I showed Bruce G. Posner's "The One-Minute Lender" (December 1984) to my friend, Luce S. A. Guse. He is one banker who still believes that "figures never lie, but liars sometimes figure."
Luce had the following advice for his fellow loan officer, the article's Titus A. Rock: Use the H-score circumspectly, for it has its pitfalls. For example, when computing the variables that involve logarithms (V7 and V9), just "tapping" the logarithmic function key may, depending on the calculator used, result in an incorrect value. Titus needs to make sure that he is computing the "log to the base 10," not the "natural log." Moreover, while most of the variables can be computed using financial statement data rounded to thousands or millions, variable V7 must be computed with tangible total assets expressed to the dollar. Luce also noted that variable V7 calls for "tangible total assets," whereas several other variables use just "total assets."
Yes, Luce can get picky; it comes from 30 years of scrutinizing business plans and loan agreements. So, the article's ragged use of terminology in describing how variable V3 is derived -- "divides by the sum of equity and retained earnings" -- irritated him a bit.
At the risk of sounding disrespectful to the academicians from Tennessee, Luce stated that he might not need multiple discriminant analysis to spot a weak company. After studying the article's sample financial data for a couple of minutes, Luce said that what he saw was a company with marginal profitability (somewhere from 0.4% to 0.7%), sluggish asset turns (sales being 1.1 times assets), and fairly hefty leverage (debt at three times equity).
In the end, Luce stated that he doubted that he would be using the H-score very extensively. He said that he has always felt that making loans was a lot like making love (to him both represent, among other things, an "entanglement"), and that one-minute lending would probably be a lot like one-minute loving . . . not very successful, or fulfilling.
INC. replies: We agree with Mr. Dougherty that there are some potential problems with statistical models. We invite him and other readers to read the article again. The "glib support" that concerns him was written with tongue in cheek.