Thursdays Prediction: 3.82 -Result- Open: 765.76 Close: 752.83 Change: -12.93 Prediction: Wrong Note: N/A Paper Trade (one ES contract, Market Orders): Buy ES @ 773.50 (9:30am) Sell ES @ 751.00 (4:00pm) Change: -22.50 Profit/Loss: ($1,125.00) Overall Profit/Loss: ($1,075.00) Next Prediction: Friday (02/27/09): Prediction: 3.77 Signal: Buy This signal should be traded by buying ES on Friday at Open (9:30am) and Exiting (selling) at Close (4:00pm).
I've been making several improvements over the last week and will start using a new version in a few days. The new version uses the same "system" but has been greatly optimized to produce better results. The new version also contains a back test feature which back tests 200 days prior. The back test shows 7 of the last 8 months to be profitable. The sharpe ratio for those 8 months is 0.69 based on a 5% risk free rate. The total points gained is 612.3 and the average monthly return is 38.27% (based on $10,000 per ES contract). The losing month (October) was a pretty big draw down of 97.38 points. Keep in mind this is all assuming the ES exactly mirrored the S&P500 which is unlikely. I was thinking about the possibility of getting it to trade from close to open, since previous tests have shown it to be more accurate on shorter predictions. For example, predicting weeks at a time didn't work so well. Thanks for the encouragement, you have an interesting idea and I will look into it. Perhaps I will make a different version that issues a prediction at close and predicts whether the market will open higher or lower.
1. By who and how? 2. Even if so, a system still can be a loser or breakeven if it guesses the small movements right and loses on big trending days.
I'll be the first to admit the results so far have been pretty bad. Before starting this thread the results were outstanding, so hopefully it gets better. If not I will keep working and trying to improve it.
OP: This is meant to be helpful, so please view this from a positive angle. First I want to say that I have probably some of your background (I hold degrees in electrical engineering, computer science and applied math. I know about the modeling parts/etc your refer to in your post). My main points are: 1. Traders think in terms of how much should they pay for a point they earn? Assume your numbers are right---You hit 80%. First problem: 80% is your prob but not theirs. Why is that? You willl know why in a moment. Suppose you tell them (and this is missing in your prediction) that they will make 3 points. Now they say to themselves to break even I need to not lose 24 points on 10 trials (2 losing, 8 winning). Which means They cannot lose 12 points on each of the two losing trials. They further want a profit, so the cannot lose say 10 points on each of the 2 losers. So far so good, but now the missing link! 2. Now the part you are most likely missing. What is the prob that the stop will get hit? Implicit in your above thinking is that it will be 20%. In fact, it will be at least twice the 20%, which is 40%. The answer may in fact be even more shocking to you! That prob tends towards one the closer to the open price to the close price is. 3. Point 2, tells us that you are assuming the wrong probs in your calculations of profitability. The prob of a loss is NOT 1 MINUS the prob of a win. It is much higher than that as explained, and the smaller your win is the higher is that probability. Now if you rerun your numbers you should be able to realize that your system under your statistics will be a LOSING system for a trader, and it will not work. That should end your experiment a as it is, as it is dead from the start. I call what I just explained the monkey trading problem. It is at the root of all the confusion people have about trading, and also I believe at the root of the people who make money in trading. They understand that a dollar earned cost more than what the next guy had assumed in his calculations, and they have other important components to make the reward part pay for their correctly calculated risk, and their profits. 4. I sympathize with you on the idea of test and experiment. I want to tell that I have been doing a test of my own here and on a blog. A test that aims at the problem of determining the top and bottom of a market trading day at the level of one minute, and to find it at the earliest time during the day. This is the heardest trading problem in the world I know of. I have posted my calls here. You can check the most recent thread for this week. I am mentiong this because even if I made those unusual predictions, even my predictions are not of sufficient importance to people here. Why? First people may already make money, and they do not care. Let us pass those. Second, why should they follow another person's call which may clash with their opinion about market and may confuse them. And Third, people here seem either to be successful or do not have the money and the appreciation to pay for what you and/or may have to offer. Therefore you are wasting your time, because if my nailing of tops and bottoms at the level of a minute is not good for them, what do you think the chances are they would be interested in the nature of work you are doing and pay you for it? 5. Last someone asked the fundamental reason behind what you do. I think you may have not addressed it correctly. Here is how I would answer that question if posed to me about what I do. My predictions are based on principles one of which is this: when the market has only ONE degree of freedom, it will move in that direction. So my job becomes to identify points in time where price has only one degree of freedom. This does not mean that all tops/bottoms have this property, but rather that price and time that has one degree of freedom is a potential top or bottom. Only a small number has that property. Then from those I have other layers to select the top and bottom among them. I just explained it roughly so that people can see the thought behind it. It is not the tools that do it, but the principles and analysis that do it. People understand want to know principles even at the conceptual level. In my case they probably would say that it makes sense to operate on the principle of one degree of freedom. Regards
Ummmm rft If one can predict the direction with 80% accuracy, it is very easy to build a system around that. Not a waste of time.
So you spent 5+ years developing this "breakthrough software" and after running it for 2 weeks live here you suddenly discover that it needs improvements & optimizations? Maybe you need to re-read something that you posted in your first post: "My goal here is to provide proof that my predictions come before the predicted trades. Too many people try to prove their systems with past results, when the only true way to prove it is with future predictions. It's very easy to make a system do well on past data." Couldn't have said it better myself.
Clearly, not working. The only reason it lost only a few points because we tanked before the open and went sideways for the rest of the day. By the way this brings up an interesting issue, the signal being wrong but still making money due to gaps. It happened for the 2nd time in less than, what 13 signals??? Another question is the wisdom of holding from open to close with this huge volatlity. Case in point: on Friday the entry (in spite of the wrong directional prediction) went 10 points in the green, no reason to give it back and end up in the red.... GTS had a good point in the post above, there is no way you perfected this system for 5 years and it performs this badly. I think you are flipping a coin. Or at least you should be....