I wish there was any set of indicators or regression equations that would be "try and true", I know of none. The only "seat of the pants" is the time variability (weighed toward more recent history, rather then multi year historical norms) of the correlation coefficients.