Loan Officers Know First: Net Tightening in the Senior Loan Officer Survey Forecasts US Business-Loan Growth Out of Sample, 1990–2026, with an Exact Clark–West Adjustment

Each quarter the Federal Reserve asks senior loan officers whether their banks have tightened or eased standards on commercial and industrial loans. I test whether the net percentage tightening for large and middle-market firms forecasts the growth of C&I loans at all US commercial banks over the following four quarters, using 146 quarterly surveys from 1990Q2 to 2026Q3 and the FRED loan series. Net tightening is strongly countercyclical: its correlation with loan growth over the four quarters ending six quarters later is minus 0.789, against plus 0.121 contemporaneously. Over 1992 to 2019 a one standard deviation rise in net tightening (21.6 points) predicts loan growth 6.4 points lower over the next year, with a HAC t statistic of minus 9.2, controlling for past loan growth; reported loan demand adds nothing (t 0.24). In recursive out-of-sample forecasts from 2003 to 2019 adding the survey cuts the mean squared error of an autoregressive benchmark by 71% (RMSE 9.14 to 4.91 points); the Clark–West statistic is 2.14. Through 2025 the cut is 66% and the Clark–West statistic 2.97. I prove the identity behind the Clark–West adjustment, which shows why a raw comparison of squared errors is biased against the larger model. When net tightening exceeds 20 points, loan growth over the next year has been negative 81% of the time, against 16% otherwise. The latest survey, for 2026Q3, shows net tightening of 0.0. A parallel test for credit-card standards finds no predictive power.