arXiv - QuantFin - Mathematical Finance最新文献_第6页

Filtered not Mixed: Stochastic Filtering-Based Online Gating for Mixture of Large Language Models 过滤而非混合：基于随机过滤的大型语言模型混合在线门控技术

arXiv - QuantFin - Mathematical Finance Pub Date : 2024-06-05 DOI: arxiv-2406.02969

Raeid Saqur, Anastasis Kratsios, Florian Krach, Yannick Limmer, Jacob-Junqi Tian, John Willes, Blanka Horvath, Frank Rudzicz

{"title":"Filtered not Mixed: Stochastic Filtering-Based Online Gating for Mixture of Large Language Models","authors":"Raeid Saqur, Anastasis Kratsios, Florian Krach, Yannick Limmer, Jacob-Junqi Tian, John Willes, Blanka Horvath, Frank Rudzicz","doi":"arxiv-2406.02969","DOIUrl":"https://doi.org/arxiv-2406.02969","url":null,"abstract":"We propose MoE-F -- a formalised mechanism for combining $N$ pre-trained\u0000expert Large Language Models (LLMs) in online time-series prediction tasks by\u0000adaptively forecasting the best weighting of LLM predictions at every time\u0000step. Our mechanism leverages the conditional information in each expert's\u0000running performance to forecast the best combination of LLMs for predicting the\u0000time series in its next step. Diverging from static (learned) Mixture of\u0000Experts (MoE) methods, MoE-F employs time-adaptive stochastic filtering\u0000techniques to combine experts. By framing the expert selection problem as a\u0000finite state-space, continuous-time Hidden Markov model (HMM), we can leverage\u0000the Wohman-Shiryaev filter. Our approach first constructs $N$ parallel filters\u0000corresponding to each of the $N$ individual LLMs. Each filter proposes its best\u0000combination of LLMs, given the information that they have access to.\u0000Subsequently, the $N$ filter outputs are aggregated to optimize a lower bound\u0000for the loss of the aggregated LLMs, which can be optimized in closed-form,\u0000thus generating our ensemble predictor. Our contributions here are: (I) the\u0000MoE-F algorithm -- deployable as a plug-and-play filtering harness, (II)\u0000theoretical optimality guarantees of the proposed filtering-based gating\u0000algorithm, and (III) empirical evaluation and ablative results using state of\u0000the art foundational and MoE LLMs on a real-world Financial Market Movement\u0000task where MoE-F attains a remarkable 17% absolute and 48.5% relative F1\u0000measure improvement over the next best performing individual LLM expert.","PeriodicalId":501084,"journal":{"name":"arXiv - QuantFin - Mathematical Finance","volume":"59 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141525941","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}

引用次数: 0

Modelling Non-monotone Risk Aversion and Convex Compensation in Incomplete Markets 不完全市场中的非单调风险规避和凸补偿建模

arXiv - QuantFin - Mathematical Finance Pub Date : 2024-06-01 DOI: arxiv-2406.00435

Yang Liu, Zhenyu Shen

{"title":"Modelling Non-monotone Risk Aversion and Convex Compensation in Incomplete Markets","authors":"Yang Liu, Zhenyu Shen","doi":"arxiv-2406.00435","DOIUrl":"https://doi.org/arxiv-2406.00435","url":null,"abstract":"In hedge funds, convex compensation schemes are popular to stimulate a\u0000high-profit performance for portfolio managers. In economics, non-monotone risk\u0000aversion is proposed to argue that individuals may not be risk-averse when the\u0000wealth level is low. Combining these two ingredients, we study the optimal\u0000control strategy of the manager in incomplete markets. Generally, we propose a\u0000wide class of utility functions, the Piecewise Symmetric Asymptotic Hyperbolic\u0000Absolute Risk Aversion (PSAHARA) utility, to model the two ingredients,\u0000containing both non-concavity and non-differentiability as some abnormalities.\u0000Significantly, we derive an explicit optimal control for the family of PSAHARA\u0000utilities. This control is expressed into a unified four-term structure,\u0000featuring the asymptotic Merton term and the risk adjustment term. Furthermore,\u0000we provide a detailed asymptotic analysis and numerical illustration of the\u0000optimal portfolio. We obtain the following key insights: (i) A manager with the\u0000PSAHARA utility becomes extremely risk-seeking when his/her wealth level tends\u0000to zero; (ii) The optimal investment ratio tends to the Merton constant as the\u0000wealth level approaches infinity and the negative Merton constant when the\u0000wealth falls to negative infinity, implying that such a manager takes a\u0000risk-seeking investment as the wealth falls negatively low; (iii) The convex\u0000compensation still induces a great risk-taking behavior in the case that the\u0000preference is modeled by SAHARA utility. Finally, we conduct a real-data\u0000analysis of the U.S. stock market under the above model and conclude that the\u0000PSAHARA portfolio is very risk-seeking and leads to a high return and a high\u0000volatility (two-peak Sharpe ratio).","PeriodicalId":501084,"journal":{"name":"arXiv - QuantFin - Mathematical Finance","volume":"142 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141254398","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}

引用次数: 0

Reinforcement Learning for Jump-Diffusions 跳跃扩散的强化学习

arXiv - QuantFin - Mathematical Finance Pub Date : 2024-05-26 DOI: arxiv-2405.16449

Xuefeng Gao, Lingfei Li, Xun Yu Zhou

引用次数: 0

A probabilistic approach to continuous differentiability of optimal stopping boundaries 最优停止边界连续可微分性的概率方法

arXiv - QuantFin - Mathematical Finance Pub Date : 2024-05-26 DOI: arxiv-2405.16636

Tiziano De Angelis, Damien Lamberton

引用次数: 0

Optimal market-neutral currency trading on the cryptocurrency platform 加密货币平台上的最佳市场中性货币交易

arXiv - QuantFin - Mathematical Finance Pub Date : 2024-05-24 DOI: arxiv-2405.15461

Hongshen Yang, Avinash Malik, Andrea Raith

{"title":"Optimal market-neutral currency trading on the cryptocurrency platform","authors":"Hongshen Yang, Avinash Malik, Andrea Raith","doi":"arxiv-2405.15461","DOIUrl":"https://doi.org/arxiv-2405.15461","url":null,"abstract":"This research proposes a novel arbitrage approach with respect to\u0000multivariate pair trading called Optimal Trading Technique (OTT). We introduce\u0000the method to selectively form a \"bucket\" of fiat currencies anchored to\u0000cryptocurrency for simultaneously monitoring and exploiting trading\u0000opportunities. To handle the quantitative conflicts that arise when receiving\u0000multiple trading signals, a bi-objective convex optimization process is\u0000designed to cater to the investor's preference between profitability and risk\u0000tolerance. This process includes tunable parameters such as volatility\u0000punishment, action thresholds. During our experiments in the cryptocurrency\u0000market from 2020 to 2022 when the market was experiencing a vigorous bull-run\u0000immediately followed by a bear-run, the OTT realized an annualized profit of\u000015.49%. We further carried out the experiments in bull, bear, and full-cycle\u0000market conditions separately, and found that OTT is capable of achieving stable\u0000profit under various market conditions. Apart from the profitability side of\u0000the OTT, the arbitrage operation provides a new perspective of trading, which\u0000requires no external shorting and never hold intermediate cryptocurrency during\u0000the arbitrage period.","PeriodicalId":501084,"journal":{"name":"arXiv - QuantFin - Mathematical Finance","volume":"7 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-05-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141171613","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}

引用次数: 0

Long Time Behavior of Optimal Liquidation Problems 最优清算问题的长期行为

arXiv - QuantFin - Mathematical Finance Pub Date : 2024-05-23 DOI: arxiv-2405.14177

Xinman Cheng, Guanxing Fu, Xiaonyu Xia

引用次数: 0

Convergence analysis of kernel learning FBSDE filter 核学习 FBSDE 滤波器的收敛性分析

arXiv - QuantFin - Mathematical Finance Pub Date : 2024-05-22 DOI: arxiv-2405.13390

Yunzheng Lyu, Feng Bao

引用次数: 0

Deep Penalty Methods: A Class of Deep Learning Algorithms for Solving High Dimensional Optimal Stopping Problems 深度惩罚方法：一类解决高维最优停止问题的深度学习算法

arXiv - QuantFin - Mathematical Finance Pub Date : 2024-05-18 DOI: arxiv-2405.11392

Yunfei Peng, Pengyu Wei, Wei Wei

引用次数: 0

Optimal information acquisition for eliminating estimation risk 消除估算风险的最佳信息获取方式

arXiv - QuantFin - Mathematical Finance Pub Date : 2024-05-15 DOI: arxiv-2405.09339

Zongxia Liang, Qi Ye

引用次数: 0

Comparative analysis of neural network architectures for short-term FOREX forecasting 短期外汇预测神经网络架构的比较分析

arXiv - QuantFin - Mathematical Finance Pub Date : 2024-05-13 DOI: arxiv-2405.08045

Theodoros Zafeiriou, Dimitris Kalles

{"title":"Comparative analysis of neural network architectures for short-term FOREX forecasting","authors":"Theodoros Zafeiriou, Dimitris Kalles","doi":"arxiv-2405.08045","DOIUrl":"https://doi.org/arxiv-2405.08045","url":null,"abstract":"The present document delineates the analysis, design, implementation, and\u0000benchmarking of various neural network architectures within a short-term\u0000frequency prediction system for the foreign exchange market (FOREX). Our aim is\u0000to simulate the judgment of the human expert (technical analyst) using a system\u0000that responds promptly to changes in market conditions, thus enabling the\u0000optimization of short-term trading strategies. We designed and implemented a\u0000series of LSTM neural network architectures which are taken as input the\u0000exchange rate values and generate the short-term market trend forecasting\u0000signal and an ANN custom architecture based on technical analysis indicator\u0000simulators We performed a comparative analysis of the results and came to\u0000useful conclusions regarding the suitability of each architecture and the cost\u0000in terms of time and computational power to implement them. The ANN custom\u0000architecture produces better prediction quality with higher sensitivity using\u0000fewer resources and spending less time than LSTM architectures. The ANN custom\u0000architecture appears to be ideal for use in low-power computing systems and for\u0000use cases that need fast decisions with the least possible computational cost.","PeriodicalId":501084,"journal":{"name":"arXiv - QuantFin - Mathematical Finance","volume":"35 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141060894","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}

引用次数: 0