Tidbits & Reminders [3]

In the previous post I mentioned how the process of removing problematic variables was itself incomplete as I removed them from all models as opposed to just the subset of models for which they give strange fitting results. I have since made the requisite changes to the previous series of posts. Removing problematic funds becomes :

 

#Removing problemtic funds for specific models only
problemRemove <- function(list.element){
	problems <- which(is.na(list.element==0) | list.element<0)
	return(list.element[-problems,])
}

model.list <- lapply(model.list,problemRemove)
subfund.names <- as.matrix(unlist(lapply(model.list,rownames)))
temp <- apply(do.call(rbind,model.list),2,unlist)

l

What counts as problematic in my case is when there are fitting problems or when the selection criteria are negative. It should be noted that there is nothing wrong with a negative AIC or SBC,I just designated negative values as problematic in this example.

 

Applying the plotting functions yields :

 

 

Let’s compare this to the case where we delete the problematic variables from all models :

 

In the latter case we deleted 4 funds across ALL models because they had model fitting issues across SOME models. In the former case we retained all funds even though some had problems in the fitting process (gaps in facetted barcharts). This evidently changes the percentage of times a particular model is the best performer.