Homework answers / question archive / You will consider a large data set of daily values between January 2, 2020 and June 3, 2022, for a total of 611 times series observations

You will consider a large data set of daily values between January 2, 2020 and June 3, 2022, for a total of 611 times series observations

Unrecognized

Share With

You will consider a large data set of daily values between January 2, 2020 and June 3, 2022, for a total of 611 times series observations. The data set contains a large number of variables stock, currency and bond markets, and the full definitions of the variables are given below. As in the first assignment, I transformed the series into returns, by taking log differences, and then, took one period lags to examine whether there are any delayed reactions. Note that the time period covers the Covid-19 global pandemic, which created high level of stress in markets and changed dynamics. Your variables of interest is RPBW, returns of the clean energy stock ETF (PBW). This is the largest and most popular clean energy investment instrument in the US. You will examine the determinants of RPBW by using returns and lagged returns of all the other variables in the data set. In other words, you will use all Rs and Ls as your potential predictors. You will consider the following methods as candidates: OLS (benchmark), Lasso, Adaptive Lasso, Elastic Net, Adaptive Elastic Net and, as a new feature, Adaptive Lasso with t(5) and Cauchy distributions. As a quick note, all penalized, and standard, regression methods are estimated assuming errors are normally distributed as default. However, normal distribution assumes no major outliers, a standard bell curve. If the data contain extreme values, a distribution like t(5) and Cauchy (which are part of stable distributions that assumes thick tails, for those of you statistically inclined) may be more appropriate and may give better predictions. This is one of the outstanding features of JMP in my view. No other program (even SAS itself) allows you to estimate penalized regressions by anything other than default normal and luckily, we don't have to know the background statistics at all, because all you have to do is change the distribution in the Model Specification platform from Normal to t(5) or Cauchy. JMP will do the rest! Follow the course methodology: - Create a cross-validation column (I did this and named it again "Holdback" column in the data file) - Estimate all candidate models and save predictions - Conduct model comparison and choose the best by Test data results (training, validation, test) - Interpret your chosen model Below are the JMP file that contains the data as well as a document with variable definition attached. PS: You will notice that in this version JMP offers a "Model Launch" option in the General Regression platform. You can use this option to conveniently change the specifications of your model; for example, from Lasso to Adaptive Lasso, or from Normal to Cauchy, without needing to go back to the original model specification page.