We employ the Brock, Dechert and Scheinkman statistic to investigate n
on-linearity in conditional mean in UK real wages, employment and outp
ut, and to fit threshold regression models to the series. The latter p
rovide reasonable representations for the non-linearity in real wages
and output but the residuals from the threshold autoregressive model f
or employment still contain a non-linear component. In a forecast comp
arison the TAR models performed better than simple autoregressions. Us
ing the lagged share of wages as an error-correction term we find evid
ence that all three series respond asymmetrically to lagged changes in
wage share and other variables. Forecasts of real wages and employmen
t derived from these models were superior to those derived from standa
rd, symmetric, ECMs.