A macroeconomic factor model names its factors in advance and takes them from the economy. Chen, Roll and Ross set the pattern in 1986 with a small set of variables, including industrial production growth, unanticipated inflation, a term-structure spread and a default spread. The factors are observable and the sensitivities are estimated by time-series regression, one asset at a time.
Why do macroeconomic factors enter as surprises?
The discipline that makes the model work is the one written into the return-generating process: each factor enters as a surprise, defined as the realised value minus the value that was expected. If GDP growth was forecast at 2.4% and came in at 3.1%, the factor value for that period is +0.7, not 3.1. Use the level and you double-count, because the expected level is already priced into E(R_i). This is the commonest error in the region and it is invisible in the output. Substituting levels for surprises on the trucking company in Worked Example 2 gives 14.18% instead of 12.86%, and neither number looks wrong on a page.
Advertisement
Once the surprises are in hand, the decomposition is arithmetic.
A synthetic US trucking company with an expected return of 9.4%, a GDP sensitivity of 1.6 and an inflation sensitivity of -0.9 meets a +0.7 GDP surprise and a -0.6 inflation surprise. Its factor contribution is 1.6(0.7) + (-0.9)(-0.6) = +1.66 points, split +1.12 from growth and +0.54 from inflation, and a firm-specific event worth +1.8 points lifts the realised return to 12.86%.
Two properties of that split are worth holding on to. The expected component is a forecast, made before the period, and it does not move when the period turns out badly. The factor component is not a judgement at all, only exposures multiplied by surprises. And the firm-specific component is the residual, which means it absorbs every real company event and also every deficiency in the factor list. A model missing an oil-price factor will report the oil move as skill.
What are the strengths and weaknesses of macroeconomic models?
Macroeconomic models have one large advantage and one large weakness. The advantage is interpretability: an exposure of 1.6 to GDP growth is a sentence a risk committee can act on. The weakness is estimation. The sensitivities come from a time-series regression on quarterly or monthly macro data, so a five-year window gives sixty monthly observations at best, the macro series are themselves revised after the fact, and the estimated b values move a lot from window to window.