1、 外文翻译 原文 Title: Financial Ratios and the Probabilistic Prediction of Bankruptcy Material Source: http:/www.jstor.org/pss/2490395 Author:James Ohlson 1 Introduction This paper presents some empirical results of a study predicting corporate failure as evidenced by the event of bankruptcy. There have b
2、een a fair number of previous studies in this field of research; the more notable published contributions are Beaver (1966; 1968a; 1968b), Altman (1968; 1973) and so on. Two unpublished papers by White and Turnbuli (1975a; 1975b) and a paper by Santomero and Vinso (1977) are of particular interest a
3、s they appear to be the first studies which logically and systematically develop probabilistic estimates of failure. The present study is similar to the latter studies, in that the methodology is one of maximum likelihood estimation of the so-called conditional logit model. The data set used in this
4、 study is from the seventies (1970-76). I know of only three corporate failure research studies which have examined data from this period. One is a limited study by Altman and McGough (1974)in which only failed firms were drawn from the period 1970-73 and only one type of classification error (miscl
5、assification of failed firms) was analyzed. Moyer (1977) considered the period 1965-75, but the sample of bankrupt firms was unusually small (twenty-seven firms). The third study, by Altman, Haldeman, and Narayanan (1977), which “updates“ the original Altman (1968)study, basically considers data fro
6、m the period 1969 to 1975. Their sample was based on fifty-three failed firms and about the same number of nonfailed firms. In contrast, my study relies on observations from 105 bankrupt firms and 2,058 nonbankrupt firms. Although the other three studies differ from the present one so far as methodo
7、logy and objectives are concerned, it is, nevertheless, interesting and useful to compare their results with those presented in this paper. Another distinguishing feature of the present study which I should stress is that, contrary to almost all previous studies, the data for failed firms were not d
8、erived from Moodys Manual. The data were obtained instead from 10-K financial statements as reported at the time. This procedure has one important advantage: the reports indicate at what point in time they were released to the public, and one can therefore check whether the company entered bankruptc
9、y prior to or after the date of release. Previous studies have not explicitly considered this timing issue. 2 Some Comments Regarding Methodology and Data Collection The fundamental estimation problem can be reduced simply to the following statement: given that a firm belongs to some prespecified po
10、pulation, what is the probability that the firm fails within some prespecified time period? No assumptions have to be made regarding prior probabilities of bankruptcy and/or the distribution of predictors. These are the major advantages. The statistical significance of the different predictors are o
11、btained from asymptotic (large sample) theory. Clearly, much can be gained by improving the data base. The evaluation of the predictive classification power of a model should be more realistic, and, more important here, the same should apply for standard tests of statistical significance. This is no
12、t to suggest that it is important to have “super accurate“ data for purposes of developing (as opposed to evaluating) a discriminatory device. It might well be that the predictive quality of any model is reasonably robust across a variety of datagathering and estimating procedures. 3 Collection of F
13、inancial Statement Data The next phase was one of actually collecting financial data for the bankrupt firms. The objective was to obtain three years of data prior to the date of bankruptcy. Each report had to include the balance sheet, income statement, funds statement, and the accountants report. I
14、n case the last available accountants report explicitly stated that the company had filed for bankruptcy, then a fourth report was collected. All reports were retrieved from the Stanford University Business School Library, which has an extensive microfilm file of 10-K reports. The relevant parts of
15、the 10-K reports were photocopied and subsequently coded. Some firms had to be deleted from the sample because no report whatsoever was available, but these were few. Other firms, again very few, were deleted because they were corporate shells and had no sales. On the other hand, no firm was deleted
16、 because of its young (exchange) age, and some firms had only one set of reports. The final sample was made up of 105 bankrupt firms. I noted that while eighteen of the 105 firms (17 percent) had accountants reports which disclosed that the company had entered bankruptcy, the fiscal year-end was pri
17、or to the date of bankruptcy. These reports were deleted and reports from the previous fiscal year were substituted. 4 A Probabilistic Model of Bankruptcy Let X, denote a vector of predictors for the i th observation; be a vector of unknown parameters, and let P(X, p) denote the probability of bankr
18、uptcy for any given X, and . P is some probability function, 0 P 1. The logarithm of the likelihood of any specific outcomes, as reflected by the binary sample space of bankruptcy versus nonbankruptcy, L( )= log(X, )+ log(1-P(X, ) is then given by: where S1, is the (index) set of bankrupt firms and
19、S2 is the set of nonbankrupt firms. For any specified function P, the maximum likelihood estimates of 1， 2 , are obtained by solving: max l( ) In the absence of a positive theory of bankruptcy, there is no easy solution to the problem of selecting an appropriate class of functions P.As a practical m
20、atter, all one can do is to choose on the basis of computational and interpretative simplicity. 5 Ratios and Basic Results For purposes of the present report, no attempt was made to develop any new or exotic“ ratios. The criterion for choosing among different predictors was simplicity. (1).SIZE = lo
21、g(total assets/GNP price-level index). The index assumes a base value of 100 for 1968. Total assets are as reported in dollars. The index year is as of the year prior to the year of the balance sheet date. The procedure assures a real-time implementation of the model. The log transform has an import
22、ant implication. Suppose two firms, A and B, have a balance sheet date in the same year, then the sign of PA PB is independent of the price-level index. (This will not follow unless the log transform is applied.) The latter is, of course, a desirable property. (2).TLTA = Total liabilities divided by
23、 total assets. (3).CLCA = Current liabilities divided by current assets. (4). OENEG = One if total liabilities exceeds total assets, zero otherwise. (5). NITA = Net income divided by total assets. (6). FUTL = Funds provided by operations divided by total liabilities. After a series of research, this
24、 paper analyzes the choice and dividing point two categories, the relationship between the afr with dividing point numerical increases, the bankrupt company discriminant accuracy lower, with the bankruptcy of the company discriminant rate is higher and higher. By 50% as dividing point in bankruptcy,
25、 model to the bankrupt company year before the discriminant 99.37%, but for correctness criterion accuracy is bankrupt company, if by 6% 32.4% as dividing point (in its samples, bankrupt company proportion of 105 / (105 + 2058) = 4.85%, according to the information provided 6% were the authors provi
26、de the most close to 4.85% a segmentation point), The corresponding to the bankrupt company discriminant 88.2% of the accuracy and correctness of the corresponding discriminant insolvent corporation is 80%. 6 Conclusions There are two conclusions which should be restated. First, the predictive power
27、 of any model depends upon when the information (financial report) is assumed to be available. Some previous studies have not been careful in this regard. Second, the predictive powers of linear transforms of a vector of ratios seem to be robust across (large sample) estimation procedures. Hence, mo
28、re than anything else, significant improvement probably requires additional predictors. 译文 应用财务比率预警财务危机 资料来源 : http:/www.jstor.org/pss/2490395 作者： 奥尔逊 1 引言 本文通过 对 财务危机 进行防范 ，但最终仍导致破产的公司实例，从实证上得到一些分析结果。在这个领域，之前已经有相当一部分研究，更值得注意并已经发表文章有 Beaver（ 1966; 1968a; 1986b） , Altman（ 1968; 1973） 等等。 White 和 Turn
29、bull（ 1975a; 1975b） 和 Santomero和 Vinso（ 1977） 写的两篇未发表的论文，让我们看起来他们似乎是第一次系统而有逻辑性地发展了预测财务危机，本文类似于后面这篇文章的研究，采用的这种方法是利用 logistic模型，最大可能地预测财务危机。 该数据集应用于该研究是在 70年代（ 1970-1976） ，在这个时期，我知道的只有三个公司进行过数据检查。一个是由 Altman 和 McGough（ 1974） 创作的有限制性的研究，文中只有失败的公司是来自 1970-1973，而且只有一种分类错误（ 失败公司的错误分类）进行了分析。 Moyer（ 1977） 认
30、为在 1965-1975，破产公司的例子不同寻常地很小（ 27个公司）。三分之一的研究，是由 Altman, Haldeman, and Narayanan（ 1977） ,更新了 Altman（ 1968） 最初的研究，基本上的数据都是从 1969-1975来的。他们的样本引用了 53家破产公司和相同数量的绩效优异的公司。而对比之下，本文的 研究是通过观察 105家破产公司和 2058家没有破产的公司。尽管就方法和客观性而言，其他的三项研究不同于目前的这一项，确实如此，但是，在本文中，很有意义并且很有实际效用地用现在的数据比较了他们的结果。 本文研究的另一个显著的特点我应该强调的是，与以前几
31、乎所有的研究相反的是失败公司的数据并不是来自穆迪手册，这些数据是间接地从 10-K 财务报表上得到的。这个过程有一个很重要的优点：这些报告及时地显示了他们向公众开放的重点，因此可以检查这个公司是在公开财务报表之前还是之后产生破产。之前的研究并没有确切地考虑时间上的问题。 2 一些关于方法和数据收集的评论 基本的估计问题可以被简化到以下状态：由于公司属于一些指定的人员，那么公司在特定的时期发生财务失败的可能性是多少呢？之前没有假设关于公司破产的可能性或者发生财务失败的概率分布。这些都是主要优势。不同预警系统的数据意义在于数据都是从渐进的理论（大样本）中得到的。 显然，提高数据库，加大数据库的 容
32、量能收获很多。对于预测模型的评估应该更加现实一点，而且，更重要的是，在这里，同样地应该应用到具有统计学意义的标准测试中。这并不是说，为了发展一项有识别能力的装置而要以拥有“超级精确”的数据为宗旨（相对于评估）。他可能意味着，财务预警模型通过一系列的数据收集和预测过程会很合理地变得更加强大。 3 财务数据的收集 下一阶段的其中一个任务是为破产公司收集财务数据。我们的目标是获得公司破产前 3年的数据。每一份报告都必须包括资产负债表、利润表、现金流量表和会计人员的报告。如果在会计师最后的有效的报告中明确声明该公司 已申请破产，那么第四份报告应该被收集。 AU 报告是来自斯坦福大学商学院图书馆里， 里
33、面有一个 容量很大、涵盖面很广 的缩微胶片 10-K 文件的报告 。相关部分被复印并且随后编码。一些公司因为没有报告可以使用而从这个样本中被删除，但是这些也很少。少数的一些其他的公司因为共同销售但是没有销售而被删除。另外一方面，没有公司由于还在起步阶段或者只有一系列的报告而被删除。 最后的样本是由破产的 105家公司组成的。我注意到 105家中的 18家公司（ 17%）的会计人员透露公司已经进入破产，是在破产日期前的一个财政年度。这些报告被删除了 ，而且被以前年度的报告替代了。 4 一种财务危机预警模型 让 X，指示一个预测 i 的向量，是一个代表未知参数的向量，然后让 P（ X，）指向财务危
34、机的可能性， P是预测的功能， 0 P 1.对数的一些可能性的精确的结果，是通过破产和非破产的两两对应的公司反应出来。 L( )= log(X, )+ log(1-P(X, ) S1代表破产的公司， S2代表没有破产的公司，作为有明确功能的 P， 1， 2用来表示最大可能的预测结果，即 max l( ) 如果 缺乏积极的破产理论，对于选取 P 的合适的功能没有简单的方法。作为一个 实际的问题，所有能做的就是选择基础的计算和简单释义。 5 比率和主要的结果 为了目前的这篇报告，不再尝试任何“新的或者特别”的比率。选择不同预测因子的标准是尽可能简单化。 （ 1） SIZE=log（总资产 /国民生
35、产总值 -平衡指标）。该指数呈为基值 100为 1968。总资产用美元作为单位。该指数的一年，是在去年的前一年资产负债表的日期。这个过程保证模型实时运行。对数的转变有一个重要的实施。假设两家公司， A和 B，在同一年有资产负债表日，那么 P（ A） -P（ B）的标志独立于价格平衡指标。（这个不会跟随除非对数改变被应用 了。）后者是，当然，一些渴望得到的财产。 （ 2） TLTA=总负债 /总资产 （ 3） CLCA=流动负债 /流动资产 （ 4） OENEG=总负债超过总资产，其他就是 0 （ 5） NITA=净收益 /总资产 （ 6） FUTL=由操作者提供的资金 /总负债 经过一系列的研
36、究，分析分割点的选择和两类误判率之间的关系，随着分割点数值的增大，对破产公司的判别正确率越来越低，而对非破产公司的判别率则越来越高。以 50%作为分割点，模型在破产前一年对非破产公司的判别正确率 99.37%，但是对破产公司的判别正确率是 32.4%，如果以 6%作为分割 点 (在其样本中，破产公司所占比例为 105/(105+2058)=4.85%，根据所提供的资料，6%是作者提供的最接近于 4.85%的一个分割点 )，对破产公司的相应判别正确率为 88.2%，对非破产公司的相应判别正确率则为 80%。 6 结论 有两个结论应该重申。采用任何一种财务危机预警模型取决与信息（财务报告）的有效性。此前的研究并没有仔细地做到这一点。其次，预测能力的线性变换向量比率（大样本）通过预测步骤看起来是有效的。因此，更大的进步需要另外的预测，才能逐步改善现状。
