In-sample/Out-of-sample Forecasts

[KevPei]

Reading 13 Time series analysis - 4.5 Comparing Forecast MOdel Performance
With regards to in/out of sample forecasts, my understanding is: Out of sample forecasts are used to actually predict future values of dependent varables; while in sample forecasts are forecasts within the range of the actual data used to build the AR model. So it seems to me that for in sample forecasts, the AR model is forecasting something that already happened. This seems pretty useless to me.
I recall from my 3rd year econmetrics class that most of the forecasting we did was out of sample forecasts (ie. predicating GDP for next quarter etc.). Yet, on CFAI text P429 under example 7, it says that “.. many articles that analysts read contain only in-sample forecast evaluations”. Can someone please carlify this for me? Is my understanding of in-sample forcasts completely off? and if not, what’s the point of in sample forecasts? Thanks!

 

[Aether]

Not sure if you’re using older material or if I’m just sleep-deprived - I don’t see the line, “…many articles that analysts read contain only in-sample forecast evaluations” in example 7.
Regardless, you just need to recognize that in-sample forecasts are not useless – they can be used to test your model’s worthiness (RMSE). Beyond that, the whole idea of forecasting is to predict the future/unknown (out-of-sample).

 

[KevPei]

@Aether Sorry, I should’ve worded it better, I meant in the paragraph below example 7. Midway through the paragraph it says ” although professional forecasters distinguish between out of sample and in sample forecasting performance, many articles that analysts read contain only in-sample forecast evaluations.”
Then in the subsequent paragraph it says “Typically, we compare the out of sample forecasting performance of forecasting models by comparing their root mean squared error (RMSE), which is the square root og the average squared error….”
So it seems to me comparing RMSEs of two models is most useful for out of sample forecasts.