I'm not particularly interested in actually doing automated trading, but I do read this section of ET and haven't seen anyone ask this question. If you were just starting out today, what would you have done differently? Are there any lessons that aspiring programmatic traders can learn that you wish you had when you were younger and less wise? Whether it's the languages you wish you had focused on, to the methods you should have used to find strategies, to which vendors don't waste your time and provide the best bang for the buck, it all goes. Divulge as little or as much as you want. IMO, it's all useful info that a lot of the readers of this forum would benefit from.
My IB Sales Rep told me I know what I'm doing (http://www.elitetrader.com/vb/showthread.php?s=&threadid=200655), I'm not sure he is right though ;-), I am a good customer though I guess Anyway, I wish I had focused less on back testing and focused more on forward testing. I listened to a recording of a guy named Howard Bandy after auto-trading for several years. I wished I had listened to him before ;-)
In order of importance: 1. Accurate clean complete data. Get this taken care of as soon as you learn of the need. 2. When dealing with data services, be sure the service allows exisiting functions to be used in created functions for the data services This is sometimes referred to as a "global variable". Most services, however, do not know what global variable means. you need to get well acquainted with tech services for your feeds. try to get past the "dumb" period ASAP. 3. Degapping. degapping occurs on two sensitive levels: a. bar to bar and b. session to session. Most of the time this is not a function services provide. If they do, then it is incomplete. In high money velocity trading it is VERY important to be able to "carry over" market boundaries under "degapped" conditions. This affords integrity of nexted fractals. A learner can do this by hand as a learning segment. Then having it provided becomes important. 4. Unlimited export/import rates. A person using a service needs the have the service be functional relative to human interaction times with markets. Usually a trader does not want to be observed while observing the markets. These two thoughts may not appear to be interconnected but they are and the sooner a person acquiring skills knows this the better. As an aside for traders who are becoming skilled. and who have found that they needed additional degrees of freedom to be effective and efficient, take a look at the contemporary contest scene and the contemporary interactive aspects of media interaction with audiences. It is all very slow and very limited with respect to real time. Today, because of the advancement of science, it is possible to make the transition from assessor to expert in very little time. To do that the person must be abole to see the markets and keep track of things to get to "know that he knows" comprehensively. Voids in the mental aspects of trading need to be filled ASAP. Discovery is the key. differentiation is the result. Sevices have to be able to carry a lot of wieght and be able to afford the trader all the information he needs within the nonstationarity of the markets. this is part of growing efficient and effective. Consistency flows mostly from the continuity of a systemmic approach. Services automate continuity for the most part. A learning trader needs to be able to add shells of more and better continuity to the core of his approach. In the past learning failure was the dominant modus of learners and potential practitioners. Today learning failure can rapidly be bypassed by a purposeful critical path of self learning IF services are there. Learning to learn and knowing how the mind works are adjuncts to fast tracking to expertise. There are many milestones along the way. By understanding that the mind is more adept than a computer, it is easy to find out the principle of anticipation is what is exacted from sensing becoming perception and the entire avoidance of prediction.
As a footnote to my prior post, it is a good idea to look at the forks in the road on the way to expertise. As a person succeeds in spending year upon year in an expert modus, he can lift up his head and look around at other proceedings to expertise by other paths. It is very heart rending to see others get snuffed out on their way or worse still to see a person get stranded at the end of a branch that ends. Some of the classic reds of people getting stranded are those done by reporters and journalists who write for money. Other classic reads are from those who are funded to do research; their research hits brickwalls at the end of branches most often. The NSF funding is the best example of research ending in unknowing failure. the primary theme of NSF published research is "throwing out the baby with the bathwater". A simple example is doing noise rduction then teting for TA patterns with "bridging" indicators. It is best to skip reads of the dissolutioned. They are just recording their trip to failure and their failure maintenance. Finally there is an early fork in the road that is almost always taken wrongly. It is the induction/deduction fork. One branch leads to probability and the other leads to non probabilistic logic and paradigm theory. Why would concerns with money not lead to logic. Logic has nothing to do with guessing. Probablity is the foundation of guessing; learn quickly to not get involved in probability. A recent documentary on the leading scammers of our time featured a retired licensed broker who lead three generations of her relatives to utter investing failure by letting one person handle their money. She read every publication of Bernie Madoff for as long as he released publications. She is mad at him for being untruthful. another person informed the SEC for a decade on Bernie's shennanigans. He figured it out in four hours in his first attempt.
incredible, next you will tell him to get a computer to backtest. lets re-cap jack hersheys most important things for automated trading 1. get good data 2. get a global variable where u can pass from window to window, timeframe to another timeframe. 3. take out degaps 4. make sure u don't skip bars I mean u really don't want to skip bars, monitor your "newly purchased computer" for it to not skip bars.
So, its okay you do not understand what a global variable is. As a trader progresses towards expertise, he needs to deal with functions of functiions. what this means to a person who is savvy is that functions are plugged into functions. A consequence is that more than one level of functions exists and is useful in REAL time. Degapping is the opposite of what you wrote. degapping is removing gaps so there is continuity of information and boundaries of that information. You do not degap nor does your service or computer have that capability. You have not gotten that far along. Certainly a lot of bars are skipped as an aspect of functionality. I may not have brought that up but you "read it" in terms of bars formed during non RTH. It is BEST to have a service that does the merging of skippable bars automatically. They come in two classes. Also, I skipped suggesting that learning traders take the effort to line up 10 to 12 very reliable leading indicators of price. Back testing has zero utility. Why/ The why is established using logic.
My understanding from when I programmed is that you set a global variable at the top so that all sub routines can have access to it rather than redefining it each time you have to pass the data. Back testing is pretty useless since it tells you what happened in the past. However, you do need to look at how the market is currently performing in the recent past to come up with something that you can forward test with a small amount of real money.
Wrong. There's no reason why the period(s) you forward test on should be more any more representative of market conditions AFTER that than the period(s) you backtest on. Even the recent past is only representative until it's not.
Jack: a question about charting and correct way to see an incompleted pattern. On chart of ES Aug. 17, at bar 19 of this chart (shows end of day only) , the book mark for the pt. 1 of the ftt was hit on the 5 min. chart. It seems that you would fold back all the preceeding bars of the incompleted trend and continue with the trend lines of the last leg of the last completed pattern, which in this case would be the red trend as shown on chart. The question has come up about whether this is correct, because if you look at the 1 minute chart, it appears that the pattern DID complete. But if you are charting a 5 mintue chart, can you jump to another time setting to see if something has completed? It seems to me if you are charting a 5 minute chart, the chart should make sense only in the 5 minute frame of reference. There appears to be disagreement among the people i have asked about this. What is the correct solution? And INTERESTINGLY: No matter how you resolve the issue of the bars, you still end up in the same place... in a r2r2b2r trend looking for the black leg. I am pretty much a newbie so I very well could have this all wrong. Apprecitate your insight... Tucson Skiman