Should we really care about building business cycle coincident indexes!

Abstract

Quite often, the goal of the game when developing new coincident indexes of the economic activity is the comparison with NBER turning points. Using Monte Carlo simulations, this note illustrates that for the USA, any random linear combination of the four coincident variables would do the job as good as other more complicated methods.

Notes

However, Harding and Pagan (2002) recall that Burns and Mitchell advocated the use of several indicators mainly because a monthly GDP was not available at that time.

Camacho and Perez-Quiros (2000) compare the forecasting performance of different procedures.

That is to say, cycles found in the level (or log level) of coincident indexes.

Notice that for the SW index for the computation of I a contraction period of three months between April and July 1989 is excluded, as well as a false trough signal in May 1967 and a false peak signal in March 1979. Only once, in March 1975, the TCB ^w built gives a one month different turning point than the dating released by TCB. This could be due to the fact that no operation is added to the simple BB rule, that the definition of the growth rate by TCB is different than the difference of the logs taken or that TCB rounds its index after the first decimal. It also makes sense that TCB and

have the same turning points. Indeed, taking a simple average of the standardized series amounts to give weights proportional to their variability.