Modeling portfolio using multiple regressions

[JoeJackson12]

Hi All,

I'm wondering about the viability of this task. I'm looking to form something similar to an investment portfolio, which aggregates different financial securities and calculates the expected aggregated return and standard deviation.

As opposed to using the historical returns and s.d. of the securities, however, I want to fit multiple regression models to sample data sets using an array of independent variables, unique to each security, to predict future returns using R Stat's

predict()

function, and obtaining projected returns and a standard error of each regression.

The purpose of this exercise is to minimize the effects of low the R-squares for each individual regression for each security. It should minimize the fact that the R-square value for the individual regressions is extremely low, thus leading to a very poor comparison between out-of-sample realized returns and those found using R's

predict()

function for each of the individual securities. Then, I can calculate a more accurate implied return for each security by looking at its weight in the aggregated portfolio, and multiplying the return of the aggregated portfolio by that weight.

My questions are (A) is this statistically viable? And (B) regardless if it's statistically viable, is there precedent to this, and is anyone aware of a function in R to do so?

Any help is much appreciated, thank you so much!

 

[noetsi]

The only way to know if it has been done before is to look at the literature or know someone in the business/economic world (because not everything they do ends up in journals I assume).

I am not really sure what you are doing, it would help to know what substantive method you are doing rather than what R code you are using as some here do not know the R code

Why does R square matter to you if you are trying to predict? It would seem something like MAPE that compares your predictions to actual results would be more useful. Forecasting that I have seen do not focus much on R square (although I am not a financial analyst I just work with time series models to some extent).

 

[JoeJackson12]

Understood. Thanks for your reply. Your suggestion to look at MAPE is helpful, and along the lines of what I'm already doing. I'll continue to think through this, and potentially reach out if I have more questions!