Sure. Here is a summary of the latest changes to games, to make their setup more intuitive for a fund manager. While at earlier times, I was differentiating between "scalping" or "hedging" entries only, now the classification is slightly more articulated and goes like this (terminology is "made up"): SCALPING ENTRIES - "Loading" - "Reversing" HEDGING ENTRIES ENTRIES - "Protecting" - "Realizing" The charts below summarize in a purely intuitive manner the meaning of these types of entries. - "Loading" <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=144034&stc=1&d=1394450376" /> - "Protecting" - "Realizing" <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=144035&stc=1&d=1394450376" /> - "Reversing" <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=144036&stc=1&d=1394450376" /> The reason why it seemed good to differentiate between some entry types is that a fund manger may feel to size differently the orders depending on the current situation. I will never stress enough that our goal is simply to <b>keep under control the PNL components (growing the G-L and bounding the Unr), in order to allow our drift to work</b>. <b>Never ever</b> interpret this as any attempt to understand or "predict" the mkt based on the past price data stream. (Let's live "prediction" to coffee ground readers, astrologists, naive users, and other various crackpots. Even if the "arrow of time" may be an illusion as B Greene recently tweeted, it's a matter of fact that humans are prevented from simultaneously perceive what we call past, present and future.) Simulations show clearly enough that there are games, which may be completely counterintuitive to a human, that can still do the job, so this kind of classification has only the purpose to make "easier" to a fund manager to create scalping/hedging games which are "understandable" to him, and therefore also more sustainable at a psychological level. In fact, if a fund manager has a feeling he is "losing control" on the trading logic, he will probably "get lost", lose the sense of his ultimate strategic goal, and, on fear, "pull the plug" (rightly so). Note that this is a fully dynamic scenario. For instance, in the last picture, when the SELL players (entries) start being closed (price going down) the still open BUY players will be those "losing", and therefore the new SELL player (entries) will automatically switch to "Hedging-Protecting" mode to "protect" them. And so on. Some rules may also be differentiated depending whether we are on the BUY or SELL side, for the purpose of construction of <b>asymmetric</b> games (like "bias").
Hi crayon851, it seems to me the session is proceeding fine, according to plans. We have been keeping a constant upward "drift" ((G-L)/L in the range 7-10%), accumulating about 10K of gains, at the same time the unrealized has been contained within a decent range. To keep it that way it is important to continue the scalping with a dynamic but smooth hedging action. Any action which is too "dramatic" (for instance any "relatively" large order) is usually highly counterproductive. Anytime you make on order which is relatively too large you kind of create a "local inefficiency" in your logic (I have experienced this several times). As it appeared in the first sessions (it was FAS in that case) it looks like the indexes relates stuff (TNA in this case) is what can be more bothering (longer upward moves). Probably, it would be good next time not to use too many of those mkt correlated components. This session appears slower as I have been using a larger distance between entries (about a 1% minimum distance, when volatility = 100%, and scaled linearly wrt volatility and position). It can probably be reduced, to grab more "noise". [I have been adding this week an "hard" limit to the minimum distance between orders, to avoid that some extremely low setting or mkt condition could create a conflict with the MM algorithms logic (I saw that with ugaz.) I have specified 2 times the avg spread as lower limit. It should provide some reasonable protection against user abuses or other malfunctions related to entry spacing and their scaling up/down logic.] <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=144138&stc=1&d=1394612342" />
a lot of that was over my head in terms of what terms youre using example FAS, gl, spread between trade( i know what spread means but not sure in the context youre using). i assume your algo is compensating for changes in market cycles? whenbyou say hedging, are you referring to your algo purchasing products that are counter to other positions? in terms of positive drift, are you referring to the assumption that the market has been increasing, regardless of the down moves? whats the objective of your algo? is it to mimic the indexes?
Here is a quick update. Load is gradually shifting on the side of the ultrashorts. The folio overall picture is staying decently in good "balance", with the G-L steadily growing. All according to our plan so far. <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=144194&stc=1&d=1394743293" Hi crayon851, Spread: Ask - Bid FAS: an ETF, see: http://www.direxionfunds.com/products/direxion-daily-financial-bull-3x-etf > when you say hedging, are you referring to your algo purchasing products that are counter to other positions? Right "hedging" entries would be those made against the group (buy or sell) of players which currently is prevalent as to losses >whats the objective of your algo? Objective is essentially to steadily grow the G-L (net gain component), while keeping under control the unrealized, within an architecture which is, by design, continuously unbalancing the gain component against the loss component.
I think I have a couples of pictures which can be of help to who is still struggling a bit to properly envision the mechanism of "loss recovery" (or "<b>Conservation of the trading information</b>"), and why here at on some posts we arrived to discuss with jb514 the (perhaps too audacious) analogy as if the stranded hedging players (say = memorized stop loss) could be seen like a sort of "artificial option" surrogate (without decay). Look a this picture, which is a <b>real $$$</b> test, made on one real accounts. This picture represents the last 100 days of VXX to this very instant (each vertical blue line indicates a new week). <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=144197&stc=1&d=1394745947" /> Look at the cyan circles and squares on top of the price curve. They represents the (still "open") buy players which served to hedge (note, for instance, that the min negative position was just -400 shares) when the price was going up. Many of their buy "peers" were luckier and could close in profit (you cannot see them in this picture, but they are shown in the picture below), but these guys unfortunately remained "stranded" above, because they never got a chance to close in profits. Never mind. We just leave them there. But <b>we don't "forget"</b> them. Imagine now the price start going up again, now we have some "protection already in place", and they are exercising they protective function. This way, in the first place 1) we are not duplicating some of the stops, and, furthermore, 2) a significant portion of the stops gets "recovered", in the neverending up and down of an instrument (and, of course, new "stops" (hedging players) get created too). If those were "memoryless" stops, there would be no way to survive in the long run. This should make hopefully a bit clearer why I was talking of of "memory-full" stops. Clearly the picture above shows only the <b>"open players"</b>. The whole set of orders looks like this (this is the same picture, but showing <b>all orders</b>, and not only the open players): <img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=144198&stc=1&d=1394745958" /> But we don't need to care about the "closed" players, as they are "gone" forever, closed in profit. So it's a virtual "player superposition" but it works just identical (algebraically speaking) as if they were traders playing on separated accounts.
Tom, sorry to drop off for a while. Thanks for the detailed responses above. I had a very busy work week and it looks like it might be a couple weeks more before I can download the software and test it. Regarding the above quote, I think what you are saying is the same as the basic argument for those who espouse market neutral strategies. What I am wondering is, do you try to predict non-directional things, like volatility? For example, when you talk about hedging/scalping too loosely or too tightly, that would seem to me to depend on volatility. Put another way, the "tightness" of entries and exits is profit maximizing when you get the maximum number of large size moves traded. Whether or not you are explicitly optimizing profit in that way when you adjust hedging, it seems it would be worth looking into. Some traders on ET talk about adjusting trade size based on volatility (e.g. the level of the VIX), maybe you could experiment with backtesting a hedging/scalping frequency that varied with the VIX?
Hi fullAutotrading, it is really evident that the trading strategy implemented in G-bot has attracted a lot of attention but not a shared (that is the keyword here) understanding of how G-bot operates. I then would like to suggest that you try to provide an algorithmic commentary to some of the graphs posted above so that it could become clearer why G-bot has made a particular choice at a given moment. For instance if I take the latest open order and single player graph (it is two graphs above), could you explain moving from left to right the logic used by Gbot in making the opening, closing, hedging and so on for each of the points in there? If the open order/single player graph is not suitable to such a description (because it only summarizes an historical situation whose points have no immediate relationship one with the other), then please do the logic explanation on a single player / any order graph. In that one the time/order/left-to-right relationship between points is governed by the logic of G-bot. It summary I believe that for a human being to be able to understand the Gbot behaviour, she/he should be able to simulate by hand step by step what is happening on the graphs that you are reporting. Until then everyone is bound to interpret what a cloud ("cloud" as those in the sky or as those displayed by Gbot) means according to their own personal feelings. To finish a minor question regarding the current trading session, how much capital is currently being traded by G-bot? Not the total in the account, but that actually used in buying assets. I understand that after the "restarting" you were adding trading capital chunks by chunks (so you were spreading the investment entries over a long period of time) to avoid that a particularly adverse temporary market condition would negatively impact on G-bot behavior. Thanks! Keep us posted!
@monkeyjoe <b>>do you try to predict non-directional things, like volatility</b> There is no form of prediction whatsoever here. You can safely remove the word "prediction" from your vocabulary when discussing in this thread without any harm to understanding of the concepts. <b>> For example, when you talk about hedging/scalping too loosely or too tightly, that would seem to me to depend on volatility</b> That's right. We do use volatility (its <b>current, realtime</b> value), for instance to space the entries and "shrink" or "inflate" (dilatation) the order cloud. Clearly, that has no relation with prediction, it's just a way to "scale the game", or else, if one did not scale the game according to volatilities, for some more volatile instruments you would have relatively "too many tight scalps and close entries" and possibly too large dd, thus unbalancing the folio or creating various other sizing problems. Vice versa, the less volatile instruments would be trading too "slowly" ("too large" scalps and "too much space" between entries), clearly in relative terms. (Also you might enter in "conflict" with the the logic of some MM algos, as you might be trading too tight and dig a large loss locally). What we want is a just a "scale invariance" property of the order cloud. Where volatility here is a scaling factor, along clearly with the tick value of the instrument. <b>> Some traders on ET talk about adjusting trade size based on volatility</b> Right. That is what the game bot is doing. When you shrink/inflate your order cloud, you are also consequently automatically adjusting your player positions (short/long) to volatility (and tick value).