1、 1 外文翻译 原文 LP Modeling for Asset-Liability Management:A Survey of Choices and Simplifications Material Source: Operations research Autheor： ManMohan S.Sodhi 1.Introduction Asset-liability management (ALM) refers to the buying and selling of securities (assets) to meet current and future payments (li
2、abilities) under uncertainty associated usually but not exclusively with interest-rate movements. It is practiced for pension-fund management and in the banking and the insurance industries primarily. Although linear programming (LP) is a exible and powerful way to model ALM using stochastic program
3、ming with recourse, there are two challenges with these dynamic models. The rst challenge is having to choose from a plethora of modeling choices with not all choices being consistent with each other or with nance theory. The second challenge is that the models are difcult to solve due to the large
4、number of scenarios that grows exponentially with the number of time periods when using standard interest-rate models or stochastic processes. As such, we rst survey modeling choices with a view to help researchers make selfconsistent choices. Next, we show that existing static models are extremely
5、simplied forms of dynamic models regarding both uncertainty and recourse variables, so there is plenty of scope to simplify dynamic models to gain solution tractability. Indeed, dynamic models that have been reported are simplications of some sort due to solution difculties, so our goal here is to p
6、rovide researchers with simplication choices. These modeling choices and solution difculties may be part of the reason for modeling missteps in the past. Many operations research (OR)-based ALM models use prices of securities different from those in the market or allow spurious arbitrage opportuniti
7、es. Rules of thumb, e.g., adding an options-adjusted spread, are sometimes incorporated to otherwise sophisticated models. At times, the focus has wavered with some researchers proposing simplications to take advantage of a 2 particular parallel computer architecture or a particular algorithm. Yet o
8、thers have proposed simplistic models to appeal to users. Researchers have not always built upon earlier successful work, e.g., the dynamic model by Bradley and Crane (1972). Happily, research is evolving in a way that builds on Bradley and Cranes work and also incorporates nance theory although the
9、 above two challenges remain. Despite these challenges, it is worthwhile to consider LP for at least four reasons. First, LP models can make specic recommendations at the individual asset level to actively leverage assets to enhance net worth (objective) rather than simply evaluating a static recomm
10、endation such as simulation approaches or providing only broad recommendations in terms of asset classes such as stochastic control methods. Second, meeting regulatory requirements and liquidity requirements (constraints) and using the output of auxiliary interest-rate and cash-ow models as input is
11、 straightforward with LP. Third, LP can incorporate imperfections including taxes, transaction costs, and regulatory or institution-specic requirements over time more exibly than other approaches. Fourth, recent advances in LP solver technology and in stochastic programming also help make the case f
12、or LP. These modeling advantages have not translated into widespread use of dynamic LP models in practice despite well-reported successes in the OR literature. According to a Financial Times columnist, practitioners view dynamic models as excessively complex with difculties of ex- plaining the downs
13、ide risk of even an optimal solution so their consensus is that much more work remains to be done before widespread adoption (Riley 2002). Banks have used advanced nancial models for pricing derivatives and there are also reports of static LP-based models having been used in the banking sector (Demb
14、o 1990, Ruffel 1986). Smaller banks needs are more modest as studied by Moynihan et al. (2002), who present a decision-support system for credit unions. As industry emphasis is on managing downside risk, it is simulation models that hold sway in industry emphasizing scenario generation, reporting, a
15、nd workow management. Some OR researchers have also moved in this direction (e.g., Mulvey et al. 2000). The insurance industry appears to lag in ALM, and hence LP modeling, as evident from a relatively recent Swiss Reinsurance Company report exhorting the life and property/casualty industry to adopt
16、 ALM practices (Swiss Reinsurance Company 2000). Dynamic models have solution tractability challenges as well requiring the use of parallel computers with as few as 4,000 scenarios (Mulvey and 3 Shetty 2004). Although Dert (1999) shows that a multistagdynamic model provides better solutions than a m
17、yopic or a simple static model, others have raised questions on the benet of dynamic models over more sophisticated static models (Zenios 1995). Therefore, for a number of reasons, it is worthwhile to explore how to simplify dynamic LP models (as done, e.g.,by Gaivoronski and de Lange 2000 using x-m
18、ix or by Mulvey et al. 1999 using hybrid securities) before we can integrate them into packaged ALM software as hybrids with simulation (as done, e.g., by Boender 1997 and by Seshadri et al. 1999). The current state-of-the-art (e.g.,Mulvey and Shetty 2004) using parallel and distributed computers ma
19、y not t the computing environment in most companies. Dembo and Rosen (1999) mention a growing interest in static methods as well as semi-dynamicstrategies in traditional areas of nance; they claim that static methods may be adequate when enterprisewide risk (subsuming ALM) is concerned with as many
20、as 400 risk factors. This paper seeks to help researchers understand simplication choices. Targeting the packaged ALM software market is not a bad idea: the 400 largest nancial institutions reportedly spent $613 million on ALM software globally in 1998 (OConnell 1999). Following a slowdown in spendi
21、ng in the 19992003 period, analysts expect similar spending of about $600 million globally in 2004expected to grow to $850 million in 2009 (Keppler 2004). As limitations of this work, we do not cover non-LP modeling approaches and refer the reader to Ziemba and Mulvey (1999) and to Zenios and Ziemba
22、 (2004). The models in this paper are stripped down to essential constraints for matching cash ows rather than industry-specic or regulatory ones, although we provide some references in the next section that are useful to understand industry-specic liabilities. We also do not provide a denitive solu
23、tion in terms of a model, but present this survey in the hope that researchers will use it to that end. Finally, we have limited the discussion to interest-rate risk as has most of the OR literature, even though credit risk (of bad loans) is quite important for banks as is exchange-rate risk. The sh
24、ortfalls in pension funds of many large companies following the stock market downfall at the turn of the millennium points to the importance of market risk as well, but we do not address this either except to refer to researchers who create joint scenarios over multiple risk factors. Still, much of
25、the discussion here is relevant regardless of whether the risk is limited to interest rates or stems from other factors as well. 4 2. Background The Financial Accounting Standards Board in the United States denes assets as probable future economic benets and liabilities as probable future sacrices o
26、f economic benets. ALM entails leveraging these assets and liabilities to enhance the net worth of the institution while quantifying the various associated risks, managing liquidity, and streamlining the management of regulatory requirements (Ong 1998). ALM has evolved from an accounting focus in th
27、e 1960s and 1970s to a centralized, integrated riskmanagement function applied to the entire balance sheet in the mid 1990s (Jarrow and van Deventer 1998) and different companies are at different rungs of this evolution (Mulvey et al. 1997, Mulvey and Ziemba 1999). In the banking industry, liabiliti
28、es are customer accounts,while assets are mortgages and personal loans. Cash ows depend on interest-rate movement, e.g., people prepay loans when interest rates drop. The failure of many banks in the early 1980s provided a compelling need to manage interest-rate risk better. As such, investors rewar
29、d more stable high-quality earnings by higher stock prices for nancial institutions (Giarla 1991). Ong (1998) and Bitner and Goddard (1992) discuss ALM practice in the banking industry. Cohen and Gibson (1978) focus on banking with Part V devoted to bond portfolio management and interest-rate risk.
30、Adamidou et al. (1993) and Ben-Dov et al. (1992) describe an optimal portfolio system at Prudential-Bache. Seshadri et al. (1999) use simulation and quadratic optimization to formulate, test, and rene the Federal Home Loan Bank of New Yorks ALM policies. In the insurance industry, assets are xed-inc
31、ome securities such as treasuries, corporate bonds, and mortgagebacked instruments. Liabilities are customer-related insurance payouts that need not be tied to interest rates. Babbel and Staking (1991) report that exposure to interest-rate risk is the strongest predictor among tested variables to ex
32、plain insolvency for property/liability insurers. Cario et al. (1994), Cario and Ziemba (1998), and Cario et al.(1998) describe the need for ALM in insurance companies and the decision-support systems they built for the Japanese insurance company Yasuda Kasai. Worzel et al. (1994) describe managing
33、a xed-income portfolio at Metropolitan Life Insurance using an index as the liability. Holmer (1994, 1999) describes an application at Federal National Mortgage Association in reference to mortgage insurance and mortgage investment. Sweeney et al. (1999) outline Falcon Asset Managements ALM approach
34、 for global insurance companies where the assets include cash and equities in multiple 5 currencies. An industry research report describes the need for ALM in the life insurance and property/casualty industry (Swiss Reinsurance Company 2000),while Gaivoronski and de Lange (2000) provide a model for
35、casualty insurers. Hoyland and Wallace (2001b) analyze a legal regulation in the Norwegian life insurance business using a dynamic ALM model. Another growing area is pension funds. Liabilities, e.g.,dened benets, are payouts to workers who are already retired or who will retire in the future. Assets
36、 usually include both xed-income securities and equities. Mulvey (1996) describes an approach to generate scenariosinterest rates, pension withdrawals, and economic scenariosfor pension-fund management at Towers Perrin; Mulvey et al. (2000) describe the overall system as additionally having a nonlin
37、ear optimization simulation model.Boender (1997), Dert (1999), and Drijver et al. (2000) consider Dutch pension funds, while Gondzio and Kouwenberg (2001) provide a decomposition-based algorithm for these implemented on a particular parallel computer. Kouwenberg (2001) discusses liability scenarios
38、in this context as well. To give a sense of relative enthusiasm for ALM software in these three industries, the global breakdown of spending was 79% in the banking sector, 4% in the insurance sector,and 17% in the pension funds sector (Keppler 2004). Some researchers have looked specically at softwa
39、re issues in ALM; for instance, Dempster and Ireland (1991) and Sodhi (1996) describe object-oriented implementations. 译文 LP 建模资产负债管理被简化的选择和调查 资料来源 : Operations research 作者： ManMohan S.Sodhi 资产负债管理（ ALM）是指购买和出售证券（资产），以履行与利率变动通常是不完全的不确定性，但当前和未来付款（负债）。这是实行了养老基金的管理和银行业和保险业为主。虽然线性规划（ LP）的是 一个灵活和强大的资产负债管
40、理方法，模型，采用有追索权的随机规划，有两个与这些“动态”模式的挑战。第一个挑战是有选择的建模选择与被选择，不是所有相互或与金融理论相一致的太多了。第二个挑战是，该模型是难以解决的情况下，由于大量的增长，随着时间的周期数成倍当使用标准的利率模型或随机过程。因此，我们首6 先调查了视图建模方法作出选择，以帮助研究人员自洽的选择。下一步，我们表明，现有的“静态”模型是非常简化的动态模型两种形式方面的不确定性和追索权的变量，所以有很多的范围，以简化动态模型来 解决可追踪性 。事实上，已报告的 动态模型，由于某种解决困难的简化，所以我们的目标是为客户提供简化的选择研究。这些模型的选择和解决方案的困难可
41、能是在过去建模失误的原因之一。许多运筹学（ OR）为基础的资产负债管理模型使用的证券在市场上的不同价格，或者允许虚假套利的机会。这一法则，例如，将“选项，调整”的传播，有时并入其他复杂的模型。有时，重点是动摇了一些研究者提出要简化并行计算机的一个特定的架构或特定算法的优势。还有一些人提出了简单的模型来吸引用户。研究人员先前并不总是建立在成功的工作，例如，由布拉德利和 Crane（ 1972）的动态模型。令人 高兴的是，研究是不断发展的方式，建立在布拉德利和起重机的工作，并结合金融理论虽然上述两个挑战依然存在。 尽管存在这些挑战，这是值得考虑至少有四个理由唱片。首先， LP 的模型可以使在个人资
42、产水平的具体建议，积极利用，以提高资产净值（目标），而不是简单的评估，如模拟方法的建议或静态随机控制方法，如只提供在资产类别方面广泛的建议。第二，满足监管要求和流动性要求（约束），并使用辅助利率作为输入现金流量模型输出与唱片简单。第三，唱片可以将包括税收，交易成本，并随着时间的推移监管或机构的具体要求比其他方法更灵活“ 沙沙”。第四，在 LP 求解器技术和随机规划的最新进展也有助于使为 LP 案件。 这些建模的优势并没有转化为实践中广泛使用的动态模型，尽管在低压或文学以及报告的成功。根据金融时报的专栏作家，从业人员认为“过于复杂”的困难与动态模型前， plaining 所以他们的共识“，甚至是
43、最佳的解决方案，下行风险”是“大量的工作有待完成” 前广泛采用 （莱利 2002 年）。银行已经使用了先进的金融衍生产品定价模型，也有静态的唱片的曾在银行界使用的模型的报告（ Dembo 1990 年， Ruffel 1986 年）。规模较小的银行需 要更多，视莫伊尼汉等人研究谦虚。 （ 2002 年），谁提出一个信用社决策支持系统。作为行业管理的重点是下跌的风险，它是模拟模型当道工业强调情景生成，报告和工作流程管理。部分或研究人员也朝着这个方向（例如，马尔维等 2000 年）。 保险行业似乎落后于资产负债管理，因此唱片造型，如从一个相对明显的瑞士再保险公司最近的报告告诫的生命和财产 /伤亡业
44、实行资产负债管理的做法（瑞士再保险公司 2000 年）。动态模型以及要求并行计算机在只有 4000 方案（马尔维和 Shetty 2004）使用的解决方案易处理的挑战。虽然 Dert（ 1999 年）显示， multistagdynamic 模型提供了比近视或一个简单的静态模型更好的解决方案，也有些人提出有关动态模型在更复杂的利益问题（ 1995 年再尼奥斯）静态模型。 7 因此，有许多原因，这是值得探讨如何简化动态唱片模型（因为这样做，例如，通过 Gaivoronski 和去修复朗格 2000 使用混合或马尔维等。使用混合型证券 1999 年）之前，我们可以整合他们将与模拟混合打包为资产负债
45、管理软件（因为这样做，例如，通过 Boender 1997 年，由 Seshadri 等 1999 年）。目前国家的最先进 的（例如，马尔维和 Shetty 2004）使用并行和分布式计算机可能不适合大多数企业的计算环境。 Dembo 和 Rosen（ 1999）提到一个静态方法，以及“半动态”在金融领域的战略传统的越来越大的兴趣，他们声称，静态方法时，可能就足够了整个企业的风险（附属于资产负债管理）与有关的多达 400个风险因素。本文旨在帮助研究人员了解简化的选择。针对包装的 ALM 软件市场是一个不错的想法： 400 最大的金融机构资产负债管理软件据说花了 6.13 亿美元 1998 年（
46、奥康奈尔 1999 年）全球。这是继在 1999-2003 年期间支出 的放缓，分析师预计约为 6 亿美元在全球同类支出 2004 年预期将增长 2009 年（开普勒 2004） 8.5 亿美元。 作为这项工作的限制，我们不包括非唱片建模方法，并参考读者 Ziemba和马尔维（ 1999 年），并再尼奥斯和 Ziemba（ 2004 年）。本文的模型拆开匹配现金流基本约束，而不是针对特定行业或监管的，虽然我们在下一节提供一些有用的提述，以了解特定行业的负债。我们也没有提供在一个模型来最终解决，但现在，希望研究人员将使用到该结束这个调查。最后，我们限制了讨论利率风险已经大部分或文学，尽管信贷风险
47、（不良 贷款）是非常重要的，因为是银行的汇率风险。在许多大公司养老基金后，在千年点转向了市场风险的重要性，以及股市的垮台不足，但我们不解决这个问题要么是指除研究人员谁创造了多重危险因素的共同方案。尽管如此，这里的大部分讨论无论是相关的风险是有限的利率或其他因素 2.背景 财务会计准则委员会在美国的定义是可能的未来经济利益和经济利益牺牲未来可能的资产负债。资产负债管理需要利用这些资产和负债，以提高该机构的净资产，而相关的各种风险量化，管理流动性，以及简化管理法规要求（翁1998 年）。资产负债管理已经从一个在 20 世纪 60 年代和 70 年代的会计集中到中央，综合“ riskmanageme
48、nt”功能适用于 20 世纪 90 年代中期（贾罗和 van Deventer 的 1998 年）和不同的公司在不同的“梯级”的这种变化对整个资产负债表（马尔维等人 1997 年，马尔维和 Ziemba 1999 年）。 在银行业，负债，客户账户，而资产抵押贷款和个人贷款。现金流量取决于利率的变动，如预付贷款人在利率下降。许多银行在 20 世纪 80 年代初的失败提供了一个迫切需要以管理利率风险的更好。因此，投资者对金融机构的奖励以较高的股票价格（ Giarla 1991）更稳定的高品质的收益。翁（ 1998）和比特纳和戈达德（ 1992）讨论了银行业资产负债管理的做法。科恩和 Gibson（
49、 1978）8 对银行集中与第五部分用于债券投资组合管理和利率风险。 Adamidou 等（ 1993 年）和 Ben - Dov 的等（ 1992）描述了一个在英国保诚，八坼最优投资组合系统。 Seshadri 等（ 1999）使用模拟和二次优化“制定，测试和完善”联邦住宅贷款纽约银行的资产负债管理政策。 在保险业，资产是“固定收益”，如国债，企业债券，证券和 mortgagebacked文书。负债是与客户相关的保险不必依赖于利率的支出。 Babbel 和放样（ 1991）的报告，暴露于利率风险是其中最强的预测指标测试变量解释财产 /责任保险公司破产。卡里诺等（ 1994 年），卡里诺和 Ziemba（ 1998），卡里诺等（ 1998 年），描述的是需要保险公司资产负债管理和决策支持系统，他们为日本安田保险公司开赛建成。 Worzel 等（ 1994）描述在大都会人寿保险管理固定收益投资组合的索引作为“责任。”霍尔默（ 1994， 1999）描述了在参考按揭保险和抵押贷款投资的一个在联邦国民抵押贷款协会的申请。斯威 尼等（ 199
