XGBoost를 활용한 리스크패리티 자산배분 모형에 관한 연구

김영훈(Younghoon Kim),최흥식(HeungSik Choi),김선웅(SunWoong Kim) 한국지능정보시스템학회 2020 지능정보연구 Vol.26 No.1

Artificial intelligences are changing world. Financial market is also not an exception. Robo-Advisor is actively being developed, making up the weakness of traditional asset allocation methods and replacing the parts that are difficult for the traditional methods. It makes automated investment decisions with artificial intelligence algorithms and is used with various asset allocation models such as mean-variance model, Black-Litterman model and risk parity model. Risk parity model is a typical risk-based asset allocation model which is focused on the volatility of assets. It avoids investment risk structurally. So it has stability in the management of large size fund and it has been widely used in financial field. XGBoost model is a parallel tree-boosting method. It is an optimized gradient boosting model designed to be highly efficient and flexible. It not only makes billions of examples in limited memory environments but is also very fast to learn compared to traditional boosting methods. It is frequently used in various fields of data analysis and has a lot of advantages. So in this study, we propose a new asset allocation model that combines risk parity model and XGBoost machine learning model. This model uses XGBoost to predict the risk of assets and applies the predictive risk to the process of covariance estimation. There are estimated errors between the estimation period and the actual investment period because the optimized asset allocation model estimates the proportion of investments based on historical data. these estimated errors adversely affect the optimized portfolio performance. This study aims to improve the stability and portfolio performance of the model by predicting the volatility of the next investment period and reducing estimated errors of optimized asset allocation model. As a result, it narrows the gap between theory and practice and proposes a more advanced asset allocation model. In this study, we used the Korean stock market price data for a total of 17 years from 2003 to 2019 for the empirical test of the suggested model. The data sets are specifically composed of energy, finance, IT, industrial, material, telecommunication, utility, consumer, health care and staple sectors. We accumulated the value of prediction using moving-window method by 1,000 in-sample and 20 out-of-sample, so we produced a total of 154 rebalancing back-testing results. We analyzed portfolio performance in terms of cumulative rate of return and got a lot of sample data because of long period results. Comparing with traditional risk parity model, this experiment recorded improvements in both cumulative yield and reduction of estimated errors. The total cumulative return is 45.748%, about 5% higher than that of risk parity model and also the estimated errors are reduced in 9 out of 10 industry sectors. The reduction of estimated errors increases stability of the model and makes it easy to apply in practical investment. The results of the experiment showed improvement of portfolio performance by reducing the estimated errors of the optimized asset allocation model. Many financial models and asset allocation models are limited in practical investment because of the most fundamental question of whether the past characteristics of assets will continue into the future in the changing financial market. However, this study not only takes advantage of traditional asset allocation models, but also supplements the limitations of traditional methods and increases stability by predicting the risks of assets with the latest algorithm. There are various studies on parametric estimation methods to reduce the estimated errors in the portfolio optimization. We also suggested a new method to reduce estimated errors in optimized asset allocation model using machine learning. So this study is meaningful in that it proposes an advanced artificial intelligence asset allocation model for the fast-developing financial markets.

결과변수와 공변량의 측정 오차를 고려한 Fay-Herriot 모형

류수락(Soorack Ryu),황진섭(Jinseub Hwang) 한국자료분석학회 2020 Journal of the Korean Data Analysis Society Vol.22 No.1

소지역의 효율적인 추정을 위해 Fay-Herriot 모형의 다양한 확장 버전이 제안되었다. 그러나 대부분의 사전 연구에서는 공변량(covariate)에만 측정오차를 고려하였으며 결과변수(outcome variable)의 측정오차(measurement error)는 고려하지 않고 있다. 이에 본 연구에서는 결과변수와 공변량의 측정오차모형을 반영한 확장된 Fay-Herriot 모형을 제안하고자 한다. 측정오차를 가지는 결과변수의 참값(true value)에 비확률성(non-stochastic)을 가정한 기능적 측정오차모형(functional measurement error model)과 측정오차를 가지는 공변량의 참값에 확률성(stochastic)을 가정한 구조적 측정오차모형(structural measurement error model)을 고려하였다. 모형 적합과 모수 추정을 위해 MCMC(Markov chain Monte Carlo) 방법의 계층적 베이지안 모형을 사용하였다. 본 연구에서 제안하는 모형의 우수성을 확인하기 위해 모의실험과 실증자료 분석을 수행하였으며, Arima et al.(2015)에서 사용한 선형적 함수와 2010 이탈리아 가계 예산 조사(2010 Italian household budget survey) 자료를 사용하였다. 모의실험과 실증자료 분석을 통해 본 연구에서 제안하는 모형의 성능이 더 우수하였다. Various extension versions for Fay-Herriot model have been proposed for efficient small area estimation. However, most previous studies have considered only covariate of measurement error model even though the outcome variable may also have measurement error model. In this paper, we propose an extended Fay-Herriot model that can reflect the measurement error model of outcome variable and covariate. We consider a measurement error model of the outcome variable and covariate. To fit the model and estimate parameters, we consider hierarchical Bayesian model approach based on Markov chain Monte Carlo method. To check the superiority of the proposed model, we use one linear function in simulation studies and we use the 2010 Italian household budget survey data for the empirical study. The results for simulation studies and the empirical study show that the proposed model has the better performance than the error model considering measurement error of covariate only.

Experimental validation of Kalman filter-based strain estimation in structures subjected to non-zero mean input

Palanisamy, Rajendra P.,Cho, Soojin,Kim, Hyunjun,Sim, Sung-Han Techno-Press 2015 Smart Structures and Systems, An International Jou Vol.15 No.2

Response estimation at unmeasured locations using the limited number of measurements is an attractive topic in the field of structural health monitoring (SHM). Because of increasing complexity and size of civil engineering structures, measuring all structural responses from the entire body is intractable for the SHM purpose; the response estimation can be an effective and practical alternative. This paper investigates a response estimation technique based on the Kalman state estimator to combine multi-sensor data under non-zero mean input excitations. The Kalman state estimator, constructed based on the finite element (FE) model of a structure, can efficiently fuse different types of data of acceleration, strain, and tilt responses, minimizing the intrinsic measurement noise. This study focuses on the effects of (a) FE model error and (b) combinations of multi-sensor data on the estimation accuracy in the case of non-zero mean input excitations. The FE model error is purposefully introduced for more realistic performance evaluation of the response estimation using the Kalman state estimator. In addition, four types of measurement combinations are explored in the response estimation: strain only, acceleration only, acceleration and strain, and acceleration and tilt. The performance of the response estimation approach is verified by numerical and experimental tests on a simply-supported beam, showing that it can successfully estimate strain responses at unmeasured locations with the highest performance in the combination of acceleration and tilt.

Experimental validation of Kalman filter-based strain estimation in structures subjected to non-zero mean input

조수진,Rajendra P. Palanisamy,김현준,심성한 국제구조공학회 2015 Smart Structures and Systems, An International Jou Vol.15 No.2

Response estimation at unmeasured locations using the limited number of measurements is an attractive topic in the field of structural health monitoring (SHM). Because of increasing complexity and size of civil engineering structures, measuring all structural responses from the entire body is intractable for the SHM purpose; the response estimation can be an effective and practical alternative. This paper investigates a response estimation technique based on the Kalman state estimator to combine multi-sensor data under non-zero mean input excitations. The Kalman state estimator, constructed based on the finite element (FE) model of a structure, can efficiently fuse different types of data of acceleration, strain, and tilt responses, minimizing the intrinsic measurement noise. This study focuses on the effects of (a) FE model error and (b) combinations of multi-sensor data on the estimation accuracy in the case of non-zero mean input excitations. The FE model error is purposefully introduced for more realistic performance evaluation of the response estimation using the Kalman state estimator. In addition, four types of measurement combinations are explored in the response estimation: strain only, acceleration only, acceleration and strain, and acceleration and tilt. The performance of the response estimation approach is verified by numerical and experimental tests on a simply-supported beam, showing that it can successfully estimate strain responses at unmeasured locations with the highest performance in the combination of acceleration and tilt.

A posteriori error estimator for hierarchical models for elastic bodies with thin domain

Cho, Jin-Rae Techno-Press 1999 Structural Engineering and Mechanics, An Int'l Jou Vol.8 No.5

A concept of hierarchical modeling, the newest modeling technology, has been introduced in early 1990's. This new technology has a great potential to advance the capabilities of current computational mechanics. A first step to implement this concept is to construct hierarchical models, a family of mathematical models sequentially connected by a key parameter of the problem under consideration and have different levels in modeling accuracy, and to investigate characteristics in their numerical simulation aspects. Among representative model problems to explore this concept are elastic structures such as beam-, arch-, plate- and shell-like structures because the mechanical behavior through the thickness can be approximated with sequential accuracy by varying the order of thickness polynomials in the displacement or stress fields. But, in the numerical, analysis of hierarchical models, two kinds of errors prevail, the modeling error and the numerical approximation error. To ensure numerical simulation quality, an accurate estimation of these two errors is definitely essential. Here, a local a posteriori error estimator for elastic structures with thin domain such as plate- and shell-like structures is derived using the element residuals and the flux balancing technique. This method guarantees upper bounds for the global error, and also provides accurate local error indicators for two types of errors, in the energy norm. Compared to the classical error estimators using the flux averaging technique, this shows considerably reliable and accurate effectivity indices. To illustrate the theoretical results and to verify the validity of the proposed error estimator, representative numerical examples are provided.

Novel state estimation framework for humanoid robot

Bae, Hyoin,Oh, Jun-Ho Elsevier 2017 Robotics and autonomous systems Vol.98 No.-

<P><B>Abstract</B></P> <P>This study proposes a new Kalman filter-based framework for humanoid robot state estimation. The conventional Kalman filter generates optimal estimation solutions only when the nominal equations of the model and measurement include zero-mean, uncorrelated, white Gaussian noise. Because a humanoid robot is a complex system with multiple degrees of freedom, its mathematical model is limited in terms of expressing the system accurately, resulting in the generation of non-zero-mean, non-Gaussian, correlated modeling errors. Therefore, it is difficult to obtain accurate state estimates if the conventional Kalman filter-based approaches are used with such inexact humanoid models. The proposed modified Kalman filter framework consists of two loops: a loop to estimate the state, and a loop to estimate the disturbance generated by the modeling errors (a dual-loop Kalman filter). The disturbance values estimated by the disturbance estimation loop are provided as feedback to the state estimation loop, thereby improving the accuracy of the model-based prediction process. By considering the correlation between the state and disturbance in the estimation process, the disturbance can be accurately estimated. Therefore, the proposed estimator allows the use of a simple model, even if it implies the presence of a large modeling error. In addition, it can estimate the humanoid state more accurately than the conventional Kalman filter. Furthermore, the proposed filter has a simpler structure than the existing robust Kalman filters, which require the solution of complex Riccati equations; hence, it can facilitate recursive online implementation. The performance and characteristics of the proposed filter are verified by comparison with other existing linear/nonlinear estimators using simple examples and simulations. Furthermore, the feasibility of the proposed filter is verified by implementing it on a real humanoid robot platform.</P> <P><B>Highlights</B></P> <P> <UL> <LI> The proposed estimator framework consists of two estimator loops. </LI> <LI> One loop for the state estimation and the other loop for the disturbance estimation. </LI> <LI> The disturbance estimated in the disturbance estimation loop is provided as feedback. </LI> <LI> The correlation between state and disturbance is calculated automatically. </LI> <LI> The proposed estimator complements the fundamental limitations of the original KF. </LI> </UL> </P>

베이지안 기법을 적용한 Incomplete data 기반 신뢰성 성장 모델의 모수 추정

박천건(Cheongeon Park),임지성(Jisung Lim),이상철(Sangchul Lee) 한국항공우주학회 2019 韓國航空宇宙學會誌 Vol.47 No.10

신뢰성 성장 시험을 수행하며 획득하게 되는 고장 정보와 누적 시험수행시간을 이용하면 신뢰성 성장 모델의 모수 추정이 가능하며, 모수 추정을 통해 해당 제품의 MTBF를 예측할 수 있다. 그러나 시험에 대한 비용, 시간 혹은 제품의 특성 등의 여러 제약으로 인해 고장 정보가 구간적으로 획득되거나, 획득한 고장 정보의 샘플 데이터(Sample Data)의 수가 작을 수 있다. 이는 신뢰성 성장 모델의 모수 추정의 오차를 커지게 하는 원인이 될 수 있다. 본 논문에서는 샘플 데이터의 수가 작을 경우 신뢰성 성장 모델의 모수 추정 시 베이지안 기법 기반의 모수 추정 방법의 적용에대해 연구를 수행하였다. 시뮬레이션 결과 신뢰성 성장 모델의 모수를 추정할 때, MLE를 적용하여 추정하는 방법보다 베이지안 기법을 적용하는 방법이 추정 정확도가 높음을 확인하였다. By using the failure information and the cumulative test execution time obtained by performing the reliability growth test, it is possible to estimate the parameter of the reliability growth model, and the Mean Time Between Failure (MTBF) of the product can be predicted through the parameter estimation. However the failure information could be acquired periodically or the number of sample data of the obtained failure information could be small. Because there are various constraints such as the cost and time of test or the characteristics of the product. This may cause the error of the parameter estimation of the reliability growth model to increase. In this study, the Bayesian method is applied to estimating the parameters of the reliability growth model when the number of sample data for the fault information is small. Simulation results show that the estimation accuracy of Bayesian method is more accurate than that of Maximum Likelihood Estimation (MLE) respectively in estimation the parameters of the reliability growth model.

A Comparative Study of Generalized Maximum Entropy Estimator for the Two-way Error Component Regression Model

전수영,진서훈 한국자료분석학회 2009 Journal of the Korean Data Analysis Society Vol.11 No.2

Recently the study of the panel data has received attention in the literature of the regression model. The model has been usually dealing with the complete data. However, in a practical manner it is rare for data to be complete. For ill-posed problems, Song and Cheon(2006) proposed a robust generalized maximum entropy estimator less sensitive to the assumption and limited situation in a panel regression model with the only individual effect. However, the time effect needs to be considered in panel data. This paper considers a two-way error component model with both individual and time effects in ill-posed problems and proposes the generalized maximum entropy(GME) estimator for the unknown parameters. This estimator is compared with a variety of existing estimators on the simulated dataset. The numerical results are in favor of the new estimator in terms of its quality when the data are ill-posed. Recently the study of the panel data has received attention in the literature of the regression model. The model has been usually dealing with the complete data. However, in a practical manner it is rare for data to be complete. For ill-posed problems, Song and Cheon(2006) proposed a robust generalized maximum entropy estimator less sensitive to the assumption and limited situation in a panel regression model with the only individual effect. However, the time effect needs to be considered in panel data. This paper considers a two-way error component model with both individual and time effects in ill-posed problems and proposes the generalized maximum entropy(GME) estimator for the unknown parameters. This estimator is compared with a variety of existing estimators on the simulated dataset. The numerical results are in favor of the new estimator in terms of its quality when the data are ill-posed.

A New Form of Nondestructive Strength-Estimating Statistical Models Accounting for Uncertainty of Model and Aging Effect of Concrete

Kee-jeung Hong,Jee-sang Kim 한국비파괴검사학회 2009 한국비파괴검사학회지 Vol.29 No.3

As concrete ages, the surrounding environment is expected to have growing influences on the concrete. As all the impacts of the environment cannot be considered in the strength-estimating model of a nondestructive concrete test, the increase in concrete age leads to growing uncertainty in the strength-estimating model. Therefore, the variation of the model error increases. It is necessary to include those impacts in the probability model of concrete strength attained from the nondestructive tests so as to build a more accurate reliability model for structural performance evaluation. This paper reviews and categorizes the existing strength-estimating statistical models of nondestructive concrete test, and suggests a new form of the strength-estimating statistical models to properly reflect the model uncertainty due to aging of the concrete. This new form of the statistical models will lay foundation for more accurate structural performance evaluation.

기업의 조세부정예측모형에 관한 연구

박현재 ( Hyun Jae Park ),김갑순(교신저자) ( Kap Soon Kim ) 한국회계학회 2016 회계저널 Vol.25 No.3

본 연구는 국내 기업의 회계자료 및 기업특성을 이용하여 조세부정 정도를 추정하는 모형을 제시하고, 예측모형의 구체적인 활용방안을 설명하였다. 구체적으로, 2000년부터 2013년까지 세무조사로 적발되어 전자공시시스템(DART)에 공시한 12월 결산 상장법인과 대응표본을 대상으로 100회 무작위추출에 의한 단계적 로짓분석(stepwise logisticregression)을 실시하여 조세부정 예측모형을 추정하였다. 최종 예측모형에는 유효세율, 부채비율, 수출비율, 매출액, 회계이익과 과세소득의 차이, 재벌여부, 감사인 규모의 7개 변수가 포함되었다. 분석결과, 국내 상장기업들은 유효세율, 부채비율, 매출액이 높을수록 조세부정 가능성이 높았으며, 반면 수출비율, 회계이익과 과 세소득의 차이가 낮을수록 조세부정 가능성이 높은 것으로 나타났다. 또한 재벌에 소속되어 있으며 외부감사인이 Non-Big4인 경우에 조세부정 가능성이 높았다. 본 연구의 예측모형은 모든 기업을 비부정기업으로 가정하는 단순전략과 비교하여 총기대오류비용을 크게 감소시킬 수 있는 것으로 나타나 비용효율적임을 제시하였다. 재무제표와 기업특성 자료만을 필요로 하는 본 연구의 예측모형은 과세당국, 학계, 외부감사인, 투자자들이 특정 기업의 조세부정 가능성을 일차적으로 평가하는데 유용할 것으로 기대된다. Tax fraud affects the national budget to reduce the tax revenues of the country. Lisowsky(2010) is, U.S. Treasury(1999) with reference to the report, it reported that tax revenues that the federal government will be lost in the tax evasion has reached 10 billion dollars a year. In Korea, there is a trend that is collected additionally the number of tax audits and the amounts collected additionally associated with increased offshore tax evasion annually. Domestic and foreign tax authorities are making various efforts to prevent such these tax fraud. U.S. Internal Revenue Service(IRS) in 2000 to tax fraud transactions(tax shelter) dedicated to agencies(the Office of Tax Shelter Analysis, OTSA) established by the tax fraud monitoring and there. Even concluded in the domestic Foreign Account Tax Compliance Act(FATCA) from September 2015 agreed to strengthen offshore tax evasion blocked by the U.S. Internal Revenue Service(IRS) through the financial information exchange. Many studies theoretically explain and empirically analyze relevance of between of corporate tax burden level, financial characteristics, corporate governance, book to tax income differences and tax avoidance. In addition, previous studies related to tax evasion has focused to analyze the differences between the various properties compared to the corporate tax fraud caught by the companies. However, can be utilized in practice, decisions regarding models and tax evasion about to estimate the extent of tax evasion of specific companies, remains a substantial portion unsolved problem. In this study, while presenting a predictive model that can determine whether to tax fraud, the specific manner of utilization of the model is trying to explore. In particular, an attempt to estimate a more sophisticated predictive models, comprehensive consideration tax avoidance and financial characteristics that it has been confirmed that affect the tax evasion, ownership and governance structure characteristics, the external auditor of the properties are in the previous research did. Specifically, the effective tax rate, company size, profitability, ratio of assets investment, debt ratio, export ratio, sales, financial constraints, discretionary accruals, research and development expenses, of the difference between profit and taxable income on eleven numbers of financial characteristic variables, ownership interest rate, foreign interest rates, jaebul, whether three of ownership and characteristic variables of the governance structure, the auditor scale of one of the external auditor of the characteristic variables to the consideration stepwise logistic regression was utilized to derive the final predictive model. In order to eliminate the estimated arbitrariness of the estimation model, 50 percent of the entire sample to predict the estimation sample in the random sampling method, and the remaining 50 percent and holdout samples. Further, in order to solve one of the reliability issues due to random sampling, 100 times and randomly, and extracted the available variables in predictive model. By using the validation sample in order to verify the suitability of the estimated prediction model, to derive the tax fraud probability of tax fraud corporate and non-fraud corporate, tax fraud probability of tax fraud company has verified whether significantly higher. Further, we is trying to verification of the economic efficiency of tax fraud estimation model by comparing the total expected cost when using the predictive model or not by the methodology of Beneish(1997), Ko and Yoon(2006). This study develops a model to detect tax fraud using a sample of firms identified as tax fraud by the Financial Supervisory Service of Korea and characteristics of corporate for the period from 2000 to 2013. We find seven useful variables: tax expense/pretax book income(ETR); debt/total asset(LEV); export proceeds/total sales(EXPORT); log(total sales)(SALES); book-tax difference/total asset(BTD); jaebul(GROUP); auditor size(BIG4). Results suggest that Korean firms are more likely to manipulate taxable income when ETR, LEV, SALES are more larger and EXPORT, BTD are more smaller and these firms are GROUP and audited by Non-Big4. Compared with a naive strategy of classifying all firms as non-manipulators, our estimation model is expected to significantly reduce costs of misclassification. Tax regulator, academic, external auditors, investor, and other users of financial statements might use our model to screen potential tax fraud, to which they can allocate additional resources for more in-depth analysis.