Skip to content
Alternative Finance

Underwriting · Lenders and funders

Loan default probability calculator

This loan default probability calculator applies a logistic scorecard: it converts your portfolio's base default rate to log-odds, shifts that figure by a fixed weight for each of six factors measured against an average file, and converts the result back to a one-year probability of default. It returns the PD, a risk grade from 1 to 5, the multiple of your portfolio average and the odds. It is for funders and lenders who want a calibratable PD rather than a points score. Watch the base rate: every output scales from it, so it must be the observed one-year default rate of an average file in your own book.

One-year probability of default
13.6%
Risk grade
3
Relative to portfolio average
1.36x
Odds of default
1 in 7.3

Logistic model: log-odds start at the portfolio base rate and shift by fixed weights per factor (coverage −1.2 per 1.0x, leverage +0.25 per turn, tenure −0.12 per year to 10, credit −0.012 per point, NSF +0.35 each, positions +0.4 each). Replace the weights with your own calibration; the structure is standard.

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.

How the logistic scorecard works

A logistic model predicts a probability by working in log-odds. Odds are probability divided by one minus probability, so a 10 percent default rate is odds of 1 to 9, and log-odds are the natural logarithm of that, −2.197. The model adds a weighted contribution from each factor to the starting log-odds and then converts back to a probability with the logistic function 1 ÷ (1 + e^−z), which always returns a value between 0 and 1. The Wikipedia article on logistic regression covers the mathematics; what matters for an underwriter is that factors add in log-odds and multiply in odds, so each weight is the log of an odds multiplier. A weight of 0.35 for one NSF means each returned item multiplies the odds of default by e^0.35, about 1.42; a weight of −1.2 per 1.0x of coverage means a full turn of extra coverage divides the odds by e^1.2, about 3.3.

This loan default probability calculator centres every factor on an average file: coverage 1.3, debt to EBITDA 2.5, four years in business, a 660 score and one open position. A file that matches the centre on every factor, with zero NSFs, gets exactly the base rate. That design makes the base rate the anchor and the weights the shape, and it means the scorecard can be dropped onto any book by changing one number, provided the book's average file looks something like the centre. If your average file has two positions and a 620 score, re-centre the weights before trusting the output.

Weights exactly as coded
FactorCentreWeight per unitOdds multiplier per unitDirection
Debt service coverage1.3−1.2 per 1.0x÷ 3.32 per 1.0x of extra coverageHigher coverage lowers PD
Debt to EBITDA2.5+0.25 per turn× 1.28 per turnHigher leverage raises PD
Years in business (capped at 10)4−0.12 per year÷ 1.13 per yearTenure lowers PD until year 10
Guarantor credit score660−0.012 per point÷ 1.13 per 10 pointsHigher score lowers PD
NSFs in last 3 months0+0.35 each× 1.42 eachEach return raises PD
Open financing positions1+0.4 each× 1.49 eachEach position raises PD

Reading the PD, grade and odds

The one-year PD is the probability that this file defaults within twelve months, on the model's assumptions. The grade bands are fixed cut-offs at 4, 8, 15 and 25 percent and exist so that a PD can be mapped to a pricing tier or a policy action without re-reading the decimal. The relative figure, PD divided by base, expresses the file in units of your average: a 1.36x file is a bit worse than average, a 0.03x file is far better, and a 4x file is one you would normally decline or price at the top of the sheet. The odds presentation, 1 in 7.3, is often the easiest to explain to a credit committee or a broker.

Two properties of the logistic form shape interpretation. First, the same shift in log-odds moves the probability more in the middle of the range than at the extremes: adding 0.35 to a 10 percent base takes it to 13.6 percent, but adding 0.35 to a 0.2 percent PD takes it only to 0.3 percent. Second, extreme inputs produce extreme outputs: a file with zero coverage and four positions returns a PD near 100 percent, which is right in spirit and useless in practice. Cap the model's output at the highest and lowest default rates your book has actually produced, and treat anything beyond them as a decline or a grade 1 rather than a precise number.

Setting the base rate

The base rate is the observed one-year default rate of an average file in your portfolio, which is not the same as the portfolio's overall default rate unless your book is evenly spread. Compute it from a seasoned vintage: of files funded 12 to 24 months ago that looked average at origination, what share defaulted within a year? For a new funder without history, start from the default rate of the product and segment you intend to serve and revise as data accumulates. The Federal Reserve's charge-off and delinquency release gives bank commercial and industrial loan rates as a floor; alternative-finance books run several times higher, and the BLS Business Employment Dynamics establishment survival data give a sense of how often young businesses close regardless of credit.

Define default before measuring it. For a daily-debit product it is typically a stated number of consecutive missed debits or a balance 90 days past due; for a term loan, 90 days past due or non-accrual. Whatever the definition, it must match the one used in the loan loss reserve calculator and the expected credit loss calculator, where the PD from this page becomes the probability input, so that the reserve and the pricing are built on the same event.

Calibrating the weights to your own book

The six weights are illustrative. Their signs and rough magnitudes reflect what small-business credit models consistently find: behavioural factors such as returned items and open positions carry large weights, personal credit carries a steady weight per point, and tenure matters most in the early years, which is why the tool caps it at ten. But the exact numbers should come from your data. The standard method is to fit a logistic regression of the default indicator on the six factors using your own funded files, then replace the weights with the fitted coefficients and the base rate with the fitted intercept. A few hundred outcomes, including a few dozen defaults, is enough for a first fit; the underwriting decision matrix builder is the lighter-weight alternative for funders who are not yet there.

  1. Assemble funded files with the six factors as measured at origination and a default flag at 12 months.
  2. Fit a logistic regression; check that each coefficient has the expected sign and that the model separates defaults from non-defaults on a holdout sample.
  3. Replace the weights and base rate in your policy document with the fitted values, and record the date and sample.
  4. Back-test quarterly: group new files by predicted grade and compare predicted with observed default rates.
  5. Re-fit when the observed rate in any grade drifts materially from the prediction, or when the product or market changes.

Banks follow formal model risk management guidance for this process, published by the Federal Reserve and the OCC, which emphasises documentation, independent validation and ongoing monitoring. A non-bank funder is not bound by it but benefits from the same discipline, particularly when the model is used to set prices that fair-lending rules under Regulation B, published by the CFPB, require to be applied consistently to business applicants.

Where the PD goes next

A PD on its own is a ranking. Multiplied by loss given default and exposure it becomes an expected loss, which is the loss premium in a price and the per-loan contribution to a reserve. The risk-based pricing calculator takes the PD directly and builds a rate from cost of funds, expected loss, operating cost and a capital charge; a file at 13.6 percent PD and 55 percent LGD carries a 7.5 percent annual loss premium, while the 40 percent file in example two carries 22 percent and is usually uneconomic at any price the merchant would accept. The grade can also be mapped to the pricing tiers in the merchant risk score calculator, which uses a points approach on overlapping inputs and should agree with this model in rank order if both are sensible.

The inputs themselves come from the rest of the underwriting workflow. Coverage after the proposed loan is produced by the business loan underwriting calculator or the DSCR calculator; debt to EBITDA requires the same EBITDA figure that the EBITDA and cash flow underwriting guide explains how to construct; NSFs come from the NSF transaction analyzer and positions from the statements and a UCC search. The guarantor's score must come from a consumer report pulled with authorization; owners can review their own files through AnnualCreditReport.com, which is useful when a score is disputed.

Questions people ask

What is a good probability of default for a small business loan?

It depends on the product and the book. The calculator's grade bands treat under 4 percent as the lowest-risk grade, 4 to 8 percent as strong, 8 to 15 percent as average for many alternative-finance books, 15 to 25 percent as weak, and 25 percent or more as the highest-risk grade. Bank commercial loan default rates are typically far lower than any of these; short-term unsecured products run higher.

Why does this loan default probability calculator start from a base rate?

Because a logistic scorecard predicts how much a file differs from average, not the absolute level. The base rate anchors the model to your portfolio's observed one-year default rate for an average file, and the six weights shift the log-odds away from it. Changing the base rate rescales every output, so it should be measured from your own seasoned vintages rather than guessed.

How much does one NSF change the probability of default?

Each returned item adds 0.35 to the log-odds, which multiplies the odds of default by about 1.42. On a 10 percent base rate that takes PD from 10 percent to 13.6 percent; a second NSF takes it to 18.3 percent. The weight is deliberately large because returned items are the most direct statement evidence that the account cannot cover its debits.

Why is tenure capped at ten years?

Because the protective effect of age flattens out: the difference between a two-year-old and a six-year-old business is large, while the difference between a twelve-year-old and a twenty-year-old business is not. The cap stops the model from assigning ever-lower PDs to very old businesses on tenure alone. Establishment survival data from the BLS show the same pattern.

Can I use the PD to set a reserve?

Yes. Multiply the PD by loss given default and the exposure to get the expected one-year loss for a file, and sum across the book or by tier. The loan loss reserve calculator does this for three tiers, and the expected credit loss calculator extends it over the remaining life of multi-year loans with discounting, which is what the CECL accounting standard requires.

Sources and further reading

Related tools and guides

More underwriting

Updated October 12, 2026 · By Alternative Finance editorial team

Next step

Turn the estimate into real offers

One request, several funding partners, side-by-side payback. Nothing is binding until you accept an offer.

  • Soft inquiry to request offers; a hard pull only if you proceed with a funder.
  • Decisions in as little as 24 hours for many products; funding often within days.
  • Use our calculators to check the true cost before you accept anything.

See what you may qualify for

Takes about a minute. Requesting offers does not affect your credit score. Free, no obligation.

We are not a lender or broker. Requests are forwarded to funding partners free of charge. How the site is paid.