Peter (and Richard)
In contrast to Richard's opinion that you have the system to beat all systems, I would like to propose that you've simply got three decent systems for reasons I'll go into below.
First of all, let me be upfront by saying that I'm a novice at system developing by most measures (5 months as a hobby) but I've thought a bit about how to interpret system stats.
I don't assume that you've got a system to beat all systems here because I need two pieces of information that you left out.
1) How much money did you start out with? I can't tell only knowing the number of $ you ended up with and the annualized rate of return (which isn't really the annualized rate of return... more on that in a minute) Without knowing this the profit of $6 mil is just fluff and has no information content.
2) How many years worth of data were you using? This is really important for reasons that are not obvious (at least to me) at first glance, and I don't mean the obvious “use lots of data” type of advice. I don't know what program you were using but I know that WOW and MS both report annualized rate of return as total % profit / years of data tested. This is really really wrong. Most notably, because you get a higher number for annualized return the more years worth of data you analyze with a constant system performance. If you have very many years of data in your test it will massively overstate your annualized return.
Here's an example:
Assume you've got a system that returns a constant 40% year in and year out. Pretty good system. But look at what the screwy annualized rate of return formula that WOW and MS give you looks like due to the fact that they are dividing the accumulated profit by the number of years studied
Constant 40% Profit per Year System
Starting with $10,000
Constant 40% Profit per Year System
_________________________________________Accumulated
______________________________Screwy_____Total %
Years_____$_______Accum_______Annuali____Profit
Elapsed___________Profit________ROR
0.......10,000
1.......14,000....4,000........40%.......40%
2.......19,600....9,600........48%.......96%
3.......27,440....17,440.......58%......174%
4.......38,416....28,416.......71%......284%
5.......53,782....43,782.......88%......438%
6.......75,295....65,295......109%......653%
7......105,414....95,414......136%......954%
8......147,579....137,579.....172%.....1376%
9......206,610....196,610.....218%.....1966%
10.....289,255....279,255.....279%.....2793%
I've never pasted from Excel into SI before so I hope the formatting comes out ok. Notice that our (by definition) 40% a year system is represented as a 279% annualized rate of return by the way these programs calculate it!!! I would bet that the majority of the people using these programs don't realize this is going on.
The way I calculate real annualized ROR from the output of these system testing routines is to put the total % return and the # of days in test into an Excel spreadsheet and use the following formula.
Real Annualized ROR =((( Total % Return +100)/100)^(1/(# Days in Test / 365)))-1
In the above example plug in 2793 for % Return and 3650 days (which is 10 years) into the above equation and it should spit out 0.4, or 40%
If you use # periods in test rather than # of days you would divide by 270 or something.
As a result, I am not going to assume that I can tell anything by your system report of 259% because it could be 40% return on a 10 year test (or less return for a longer test). Richard will tell us that it is misleading to look at the return of a system, that there are other measures that should be examined, and he's right I guess. Personally, I'd like to ask Richard to take us by the hand and give us a tutorial on evaluating a few system results.
I do have a question for you guys though. It looks like your tests are conducted over a body of stocks rather than just one. You get a composite system test using multiple stocks. I haven't figured out how to do that with WOW or MS. Can anyone help me with that? I have been thinking about doing this in a labor intensive way by taking trade data from single stock system tests and compiling them in Excel. I think that you guys can lead me to a better way.
I also have a question for Richard. You suggest that Peter should bottle and sell his system. Hypothetically, let's say a person had systems that spit out very similar (or better) stats than what Peter showed. How would you suggest one should go about making money from such a system other than simply trading with it. I know that one option would be to try to convince people to let me invest their money using my secret system for a fee. I suppose one could try to convince many people to pay a couple of thousand to get the formulas and the right to use the system themselves. A third option may be to post past results on a web site and try to get people to pay a subscription to get up to date trade advice on specific stocks. I'm not very optimistic about being able to convince anyone that I can manage their money better than the “established” firms. If you were this hypothetical person, how exactly (lots of detail please) would you “bottle and sell it”? It should be a fun exercise to think about at the least. |
