Help - coefficients have wrong signs!

[Pete]

Hi guys

I'm a struggling with an econometric model. I am trying to predict tenant demand for office space based on a selection of 5 lagged economic variables. The F-Stat is ok, and the P values are all less than 0.05. I also achieve a reasonable r2 of 0.83.

My issue is that 3 of my variables have coefficients with the "wrong sign". For example I use consumer confidence, business confidence and GDP data - all of which reduce office space demand as they get higher according the the regression, which makes no sense to me at all!

I am self taught so any help would be appreciated.

Many thanks

[trinker]

Sometimes our expectations are wrong. What if those three variables increase over seas job exportation or are a product of online sales that don't have office space needs or...

Could you tell us more? What are your IVs and DVs? What was your research question? I assume you did regression; if so what type of regression? Was this a time series? What is the t-stat and pvals on the weights?

[geosun]

If its just a problem with the correlations going in the wrong direction, then all you have to do to change that is fiddle with the + and - till it looks right.
but you are going to have to run the test again with the new equation you came up with.

[bryangoodrich]

I have to go with trinker. What is your model? Is there any prior research you're basing this on? In other words, is there already existing theory that suggests these coefficients should come out a certain way? If so, how did they model these variables? How are you representing the variables in your model? E.g., we talk about "inflation" in economics, but we have to model with something, and that choice is critical (do we use the CPI or some other index?). How are you representing "Demand for office space?" You also say you're using lag variables, so this is a time series model? I would be cautious about interpreting the r-squared, as they tend to be high in time series from what I recall. Time series is not my area, though.

[noetsi]

This might be useful

http://www.stat.columbia.edu/~gelman...wrong_sign.pdf

[Pete]

Thanks for your help so far. I must admit I am not an academic and work in a large real estate company in Australia that specialises in Office buildings.

I am looking to build a regression model that helps me predict office space demand (known as office space absorption, or changes in how much office space is occupied within a period).

The variables I am using are:

The ASX all ordinaries movement - a share index
Business conditions index level - an index produced by NAB one of Australia's banks
Business confidence index level - see above
Consumer condidence index level - an index produced by another bank Westpac
Australia's GDP movement - obtained from the Australian Bureau of Statistics

All data is lagged 6 months; for example in my data sheet Dec-97 Aus GDP is acutall Jun-97 GDP.

Because the data I'm using is lagged, I'm hoping to get a view of 6 months of future absorption by using this model.

Thanks again for your help everyone.

Pete

[noetsi]

Good luck. Be careful doing regression with lagged variables. It adds a variety of problems (of which I unfortunately am no expert; I just know the warnings that are commonly made about this).