基于深度学习的VaR测算研究

发布时间：2018-05-18 09:46

本文选题：VaR + 深度学习　；参考：《兰州财经大学》2017年硕士论文

【摘要】：在美国次贷危机的影响下,全球经济遭遇重创。尽管次贷危机现在已经渐渐远离,但它所产生的伤害仍然继续,使得人们不得不反思其背后的原因,金融风险管理就是在这个时候逐渐被重新强调起来的。金融市场风险是由金融资产未来波动的不确定性引起的。由于金融资产的波动会带动其价值的波动,这些波动一方面造就了金融市场的活跃性和流动性,使得各种经济资产表现为价值运动,但另一方面也会导致经济过渡虚拟,风险和不确定性将被无限放大,给投资人、企业、社会、国家带来巨大损失,甚至引发金融危机。金融风险管理就是要通过各种技术手段找出各个投资组合的最大可能损失,并在此基础上进行分析与决策,从而维护金融市场健康稳定的发展。金融风险度量(Financial risk metrics)是金融风险管理中的核心与根本,是金融风险管理的最优先问题,它对金融风险管理起着杠杆的作用。传统的金融风险度量方法是以国外学者发明的波动率方法为代表,通过测算金融资产收益率的方差或标准差来度量风险大小。由于它只描述了金融资产收益的偏离程度,不能对偏离的方向和损失水平进行说明,导致其应用有限,因而不再很好的适应快速变化的金融发展。VaR作为一种新的金融风险度量工具问世,打破了以波动率方法为代表的传统度量方法的统治地位。它通过对金融风险进行定量的计算,从而有效的进行风险分析,更直观的揭露风险,因而在对金融市场风险度量上得到了广泛的应用,同时对金融风险的量化管理也起到了显著的效果。这使得VaR迅速成为标杆,并普遍的被应用于金融市场风险的度量中。尽管VaR的研究历史悠久。但现有的关于VaR计算方法改进的研究并不多,大部分都集中于研究VaR在各个领域中的应用。尤其是我国对VaR的研究起步相对较晚,其中较多的研究是基于国外已经成熟的研究成果,从它们的概念、原理、方法以及运用VaR方法进行实证研究等方面来说明,鲜有学者对VaR计算方法提出架构,因而忽视了基于VaR的风险测量中存在的一些缺陷。VaR方法是通过对金融资产过去的收益特征进行统计分析来估算未来可能发生的最大损失。因此,在计算VaR的过程中其精度的高低依赖于对所研究的金融资产收益率的分布的假设和对其方差的估计。这意味着,基于VaR的风险测量方法存在着对样本数据特征的认识不足问题,将导致风险测量的不准确,甚至产生较大的偏差。同时科技的进步,金融市场的不断变革,使得人工智能在金融分析管理中越来越重要,并引起了学者的高度关注。近几年,利用深度学习处理大数据更是掀起了一股技术方法创新浪潮。给量化金融市场风险上增添了强有力的工具,突破了金融风险度量的盲区。股票市场是预测未来实体经济发展和调动资金流向的重要场所,也是金融市场重要的极重要的一部分。因为股票市场不仅仅是募集资金的场所,它也是是公众投资理财的重要渠道。股票市场中的股票作为融资理财的凭证,与人们的经济活动息息相关。而且,股票投资的本身也是进行风险投资。因而,股票市场的波动性可以反映出金融风险的波动性,且可以用它来研究金融风险的度量方法。综上,研究股票市场的波动性有着代表性意义。因此,本文以我国的股票市场为例,在已有文献的基础上,针对目前VaR方法存在的缺陷,提出了基于深度学习的VaR测算。首先对传统意义上的损失进行改进,使用预期损失,从而更加符合现实中人们对损失的多样化定义。其次,分别对股票收益率数据建立ARCH族模型以及对预期损失建立深度人工神经网络模型,进而对VaR进行更加精确的预测。经实证发现,在深度学习下的VaR计算比ARCH族模型下的VaR计算更加精确。说明基于深度学习的VaR计算具有更好的实用性。
[Abstract]:Under the impact of the American subprime crisis, the global economy has been hit hard. Although the subprime crisis is now getting far away, the damage it produces continues to cause people to reflect on the reasons behind it. Financial risk management is gradually reemphasized at this time. The risk of financial markets is the future of financial assets. The volatility of volatility causes the volatility of the value of financial assets. These fluctuations, on the one hand, create the activity and liquidity of the financial market, and make all kinds of economic assets as a value movement, but on the other hand, the economic transition will also lead to the virtual economic transition, and the risks and uncertainties will be magnified indefinitely. The enterprises, the society and the state bring huge losses and even lead to the financial crisis. Financial risk management means to find out the greatest possible losses of each portfolio by various technical means, and to analyze and make decisions on this basis so as to maintain the healthy and stable development of the financial market. The financial risk measurement (Financial risk metrics) is the finance. The core and fundamental of risk management is the most important problem in the management of financial risk. It plays a lever role in the management of financial risk. The traditional method of financial risk measurement is represented by the method of volatility invented by foreign scholars, measuring the risk by measuring the variance or standard deviation of the yield of financial assets. Because it is only described. The deviation degree of the financial assets income can not be explained in the direction of deviation and the level of loss, which leads to the limited application of the financial development, which is no longer good to adapt to the rapid change of financial development.VaR as a new financial risk measurement tool, breaking the dominant position of the traditional measurement method represented by the method of volatility. The quantitative calculation of financial risk, so as to effectively carry on the risk analysis, more intuitively expose the risk, has been widely used in the financial market risk measurement, and also has played a significant effect on the quantitative management of financial risk. This makes VaR quickly become a benchmark and is widely used in the financial market. In the measurement of risk, although VaR has a long history of research, there are few existing researches on the improvement of VaR computing methods. Most of them focus on the study of the application of VaR in various fields. Especially, the research on VaR is relatively late in our country. Many of them are based on the mature research results abroad, from their concepts. Theory, method and empirical research on the use of the VaR method show that few scholars have put forward the framework of VaR calculation method, so that some defects in the risk measurement based on VaR are ignored.VaR method is to estimate the possible maximum loss in the future by statistical analysis of the past income characteristics of financial assets. In the process of calculating VaR, its accuracy depends on the hypothesis of the distribution of the yield of the financial assets studied and the estimation of its variance. This means that the risk measurement method based on VaR has a lack of understanding of the characteristics of the sample data, which will lead to the inaccuracy of the risk measurement and even a larger deviation. The progress and the continuous change in the financial market make the artificial intelligence more and more important in the financial analysis and management, and have aroused the high attention of the scholars. In recent years, the use of deep learning to deal with large data has set off a wave of technological innovation. It has added a powerful tool to quantify the risk of the financial market and broke through the financial risk. The stock market is an important place to predict the development of the future real economy and to mobilize the flow of funds. It is also an important part of the financial market. Because the stock market is not only a place to raise funds, it is also an important channel for public investment in financial management. Stock market shares are used as a voucher for financing and financing. It is closely related to people's economic activities. Moreover, the stock investment itself is also a risk investment. Therefore, the volatility of the stock market can reflect the volatility of the financial risk, and can be used to study the measurement of financial risk. As an example of the stock market, on the basis of the existing literature, in view of the defects existing in the current VaR method, a VaR calculation based on depth learning is proposed. First, the loss in the traditional sense is improved and the expected loss is used, which is more consistent with the diversified definition of the loss in reality. Secondly, the stock return data is set up to be ARCH respectively. The model of the family and a deep artificial neural network model for the expected loss are set up to make a more accurate prediction of the VaR. It is found that the VaR calculation under the depth learning is more accurate than the VaR calculation under the ARCH model. It shows that the VaR calculation based on the depth learning is more practical.
【学位授予单位】：兰州财经大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F224;F831

【参考文献】