topic Poor Regression Model Predicts Correct in Statistical Procedures

Poor Regression Model Predicts Correct

Toni2 — Sun, 07 Mar 2021 19:36:52 GMT

Hi all,

I wonder if someone can help on this. I have run with OLS this model

y=1.0527x - 0.082

where

y= IOS (percentages)

x = Bank Rate (percentages)

When i plot y versus x and add the line the model doesn't fit the data due to low R2 and the outliers (please see the graph in the attached file)

However, when I see the back testing (yellow line in the graph in the attached line) the model almost perfectly fits the data which i am struggled to understand or explain

If the model is not correct how can correctly predict the data ?

My first thought is that since the Bank Rate is stable we might have violation of OLS assumptions but again how the back testing is almost correct?

Thanks,

Antonis

Re: Poor Regression Model Predicts Correct

PGStats — Sun, 07 Mar 2021 21:30:16 GMT

The R2 looks bad because you don't have enough contrast in your data (in the second half of the time series), which makes the error (the unexplained part of the signal) look large compared to the part of the signal explained by your model. You should be able to make a better assessment of the fit if you replot the second half of the time series and rescale the y axis.

Re: Poor Regression Model Predicts Correct

Toni2 — Mon, 08 Mar 2021 20:23:49 GMT

Hi, thank you for your response. I am going to have a look on this.

Re: Poor Regression Model Predicts Correct

gcjfernandez — Tue, 09 Mar 2021 00:01:24 GMT

Your data is autocorrelated (timeseries) with interruption (significant time shifts). This violate OLS assumptions. Treat this as a time series problem. Use PROC ARIMA or PROC ESM from SAS ETS.