| Index | Recent Threads | Who's Online | User List | Search |
|
|
 |
Portfolio123 » List all forums » Forum: Ranking Systems and Factors/Functions » Thread: Calculation of Overall Rank for a Ranking System |
|
Total posts in this thread: 3 |
[Request a Feature]
[Post new Thread] [Add To My Favorites] [Watch this Thread] |
| Author |
|
|
hyper
Advanced Member
|
I wanted to understand how an overall rank value was actually calculated based on the individual ranks of the constituent factors and their corresponding weights in a ranking system. I haven't been able to find the mathematical calculation anywhere. I looked through the documentation, however the calculation process is only described in a few vague sentences. This is such a fundamental concept to this entire web site and business, yet it seems that we all simply take it for granted. So I tried to derive the overall rank value myself in a spreadsheet. The values I got were a little different. However, more surprisingly, was that the rank position for almost 600 tickers out of 8005 were off by as much as 4 positions. If you look at the attached image, you can see for example that TOF appears at position 3284 based on Portfolio123's ranks, however based on my weighted sums, it appears in position 3288. I would have thought that even though the sum of weighted values that I calculated may be a little different to Portfolio123's ranks, that at least the order when sorted from highest sum to lowest would be the same. They are not - almost 7.5% of tickers are in a different position, and I'd like to know why. I think the documentation should include exactly how the ranks are calculated. Can someone explain the discrepancies that I've discovered? ---------------------------------------- 
rank_calculation_discrepancies.gif
(126883 bytes)
(Download Count: 104)
---------------------------------------- [Edit 1 times, last edit by hyper at May 12, 2008 2:01:30 AM] |
||
|
|
marco
|
It's hard to pinpoint the cause in your example. The small differences sure appear like rounding errors. If you are downloading the data from P123, the 2 decimal places of precision is probably the cause. The best thing to do to replicate the ranking exactly is to try it with a much smaller universe so it's manageable. Try running your test with DOW 30. Ranking is just a bunch of sorts, then assigning percentile values from 100-0. Composite nodes sum all the weighted percentile scores then re-normalize from 100-0. NA's are dropped to the bottom and all get the next lowest score. For example: If you have 5 values and 2 of them are NA, then the first 3 get 100,80,60. The 2 NA's both get 40. This is done so that factors with many NA's (like PE) don't over-penalize the overall score. ---------------------------------------- Portfolio123 Staff |
||
|
|
garryp
Advanced Member UNITED STATES Joined: Feb 15, 2008 Posts: 104 Status: Offline |
So... If I use a prescreen that returns only 5 stocks, they will be ranked 100, 75, 50, 25, 0? ---------------------------------------- -- Garry P. |
||
|
| [Show Thread Printable Version] [Post new Thread] |
