I have two general questions regarding "in-sample fitting vs. out-of-sample backtesting" kind of analyses. Is there any relationship between the length of the data collected for in-sample fitting ($a$) and the length of the data reserved for out-of-sample performance ($b$)?
Should $a$ be longer than $b$? If yes by how much (in absolute or relative terms)? Or is it okay to have a longer $b$ than $a$? Is the 50:50 rule discussed in the following link justified? Is it used in academia or by practicioners? What is the ideal ratio of in-sample length to out-of-sample length?
Let's now assume you somehow determined that $b$ is too long for $a$. Would you extend $a$ and reduce $b$ keeping $a+b$ unchanged? Or just reduce $b$, and thus reducing the length of the testing period?
