Do economic variables improve bond return volatility forecasts?

- In-sample forecasting ability concentrates on short maturities and turbulent times.
- Out-of-sample performance is poor, but forecast combination improves a bit.
- Poor out-of-sample result is mainly due to overfitting.

This paper explores whether various economic variables improve monthly bond return volatility forecasts using the 1963-2012 data. In-sample analysis indicates that stock return or Federal Funds rate difference Granger causes bond volatility of all maturities. The forecasting ability of other variables mainly appears at the short end of the term structure or during the relatively turbulent time. Out-of-sample analysis suggests little evidence of forecast improvement, though forecast combination does improve the performance. Decomposing the out-of-sample forecasts indicates that the poor performance is primarily attributed to overfitting, and variable reduction by principal components does not change the results.