Formula
Starting log-odds z₀ = ln(Base ÷ (1 − Base)), where Base is the Portfolio base default rate as a decimal.
z = z₀ − 1.2 × (Debt service coverage − 1.3) + 0.25 × (Debt to EBITDA − 2.5) − 0.12 × (min(Years in business, 10) − 4) − 0.012 × (Guarantor credit score − 660) + 0.35 × NSFs in last 3 months + 0.4 × (Open financing positions − 1).
One-year probability of default = 1 ÷ (1 + e^−z). Relative to portfolio average = PD ÷ Base. Odds of default = 1 in (1 ÷ PD).
Risk grade: 1 below 4%, 2 from 4% to 8%, 3 from 8% to 15%, 4 from 15% to 25%, 5 at 25% or above.
Worked examples
Example 1: the default file is average on five factors and has one NSF
Inputs: Portfolio base default rate 10%, Debt service coverage 1.3, Debt to EBITDA 2.5, Years in business 4, Guarantor credit score 660, NSFs in last 3 months 1, Open financing positions 1. Starting log-odds: z₀ = ln(0.10 ÷ 0.90) = ln(0.1111) = −2.197. The scorecard centres each factor on an average file: coverage of 1.3, leverage of 2.5, four years, a 660 score and one position all produce a zero shift because each equals its centre. Coverage: −1.2 × (1.3 − 1.3) = 0. Leverage: 0.25 × (2.5 − 2.5) = 0. Tenure: −0.12 × (4 − 4) = 0. Credit: −0.012 × (660 − 660) = 0. NSFs: 0.35 × 1 = +0.35. Positions: 0.4 × (1 − 1) = 0. So z = −2.197 + 0.35 = −1.847. PD = 1 ÷ (1 + e^1.847) = 1 ÷ (1 + 6.341) = 0.1362, shown as 13.6%. That is grade 3, 1.36 times the portfolio average of 10 percent, and odds of 1 in 7.3. The whole uplift comes from the single returned item, which illustrates the weight: one NSF multiplies the odds of default by e^0.35, about 1.42.
Example 2: a weaker file across the board
Inputs: base rate 10%, coverage 1.1, debt to EBITDA 3.5, 2 years in business, score 630, 2 NSFs, 1 open position. z₀ = −2.197 as before. Coverage: −1.2 × (1.1 − 1.3) = −1.2 × −0.2 = +0.24. Leverage: 0.25 × (3.5 − 2.5) = +0.25. Tenure: −0.12 × (2 − 4) = +0.24. Credit: −0.012 × (630 − 660) = −0.012 × −30 = +0.36. NSFs: 0.35 × 2 = +0.70. Positions: 0.4 × (1 − 1) = 0. Sum of shifts = 0.24 + 0.25 + 0.24 + 0.36 + 0.70 = +1.79. z = −2.197 + 1.79 = −0.407. PD = 1 ÷ (1 + e^0.407) = 1 ÷ (1 + 1.502) = 0.400, shown as 40.0%. Grade 5, four times the portfolio average, odds of 1 in 2.5. No single factor is dramatic: coverage is a little thin, leverage a turn higher, the business a little younger, the score 30 points lower and two items returned rather than one. In log-odds they add; in probability they compound, and a file that looks only moderately weak on each line is four times as likely to default as the average.
Example 3: a strong file in a lower-risk book
Inputs: base rate 6%, coverage 2.0, debt to EBITDA 1.0, 12 years in business, score 760, 0 NSFs, 0 positions. z₀ = ln(0.06 ÷ 0.94) = ln(0.0638) = −2.752. Coverage: −1.2 × (2.0 − 1.3) = −0.84. Leverage: 0.25 × (1.0 − 2.5) = −0.375. Tenure: years are capped at 10, so −0.12 × (10 − 4) = −0.72. Credit: −0.012 × (760 − 660) = −1.2. NSFs: 0. Positions: 0.4 × (0 − 1) = −0.4. Sum of shifts = −3.535. z = −2.752 − 3.535 = −6.287. PD = 1 ÷ (1 + e^6.287) = 1 ÷ (1 + 537.3) = 0.0019, shown as 0.2%. Grade 1, 0.03 times the portfolio average, odds of 1 in 538. Two cautions apply. The credit weight dominates: 100 points above the centre contributes −1.2, as much as the other four favourable factors combined. And a PD of 0.2 percent is below what most small-business books ever observe; the scorecard extrapolates linearly in log-odds and should be floored in practice at a minimum PD the funder has actually seen, which is one of the reasons the weights are meant to be replaced with a calibration from your own data.