algorithm - Strategy for building best fit multiple regression model with time lagged variables -
i building multiple regression model - wrapped in function - 1 dependent variable , dozen independent variables. reason why building function need analysis approximately 75 different datasets.
the challenge independent variables correlate better dependent variable when lagged in time. unfortunately, not time lags same each variable , determine optimal mix of time lags each variable while getting optimum adjusted r^2 value multiple regression model. moreover, after building initial model try reduce model using step(modelbase, direction="both") function on model.
in approach have time lag independent variables same number of weeks. results in best possible model independent variables have same time lag, believe (with valid hypothesis supporting this) there better model out there when differ time lag each independent variable. question best strategy determine best fit model without making number of options huge. if want determine between 0 , 20 weeks time lag in weekly steps 12 independent variables trying find match between 4.096e+15 variables (=20^12).
i can imagine reducing problem following strategy: start finding best fit model 1 independent variable @ different time lags. second step add second independent variable different time lags , find best model 2 independent variables second tried @ different time lags while first kept constant. add third variable take similar approach second keeping first 2 variables constant , change try third different time lags. tells me strategy might decent approach, there might better overall model contains not optimal variables each individual independent variable.
is there shine light on how tackle challenge?
Comments
Post a Comment