Direction of Change Predictability/ Probit Model

[fredrickt]

Hello,

I am currently doing my final year project and i am looking into directional change predictability of excess returns on the S+P500.

I have used a GARCH model using past excess returns to produce a set of conditional variance series as a measure of risk. However the variables in the models themselves show little forecastability for excess returns. Am i doing something wrong here?

My main problem however lies with the probit model. I generated my binary values in excel using the IF function before importing them into eViews(0 if negative 1 if positive), is this the correct method or should they be generated in Eviews? Using a variety of variables such as short/long interest rates, gdp, term structure i have then generated a probit model. All the resulting models seem to fluctuate around the 0.4 level, is this normal? i expected it to be more like 0.5.

I have attached my basic workfile if that may be of any use.

Thank you in advance for any advice given.

[trubador]

GARCH models deal with the variance part and operate on the residuals. Therefore they are not directly related to mean equation, which is used to generate the forecasts of the dependent variable.
There is nothing wrong with recoding the variable in Excel, though you do not need to (try @recode in EViews). Interpretation of probit models are not the same as traditional regression models. The results seem to fluctuate around 0.4 level, because it is more-or-less equal to the mean of your binary dependent variable. "Expectation-Prediction Evaluation" and "Goodness-of-Fit test" would be much more useful (go to Equation-->View).