Where synthetic losses hide in charge-off math.

A synthetic identity produces a loss that is, on paper, indistinguishable from ordinary credit risk. That single fact is why the problem is hard, and it is an accounting fact before it is a fraud fact.

Two losses, one line item

Consider two accounts that both end in a charge-off. The first is a real member who lost a job and stopped paying. The second is a synthetic that was cultivated for two years and then drained in a week. Both land in the same place: an unpaid balance, aged past the threshold, written off against the allowance. The general ledger cannot tell them apart, because at the moment of loss they look identical. The difference is entirely in the history that preceded them, and history is not what the loss line records.

Why the ratios stay calm

The numbers a board watches are designed to summarize, and summarizing is exactly what hides this. A few points worth sitting with:

• Charge-off ratios are portfolio-level. A handful of cultivated synthetics inside a large, healthy book barely move a basis point. The signal is real but it is diluted below the resolution of the metric.
• Non-performing-asset ratios can look clean during cultivation. A cultivating synthetic is, by design, performing. It pays on time. It is not delinquent. It becomes non-performing only at the very end, briefly, on its way to charge-off.
• Vintage curves blur fraud into credit. When cultivated losses land months or years after origination, they smear across vintages and read as normal seasoning rather than a distinct population.
• Recoveries never come. Ordinary defaults produce some recovery over time. A synthetic has no one behind it to pursue, so recovery is near zero, but that shows up as a slightly worse loss-given-default, not as a flag that says "this borrower never existed."

The category error

None of this is an accounting failure. The books are correct. The error is one of category: a loss that was fraud is filed as credit risk, and once it is filed that way, every downstream process treats it as credit risk. Model tuning, reserve setting, underwriting adjustments, all of them absorb the synthetic loss as a small worsening of credit performance and adjust for it as if the answer were tighter underwriting. Tighter underwriting does not stop a borrower who is manufactured to pass it.

What actually separates them

If the loss line cannot distinguish the two accounts, something else has to, and the only thing that can is the behavior in between. A real member's history is messy and coherent. A synthetic's history is thin, optimized, and eventually shows credit appetite running ahead of any genuine economic base. That separation is not visible in a ratio. It is visible in a replay of the account's own timeline, which is a different exercise from watching the portfolio's summary statistics stay calm.

The practical takeaway is uncomfortable: a portfolio can look healthy on every number a synthetic is engineered not to disturb, and still be carrying a fraud book that will only ever announce itself as ordinary loss.