Best analysis method for three interrelated longitudinal series I have data which consist of three longitudinal series of financial data.  The hypothesis is that two of the series are "caused" (in a loose sense) by the third.  It is fine to investigate this as two separate hypotheses.  All the series are about the relationship between two specific countries.  There are data for about 20 quarters on one of the series, and about 80 months on the other two series.
A little exploratory work reveals that none of the series are stationary, and that there is no apparent seasonality.  
So, how best to analysis these hypotheses?  Can I simply try regressions with different lags, or should I do a more formal time series analysis, or should I do something else?
Peter
 A: In a perfect world we could form a Vector Arima Model which would have two endogenous and three exogenous variables. The two equations would also include any necessary deterministic inputs such as Level Shifts , Local Time Trends , Seasonal Pulses and Pulses as necessary. Since you have relaxed to predict/analyse the two endogenous series simultaneously we can proceed with two Transfer Functions. These two Transfer Functions will be formed to fully utilize any contemporaneous or lag structure in each of the two potential causal series. Furthermore omitted stochastic structure will be proxied by including an ARIMA component. Additionally any and all omitted deterministic structure ( e.g. a law change ) will be proxied by employing Intervention Detection schemes to suggest Level Shifts/Local Time Trends/Seasonal Pulses ( e.g a June effect )/Pulses. Care would be taken to ensure that the parameters of said model did not vary over time or that non-constant error variance biased the results. A Transfer Function is often referred to as an ARMAX Nodel, consider  The series Y being predicted by series B and an error term E. 
