Yesterday, I looked at the predictive powers of P/E and P/B on the indices.

I mentioned the stylistic facts of how P/B outperformed P/E and also that the 2 year window seemed to be the best one. I spent today redoing the thing in Matlab – Here is a visualization of those stylistic facts:

so far so good.

For today’s post I wanted to augment this analysis with momentum/mean reversion information and see if we could improve those r-squareds. Turns out we can:

In the chart below I added an “AR” (for autoregression) term to the regression model.

• For simplicity the AR period was set the same as the look-forward period,
• so for example for the AR factor on the HSI we used the cumulative return over the past 2 years to predict the cumulative return over the next 2 years.
• Obviously a ...