Example: Credit card 90+ delinquency (2-quarters ahead)

NY Fed HHDC credit-card transition into 90+ days late from bank card delinquency, the card rate, and consumer debt service at t−6 months

This model estimates Quarterly share of credit-card balances newly 90+ days late two quarters ahead (six months) using the New York Fed Household Debt and Credit report and FRED prints dated six months earlier. The New York Fed series is quarterly; the six-month horizon is two HHDC quarters.

• DRCCLACBS: commercial-bank credit-card delinquency rate for the quarter six months before the quarter being forecasted
• TERMCBCCALLNS: commercial-bank credit-card interest rate (all accounts) for that same month
• CDSP: consumer debt-service payments as a percent of disposable personal income for that same quarter

Our regression model is

CC90_estimated(t) = DRCCLACBS * 1.625873220 + TERMCBCCALLNS * 0.206456254 + CDSP * -0.054947388 + -2.072265988
and yields R^2 = 95.6% as of April 2026 (n = 90 quarters) in-sample on the latest paired actuals and estimates.

Coefficients are ordinary least squares on every complete lagged pair from 2004 through the latest actual, including the COVID and post-COVID years. That full-sample fit is a better statistical model of the printed series than one built only on pre-COVID data. A regression trained on the first 70% of paired samples (through 2019) and scored on the last 30% (2019–2026) does worse on that later window than predicting the holdout mean. The R² above is in-sample on the full window, including the period shown as a forecast; it is not an out-of-sample validation score.