Hi Tom & PocketChange. Thanks for expanding on the subject PocketChange. I understand what you are saying. I know it has to be a portfolo to get the effect of diversity. This issue I need to resolve is that I don't know whether your "random" data has any bearing on "real life"... I am sure you've done your homework and that it does, BUT I would like to see what your "trained" on Random data portfolio would do over past year. I.e. Select you current best parameters and instruments - and run them over a years worth of Live. You might say that this has no value. BUT if I had invested in your system a year ago - it's the performance I would have got... Do you see what I am getting at? I think the idea of using the random data is intriguing, but I want to compare it to what I know. Moon
Of course it has value... my point was to backtest the folio of instruments with the global take profit and close all rules. Your results will then be spot on to real life. I don't really consider what Tom has developed as a system for retail investors but as an algorithmic trading tool for hedge funds and professional traders.
PocketChange, Unfortunately - I do not have a year's data - so I cannot run the Bot myself. It would be great if another user (or Tom) would run this and publish a total equity chart for the year.
I think you are missing the point. You have to select a folio of instruments and determine the scalp size relative to your risk appetite. Run the program against historic tick data and you will get results just for your case. If we both run tradesets of ES, 6E, CL and I have my scalp size set to $100 and yours is $500 the results will be completely different. The random testing Tom refers to is for purposes of testing the robustness of the base algo. Testing different cases against different data sets... 1000+ cases. The general idea is to eliminate curve fitting when tuning the base algo parameters. ES has been on a tear this past year. Tuning an algo against this one bullish data set will get crushed on future sideways and bearish price action. Assume you want to comprehensively test a years worth of tick data on a randomized folio of 3 instruments out of suite of 10 assessing 5 different scalp increments... 3600 data runs each generating unique results. They all will be profitable based on the algo but each will experience different depths of pain in terms of draw down and unrealized losses. If the draw down is beyond your ability to finance and margin the game is over for you.
Hi Moon, I am just in part responding to your wish, by providing the results of the bot backtested on the 2 weeks just before and including the flash crash. This data is not mine but was given to me by PocketChange some time ago. This is the bot behavior as is now, "default" strategy, with "default" scalp size. Remember what happened in that day. Most firm just gave up. This is a folio with 4 of the the wildest beasts. This is <b>backtest</b>. Note that all the other pictures you have seen on this thread <b>are not backtest</b> but live trading. <img src="http://www.elitetrader.com/vb/attachment.php?s=&postid=3123703" width="1200"> Clearly, this is just an occurence, and as pocketchange said perfectly, results depends on the user settings. Will show also some trades in details. Tom
this is RLM crash in detail, just to see the action (the hedging mechanism can be changed at will: this is the current default)
This crash is just a joke compared to what happens with random data (Geometric Brownian Motion, etc.). So you understand why i only trust this kind of <b>robustness analysis</b> done with random generators. Personally, I would never trade anything just because I see some history of good results. I want a much stronger "reassurance" based on a sufficient variety of wild situations. And i want to know the max DD with a given set of rules, because i need to know if my capital will suffice. Tom
Thanks for that. In the RLM, are the 3 red triangles at the bottom the trendfollowing hedge? PS: Going to bed - will pick this up tomorrow. Moon
There is somewhere, i think in the previous thread, a legend for <b>order symbols</b>. These symbols are made up. I use: blue = BUY maroon = SELL circle = global position was zero before order square = global position is zero after order triangle = global abs position was not zero before order ( pointing upward = signed position being increased, pointing downward = signed position being decreased) the 2 numbers near each simbol are fill price and order size. Pocketchange is an authority and i had the honor to first meet him just here some years ago when we were discussing here some strategy aspects, which also included the use of <b>multiple strangles of options</b>. [Since there i even had to change my nick on ET because, in one of my long absence for code development, i even forgot what it was! ;-)) ] At that time, the bot would still send orders simultaneously to different accounts in order to hedge. In time i have developed an <b>"overlay" engine</b> which allows to easily "project" on one single account all the activity of any number of different "traders" (coordinated by the bot). I am happy he is participating in the discussion because my knowledge of English (not a mothertongue for me, unfortunately) would probably not allow me to be effective as he is in any explanation. This is many years of unbelievably hard work and growing up with it, so it's normal if some concepts have to be "digested". Pocketchange is a real top quant and his advice i consider <b>invaluable</b> I like how he summarized in his post many aspects of this works in a very professional way. I agree with him that this bot is not intended for a retail trader (provided that there really is such a thing like a retail "trader" ). <b>Curve fitting</b> (in an broad sense) is the <b>first and foremost enemy of a quant or strategist</b>. The computer will always find its incredibly subtle ways to fool you, and its a fierce fight. Not even random data, in my opinion, can be considered at all "overfit free" ! )) As to the question "I don't know whether your "random" data has any bearing on "real life"... " this is clearly a question that no one can answer. On the other hand i might provide you with any tickdata and nobody would be able to say if that tickdata was <b>randomly generated or was "real"</b>, nor he would even be able to guess the timespan, if i dont tell him (<b>selfsimilarity</b>). Working only with just some past data can be misleading and frustrating, unless one chooses carefully the tickdata to ensure some variety of situations and adopt entry randomization, resampling and other forms of what i call <b>"robustness analysis"</b>. But still this won't be really enough ... Tom
Hi Tom, Yes I figured PocketChange was a serious trader. Can you point me at the the part in your thread where it explains either how to reset the "mark" around which you trade - or how to cope with massive positions. I am testing a ShortOnly tarder in USDCHF but as that went up over the last year - my position is enormous!