Welcome,
Guest



I'd like to be able to rank by value composites (either VC1 and VC2, or even better custom Value composites)
This would require ability to percentagerank on several columns, then being able to adding these together and produce a new percentagerank based on the calculated sums. The ideal solutio would be a fully flexible one, but this would require a more advanced interface and som coding. Adding a VC1 and a VC2 condition/screen would be a good start. As before, this is very easy to do in excel, so a full excel export would solve this issue. Thank you for a good prouct 
erik


We do not support percentile ranking, but you can sort by a custom column in either ascending or descending order.
Instead of "percentile ranking from several columns", you may want to consider creating a composite factor that is calculated from the values of multiple columns. You can then sort by that new custom column. 


Thank you for your answer. I'm sorry to hear that percentile ranking will is not supported and that the 50 row limit can not be lifted.
As told before I am very interested in screening for companies that has a good combination of ratios, possibly custom rations , but to start with James O’Shaughnessy rations Value Composite One and Value Composite Two ration from www.amazon.com/WhatWorksStreetFourthEdition/dp/0071625763 This is a strategy that has backtested very well in several academic papers and that I think a lot of people would be interested in. If you are not familiar with the O’Shaughnessy value composites, the idea is to combine several ratios in to one rank. As the ratios will differ in absolute value, percentile rank is used to make each ratio "neutral", to weigh as much as the other. I hope you understand the concept. From your post I get the feeling that this actually can be done in the screener, but if you do not support percentile ranking I do not understand how. Maybe I've missed something? Could you please point me in the right direction by poining me to a tutorial or give an example of how to do this with a composite factor? Again, the percentile rank is important to make the ration neutral to one another. If this can not be done, could it be done in the back end? And if so, would you then consider adding Value Composite One and/or Value Composite Two as a conditioning to screen for? 
erik


It's not the same, but it might be a pragmatic compromise to map values of each column into a variable with values from 0 to 1. Then you can add up those variables to get a fair total rank.
Let's say you want to map P/B. One approach would be to decide on an interval for those values, anything outside is set to interval limits using MAX() or MIN(). Using simple calculations, you then map that onto 0 to 1 and if you choose a decent interval the distribution should be "ok". There are probably more sophisticated mathematic formulas to handle this, so that you get a good distribution and "rank". What do you think of that idea as a workaround? 
Claes
The following user(s) said Thank You: testardetta


Thank you wwwclaes for the suggestion. I understand that your idea is to do the follow calculation:
If P/B is calculated to be between x and y where x>0 and y>x I do the following 1) Subtract x from each calculated P/B value to get 0 as lowest value 2) Davide all calculated values (P/Bx) with (YX) to get 1 as highest value I now have all values normalized between 0 and 1 and can add them together. This certainly is an option that probably would give similar results as the original method with percentile rank. I might give it a go, but would much rather use the original method as described by O’Shaughnessy. The original method is what is used in all academia and backtests. It is quite uncertain what a variation like this one would do for performance (could be positive or negative, but I do not what to gamble) If you have another suggestion, or if I misunderstood you, please reply again. All suggestions and ideas are welcome. The best solution would of course be if admin provided the functionality or added percentile rank in some way 
erik


Yes, that was that I was proposing. It is certainly not the same thing, but it is doable right now in screener.co. The biggest drawback is probably that the distributions may be uneven. E.g. if you choose intervals so that most values end up near 1 for P/B and near 0 for P/E, then P/B will have a much higher influence on the total ranking.
I might try something similar myself and will let you know if I end up with a result. Likewise, I'm interested if you try it out yourself. 
Claes

