HSBC Deploys Machine Learning Model to Forecast 10-Year Treasury Yield Direction at 65 Percent Accuracy

The model processes macroeconomic inputs including inflation prints, labor market data, and Fed balance sheet metrics to generate binary forecasts for yield changes. Backtested results show outperformance versus random walk benchmarks during the 2022-2023 tightening cycle, when yields rose from 1.5 percent to above 4 percent. Treasury market participants use the output to adjust duration exposure ahead of auctions and FOMC meetings.

US fiscal deficits above 6 percent of GDP have increased net Treasury supply, raising the incentive for foreign official holders to rebalance reserves. The model implicitly captures this supply pressure through yield curve steepness variables. Primary records from Treasury International Capital data confirm reduced Chinese and Japanese holdings as a share of outstanding debt since 2021, consistent with higher required yields to clear auctions.

Yield forecasts carry direct implications for dollar funding costs and the relative attractiveness of US assets versus euro or yen alternatives. Sustained model signals for higher yields would tighten the fiscal space available for defense outlays and sanctions enforcement programs that rely on dollar dominance.

Next data releases on core PCE and monthly Treasury refunding announcements will serve as live tests for the model's signal persistence through year-end.