I am attempting to estimate a linear model as: $$ y = a +bX +e $$ I have a series of annual returns and I would like to estimate the effect of volatility on losses. My Null hypothesis is that volatility in period $t$ should have no linear relationship with losses in period $t+1$. So in actual fact, my model is: $$ \min[0,y]t = a+bX_{t-1} +e $$ where $x$ is the standard deviation of $y$ in period $t-1$.
My questions are as follows:
- What is the best way to test if my model is correctly specified?
- Given that $X$ is rolled over in each period, should I be worried on the impact of serial correlations?
- In the case of serial correlations, can this be better reduced by using higher lags of $X$?
