Why MAE and RMSE are the same

[chtsy]

Hello all,

I am doing recursive window to forecast stock return. After obtain predictions, I use RMSE, MAE, MAPE, Theil to evaluate the performance.

But I found that the results of MAE and RMSE are totally the same, and some of Theil value is 1.

I do not know whether these are correct? Could anyone can help me check it.

Code: Select all

%y = "stock" 'y is stock return %x1 = "div2" 'x1 is dividend yield %x2 = "sir" 'x2 is short-term interest rate %x3="ipg" 'x3 is industrial production growth 'stock regression 1 step ahead for !i=0 to 124-1 'estimate the regression 124 times smpl 1980m2 2004m12+!i 'recursive-sample starts at same place 'sample is extended by 1 obs each time equation stkreg1s{!i}.ls {%y} c {%x1}(-1) {%x2}(-1) {%x3}(-1) stkreg1s{!i}.makeresid stkreg1sres smpl 2005m1+!i 2015m04 'set forecast sample stkreg1s{!i}.forecast(f=na) tmp_stkreg1sfcst series stkreg1sfcst stkreg1sfcst = tmp_stkreg1sfcst next 'forecast evaluation %y="stock" scalar mape = @mape(stock, stkreg1sfcst) scalar mae = @mae(stock, stkreg1sfcst) scalar theil = @theil(stock, stkreg1sfcst) scalar RMSE = @rmse(stock, stkreg1sfcst)

By the way, which one of the following two is correct?
scalar RMSE = @rmse(y, yfcst)
scalar RMSE = @rmse(yfcst, y)

Thank you very much!

[EViews Gareth]

You are computing the forecast evaluation statistics over a single value. That's what causes them to be identical.

For RMSE it doesn't matter which order you put the series in.

[chtsy]

Thanks for your reply.

This is a recursive window forecasting. If I want to forecast 1 step ahead from 2005m01 to 2015m12, then how to modify the program?

Will the forecast in menu have the same result of programming?

Besides that, I use the data from prediction and compare it with actual data and computing RMSE and MAE manually, the result is totally different with the one programming generated. Driven crazy...

Thanks again.

[EViews Gareth]

Change the sample before calculating the evaluation statistics.

[chtsy]

The result I obtained from the programming is a little strange, can I do the forecast from the menu after finish the regression. It is much simpler and the result seems to be more satisfied.

[EViews Gareth]

If you want to.