Example: Bank real-estate loans (6-month ahead)

Commercial-bank real-estate loans from the 30-year mortgage rate, Case-Shiller, unemployment, the 10y–2y curve, the funds rate, and real income

This model estimates commercial-bank real-estate loans six months ahead from monthly FRED prints dated six months earlier. The funds rate and real disposable income are included with the four housing-market prints. Coefficients are fit through December 2023 and then frozen. MBA application indexes are not on FRED; real-estate loans outstanding are the public dollar series used here.

• MORTGAGE30US: 30-year fixed mortgage rate for the month six months before the month being forecasted
• CSUSHPISA: S&P/Case-Shiller U.S. National Home Price Index for that same month
• UNRATE: civilian unemployment rate for that same month
• T10Y2Y: 10-year minus 2-year Treasury yield spread for that same month
• FEDFUNDS: effective federal funds rate for that same month
• DSPIC96: real disposable personal income (chained 2017 dollars) for that same month

Our regression model is

REALLN_estimated(t) = MORTGAGE30US * 93.060225731 + CSUSHPISA * 3.143622861 + UNRATE * 14.001037225 + T10Y2Y * -74.098679548 + FEDFUNDS * -38.887437892 + DSPIC96 * 0.307747193 + -1276.362107599
and yields R^2 = 87.7% as of December 2023 (n = 240 months) in-sample on paired actuals through 2023Q4.
On later paired actuals the same frozen coefficients yield R^2 = -5.1% as of August 2026 (n = 32 months) out of sample.

Coefficients are ordinary least squares on complete lagged pairs from 2004 through 2023Q4. Later dates, including the period shown as a forecast, use those frozen coefficients. On the leftover paired actuals that frozen fit beats predicting the holdout mean. The first R² above is in-sample through 2023Q4; the second is out-of-sample on later paired actuals.