THEORETICAL BACKGROUND理論背景A wide range of prediction models can be found in literature review. These models use different approaches to evaluate company's performance and predict its failure. A brief chronological and methodological overview of the most important papers is presented below, describing researches that were pioneering (at least for the applied methodology) only at the time and those that are either still widely used or left a mark and provided incentives for other scientists.在文獻(xiàn)查閱中，可以查到各式各樣的預(yù)測(cè)模型。這些模型使用不同的方法評(píng)估公司績(jī)效并預(yù)測(cè)其破產(chǎn)。以下按照年代順序和分析方法分類，簡(jiǎn)要介紹一些最主要的理論。這些研究有的在當(dāng)時(shí)是開拓性的（至少應(yīng)用方法如此），但僅領(lǐng)一時(shí)風(fēng)騷；一些仍在被廣泛使用，或者轉(zhuǎn)瞬即逝，但激發(fā)了其他的科學(xué)家。The financial statements analysis dates to the 1920s, when F. Donaldson Brown joined Du Pont Company and in period from 1920 to 1924 developed a model for financial performance measurement. This deductive ratio, known as Du Pont Model analyzes the profitability of a company using traditional financial ratios. It integrates company's profitability and asset management, measuring its liquidity, leverage and efficiency. Return on investment is calculated by interacting different financial ratios from the aforementioned areas. The main strengths of Du Pont model are its simplicity and its linkage to compensation schemes. On the other hand, its main disadvantage is that is based on accounting numbers from balance sheet and income statement which are not reliable because of accrual concept. Du Pont Model remained the dominant form of financial analysis until the 1970s.財(cái)務(wù)決算分析始于20世紀(jì)20年代，F(xiàn).Donaldson Brown入主杜邦公司之后，在1920年到1924年之間建立了一個(gè)用于財(cái)務(wù)狀況評(píng)估的模型。這個(gè)被稱作“杜邦模型”的演繹比率，利用傳統(tǒng)的財(cái)務(wù)比率分析了公司的盈利性。它綜合了公司的盈利性和資產(chǎn)管理，測(cè)量它的流動(dòng)性、杠桿和效率。并通過(guò)交叉引用以上領(lǐng)域的相關(guān)財(cái)務(wù)比率計(jì)算出投資回報(bào)率。杜邦模型最大的優(yōu)點(diǎn)是它的簡(jiǎn)單和它與薪資規(guī)劃的聯(lián)結(jié)。另一方面，它最大的缺點(diǎn)是它是基于資產(chǎn)負(fù)債表和損益表中的會(huì)計(jì)數(shù)字的。而這些數(shù)字，由于利息觀念的存在，是不可靠的。杜邦模型直到20世紀(jì)70年代仍是財(cái)務(wù)分析的主流。One of the first researches in the area of ratio analysis and bankruptcy prediction, applying statistical methods was conducted by Beaver (1967) in 1966. He applied univariate analysis, using each of 14 accounting ratios separately, and found the difference in financial ratios structure between failed and non-failed companies for as long as five years prior to failure. He concluded that the cash flow to debt ratio was the best single ratio predictor, with a 78% accuracy in predicting bankruptcy. Discussants criticized his model because of the statistical methods used. 最早涉足比率分析和破產(chǎn)預(yù)測(cè)領(lǐng)域的一位研究者，用的是統(tǒng)計(jì)方法。由Beaver（1967）在1966年實(shí)施。他應(yīng)用了單變量分析，分別使用14個(gè)會(huì)計(jì)比率，從而至少在破產(chǎn)的五年之前，就找出了破產(chǎn)與未破產(chǎn)公司之間財(cái)務(wù)比率結(jié)構(gòu)的差別。他總結(jié)出，現(xiàn)金流與負(fù)債比率是最好的單比率預(yù)測(cè)器——在破產(chǎn)預(yù)測(cè)中有78%的正確率。而評(píng)論者則因?yàn)樗褂玫慕y(tǒng)計(jì)方法而批評(píng)他的模型。Altaian (Altaian and Hotchkiss, 2006) precluded this statistical limitation and applied multivariate discriminant analysis to construct his Z-score in 1968, probably the most famous model to the present day. The Zscore model claimed to predict bankruptcy correctly in 95% of the cases one year prior to defaulting, and in 83% of the cases two years in advance and rapidly decreased for longer time observations. However, in the 1980s the theorists, mainly econometricians, began to stress the statistical methodology applied as a major deficiency of the Zscore model. They claimed that linear discriminant analysis follows a normal distribution of financial ratios and equal structure of variance and covariance between paired good and bad companies, which is often not the case in practice. Altaian阿勒坦恩（Altaian阿勒坦恩和Hotchkiss霍奇基斯, 2006）排除了這種統(tǒng)計(jì)上的局限，應(yīng)用多變量判別分析，在1968年建立起他的Z-score理論——也許是現(xiàn)今最有名的模型。Z-score模型宣稱，其提前一年預(yù)知破產(chǎn)的準(zhǔn)確率在95%，提前兩年預(yù)知的正確率為83%，但在更長(zhǎng)時(shí)間觀測(cè)中，正確率會(huì)急劇下降。但是，在20世紀(jì)80年代，理論家們——主要是經(jīng)濟(jì)學(xué)家，將Z-score模型中的統(tǒng)計(jì)方法視為其重大缺陷。他們認(rèn)為在分析成對(duì)的優(yōu)秀與不良公司時(shí)，線性判別分析遵從財(cái)務(wù)比率的正態(tài)分布，以及方差和協(xié)方差的平衡結(jié)構(gòu)。但現(xiàn)實(shí)往往不是如此。