{"title":"효율적 위기예측을 위한 패널자료의 선택: 신호접근모형을 중심으로 (Crisis Prediction and Choice of Panel Data: The Case of Signal Extraction Model)","authors":"Kyungsoo Kim","doi":"10.2139/ssrn.3019737","DOIUrl":null,"url":null,"abstract":"<b>Korean Abstract:</b> 본 연구는 효율적인 외환위기예측을 위한 표본자료의 선택방안을 찾는 데 목적이 있다. 이를 위해 외환위기를 겪었던 24개국을 동아시아, 중남미, 유럽, 중동·아프리카 등 4개 지역으로 구분하고 다시 15개 소그룹으로 패널자료를 구성하였다. 15개 패널 자료별로 신호접근법에 따른 외환위기예측모형을 구축, 동아시아 외환위기기간을 표본외예측으로 하여 외환위기 예측성과를 평가하였다. 상식적 견해와 달리 동아시아를 제외하면 특정지역만을 대상으로 구축한 모형의 예측성과는 다른 지역을 포함한 모형보다 저조하였다. 한편 동아시아 지역만을 대상으로 구축한 예측모형은 표본외예측에서 잡음/신호비율이 작은 대신 신호회수가 적고 종합위기지수로부터 구한 위기확률이 낮았다. 그것뿐만 아니라 경보를 작동한 지표의 유형이 대부분 비금융부문에 속해 동아시아 외환위기의 공동요인으로 평가되는 취약한 금융에 대한 유용한 정보를 제공하지 못하였다. 확장된 표본기간에서 이들 모형은 지표의 최적임계치와 예측력에 큰 변화가 있고 유효지표가 뒤바뀌어 모형이 안정성 측면에서 취약하였으며 그 결과 표본외예측에서 제2종오류가 과다하게 발생하였다. 이상의 결과는 비록 표본 예측력이 떨어져도 광범위한 패널자료에 의존하는 예측모형의 표본외 예측력이 더 우수하다는 함의를 가진다.<br><br><b>English Abstract:</b> The purpose of this study is to choose acceptable panel data for crisis prediction. According to common sense view it would be best to use panel data of East Asian countries when it comes to predict crises in these countries. Contrary to that view it is not, however. The paper considers 15 combinations of panel data. These panel data are composed by maximum four regional groups of 24 crisis-ridden countries such as East Asia, Latin America, Europe, and Middle East and Africa. And then the paper builds 15signal extraction models (SEM) based on each combination of panel data and assesses the predictability of currency crisis. SEM based on panel data composed of each individual region does not perform well except East Asia. East Asian countries, however, although it has the lowest noise signal ratio, has given least warning signals and the lowest probability of crisis associated with the crisis composite index. Furthermore, most indicators alarmed are non-financial, which fails to provide the useful information such that financial fragility is the common cause of the crisis. When sample period is extended there’s a big change in both indicator’s optimal threshold level and the noise signal ratio, and even effective indicators. As a result, all models of the panel data based on each individual region have serious type II error problem in out-sample forecast. The implication is that SEM based on broader panel data even though it should be inferior in in-sample forecast turns out to be superior in out sample forecast essentially because it has more case of crises.","PeriodicalId":445951,"journal":{"name":"ERN: Forecasting & Simulation (Prices) (Topic)","volume":"86 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2006-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"ERN: Forecasting & Simulation (Prices) (Topic)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.2139/ssrn.3019737","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Korean Abstract: 본 연구는 효율적인 외환위기예측을 위한 표본자료의 선택방안을 찾는 데 목적이 있다. 이를 위해 외환위기를 겪었던 24개국을 동아시아, 중남미, 유럽, 중동·아프리카 등 4개 지역으로 구분하고 다시 15개 소그룹으로 패널자료를 구성하였다. 15개 패널 자료별로 신호접근법에 따른 외환위기예측모형을 구축, 동아시아 외환위기기간을 표본외예측으로 하여 외환위기 예측성과를 평가하였다. 상식적 견해와 달리 동아시아를 제외하면 특정지역만을 대상으로 구축한 모형의 예측성과는 다른 지역을 포함한 모형보다 저조하였다. 한편 동아시아 지역만을 대상으로 구축한 예측모형은 표본외예측에서 잡음/신호비율이 작은 대신 신호회수가 적고 종합위기지수로부터 구한 위기확률이 낮았다. 그것뿐만 아니라 경보를 작동한 지표의 유형이 대부분 비금융부문에 속해 동아시아 외환위기의 공동요인으로 평가되는 취약한 금융에 대한 유용한 정보를 제공하지 못하였다. 확장된 표본기간에서 이들 모형은 지표의 최적임계치와 예측력에 큰 변화가 있고 유효지표가 뒤바뀌어 모형이 안정성 측면에서 취약하였으며 그 결과 표본외예측에서 제2종오류가 과다하게 발생하였다. 이상의 결과는 비록 표본 예측력이 떨어져도 광범위한 패널자료에 의존하는 예측모형의 표본외 예측력이 더 우수하다는 함의를 가진다.
English Abstract: The purpose of this study is to choose acceptable panel data for crisis prediction. According to common sense view it would be best to use panel data of East Asian countries when it comes to predict crises in these countries. Contrary to that view it is not, however. The paper considers 15 combinations of panel data. These panel data are composed by maximum four regional groups of 24 crisis-ridden countries such as East Asia, Latin America, Europe, and Middle East and Africa. And then the paper builds 15signal extraction models (SEM) based on each combination of panel data and assesses the predictability of currency crisis. SEM based on panel data composed of each individual region does not perform well except East Asia. East Asian countries, however, although it has the lowest noise signal ratio, has given least warning signals and the lowest probability of crisis associated with the crisis composite index. Furthermore, most indicators alarmed are non-financial, which fails to provide the useful information such that financial fragility is the common cause of the crisis. When sample period is extended there’s a big change in both indicator’s optimal threshold level and the noise signal ratio, and even effective indicators. As a result, all models of the panel data based on each individual region have serious type II error problem in out-sample forecast. The implication is that SEM based on broader panel data even though it should be inferior in in-sample forecast turns out to be superior in out sample forecast essentially because it has more case of crises.
选择有效预测危机的面板数据:基于信号访问模型(Crisis Prediction and Choice of Panel Data: The Case of Signal Extraction Model)
Korean Abstract:本研究的目的是寻找有效预测外汇危机的标本资料的选择方案。为此,将经历金融危机的24个国家分为东亚、中南美、欧洲、中东和非洲等4个地区,再由15个小组组成小组资料。按照15个小组的资料,建立了根据信号接近法的外汇危机预测模型,以东亚外汇危机期间为样本外预测,评价了外汇危机预测成果。与常识性见解不同,除了东亚以外,只以特定地区为对象构建的模型预测成果低于包括其他地区的模型。另外,只以东亚地区为对象构建的预测模型在标本外预测中,杂音/信号比率小,但信号次数少,从综合危机指数中求出的危机概率较低。不仅如此,启动警报的指标类型大部分属于非金融部门,没能提供被评价为东亚金融危机共同因素的脆弱金融的有用信息。在扩展的采样期间,这些模型的指标最佳临界值和预测能力有很大变化,有效指标颠倒,模型在稳定性方面脆弱,结果导致采样外预测出现过多的第二种误流。以上结果的含义是,虽然标本预测能力较差,但依赖广泛面板资料的预测模型的标本外预测能力更优秀。英语Abstract: The purpose of this study is to choose acceptable panel data for crisis prediction。According to common sense view it would be best to use panel data of East Asian countries when it comes to predict crises in these countries。Contrary to that view it is not, however。The paper considers 15 combinations of panel data。“These panel data are composed by maximum four regional groups of 24 crisis-ridden countries such as East Asia, Latin America, Europe, and Middle East and Africa”And then the paper builds 15signal extraction models (SEM) based on each combination of panel data And assesses the predictability of currency crisis。SEM based on panel data composed of each individual region does not perform well except East AsiaEast Asian countries, however, although it has the lowest noise signal ratio, has given least warning signals and the lowest probability of crisis associated with the crisis composite index。Furthermore, most indicators alarmed are non-financial, which fails to provide the useful information such that financial fragility is the common cause of the crisis。When sample period is extended there ' sa big change in both indicator ' s optimal threshold level and the noise signal ratio, and even effective indicators。As a result, all models of the panel data based on each individual region have serious II error problem in out-sample forecast。The implication is that SEM based on broader panel data even though it should be inferior in-sample forecast turns out to be superior in out sample forecast essentially because it has more case ofcrises。