hello,anyone who can give me a simple Regression model or equation on forecasting using both Cross-sectional Data and Time-series Data?the thing is that i can do the forecast based in time series analysis only, but hv no idea abt Cross-sectional Data, just wanna know the model structure... tks a lot!

Do you mean Panel Data or two step regressions a'la Fama + MacBeth?K

well, by means of regression analysis, the data used to forecast the futures price could be the factors like oil prices, USD index, spot prices, etc.not sure abt the data type or sth like u said, sorry abt that...maybe u d like to show me the most basic one for the form of forecast model

Simple would be a multivariate linear regression with dummies for seasonality (oil etc). don't foget to use and adjusted R^2.Alk

dont forget to check multi collinarity. For forcasting you should also look at update rules like adapting forecast errors.

QuoteOriginally posted by: GP03hello,anyone who can give me a simple Regression model or equation on forecasting using both Cross-sectional Data and Time-series Data?the thing is that i can do the forecast based in time series analysis only, but hv no idea abt Cross-sectional Data, just wanna know the model structure... tks a lot!Cross sectional models dont have the element of time in their data. For example you would like to have an equation relating the annual household income and the amount spent on a particular product. this regression would beC=a+b*I+eC and I are data collected from various households, of various income bruckets, not time series of data from the same household. HTH

GP03,Here is a simple example in the context of modeling long run home prices. Let's conjecture that there is a long run relationship between income and home prices. We have monthly data for the 50 US states from Jan 1980 to Dec 2008. The regression model is of the formP(t,i) = alpha + beta(i) Y(t,i) + e(t,i) for i = 1,2...50 and t = 1,..,348Note: if I am not mistaken, we have 29 years of data, so 29 X 12 = 348 monthly observations.Here are the two dimensions of the data:1. Time series - for every state we have a time series of home prices and state income2. Cross Section - for every point in time, t, we have home prices and state income across 50 states.To allow for the fact that there are differences across states, you may want to allow each state to have a separate intercept, i.e. the model isP(t,i) = alpha(1)*D(1) + ... + alpha(50)*D(50) + beta(i) Y(t,i) + e(t,i)In the above regression, we are allowing each state to have its own intercept and own slope parameter, so it boils down to OLS on each separate state. In a panel regression, we may want assume that beta is the same across states, so instead we use the modelP(t,i) = alpha(1)*D(1) + ... + alpha(50)*D(50) + beta Y(t,i) + e(t,i)i.e. we got rid of the subscript i on beta. This should get you started. As someone else mentioned, you may want to google for panel regression. There is a decent primer athttp://www.nyu.edu/its/pubs/connect/fall03/yaffee_primer.html

obeelde,thanks so much for ur good example and clear explanation!will go through it later step by step