Statistical

models opposed to approach of Judgmental models is the one of statistical modelling,

which advocates superiority of quantitative data in establishing underlying

causal relationship between the probability of default and cause factors.

Statistical models are governed by statistical methods, considering many

factors simultaneously, thus calculating and analysing multivariate correlation

in order to identify most powerful factors and produce statistically derived

weights to be used in consequent scoring model.

Use of

statistical models in collection process advocates the emphasis placed on

inherent risk characteristic of the borrower as opposed to aging items. Why is

statistical model preferential to judgmental in such a case? Reason is that the

latter informs on quality of the risk separating lowest risk accounts from

highest ones, while statistical model quantifies the risk by informing you on

the probability of default and, therefore, associated expected loss, i.e. value

of the risk (Driving Internal Collection Results With Statistical-based Credit

Scoring, 2010).

One of the

most important and prominent contributions made to the field was the

observation by Beaver (1967) that there are several financial ratios, which

differ significantly between failed and nonfailed firms, in particular cash

flow/net worth and debt/net worth (Falkenstein, Boral, & Carty,

RiskCalc(TM) For Private Companies: Moody’s Default Model, 2000). In short,

differences in such ratios for viable and bankrupt companies increase as time

to default shortens – as failure neared, firms became more dissimilar.

One step

further was taken by Altman (1968), who reasoned it optimal to create a

multivariate model with several such important explanatory variables that had

low correlation as compared to a single variable model in predicting solvency

of a firm. Quantitative data does not do the work if is used as individual

ratios, as that would constrain the analyst to compare the stats sequentially

and likely end up with ambiguous result. Multivariate model is therefore useful

tool in this field. Altman was the first one to use a statistical model to

predict default probabilities of companies, estimating the Z-score model

through the application of discriminant analysis, which subsequently underwent

several refinements and became a benchmark in academic literature. Although

popular, Z-score failed to establish itself as the ultimate tool in determining

credit quality of a borrower. Reason is that the model does not perform well,

scoring roughly equivalent to a simple univariate ratio benchmark as

liabilities/assets (Falkenstein, Boral, & Carty, RiskCalc(TM) For Private

Companies: Moody’s Default Model, 2000).