Effect of SME and Chaebol Loan Contracts on Debt Covenant Hypothesis: Evidence from Korea

( Sun Min Kang ),( In Tae Hwang ),( Kang Sung Hur ) 한국회계학회 2017 회계학연구 Vol.42 No.2

This study provides an in-depth analysis of the covenant-based hypothesis that focuses on whether and how correlations between debt ratios and earnings management (measured by discretionary accruals) are affected by the size and nature of firms. We performed a pooled ordinary least squares regression using a sample of 10,983 firms listed in Korea from 2002 through 2012. In the Korean financial market, SMEs and firms belonging to chaebols (conglomerates) are markedly different in terms of their bargaining power with financial institutions. The results suggest that SMEs have a larger portion of secured loans and that chaebols can use their credit in the bond market to access alternative sources of financing. Accordingly, neither SMEs nor chaebols are motivated to make income-increasing choices for larger discretionary accruals to relax their debt constraints. In contrast, a large firm with a higher debt ratio is likely to make accounting choices to increase reported earnings, which results in more discretionary accruals. The debt agreements between lenders (financial institutions) and borrowers (firms) have different implications in the Korean financial market depending on a firm`s size and nature.

우리말의 자동 음소분할에 관한 연구

권호민,한학용,허강인 동아대학교 정보기술연구소 2003 情報技術硏究所論文誌 Vol.11 No.1

In this paper we used Support Vector Machines (SVMs) recently proposed as the learning method, one of Artificial Neural Network, to divide continuous speech into phonemes, an initial, medial, and final sound, and then, performed continuous speech recognition from it. Decision boundary of phoneme is determined by algorithm with maximum frequency in a short interval. Recognition process is performed by Continuous Hidden Markov Model(CHMM), and we compared it with another phoneme divided by eye-measurement. From experiment we confirmed that the method, SVMs, we proposed is more effective in an initial sound than Gaussian Mixture Models(GMMs).

패턴구성에 따른 RPNN의 음성인식 성능비교

김주성,한학용,허강인 東亞大學校 附設 情報通信硏究所 1997 情報通信硏究所論文誌 Vol.5 No.1

In this paper, we use th RPNN as a non-linear predictor according to the time variation of a speech without an alignment procedure. We perform the recognition experiment of Korean 100 syllables based on RPNN In the experiment, we compare the recognition rates, by increasing the prediction order and the number of hidden(recursive) units, and by composing the pattern. When the prediction order is 3-rd order and the hidden units are 10 units, the recognition rates show better result than others. And when we compose the pattern as the method 2, they are far improved than the method 1. In the method 1, the highest recognition rate is 82.40% at the 4-th prediction order and the 10 hidden units. And in the method 2, it is 84.90% at the 3-rd prediction order and the 10hidden units, and it is improved at 2.50% than the method 1

Third-Generation P2Y₁₂ Inhibitors in East Asian Acute Myocardial Infarction Patients: A Nationwide Prospective Multicentre Study

Kang, Jeehoon,Han, Jung-Kyu,Ahn, Youngkeun,Chae, Shung Chull,Kim, Young Jo,Chae, In-ho,Hur, Seung-Ho,Seong, In-Whan,Chae, Jei-Keon,Cho, Myeong Chan F K SCHATTAUER VERLAGSGESELLSCHAFT MBH 2018 Thrombosis and Haemostasis Vol.118 No.3

<P>Third-generation P2Y(12) inhibitors (prasugrel, ticagrelor) are recommended in acute myocardial infarction (AMI). We aimed to evaluate the efficacy and safety of third-generation P2Y(12) inhibitors in East Asian AMI patients. From the Korean AMI Registry, 9,355 patients who received dual antiplatelet agent (aspirin with clopidogrel [AC], 6,444 [70.5%] patients; aspirin with prasugrel [AP], 1,100 [11.8%] patients; or aspirin with ticagrelor [AT], 1,811 [19.4%] patients) were analysed. In-hospital endpoints were all-cause mortality or bleeding events during admission and 1-year endpoints were major adverse cardiac and cerebrovascular events(MACCE) andmajor bleedingevents. Regardingin-hospital events, AP andAT showed similar all-causemortality rates but higher bleeding event rates comparedwith AC. This trend was extended to 1-year endpoints; Cox regression analysis showed that thirdgeneration P2Y(12) inhibitors had significantly higher bleeding risk (AP vs. AC: hazard ratio [HR], 2.14; 95% confidence interval [CI], 1.53-2.99; p < 0.001; AT vs. AC: HR, 2.26; 95% CI, 1.73-2.95; p < 0.001). A propensity scorematched tripletof 572 patients showed similar 1year MACCE and higher bleeding events with third-generation P2Y(12) inhibitors (2.1 vs. 2.6 vs. 2.1%, p = 0.790 for MACCE and 3.1 vs. 8.0 vs. 8.0%, p < 0.001 for bleeding events, in AC, AP andAT groups, respectively). Inverse probabilityweighted regression analysis and pooled analysis after randomly imputing missing variables showed consistent results. Collectively, prasugrel and ticagrelor showed similar rates of 1-year MACCE, but a higher rate of bleeding events, compared with clopidogrel in Korean AMI patients. Further studies are warranted to adapt Western guidelines on third-generation P2Y(12) inhibitors for East Asians.</P>

회귀구조에 따른 예측신경망의 음성인식 성능비교

한학용,김주성,허강인 東亞大學校 附設 情報通信硏究所 1998 情報通信硏究所論文誌 Vol.6 No.1

In this paper, we compare the recognition rate between Recurrent Prediction Neural Networks and CHMM according to increasing the prediction order, the number of hidden units for RPNN and the state for CHMM. We perform the recognition experiment of Korean 100 syllables and ETRI's SAMDOLI Speech DB. In the DIGIT syllables, the recognition rate of Neural Networks, 98.5%, similar to CHMM, 99%, of the 5 state numner. In the 100 syllabes, the Jordan's RPNN which based on the number of hidden units is larger than 25 units shows the higher reocgnition rate than CHMM of the 5 state number. When CHMM based on the 6 and 7 state, RPNN's recognition rate is lower than CHMM.

상태수 변화에 따른 NNPHMM의 성능비교

김창근,김상범,김수훈,허강인 東亞大學校 附設 情報通信硏究所 1997 情報通信硏究所論文誌 Vol.5 No.1

In this paper, we compose neural predictive HMM(NNPHMM) to provide the dynamic feature of the speech pattern for the HMM. The NNPHMM is the hybrid network of neural network and the HMM. The NNPHMM trained to predict the future vector, varies each time. It is used instead of the mean vector in the HMM. In the experiment, we compared the recognition abilities of the one hundred Korean syllables according to the variation of state number and prediction order. The state number of the NNPHMM increased from 4 to 6. And the prediction orders increased from 2nd to 4th order. The NNPHMM in the experiment is composed of multi-layer perceptron with one hidden layer and CMHMM. As a result of the experiment, the average recognition rates are 85.2%, 85.6% and 85.5% when the state number is 5, the prediction order is the 2nd, 3rd, 4th, and the hidden layer is 10 dimensions, respectively.

독립성분분석과 SVM을 이용한 강인한 화자식별시스템

박정원,김평환,김창근,허강인 동아대학교 정보기술연구소 2003 情報技術硏究所論文誌 Vol.11 No.1

In this paper, we propose effective speaker verification method using SVM(support vector machines) and ICAhdependent component analysis). In general, SVM is classification method which classify two class set by finding voluntary nonlinear boundary in vector space and possesses high classification performance under few training data number. In this paper, we compare existing algorithms with SVM, and we compare verification performance of feature parameters using ICA, PCA(principa1 component analysis), MFCC(mel frequency cepstrum coefficient) for SVM speaker verification system. In result, ICA feature parameter showed the highest verification performance because of superior linear classification.

음성인식기 구현을 위한 잡음에 강인한 음성구간 검출기법

김창근,박정원,허강인 동아대학교 정보기술연구소 2003 情報技術硏究所論文誌 Vol.10 No.2

One of the most important subjects, in the implementation of real time speech recognizer, is to design both reliable VAD(Voice Activity Detection) and suitable speech feature vector. But, because it is difficult to calculate reliable VAD in the environment having surrounding noise, designing suitable speech feature vector may not be obtained. Solving this problem, in this paper, we implement not only short time power spectrum which is generally used but also two additive parameters, the comparison measure of spectrum density having robust property in noise and linear discriminant function using linear regression, perform VAD by using the combination of each parameter having apt weight in other magnitudes of surrounding noise and confirm that proposed parameters have a robust characteristic in circumstances having surrounding noise by using DTW(Dynamic Time Warping) in recognition experiment.

항공기 자세회복을 위한 자동회복장치 설계 및 검증에 관한 연구

김종섭(Chong-sup Kim),조인제(In-je Cho),강임주(Im-ju Kang),허기봉(Gi-bong Hur),이은용(Eun-yong Lee) 제어로봇시스템학회 2008 제어·로봇·시스템학회 논문지 Vol.14 No.3

Relaxed static stability(RSS) concept has been applied to improve aerodynamic performance of modem version supersonic jet fighter aircraft. Therefore, flight control system are necessary to stabilize an unstable aircraft and provides adequate handling qualities. Also, flight control systems of modem version aircraft employ a safety system to support emergency situations such as a pilot unknown attitude flight conditions of an aircraft in night flight-testing. This situation is dangerous because the aircraft can lose if the pilot not take recognizance of situation. Therefore, automatic recovery system is necessary. The system called the "Pilot Activated Recovery System" or PARS, provided a pilot initiated automatic maneuver capable of an aircraft recoveries in situations of unusual attitudes. This paper addresses the concept of PARS and designed using nonlinear control law design process based on model of supersonic jet trainer. And, this control law is verified by nonlinear analysis and real-time pilot evaluation using in-house software. The result of evaluation reveals that the PARS support recovery of an aircraft unusual attitude and improve a safety of an aircraft.

CFG를 이용한 제어문 음성 인식에 관한 연구

김상범,김주성,김수훈,허강인 東亞大學校 附設 情報通信硏究所 1996 情報通信硏究所論文誌 Vol.4 No.1

In this study, we perform continuous speech recognition system using context free grammar. We use DDCHMM as train model, that is able to express time-transient-variation very well. Using one pass DP method as recognition algorithm which is attempted by expanding dynamically a finite automaton from a part of suggesting context-free-grammar of recognition process. We can increase the recognition rates of control sentence speech recognition. Car control command, robot control command, and newspapers editorial continuous speech are made to test recognition rates of control sentence speech using context-free-grammer.