End of our second week. Here is a quick update. All is proceeding fine so far. Also the dd has been reabsorbed. We are still using very little margins. Next week I will load new updated games, with some conceptual changes, which I will explain in the next posts. <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=142908&stc=1&d=1391817417" />
Here is an update at the end of the third week. All is proceeding ok so far. We are quietly scalping and growing up out "net gain" (the net difference between "Gain" and "Loss") while the PNL is still fluctuating up and down about zero. Four layers are currently experiencing some DD: TNA, UPRO, ERX, SLV. Since they are all short, I have just open, for each one, another layer with a "long only" position constraint. I am planning to get rid of UVXY as soon as possible, because it appears to be often too problematic when reversing, because too often it is not shortable. Also, it is a perfect replica of VXX, so there is no point in wasting our time with it. In a next post will discuss some changes I have just introduced in the scalping/hedging games and their rationale. Note that I have also made some cosmetic change to the PNL display, and now I am showing, by default, PNL, Unrealized and, in dotted green, directly the Gain-Loss ("G-L") net difference. <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=143188&stc=1&d=1392425304" />
Hi Fullautotrading, thank you a lot for you continuous updating. I am very eager to read about the novel game rules that you have developed recently. Please let us read about them soon. Thanks!
Hi AItrader, thank you. I have in mind a general reorganization of the player architecture. (The changes I will describe in this post are actually a prelude to a massive systematization of the various concept matured so far with experience.) Let me explain some changes I have introduced to make the scalping hedging games more intuitively appealing. First of all, let's say that these changes might be considered almost "irrelevant" respect to the general principle of our approach, which is to <b>systematically transform the hedging action into the future profits which create our edge</b>. However, giving a good "intuitive" structure to the games, may help a lot our intuition to create scalping /hedging games which are more effective in making grow the Gain at a faster rate respect to the Loss, and therefore increase the performances ($$$/Risk) over a given period. Also, when the game is "intuitive" to the fund manager, it is easier to follow it, and therefore to more easily understand and embrace the risk that we must necessarily face every time we trade ("psychological" side of the game). The main change I have introduced is a "classification" of entry types. While before I was differentiating between "scalping" and "hedging" entry, now within each of those 2 categories, I have defined 2 types. The following picture shows in intuitive terms the idea, with reference to the sell scalping side (for the buy scalping side clearly, just invert buys with sells). So we have: <b>"Scalping"</b> entry - Loading - Reversing <b>"Hedging"</b> entry - Protecting - Realizing <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=143217&stc=1&d=1392547750" /> In reality I am still pondering the actual analytic conditions which determine these different types of entries, and I will be refining them in the next weeks, however the general ideas are as follows. Assume you start with a short entry, that would be a "Scalping Loading" type of entry. Then, say that you experience some significant DD and some buys occurs to stop the bleeding. That would be an "Hedging Protecting" entry. Then assume that the price start reversing. The possible sells you make are, in this case, "Scalping Reversing" entries, while the buys that you make are "Hedging Realizing" entry types. The reason why I like to differentiate these types of entries is because, is that it is intuitively nice to have different entry and sizing rules under these different conditions. Clearly, it remains the problem of analytic determinations of these conditions and this can be refined by actually studying the game in action until we find the most suitable conditions. For the moment I have defined some basic conditions which should provide a good starting point.
Hi Tom I haven't gone through the entire thread but I wanted to encourage you to keep showing the full account PNL history. I know you posted something from the start of the year. If that is when you want the period to begin then leave that as a line in the sand. I seem to only see snapshots of 4 or 5 day periods. This isn't an exercise in trying to embarrass you but rather help your system build credibility. We all make mistakes and everyone goes through DD, leave you record in place with mistakes and all that way people who aren't following you blow by blow can ask questions or read about why there was such PNL behaviour on a certain date or period. Otherwise this thread feels like a scientific experiment that keeps resetting. This way the equity curve will rather express the evolution of your algo development journey. If I am mistaken and you do have a recent snapshot of the full folio PNL forgive me. I wish you the best of luck. Mick
Hi Mike, thank you for your advice. "this thread feels like a scientific experiment" is of course something that makes me happy, as <b>it is</b> (also) a scientific experiment. I am jut sharing with you a continuous journey of construction About the "keep resetting", when you will have time to read all the thread, you will see that I have shown the full and continuous history and equity curve, without missing any significant event or DD, up to <b>just 1</b> "reset". (Actually the DD dynamics and issues is what I am most interested in.) I have explained why I preferred to reset at that point: the changes in the game structure and program were indeed very large (including new devices to deal with some problems with order placement under blocking due to hitting the initial margin), and I preferred to start a "clean" session. So just 1 reset, and in the spirit of the scientific construction. So it's a little early to talk of a continuous "resetting". (And probably there should be no need of more resets.) The equity curve I show is in any case continuous and showing the whole history, even if I have not been posting an "update" each and every day. Sometimes there aren't significant changes from a day to the next, and in that case I feel like annoying you with a new screenshot everyday. However, if you prefer me to update the PNL screenshot everyday, I will post it on a daily basis in any case. It's no problem at all, of course. Here is a quick update: <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=143340&stc=1&d=1392885178" /> From a conceptual point of view, it is important to understand that the <b>conservation of trading information</b> is exactly the source of our edge. And therefore resetting a trading session is, in principle, a <b>full violation</b> of the methodology itself. For instance, many people have asked me if they can "daily trade", in the sense that they want to be completely "flat" at the end of each session. The answer is that (while I personally, most of the time, I would not feel the need to do so), they <b>can</b>, provided that they do that by <b>"keeping the trading information"</b>, that is simply neutralizing the currency position with an "offsetting player", and, then, closing it the next day, when resuming trading, so that the application can continue its logic of "loss recovery". Instead, closing all positions each day like with an ordinary "memory-loss" stop, even in profit, would be an open violation of our trading methodology, and would lead, in the long term, to an unprofitable outcome. In fact, <b>as with (almost) all strategies which do not have "loss recovery" mechanisms</b>, the result will be an equity curve that, respect to the probabilities involved, will be essentially fluctuating around zero (even if may do so with large variance), but with a significant negative drift, due to commissions, spread, interests and other trading costs. (The same strategies might however show positive results in backtest, of course merely as result of a process of curve fitting.)
Thanks for the feedback. Good luck. Maybe in coming weeks you will start uploading performance once a week on our platform. You can simply upload an IB XML flex query which will simply display your equity curve. That way a 3rd party company (ours) that has allocated more than $41 million to managers such as yourself can watch your performance track record unfold. I haven't mentioned our company as I don't wish to be accused of shameless marketing but Tom knows we are more than just casual observers we have invested $100's of thousands as an investor once with the GBot. Regards Mike
Hi Mike, well there are in general many websites which allow people to upload their equity curves. I personally feel this kind of approach to be meaningless for a precise reason that I will explain below. In my view, <b>if there is no reason for an edge, then there is no edge.</b> This is why to me is essential that "my investors" and scientific sponsors <b>fully understand the methodology</b>, first of all. Assume I have large funds and one person comes to me with an amazing equity curve (let's assume it be "real", and not backtesting, which of course would not make any sense at all). Would that alone lead me to allocate $$$ on his approach ? Of course not. We all know how large the variance of the phenomenon is, and there may be many ways to obtain a positive piece of equity curve. What I need to see (1) is the <b>reason</b> of the edge. Not some crackpotical argument or self-deception based no rational arguments ("prediction" and so on). And also (2) a meaningful and powerful <b>implementation</b> of the methodology (and we know that the mere implementation and testing of a trading engine itself takes several years of hard work, and often painful losses, for mere technical issues and low level details). Finally (3), there is a need of actual good <b>management</b> of the methodology within real world constraints, and that is mostly a matter of experience (and, of course, also intelligence of the manager, as <b>even the perfect tool can always be used in the wrong manner</b>). So it's a stack of issues which <b>all</b> need to be addressed and solved successfully, and that is why, if done as it should, trading is a science and an art, requiring immense efforts. On top of all that, you need to find the right capital to feed the trading machine, and it's certainly not a matter of $100K, or one might find himself run at the first issue or DD. It's a large investment involving a combination of both human and tangible monetary capital, and, in a way, what gives dignity to a process which might otherwise appear merely speculative to a superficial observer.
End of week. Here is a quick update. Not much happened since last update, we are quietly scalping and growing the "net gain" ("G-L"). <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=143401&stc=1&d=1393029744" /> On the development side, I am working on a general reorganization of the internal architecture which manages the players. Some preliminary test indicates a good increase of the speed of the trading engine (about 2x in simulation). (I am going to test it for a few weeks on my machine however before moving the production account to the new version.)
Folio update: <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=143558&stc=1&d=1393370069" />