forecasting using regression: do i have to take a difference?

im using regression to forecast a variable. so i take my independent variables and i lag them by one or two months and then use that to predict next month’s values of my dependent variable. can someone plz tell me if i need to take a difference of both my indep and dep variables before doing the regression? or can i just leave the variables as is? i always thought i could just do a regression on the variables and then use my predicted equation to predict next month’s but someone is telling me you always have to take a difference first. thanks!

If the variables are I(1) you have to make sure they are I(0). You might want to pick up a statistics book that covers the time series topics. For financial data, this usually means you have to convert them to log differences/percent changes.

Just think about which you are modelling. If you differnce everything, you are saying that the change in month T depends on previous changes in months T-1, T-2, etc. If you don’t difference them you are saying the value in Month T depends on month T-1 and Month T-2. Both are reasonable in various situations.

A few thoughts in addition to what jmh and JoeyDVivre have said. It’s important to know whether your processes are stationary or not. You can think of a stationary process as a process that is at least bounded (for example, random walk is not stationary whereas shocks used in random walk are). Typically regression is used to express stationary processes as linear combinations of some other stationary processes. When you do that, all standard regression theory applies (use R^2 to assess goodness of fit, t-stats to assess significance of individual factors, etc). Under some conditions (read about co-integration), you can regress non-stationary processes using other non-stationary processes. You should not use regression in any other cases. Never regress a non-stationary process on a stationary (for example, regress stock prices on bond yields) or vice versa since there is no theoretical foundation for that.

thanks guys! jmh sorry what does I(1) and I(0) stand for?

jimjohn Wrote: ------------------------------------------------------- > thanks guys! jmh sorry what does I(1) and I(0) > stand for? http://en.wikipedia.org/wiki/Order_of_integration

They are economics terms related to comparing the Chinese and US economies. The first is pronounced “I won” and the second is pronounced “I owe”. hahahahahahahaha… maybe a little funny?

how about: I(1) - cyclops I(0) - blind