors on the cost-effectiveness of screening for colorectal cancer (CRC) and esophageal adenocarcinoma (EAC). METHODS: Twenty different CRC screening strategies and two EAC screening strategies were evaluated using microsimulation. Average health-related expenses, disutility and mortality by age for the U.S. general population were estimated using surveys and lifetables. First, we evaluated strategies under default assumptions, with average mortality, and no accounting for health-related costs and disutility. Then, we varied costs, disutility and mortality between 100% and 150% of the estimated population averages, with 125% as the best estimate. Primary outcome was the incremental cost per quality-adjusted life-year (QALY) gained among efficient strategies. RESULTS: The set of efficient strategies was robust to assumptions on future costs, disutility and mortality from other causes of morbidity. However, the incremental cost per QALY gained increased with higher assumed values. For example, for CRC, the ratio for the recommended strategy increased from $15,600 with default assumptions, to $32,600 with average assumption levels, $61,100 with 25% increased levels, and $111,100 with 50% increased levels. Similarly, for EAC, the incremental costs per QALY gained for the recommended EAC screening strategy increased from $106,300 with default assumptions to $198,300 with 50% increased assumptions. In sensitivity analyses without discounting or including only above-average expenses, the impact of assumptions was relatively smaller, but best estimates of the cost per QALY gained remained substantially higher than default estimates. CONCLUSIONS: Assumptions on future costs, utility and mortality from other causes of morbidity substantially impact cost-effectiveness outcomes of cancer screening. More empiric evidence and consensus are needed to guide assumptions in future analyses.

A community-powered search of machine learning strategy space to find NMR property prediction models.

The rise of machine learning (ML) has created an explosion in the potential strategies for using data to make scientific predictions. For physical scientists wishing to apply ML strategies to a particular domain, it can be difficult to assess in advance what strategy to adopt within a vast space of possibilities. Here we outline the results of an online community-powered effort to swarm search the space of ML strategies and develop algorithms for predicting atomic-pairwise nuclear magnetic resonance (NMR) properties in molecules. Using an open-source dataset, we worked with Kaggle to design and host a 3-month competition which received 47,800 ML model predictions from 2,700 teams in 84 countries. Within 3 weeks, the Kaggle community produced models with comparable accuracy to our best previously published 'in-house' efforts. A meta-ensemble model constructed as a linear combination of the top predictions has a prediction accuracy which exceeds that of any individual model, 7-19x better than our previous state-of-the-art. The results highlight the potential of transformer architectures for predicting quantum mechanical (QM) molecular properties.

DPP: Deep predictor for price movement from candlestick charts.

Forecasting the stock market prices is complicated and challenging since the price movement is affected by many factors such as releasing market news about earnings and profits, international and domestic economic situation, political events, monetary policy, major abrupt affairs, etc. In this work, a novel framework: deep predictor for price movement (DPP) using candlestick charts in the stock historical data is proposed. This framework comprises three steps: 1. decomposing a given candlestick chart into sub-charts; 2. using CNN-autoencoder to acquire the best representation of sub-charts; 3. applying RNN to predict the price movements from a collection of sub-chart representations. An extensive study is operated to assess the performance of the DPP based models using the trading data of Taiwan Stock Exchange Capitalization Weighted Stock Index and a stock market index, Nikkei 225, for the Tokyo Stock Exchange. Three baseline models based on IEM, Prophet, and LSTM approaches are compared with the DPP based models.

The predictive power of stock market's expectations volatility: A financial synchronization phenomenon.

We explore the use of implied volatility indices as a tool for estimate changes in the synchronization of stock markets. Specifically, we assess the implied stock market's volatility indices' predictive power on synchronizing global equity indices returns. We built the correlation network of 26 stock indices and implemented in-sample and out-of-sample tests to evaluate the predictive power of VIX, VSTOXX, and VXJ implied volatility indices. To measure markets' synchronization, we use the Minimum Spanning Tree length and the length of the Planar Maximally Filtered Graph. Our results indicate a high predictive power of all the volatility indices, both individually and together, though the VIX predominates over the evaluated options. We find that an increase in the markets' volatility expectations, captured by the implied volatility indices, is a good Granger predictor of an increase in the synchronization of returns in the following month. Estimating, monitoring, and predicting returns' synchronization is essential for investment decision-making, especially for diversification strategies and regulating financial systems.

Intelligent prediction of RBC demand in trauma patients using decision tree methods.

BACKGROUND: The vital signs of trauma patients are complex and changeable, and the prediction of blood transfusion demand mainly depends on doctors' experience and trauma scoring system; therefore, it cannot be accurately predicted. In this study, a machine learning decision tree algorithm [classification and regression tree (CRT) and eXtreme gradient boosting (XGBoost)] was proposed for the demand prediction of traumatic blood transfusion to provide technical support for doctors. METHODS: A total of 1371 trauma patients who were diverted to the Emergency Department of the First Medical Center of Chinese PLA General Hospital from January 2014 to January 2018 were collected from an emergency trauma database. The vital signs, laboratory examination parameters and blood transfusion volume were used as variables, and the non-invasive parameters and all (non-invasive + invasive) parameters were used to construct an intelligent prediction model for red blood cell (RBC) demand by logistic regression (LR), CRT and XGBoost. The prediction accuracy of the model was compared with the area under the curve (AUC). RESULTS: For non-invasive parameters, the LR method was the best, with an AUC of 0.72 [95% confidence interval (CI) 0.657-0.775], which was higher than the CRT (AUC 0.69, 95% CI 0.633-0.751) and the XGBoost (AUC 0.71, 95% CI 0.654-0.756, P < 0.05). The trauma location and shock index are important prediction parameters. For all the prediction parameters, XGBoost was the best, with an AUC of 0.94 (95% CI 0.893-0.981), which was higher than the LR (AUC 0.80, 95% CI 0.744-0.850) and the CRT (AUC 0.82, 95% CI 0.779-0.853, P < 0.05). Haematocrit (Hct) is an important prediction parameter. CONCLUSIONS: The classification performance of the intelligent prediction model of red blood cell transfusion in trauma patients constructed by the decision tree algorithm is not inferior to that of the traditional LR method. It can be used as a technical support to assist doctors to make rapid and accurate blood transfusion decisions in emergency rescue environment, so as to improve the success rate of patient treatment.

Predicting postoperative opioid use with machine learning and insurance claims in opioid-naïve patients.

BACKGROUND: The clinical impact of postoperative opioid use requires accurate prediction strategies to identify at-risk patients. We utilize preoperative claims data to predict postoperative opioid refill and new persistent use in opioid-naïve patients. METHODS: A retrospective study was conducted on 112,898 opioid-naïve adult postoperative patients from Optum's de-identified Clinformatics® Data Mart database. Potential predictors included sociodemographic data, comorbidities, and prescriptions within one year prior to surgery. RESULTS: Compared to linear models, non-linear models led to modest improvements in predicting refills - area under the receiver operating characteristics curve (AUROC) 0.68 vs. 0.67 (p < 0.05) - and performed identically in predicting new persistent use - AUROC = 0.66. Undergoing major surgery, opioid prescriptions within 30 days prior to surgery, and abdominal pain were useful in predicting refills; back/joint/head pain were the most important features in predicting new persistent use. CONCLUSIONS: Preoperative patient attributes from insurance claims could potentially be useful in guiding prescription practices for opioid-naïve patients.

Index tracking strategy based on mixed-frequency financial data.

To obtain market average return, investment managers need to construct index tracking portfolio to replicate target index. Currently, most literatures use financial data that has homogenous frequency when constructing the index tracking portfolio. To make up for this limitation, we propose a methodology based on mixed-frequency financial data, called FACTOR-MIDAS-POET model. The proposed model can utilize the intraday return data, daily risk factors data and monthly or quarterly macro economy data, simultaneously. Meanwhile, the out-of-sample analysis demonstrates that our model can improve the tracking accuracy.

Predicting corporate credit risk: Network contagion via trade credit.

Trade credit is a payment extension granted by a selling firm to its customer. Companies typically respond to late payments from their customers by delaying payments to suppliers, thus generating a ripple through the transaction network. Therefore, trade credit is as a potential vehicle of propagation of losses in case of default events. The goal of this work is to leverage information on the trade credit among connected firms to predict imminent defaults of firms. We use a unique dataset of client firms of a major Italian bank to investigate firm bankruptcy between October 2016 to March 2018. We develop a model to capture network spillover effects originating from the supply chain on the probability of default of each firm via a sequential approach: the output of a first model component on single firm features is used in a subsequent model which captures network spillovers. While the first component is the standard econometrics way to predict such dynamics, the network module represents an innovative way to look into the effect of trade credit on default probability. This module looks at the transaction network of the firm, as inferred from the payments transiting via the bank, in order to identify the trade partners of the firm. By using several features extracted from the network of transactions, this model is able to predict a large fraction of the defaults, thus showing the value hidden in the network information. Finally, we merge firm and network features with a machine learning model to create a 'hybrid' model, which improves the recall for the task by almost 20 percentage points over the baseline.

Vertical tank capacity measurement based on Monte Carlo method.

Vertical tanks are commonly used appliances for liquids, and its capacity is very important for quantitative liquid ratio and liquid trade. In order to measure the capacity of vertical tanks more conveniently, this paper proposes a vertical tank capacity measurement method based on Monte Carlo Method. The method arranges a plurality of sensor points on the inner surface of the tank, and then performs Monte Carlo tests by generating a large number of random sample points, and finally calculates the capacity by counting the sample points that meet the criterion. The criterion for whether a sample point is located in the tank, which is the core issue, is established with the coordinates of sensor points and the distance between different sensor points along the surface of the tank. The results show that the absolute error of the measurement results of the proposed method does not exceed ±0.0003[m3], and the absolute error of capacity per unit volume has a linear relationship with the height of the vertical tank, and has little effect with the radial size of the vertical tank.

A versatile web app for identifying the drivers of COVID-19 epidemics.

BACKGROUND: No versatile web app exists that allows epidemiologists and managers around the world to comprehensively analyze the impacts of COVID-19 mitigation. The http://covid-webapp.numerusinc.com/ web app presented here fills this gap. METHODS: Our web app uses a model that explicitly identifies susceptible, contact, latent, asymptomatic, symptomatic and recovered classes of individuals, and a parallel set of response classes, subject to lower pathogen-contact rates. The user inputs a CSV file of incidence and, if of interest, mortality rate data. A default set of parameters is available that can be overwritten through input or online entry, and a user-selected subset of these can be fitted to the model using maximum-likelihood estimation (MLE). Model fitting and forecasting intervals are specifiable and changes to parameters allow counterfactual and forecasting scenarios. Confidence or credible intervals can be generated using stochastic simulations, based on MLE values, or on an inputted CSV file containing Markov chain Monte Carlo (MCMC) estimates of one or more parameters. RESULTS: We illustrate the use of our web app in extracting social distancing, social relaxation, surveillance or virulence switching functions (i.e., time varying drivers) from the incidence and mortality rates of COVID-19 epidemics in Israel, South Africa, and England. The Israeli outbreak exhibits four distinct phases: initial outbreak, social distancing, social relaxation, and a second wave mitigation phase. An MCMC projection of this latter phase suggests the Israeli epidemic will continue to produce into late November an average of around 1500 new case per day, unless the population practices social-relaxation measures at least 5-fold below the level in August, which itself is 4-fold below the level at the start of July. Our analysis of the relatively late South African outbreak that became the world's fifth largest COVID-19 epidemic in July revealed that the decline through late July and early August was characterised by a social distancing driver operating at more than twice the per-capita applicable-disease-class (pc-adc) rate of the social relaxation driver. Our analysis of the relatively early English outbreak, identified a more than 2-fold improvement in surveillance over the course of the epidemic. It also identified a pc-adc social distancing rate in early August that, though nearly four times the pc-adc social relaxation rate, appeared to barely contain a second wave that would break out if social distancing was further relaxed. CONCLUSION: Our web app provides policy makers and health officers who have no epidemiological modelling or computer coding expertise with an invaluable tool for assessing the impacts of different outbreak mitigation policies and measures. This includes an ability to generate an epidemic-suppression or curve-flattening index that measures the intensity with which behavioural responses suppress or flatten the epidemic curve in the region under consideration.