I am constructing a Vector autoregression model and I have used AIC to find how many lags I should use. Does 7 lags seem unreasonable? I am trying to find the impact the property market has had on the economy and my variables are GDP, inflation, employment rates, housing prices, number of housing complete, long-term interest rates and mortgage rates. How many lags makes economic sense?
$\begingroup$ I am assuming you are using monthly data, so 7 seems fine. But for robustness sake, you could look at other lag lengths. Maybe 1,2,3,6,12 and 7 due to best AIC... $\endgroup$– AkavallJun 8, 2012 at 13:32
For pure modeling, you can use the partial auto-correlation function to have a sense of how many past observations have an influence on the outcome. Note that this will not necessarily improve the forecast error, if your goal is prediction.
You can look into ARIMA models for more information on time-series analysis.