"C.Rothe" <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>... > Hello, > > I could need some advice on the following problem: I have measurements of > certain similar economic variables in a fixed region over time (i.e. I have > unemployment > rates for different age groups in a country over the last 33 years). The > data set > also includes a number of explanatory variables (both economic and > socio-demographic). > > In the end, I want to be able to make statements like "Policy A will reduce > youth > unemployment, whereas Policy B affects all types of unemployment to the same > extend". > > Can anyone suggest a suitable class of models to approach this problem? > Linear > regression does not seem appropriate since I'm looking at time series data. > Somebody suggested to look at panel data analysis, but I'm not too sure > about > this either since the values of the explanatory variables would be the same > for all > groups. Does anyone have a better idea? > > Thanks > > Chris
Chris, You are quite correct regression , under the assumption of uncorrelated errors with constant mean and constant variance is probably an incorrect approach... but it is a start and then one can identify needed lags ( and leads ) of the user-suggested causal variables and an appropriate ARIMA augementation to proxy unspecified (omitted !) stochastic causal input series AND empirically identified Intervention Variables reflecting omitted Deterministic structure such as Pulses, Seasonal Pulses, Level Shifts and Local Time Trends . Additionally one can determine the need for power transformations of Y via Box-Cox tests and/or structural break points in the variance. There are approaches that deal directly with the model identification phase thus bypassing rounds and rounds of iterations in adding potentially needed structure and/or deleting unnecessary model structure thus achieving parsimony... these procedures are referred to as Transfer Function MOdel Identification ... AUTOBOX/FREEFORE does this automatically or you can suggest your own model ..Please feel free to download a Freeware version http://www.autobox.com/freef.exe or use TUCOWS or DOWNLOAD.COM The model that you are looking for is a TRansfer Function Model with Intervention Detection. If I can help, send me the data in a spread-sheet and I will send you back \an analysis using FreeFore .... all for free ! regards Dave Reilly Automatic Forecasting Systems 215-675-0652 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
