Despite the eye-catching advances in the area, deploying Deep Reinforcement Learning (DRL) in financial markets remains a challenging task. Model-based techniques often fall short due to epistemic uncertainty, whereas model-free approaches require large amount of data that is often unavailable. Motivated by the recent research on the generation of realistic synthetic financial data, we explore the possibility of using augmented synthetic datasets for training DRL agents without direct access to the real financial data. With our novel approach, termed synthetic data augmented reinforcement learning for trading (SDARL4T), we test whether the performance of DRL for financial trading can be enhanced, by attending to both profitability and generalization abilities. We show that DRL agents trained with SDARL4T make a profit which is comparable, and often much larger, than that obtained by the agents trained on real data, while guaranteeing similar robustness. These results support the adoption of our framework in real-world uses of DRL for trading.
{"title":"Synthetic Data Augmentation for Deep Reinforcement Learning in Financial Trading","authors":"Chunli Liu, Carmine Ventre, M. Polukarov","doi":"10.1145/3533271.3561704","DOIUrl":"https://doi.org/10.1145/3533271.3561704","url":null,"abstract":"Despite the eye-catching advances in the area, deploying Deep Reinforcement Learning (DRL) in financial markets remains a challenging task. Model-based techniques often fall short due to epistemic uncertainty, whereas model-free approaches require large amount of data that is often unavailable. Motivated by the recent research on the generation of realistic synthetic financial data, we explore the possibility of using augmented synthetic datasets for training DRL agents without direct access to the real financial data. With our novel approach, termed synthetic data augmented reinforcement learning for trading (SDARL4T), we test whether the performance of DRL for financial trading can be enhanced, by attending to both profitability and generalization abilities. We show that DRL agents trained with SDARL4T make a profit which is comparable, and often much larger, than that obtained by the agents trained on real data, while guaranteeing similar robustness. These results support the adoption of our framework in real-world uses of DRL for trading.","PeriodicalId":134888,"journal":{"name":"Proceedings of the Third ACM International Conference on AI in Finance","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127568128","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
A. Alden, Carmine Ventre, Blanka Horvath, Gordon Lee
Neural networks hold out the promise of fast and reliable derivative pricing. Such an approach usually involves the supervised learning task of mapping contract and model parameters to derivative prices. In this work, we introduce a model-agnostic path-wise approach to derivative pricing using higher-order distribution regression. Our methodology leverages the 2nd-order Maximum Mean Discrepancy (MMD), a notion of distance between stochastic processes based on path signatures. To overcome the high computational cost of its calculation, we pre-train a neural network that can quickly and accurately compute higher-order MMDs. This allows the combination of distribution regression with neural networks in a computationally feasible way. We test our model on down-and-in barrier options. We demonstrate that our path-wise approach extends well to the high-dimensional case by applying it to rainbow options and autocallables. Our approach has a significant speed-up over Monte Carlo pricing.
{"title":"Model-Agnostic Pricing of Exotic Derivatives Using Signatures","authors":"A. Alden, Carmine Ventre, Blanka Horvath, Gordon Lee","doi":"10.1145/3533271.3561740","DOIUrl":"https://doi.org/10.1145/3533271.3561740","url":null,"abstract":"Neural networks hold out the promise of fast and reliable derivative pricing. Such an approach usually involves the supervised learning task of mapping contract and model parameters to derivative prices. In this work, we introduce a model-agnostic path-wise approach to derivative pricing using higher-order distribution regression. Our methodology leverages the 2nd-order Maximum Mean Discrepancy (MMD), a notion of distance between stochastic processes based on path signatures. To overcome the high computational cost of its calculation, we pre-train a neural network that can quickly and accurately compute higher-order MMDs. This allows the combination of distribution regression with neural networks in a computationally feasible way. We test our model on down-and-in barrier options. We demonstrate that our path-wise approach extends well to the high-dimensional case by applying it to rainbow options and autocallables. Our approach has a significant speed-up over Monte Carlo pricing.","PeriodicalId":134888,"journal":{"name":"Proceedings of the Third ACM International Conference on AI in Finance","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128597787","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Shibal Ibrahim, Wenyu Chen, Yada Zhu, Ping Chen, Yang Zhang, R. Mazumder
Financial time series forecasting is challenging due to limited sample size, correlated samples, low signal strengths, among others. Additional information with knowledge graphs (KGs) can allow for improved prediction and decision making. In this work, we explore a framework GregNets for jointly learning forecasting models and correlations structures that exploit graph connectivity from KGs. We propose novel regularizers based on KG relations to guide estimation of correlation structure. We develop a pseudo-likelihood layer that can learn the error residual structure for any multivariate time-series forecasting architecture in deep learning APIs (e.g. Tensorflow). We evaluate our modeling and algorithmic proposals in small sample regimes in real-world financial markets with two types of KGs. Our empirical results demonstrate sparser connectivity structures, runtime improvements and high-quality predictions.
{"title":"Knowledge Graph Guided Simultaneous Forecasting and Network Learning for Multivariate Financial Time Series","authors":"Shibal Ibrahim, Wenyu Chen, Yada Zhu, Ping Chen, Yang Zhang, R. Mazumder","doi":"10.1145/3533271.3561702","DOIUrl":"https://doi.org/10.1145/3533271.3561702","url":null,"abstract":"Financial time series forecasting is challenging due to limited sample size, correlated samples, low signal strengths, among others. Additional information with knowledge graphs (KGs) can allow for improved prediction and decision making. In this work, we explore a framework GregNets for jointly learning forecasting models and correlations structures that exploit graph connectivity from KGs. We propose novel regularizers based on KG relations to guide estimation of correlation structure. We develop a pseudo-likelihood layer that can learn the error residual structure for any multivariate time-series forecasting architecture in deep learning APIs (e.g. Tensorflow). We evaluate our modeling and algorithmic proposals in small sample regimes in real-world financial markets with two types of KGs. Our empirical results demonstrate sparser connectivity structures, runtime improvements and high-quality predictions.","PeriodicalId":134888,"journal":{"name":"Proceedings of the Third ACM International Conference on AI in Finance","volume":"134 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132919304","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yue Leng, E. Skiani, William Peak, E. Mackie, Fuyuan Li, Thwisha Charvi, Jennifer Law, Kieran Daly
The financial domain naturally contains multiple different types of entities such as stocks, product categories, investment participants, intermediaries, and customers, and interactions between these entities. This paper introduces a Graph-based Collaborative Filtering Category Recommendation (GCFCR) system as a first step in modelling the financial domain as an inter-connected, heterogeneous, dynamic system of nodes and edges. The goal of this paper is to identify customer interest based on the neighborhood of each node and make personalized suggestions or identify relevant content for each customer. Matching relevant products and services to customers is a key foundation of building and maintaining strong customer relationships, facilitating more personalized marketing which can ultimately result in increased customer activity, trust, and revenue. In this paper, we run a set of experiments to compare different recommendation techniques, concluding that the proposed GCFCR approach outperforms in this real-life application.
{"title":"Customer-Category Interest Model: A Graph-Based Collaborative Filtering Model with Applications in Finance","authors":"Yue Leng, E. Skiani, William Peak, E. Mackie, Fuyuan Li, Thwisha Charvi, Jennifer Law, Kieran Daly","doi":"10.1145/3533271.3561757","DOIUrl":"https://doi.org/10.1145/3533271.3561757","url":null,"abstract":"The financial domain naturally contains multiple different types of entities such as stocks, product categories, investment participants, intermediaries, and customers, and interactions between these entities. This paper introduces a Graph-based Collaborative Filtering Category Recommendation (GCFCR) system as a first step in modelling the financial domain as an inter-connected, heterogeneous, dynamic system of nodes and edges. The goal of this paper is to identify customer interest based on the neighborhood of each node and make personalized suggestions or identify relevant content for each customer. Matching relevant products and services to customers is a key foundation of building and maintaining strong customer relationships, facilitating more personalized marketing which can ultimately result in increased customer activity, trust, and revenue. In this paper, we run a set of experiments to compare different recommendation techniques, concluding that the proposed GCFCR approach outperforms in this real-life application.","PeriodicalId":134888,"journal":{"name":"Proceedings of the Third ACM International Conference on AI in Finance","volume":"10890 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133405700","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Felix Prenzel, R. Cont, Mihai Cucuringu, Jonathan Kochems
Classical models for order flow dynamics based on point processes, such as Poisson or Hawkes processes, have been studied intensively. Often, several days of limit border book (LOB) data is used to calibrate such models, thereby averaging over different dynamics - such as intraday effects or different trading volumes. This work uses generative adversarial networks (GANs) to learn the distribution of calibrations – obtained by many calibrations based on short time frames. The trained GAN can then be used to generate synthetic, realistic calibrations based on external conditions such as time of the day or volatility. Results show that GANs easily reproduce patterns of the order arrival intensities and can fit the distribution well without heavy parameter tuning. The synthetic calibrations can then be used to simulate order streams which contain new dynamics such as temporary drifts, different volatility regimes, but also intra-day patterns such as the commonly observed U-shape that reflects stylized behaviour around open and close of market hours.
{"title":"Dynamic Calibration of Order Flow Models with Generative Adversarial Networks","authors":"Felix Prenzel, R. Cont, Mihai Cucuringu, Jonathan Kochems","doi":"10.1145/3533271.3561777","DOIUrl":"https://doi.org/10.1145/3533271.3561777","url":null,"abstract":"Classical models for order flow dynamics based on point processes, such as Poisson or Hawkes processes, have been studied intensively. Often, several days of limit border book (LOB) data is used to calibrate such models, thereby averaging over different dynamics - such as intraday effects or different trading volumes. This work uses generative adversarial networks (GANs) to learn the distribution of calibrations – obtained by many calibrations based on short time frames. The trained GAN can then be used to generate synthetic, realistic calibrations based on external conditions such as time of the day or volatility. Results show that GANs easily reproduce patterns of the order arrival intensities and can fit the distribution well without heavy parameter tuning. The synthetic calibrations can then be used to simulate order streams which contain new dynamics such as temporary drifts, different volatility regimes, but also intra-day patterns such as the commonly observed U-shape that reflects stylized behaviour around open and close of market hours.","PeriodicalId":134888,"journal":{"name":"Proceedings of the Third ACM International Conference on AI in Finance","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116661433","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We propose an architecture for multivariate time-series prediction that integrates a spatial-temporal graph neural network with a filtering module which filters the inverse correlation matrix into a sparse network structure. In contrast with existing sparsification methods adopted in graph neural networks, our model explicitly leverages time-series filtering to overcome the low signal-to-noise ratio typical of complex systems data. We present a set of experiments, where we predict future sales volume from a synthetic time-series sales volume dataset. The proposed spatial-temporal graph neural network displays superior performances to baseline approaches with no graphical information, fully connected, disconnected graphs, and unfiltered graphs, as well as the state-of-the-art spatial-temporal GNN. Comparison of the results with Diffusion Convolutional Recurrent Neural Network (DCRNN) suggests that, by combining a (inferior) GNN with graph sparsification and filtering, one can achieve comparable or better efficacy than the state-of-the-art in multivariate time-series regression.
{"title":"Network Filtering of Spatial-temporal GNN for Multivariate Time-series Prediction","authors":"Yuanrong Wang, T. Aste","doi":"10.1145/3533271.3561678","DOIUrl":"https://doi.org/10.1145/3533271.3561678","url":null,"abstract":"We propose an architecture for multivariate time-series prediction that integrates a spatial-temporal graph neural network with a filtering module which filters the inverse correlation matrix into a sparse network structure. In contrast with existing sparsification methods adopted in graph neural networks, our model explicitly leverages time-series filtering to overcome the low signal-to-noise ratio typical of complex systems data. We present a set of experiments, where we predict future sales volume from a synthetic time-series sales volume dataset. The proposed spatial-temporal graph neural network displays superior performances to baseline approaches with no graphical information, fully connected, disconnected graphs, and unfiltered graphs, as well as the state-of-the-art spatial-temporal GNN. Comparison of the results with Diffusion Convolutional Recurrent Neural Network (DCRNN) suggests that, by combining a (inferior) GNN with graph sparsification and filtering, one can achieve comparable or better efficacy than the state-of-the-art in multivariate time-series regression.","PeriodicalId":134888,"journal":{"name":"Proceedings of the Third ACM International Conference on AI in Finance","volume":"162 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123421879","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Given their strong performance on a variety of graph learning tasks, Graph Neural Networks (GNNs) are increasingly used to model financial networks. Traditional GNNs, however, are not able to capture higher-order topological information, and their performance is known to degrade with the presence of negative edges that may arise in many common financial applications. Considering the rich semantic inference of negative edges, excluding them as an obvious solution is not elegant. Alternatively, another basic approach is to apply positive normalization, however, this also may lead to information loss. Our work proposes a simple yet effective solution to overcome these two challenges by employing the eigenvectors with top-k largest eigenvalues of the raw adjacency matrix for pre-embeddings. These pre-embeddings contain high-order topological knowledge together with the information on negative edges, which are then fed into a GNN with a positively normalized adjacency matrix to compensate for its shortcomings. Through comprehensive experiments and analysis, we empirically demonstrate the superiority of our proposed solution in a Bitcoin user reputation score prediction task.
{"title":"Eigenvector-based Graph Neural Network Embeddings and Trust Rating Prediction in Bitcoin Networks","authors":"Pin Ni, Qiao Yuan, Raad Khraishi, Ramin Okhrati, Aldo Lipani, F. Medda","doi":"10.1145/3533271.3561793","DOIUrl":"https://doi.org/10.1145/3533271.3561793","url":null,"abstract":"Given their strong performance on a variety of graph learning tasks, Graph Neural Networks (GNNs) are increasingly used to model financial networks. Traditional GNNs, however, are not able to capture higher-order topological information, and their performance is known to degrade with the presence of negative edges that may arise in many common financial applications. Considering the rich semantic inference of negative edges, excluding them as an obvious solution is not elegant. Alternatively, another basic approach is to apply positive normalization, however, this also may lead to information loss. Our work proposes a simple yet effective solution to overcome these two challenges by employing the eigenvectors with top-k largest eigenvalues of the raw adjacency matrix for pre-embeddings. These pre-embeddings contain high-order topological knowledge together with the information on negative edges, which are then fed into a GNN with a positively normalized adjacency matrix to compensate for its shortcomings. Through comprehensive experiments and analysis, we empirically demonstrate the superiority of our proposed solution in a Bitcoin user reputation score prediction task.","PeriodicalId":134888,"journal":{"name":"Proceedings of the Third ACM International Conference on AI in Finance","volume":"623 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123076642","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
D. Zhou, Ajim Uddin, Xinyuan Tao, Zuofeng Shang, Dantong Yu
Understanding bond (debt) valuation and predicting future prices are of great importance in finance. Bonds are a major source of long-term capital in U.S. financial markets along with stocks. However, compared with stocks, bonds are understudied. One main reason is the infrequent trading in the secondary market, which results in irregular intervals and missing observations. This paper attempts to overcome this challenge by leveraging network information from bond-fund holding data and proposes a novel method to predict bond prices (yields). We design the temporal bipartite graph neural networks (TBGNN) with self-supervision regularization that entails multiple components: the bipartite graph representation module of learning node embeddings from the bond and fund interactions and their associated factors; the recurrent neural network module to model the temporal interactions; and the self-supervised objective to regularize the unlabeled node representation with graph structure. The model adopts a minibatch training process (Minibatch Stochastic Gradient Descent) in a deep learning platform to alleviate the model complexity and computation cost in optimizing different modules and objectives. Results show that our TBGNN model provides a more accurate prediction of bond price and yield. It outperforms multiple existing graph neural networks and multivariate time series methods: improving R2 by 6%-51% in bond price prediction and 5%-70% in bond yield prediction.
{"title":"Temporal Bipartite Graph Neural Networks for Bond Prediction","authors":"D. Zhou, Ajim Uddin, Xinyuan Tao, Zuofeng Shang, Dantong Yu","doi":"10.1145/3533271.3561751","DOIUrl":"https://doi.org/10.1145/3533271.3561751","url":null,"abstract":"Understanding bond (debt) valuation and predicting future prices are of great importance in finance. Bonds are a major source of long-term capital in U.S. financial markets along with stocks. However, compared with stocks, bonds are understudied. One main reason is the infrequent trading in the secondary market, which results in irregular intervals and missing observations. This paper attempts to overcome this challenge by leveraging network information from bond-fund holding data and proposes a novel method to predict bond prices (yields). We design the temporal bipartite graph neural networks (TBGNN) with self-supervision regularization that entails multiple components: the bipartite graph representation module of learning node embeddings from the bond and fund interactions and their associated factors; the recurrent neural network module to model the temporal interactions; and the self-supervised objective to regularize the unlabeled node representation with graph structure. The model adopts a minibatch training process (Minibatch Stochastic Gradient Descent) in a deep learning platform to alleviate the model complexity and computation cost in optimizing different modules and objectives. Results show that our TBGNN model provides a more accurate prediction of bond price and yield. It outperforms multiple existing graph neural networks and multivariate time series methods: improving R2 by 6%-51% in bond price prediction and 5%-70% in bond yield prediction.","PeriodicalId":134888,"journal":{"name":"Proceedings of the Third ACM International Conference on AI in Finance","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133583050","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Blockchain protocols’ main differentiator is their purported decentralization that unlocks various information technology applications that were supposedly impossible beforehand. The key promise is that incentive-driven participation of a large set of interested parties can lead to decentralized protocol states where no single operator can be a “single point of failure.” Despite this promise, there is little systematic analysis of decentralization in blockchain systems and the sporadic theoretic and empirical investigations that exist paint a rather negative picture due to resource “pooling behaviors” that are impossible to prevent in the “permissionless” setting of such protocols where parties have no designated identities. Motivated by this, in this paper we study the Nash dynamics of pooling in the context of Proof of Stake systems, following an agent-based modeling approach. Our focus is the Cardano blockchain as it features a number of attractive characteristics making it conducive to an in-depth analysis. We aim to answer the question of whether the incentive mechanism employed is capable of promoting decentralization. To this end, we present a simulation engine that enables strategic agents to engage in a number of actions empirically observed in the real-world deployment of the system. The engine simulates the “stake pool operation and delegation game" via successive agent actions that improve their utility as more information about their environment becomes evident in the course of the simulation. We investigate convergence to equilibrium states, and we measure various decentralization metrics in these states, such as the Nakamoto coefficient, which asks how many independent entities exist that collectively command more than of the system’s resources. Our results exemplify the ability of the incentive mechanism to steer the system towards good equilibria and also illustrate how the decentralization features of such equilibria are affected by different choices of the parameters used in the mechanism and the distribution of stake to participants.
{"title":"Decentralization Analysis of Pooling Behavior in Cardano Proof of Stake","authors":"Christina Ovezik, A. Kiayias","doi":"10.1145/3533271.3561787","DOIUrl":"https://doi.org/10.1145/3533271.3561787","url":null,"abstract":"Blockchain protocols’ main differentiator is their purported decentralization that unlocks various information technology applications that were supposedly impossible beforehand. The key promise is that incentive-driven participation of a large set of interested parties can lead to decentralized protocol states where no single operator can be a “single point of failure.” Despite this promise, there is little systematic analysis of decentralization in blockchain systems and the sporadic theoretic and empirical investigations that exist paint a rather negative picture due to resource “pooling behaviors” that are impossible to prevent in the “permissionless” setting of such protocols where parties have no designated identities. Motivated by this, in this paper we study the Nash dynamics of pooling in the context of Proof of Stake systems, following an agent-based modeling approach. Our focus is the Cardano blockchain as it features a number of attractive characteristics making it conducive to an in-depth analysis. We aim to answer the question of whether the incentive mechanism employed is capable of promoting decentralization. To this end, we present a simulation engine that enables strategic agents to engage in a number of actions empirically observed in the real-world deployment of the system. The engine simulates the “stake pool operation and delegation game\" via successive agent actions that improve their utility as more information about their environment becomes evident in the course of the simulation. We investigate convergence to equilibrium states, and we measure various decentralization metrics in these states, such as the Nakamoto coefficient, which asks how many independent entities exist that collectively command more than of the system’s resources. Our results exemplify the ability of the incentive mechanism to steer the system towards good equilibria and also illustrate how the decentralization features of such equilibria are affected by different choices of the parameters used in the mechanism and the distribution of stake to participants.","PeriodicalId":134888,"journal":{"name":"Proceedings of the Third ACM International Conference on AI in Finance","volume":"67 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130519560","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Motivated by practical considerations in machine learning for financial decision-making, such as risk-aversion and large action space, we initiate the study of risk-aware linear bandits. Specifically, we consider regret minimization under the mean-variance measure when facing a set of actions whose reward can be expressed as linear functions of (initially) unknown parameters. We first propose the Risk-Aware Explore-then-Commit (RISE) algorithm driven by the variance-minimizing G-optimal design. Then, we rigorously analyze its regret upper bound to show that, by leveraging the linear structure, the algorithm can dramatically reduce the regret when compared to existing methods. Finally, we demonstrate the performance of the RISE algorithm by conducting extensive numerical experiments in a synthetic smart order routing setup. Our results show that the RISE algorithm can outperform the competing methods, especially when the decision-making scenario becomes more complex.
{"title":"Risk-Aware Linear Bandits with Application in Smart Order Routing","authors":"Jingwei Ji, Renyuan Xu, Ruihao Zhu","doi":"10.1145/3533271.3561692","DOIUrl":"https://doi.org/10.1145/3533271.3561692","url":null,"abstract":"Motivated by practical considerations in machine learning for financial decision-making, such as risk-aversion and large action space, we initiate the study of risk-aware linear bandits. Specifically, we consider regret minimization under the mean-variance measure when facing a set of actions whose reward can be expressed as linear functions of (initially) unknown parameters. We first propose the Risk-Aware Explore-then-Commit (RISE) algorithm driven by the variance-minimizing G-optimal design. Then, we rigorously analyze its regret upper bound to show that, by leveraging the linear structure, the algorithm can dramatically reduce the regret when compared to existing methods. Finally, we demonstrate the performance of the RISE algorithm by conducting extensive numerical experiments in a synthetic smart order routing setup. Our results show that the RISE algorithm can outperform the competing methods, especially when the decision-making scenario becomes more complex.","PeriodicalId":134888,"journal":{"name":"Proceedings of the Third ACM International Conference on AI in Finance","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115400088","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：