Wow. I thought we were making progress, but i am very disappointed with your replies. Maybe I actually have overestimated you and your capacity to handle these concepts. I will have to think about whether it's worth trying to continue this discussion or not... I have been at the office all day and my brain is fried... so I'm leaning toward "not". Someone who doesnt understand basic statistics telling me I have work left to do on my pairs system... or that he would throw a system out because of a number Wealth Lab made up... hmmm... probably not possible to have a constructive conversation with that person...
Does that mean it's my turn? Anyways... I figured that bwolinksy may ask for the MAE and MFE so I added 2 columns for the OpenDD / OpenRun-up (I usually don't follow the "fill efficiency" but I figured it can help bwolinsky...) for each trade. Format is tab-delimited: Actual PL - OP/DD - OP/RU
That's right, Gann, it is your turn. Profit Percentiles Smallest 1% -.1 -.1 5% -.09 -.1 10% -.08 -.1 Obs 50832 25% -.04 -.1 Sum of Wgt. 50832 50% .03 Mean .0251957 Largest Std. Dev. .0725205 75% .09 .15 90% .13 .15 Variance .0052592 95% .14 .15 Skewness -.003583 99% .15 .15 Kurtosis 1.802538 MAE Percentiles Smallest 1% -.12 -.18 5% -.1 -.17 10% -.09 -.17 Obs 50832 25% -.05 -.16 Sum of Wgt. 50832 50% .01 Mean .0093378 Largest Std. Dev. .0737714 75% .07 .15 90% .11 .15 Variance .0054422 95% .12 .15 Skewness -.0117744 99% .14 .15 Kurtosis 1.878039 MFE Percentiles Smallest 1% -.09 -.1 5% -.07 -.1 10% -.06 -.1 Obs 50832 25% -.02 -.1 Sum of Wgt. 50832 50% .04 Mean .0409791 Largest Std. Dev. .0736008 75% .1 .21 90% .14 .23 Variance .0054171 95% .15 .27 Skewness -.0018808 99% .17 .32 Kurtosis 1.872599 I'm having trouble understanding what I'm looking at. Your description goes: Actual PL - OP/DD - OP/RU I'm missing why there would be negative values for your OP/RU and positive values for your OP/DD? Can you try to explain that? I've always thought of MAE and MFE as hard values where MAE would be given always below 0, and MFE always above 0. Perhaps there are some data manipulations I need to do to understand what you're calculating? I'm thinking about replacing any OP/DD>0 just as 0, and any OP/RU as 0 as well. If this was not your intention, perhaps you are showing slippage this way, where there was never any MAE or MFE, possibly? The data seem a bit more reasonable when I make that manipulation: Profit Percentiles Smallest 1% -.1 -.1 5% -.09 -.1 10% -.08 -.1 Obs 50832 25% -.04 -.1 Sum of Wgt. 50832 50% .03 Mean .0251957 Largest Std. Dev. .0725205 75% .09 .15 90% .13 .15 Variance .0052592 95% .14 .15 Skewness -.003583 99% .15 .15 Kurtosis 1.802538 MAE Percentiles Smallest 1% -.12 -.18 5% -.1 -.17 10% -.09 -.17 Obs 50832 25% -.05 -.16 Sum of Wgt. 50832 50% 0 Mean -.0272466 Largest Std. Dev. .0379016 75% 0 0 90% 0 0 Variance .0014365 95% 0 0 Skewness -1.122019 99% 0 0 Kurtosis 2.921274 MFE Percentiles Smallest 1% 0 0 5% 0 0 10% 0 0 Obs 50832 25% 0 0 Sum of Wgt. 50832 50% .04 Mean .0555341 Largest Std. Dev. .0561523 75% .1 .21 90% .14 .23 Variance .0031531 95% .15 .27 Skewness .5384886 99% .17 .32 Kurtosis 1.908338 From this, I'm going to try to construct a histogram, and find an optimal profit target and stop only. I may only find a stop, but if these values have already been optimized it is not likely that I could find better fits for the data. My goodness, where are my manners? You've been waiting a long time, Gann. So I'll analyze just the profit distribution from the summary statistics below: Profit Percentiles Smallest 1% -.1 -.1 5% -.09 -.1 10% -.08 -.1 Obs 50832 25% -.04 -.1 Sum of Wgt. 50832 50% .03 Mean .0251957 Largest Std. Dev. .0725205 75% .09 .15 90% .13 .15 Variance .0052592 95% .14 .15 Skewness -.003583 99% .15 .15 Kurtosis 1.802538 (BTW you have one hellaciously large dataset. When constructing new values, I get this message: no room to add more variables due to width An attempt was made to add a variable that would have increased the memory required to store an observation beyond what is currently possible. You have the following alternatives: 1. Store existing variables more efficiently; see help compress. 2. Drop some variables or observations; see help drop. (Think of Stata's data area as the area of a rectangle; Stata can trade off width and length.) 3. Increase the amount of memory allocated to the data area using the set memory command; see help memory. ) So I guess I'm going to have to use only the description, I guess as we were always doing, neither of the three options enabled me to manipulate this data beyond the changes I had made. Profit Percentiles Smallest 1% -.1 -.1 5% -.09 -.1 10% -.08 -.1 Obs 50832 25% -.04 -.1 Sum of Wgt. 50832 50% .03 Mean .0251957 Largest Std. Dev. .0725205 75% .09 .15 90% .13 .15 Variance .0052592 95% .14 .15 Skewness -.003583 99% .15 .15 Kurtosis 1.802538 I'm sure first you think I'm going single you out with the negative skewness of -0.003583. This I don't believe is significantly different than 0, so there's actually no skewness to address. The kurtosis is quite peaked, and I noticed your average profit being 0.0251957 or 2.51957%, which is excellent for a 50832 trade dataset. I think we'll see large profits from you in the future. So 95% of the trades approximately fall between 2.519+/-7.25%*1.96= a confidence interval of 16 3/4's percent and minus 11 3/4's percent. That does show me where the skewness is, and appears that your losses, while slightly less, have a greater impact on your profits. With a win percent rate of 0.580146364=58%, for a large dataset and given the large average profit, I would say this is a very good system. Kudos, but I don't think you needed me to tell you that, and I do appreciate your participation. My next post, I will try to see if there's some way to optimize this data, but I think it may be optimized already. If it is, please tell me, and I won't try to.
ahhh... I coded the Open Position Run Up and Drawdown wrong and found it... I was generating a execution efficiency / liquidity risk... (In simple terms, it's the max/min price of execution likelihood based on DOM liquidity...) I'll fix the output app. and regenerate the txt, later on. Just ignore column 2 and 3 for now... - FYI. It's never optimized. Calibrated? Obviously, yes. (It may mean the same thing to you... but for me it isn't) ------------------------------------------------------------ So... Bwolinsky... admit it... - I'm a better Strategy Developer (Systematic Trader) than you. - My models are way better than yours. I WANNA HEAR YOU SAY IT!!!!
Gann, could you describe what your system is doing? Please don't tell me it's based on signs of the moon...lol. Just a joke. I see a very good investment basis in your system. I was thinking I would be much better enabled to help each one of our profit distribution sharers if I knew more about the time periods involved, dates would be extremely helpful, and especially APR and DD. Realizing, Gann, that you most likely gave me a chronological profit distribution, and perhaps, Mike and Talon did, meaning I'll have to go back and look, I actually have a lot of suspicion about the graph. My final judgement is that <i>it's too good to be true.</i> I find the lack of volatility a sign of statistical impossibility. You can prove me wrong if you can show me where the drawdown is, because I don't ever accept "no drawdown" as an answer. Could you provide your system statistics? Specifically APR, Sharpe, Drawdown, and profit factor? I believe it's your original research, but I don't find it likely that it is possible whatever it may be based on, and that's my subjective judgement by the shape of the graph. So, truthfully, is this a real distribution? Are these trades in chronological order? Did you actually trade this? If so, what were the system statistics. I wish I could graph it for you, but after compiling the data, I see it is unlikely that this is a real system, and I'm not saying that just to spite you, Gann. I didn't know what I was looking at until I saw the graph. The graph I've seen many times before. A kind of smooth second-half U-Shape with zero drawdown. Come on, Gann, please be real with me. I can tell by the graph it's not real, and I do see a lot of evidence of curve fitting. I don't believe I'm going to try to find the optimal stops and targets because they're so optimized as to suggest statistical impossibility, as I have said. If anyone is wondering what I'm looking at, please compile Gann's tabris .txt system into excel by saving the file, opening as a tab delimited, and graphing the trades chronologically. This kind of simulation I'm always wary of, and if he can prove it is legimiate, I would say it's the best I've ever seen, but without evidence beyond this, it does suggest a curve fit, and I didn't see that in Talon's or Mike's datasets.
Please see my reply above. Your system as graphed is statistically impossible, please understand perfectly, in my honest opinion, so I would not say that just yet unless you can prove these trades actually happened, with system statistics in the realm of possibility. If you can do that with a simple TS performance summary, then, yeah, it's one of the best I've ever seen, but I still have my doubts.
1. It's a live result. I dunno how you expect me to generate a TS or WL report with that... The posted 50832 trades are AS-IS. 2. You seemed interested in institutional level trading models. So provided it. I never asked for you to help me. All I'm doing is bragging about how great I am as a developer. I seriously thought that your claims about being the best developer was funny considering that you've never seen REAL models that are traded internally within banks and Quant funds. 3. It's chronological but I won't provide any dates and symbols. Reason is simple, it's part of the business to be good at it reverse-engineering. Obviously, it's alot easier to take someone's trade log than exposing a tendency from the market. (Though, I have to admit that you'll never achieve it through Wealth-Lab... but you're not the only person in ET...) 4. You don't have to believe me, as a matter of fact, it's good to be skeptical. It's a bit funny because you're now acting like a discretionary trader who doesn't believe that computers can trade and make money. Maybe another analogy are those loser retail guys claiming that Goldman Sachs made all their profits by cheating in the market. 5. Just keep in mind... If you are really serious about this business, you should really move to Chicago or NYC and start getting exposed to the industry. I'd say, 2 months at a Tier 1 Prop. Group / HF, you'll believe me. (Though.... I must admit that it'll be tough for you to join one... but you can always try... plus, you make better connections living in the city) Sincerely, I find your basis to be very naive in regards to the industry. Also, learn how to program. ------------------------------------------- If "I don't believe you" is your answer... I'll let you go on with your thread. But seriously??? "I don't believe it's possible???" C'mo-------n... That's really boring... (Magna... I've been very nice. I was expect this to be fun but without profanity??? It's not really that fun... )
Lol.. no it's not really that fun. It should have been clear with 60 seconds work that you sent him a uniform distribution... nevermind where it came from, etc... and he talks about kurtosis and skew. If I was him I would stop bragging about the quality of my Centre College education and math minor.... seriously... is this guy for real? I have completely, 100%, given up on this thread. The kid can't handle it. However... TSGannGalt gave him the best advice possible: move to Chicago or New York and get in an environment (hedge fund, bank trading desk (though you won't actually do much trading) or prop shop) where you're surrounded by people you can learn from. When people evaluate you for these jobs they will look at mental potential and personality... the arrogance you have demonstrated here I hope is just a show for the internet (but alas probably is not since you attach your face and name to it... it's probably you... in which case anyone interviewing will see it and you'll be done before you start.) I can also guarantee you wouldn't make it through our interview process... not because of lack of education (I seriously hired someone who didn't know what a bid / ask spread was) but because of lack of critical thinking skills... and the ultimate death knell for a trader... an inability to re-assess what you're doing and admit you might be wrong. I'm done here... sorry we weren't able to accomplish more... when you're looking at no P&L from the QID QLD thing in a few years BoWo come back to this thread and at least you'll understand why it happened... but by that time you'll be a few years older and a lot of water under the bridge and not so easy to seize new opportunities. I hope I'm wrong about you, but I'm sure I'm not wrong about your systems or system development ability.
It's not "I don't believe you", more than "it is just too good." I think there's a difference. Without question, it's an excellent performance summary. And I do think "calibrated" and "optimized" are the same thing. I do know how to program, Gann. Python, SAS, STATA, MIPS, WLP, TS, .NET, C# for NT. I got them all. I got a program to work in JAVA from E-signal, too. What I do with python gets me exposure as a kind of internet trolling bot, but that's another story that I won't go into. I wish you could describe your system. Seriously, if you described what your system is doing, I would know from that if it was a real distribution, so how about it? I need a little more convincing, because the plot of your trades is so attractive I'm concerned about the lack of volatility in the curve. Please try and continue the dialogue. Your system is the most intriguing I've looked at. Have you thought about becoming a vendor? I guess with a system like that, extra money to a few thousand a month you don't need, and that's about what you could expect.-- Beau