Trading algorithmically a folio without stops (with IB), real $$$

[jb514]

Not to kick you while you're down but early I brought up some points that you disagreed with. I was arguing that you are in fact synthetically trading volatility. One interesting point is that a lot of commodities, since they are traded overnight, often trade in a tight range and gap overnight. Since you are trying to make more money scalping(short term volatility) than you lose when the instrument trends(long term volatility), you wouldn't want a symbol that makes its biggest moves overnight when you can't scalp it.

SLV is a good example. There are a lot of days where it trades in a tight range while it makes the big moves overnight. You'd want symbols that make the biggest moves intraday where you can scalp them. Look at equities. There are a lot of very inefficient symbols with room to provide liquidity...

 

[fullautotrading]

.... you wouldn't want a symbol that makes its biggest moves overnight when you can't scalp it.
...
SLV is a good example.

...
You'd want symbols that make the biggest moves intraday where you can scalp them. Look at equities. ...

Hi jb514,

Yes, undoubtedly the choice of a good folio is a central issue. I feel that most of the time they have to be actually "tried out" to be able to make a selection. Of course, the suitability, may also depend on the features of the scalping/hedging game one has created. In principle one could devise different games for different instruments, or tune specifically rules and parameters for each one instrument, or a class of instruments with similar dynamics.

About gaps, so far I was not much bothered by them, maybe because they can be both in favor or against, and on the long period may tend to neutralize their effect. SLV, so far, did not give too much problems (but it might in future), maybe because it trades in "extended" hours too. Not quite like a futures, but in fact it is possible to trade it I think from 4am to 8pm. (There is an handful of instruments which are like that). What I found bothering so far is stuff which is either too "slow", or, on the other extreme, too crazy and wild (like the gas in our case).
(Of course if one is uncomfortable with some instrument going overnight, or maybe there is a dividend issue, the corresponding position might be momentarily closed by opening an opposite player, and possibly undone when mkt reopens.)

If you have more candidates to suggest for the folio, do let me know, we can put them under watch for a while, and in case give them a try.

 

[TskTsk]

> I see now that options are only used to hedge deltas (thetas neutralized)

That's right. Since, by assumption, we don't want to allow ("memory-less") losses, we are not allowing paying time decay. In case of assignment of some option, the corresponding position will be transformed ("fill projection" facility) into an ordinary player and will only close in profit (as, by definition, all players).

> Does it transmit one and one order

That's right. The app never leaves "presubmitted" orders (in automatic mode): all automated orders are either executed "immediately" (say, within a given reasonably short timeout) or they are discarded (ie., there will never be "several active limit orders ", but only one).

[There is actually the possibility to "presubmit" orders too, but that is a discretionary feature. Or, there is the possibility to "enqueue" orders which will be followed automatically by the app in the future, but this is also another discretionary feature, which we can ignore, for simplicity, for the moment.]

All the colorful squares or circles you see in the "order cloud" (in "order-view mode") are only (already) executed orders (blue=Buy, red=Sell).

Each order placed is actually the result of the activity of the logic layer created by the superposition of the "virtual players". In practice each player will have its own requests like for instance, "open" some positions, or "close" some positions, and so on. All these multiple requests are "pooled" (algebraically) and channeled into a unique physical order each time (say at each tick, or in between ticks). Then, on execution, or partial execution, the possible fill is "redistributed" to the multiple players accordingly to their requests, thus changing their virtual states.

If it can be more expressive, you can think of this like (mathematically) equivalent to having multiple independent "real" individual traders (they will be "partners") working on different and distinct accounts (possibly using the same financial instrument), but talking continuously to each other, and coordinating their actions for reciprocal "protection", while nevertheless also pursuing an "individual" profit, whenever possible, but assuming they are sharing a financial interest in maximizing the overall profit (say, they will share the sum of individual PNLs) and minimizing the overall risk.

Thanks, that clears it up a bit.

 

[fullautotrading]

Here is a quick trading update.
Today has been the first day of the process of "cleaning up" the mess of yesterday.

I have begun first of all by removing all layers with positive PNLs and started testing some mechanism for stricter hedging (for wilder instruments). Tomorrow, I should continue and remove all the messed up layers, including those with negative PNLs.

In principle, of course, we would not want to do this in production, with a well consolidated scalping/hedging game, in order not to lose the "past trading information", which is the essence of the method (and the "net gain" accumulated). However, since we are in an experimental stage, and I want to start testing some changes to scalping/hedging game, it's more convenient to just start with "clean" layers, instead of messing on the existing ones.

So basically we will "set back", gradually restarting the process as we did in the very beginning.

Tomorrow, when finished removing all layers (except perhaps a very few which can be kept), I will provide the complete list of them with the respective PNLs.

In the meantime, I have added several features which when coping with these difficulties have appeared to useful. Such for instance the possibility to selectively close all the positive Buy or Sell players. Also, with the occasion, I have added a new order sizing mode and have perfected the layer closing procedure (close all) and some details on dealing with orders rejected due to hitting the initial margins.

For the future, the following ETFs instruments certainly will not be used:
DRN, SRS, EEM, IYR, TWM (too "slow", and essentially strongly correlated with indexes, they tie up margins for long time, with scarce or no results). Also several instruments which are mere repetition of mkt indexes should be avoided. TVIX and SVXY may be other ones to avoid, as VXX can do much better the job.

The next challenge is to be able to make a game capable to cope with the crazy runners like UGAZ, GASX and similar (which have essentially caused the problems), but that is a job to be made in small pills. I will paper trade those instruments separately for a while to perfect a suitable scalping/hedging game for them, before resuming the fight in our live folio.

<img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=142445&stc=1&d=1390523417" />

 

[fullautotrading]

<b>Plan for next test</b>

I have shut down everything on the current test. Just after the clean up there was a very large mkt move (which actually could have completely settled the FAS situation).
I just re-opened 1 single layer on VXX, in the attempt to exploit something of that move. Anyway, <b>let's move on and focus</b>.

From Monday I will start a new test organized as follows:

*Instance 1 of the app - <b>Real $$$</b> - We gradually restart trading a clean folio

*Instance 2 - Real $$$ - I will just trade the last VXX layer and then discard completely the instance as soon it closes its players

*Instance 3 - PaperT acct - I will trade only some layers over there to test specific games for "wild runners". The layers will be UGAZ, GASX, FAS (as etfs) and NG, HH, SI (as futures). [If you have other candidates suitable for this section let me know, of course]

Based on how the instance 3 goes, we might consider reintroducing those instruments in the real $$$ folio.

For Instance 1, the new folio. I will start with a few instruments. We will on particular:
- avoid too many replication of same instrument
- avoid too large orders (I have added a constraint to the app, for this)


<b>Games</b>

From the previous experience, I gather that there a need of differentiating among games, as "one game fit all", seem either a bit difficult, or in any case inefficient. While I have added a mechanism to automatically <b>scale the whole scalping/hedging game with volatility</b>, I think there is still need of some (at least basic) <b>structural</b> diversification.

So will start making <b>2 simple classes</b> (later we may be differentiating more) and use 2 different games for the 2 classes.

- <b>First game</b> (let's call it for now "<b>Bias</b>"): for <b>leveraged bear ETFs</b>. This game will be applied to instruments such as ERY, ZSL, VXX, TZA, GLL, FAZ, TBT .... and similar.
The game will be accompanied with a "short only constraint" on the global position (in future we may reevaluate that)

- <b>Second game</b> (let's call it for now "<b>Shield</b>"): for <b>leveraged bull ETFs or more general use </b>. This game will be applied to instruments such as ERX, UCO, TNA ... and similar.

The current ideas it to give the first game ("Bias") a less hedging action and allow it for more unilateral DD, as we try to exploit also the <b>intrinsic drift of the instruments</b> and the decay architecturally built in those special instruments.
The second game ("Shield") would have a stronger hedging action, which will be obtained by overlaying opposite players so that the position would be continuously neutralized or "inverted" depending on direction, and we grab mainly the inner fluctuations.

Finally, there will be a <b>3rd game</b> (name to be given) which we will be building up to deal with the more "crazy runners" (ugaz, ng...), and I will adjust it dynamically on paper, until I get it's doing a good job. After that, we might move some of those instruments on our real $$$ folio (instance 1) and start trading them with small packets.

 

[fullautotrading]

First day of our new test.
I started 14 ETFs among those which seemed to do better based on the past experience. I also divided them into 2 groups, assigning 2 different scalping /hedging games ("Bias" and "Shield" as mentioned in the previous post):

They are:

"Shield"
ERX STK SMART DIREXION DAILY ENERGY BUL 3X
UCO STK SMART PROSHRE ULT DJ-UBS CRUDE OIL
TNA STK SMART DIREXION DLY SM CAP BULL 3X
SLV STK SMART_USD ISHARES SILVER TRUST

"Bias"
DGAZ STK SMART VELOCITYSHARES 3X INVERSE NA
DGLD STK SMART VELOCITYSHARES 3X INVERSE GO
ERY STK SMART_USD DIREXION DLY ENERGY BEAR 3X
FAZ STK SMART DIREXION DAILY FINL BEAR 3X
GLL STK SMART_USD PROSHARES ULTRASHORT GOLD
TBT STK SMART PROSHARES ULTRASHORT 20+Y TR
TZA STK SMART DIREXION DLY SM CAP BEAR 3X
UVXY STK SMART PROSHARES ULTRA VIX ST FUTUR
VXX STK SMART_USD IPATH S&P 500 VIX S/T FU ETN
ZSL STK SMART PROSHARES ULTRASHORT SILVER

Later will add more layers, but for now I wish to follow these, to tune some parameters of the new games (especially the minimum distances between entries, and the order sizing rules).

In the meantime I had some ideas to make easier the use of different games for classes of instruments (later I will explain the changes I have in mind).

I also changed a bit the display of the PNL, making optional the display of Gain and Loss, but adding the "NetGain component".
This way, by default it's shown: PNL (cyan), "NetGain comp" in green and unrealized in orange. I think this way is better, as the PNL is less "flattened", and actually we care about the difference between "Gain" and "Loss".

<img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=142544&stc=1&d=1390862896" />

 

[fullautotrading]

<b>Game management reorganization</b>

While the folio is quietly scalping, these last 2 days I have been working on some architectural changes regarding the "management" of the "games".(These changes will be reflected in the next app. update).

So far we had a "shared game" for the folio and, then, for each instrument it was possible to specify a "specific" scalping/hedging game.

Now, in light of experience, it seems more useful to have a different organization. In fact, it appears clear that is is more convenient to classify instruments in a few categories and then assign to each category a specific game.

Following this perspective, the new game organization is as follows.

Games are defined separately and placed in a list. The manager can anytime add/remove/edit them. Then, for each instrument it is possible to select the game it will use.

If a game is changed, all the instruments which are using that game, will be affected.
For instance, if I have assigned the game called "Bias" to the Ultrashort ETFs, when I modify a rule of the game (for instance the minimum distance% between entries), all the instruments which are using that game, with benefit of the change.

This organization appears to be a simplification. In fact in the previous approach, each
instrument had to be changed singularly, which was kind of annoying, and error prone.

<img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=142600&stc=1&d=1390994525" />

 

[fullautotrading]

Here is a quick update for our 4th day.
I just added to the initial 14 instruments a replica of vxx, dgaz, uvxy, and just a layer to watch sco (not activated yet). (After all, even though there are hundreds of ETFs, it does seems that the number of instruments decently tradable by scalping is not that large.)

So far we have been using a small fraction of the margins (and will be gradually increasing). Looks like the new games do a better work at hedging (respect to when we started this journey) and it's particularly good to have imposed a maximum size to the orders.

About "cloning" the instruments, I believe that it is better to "clone" than to increase the trading packet of a single layer, because, this way we still trade with a larger size, but we can take advantage of a possible reduction of the real position in the account (that is take advantage of the "accidental" hedging effects caused by overlaying long/short positions).

I am working now on some new ideas to "refine" the entries and make the games more effective or, at least, intuitively appealing. In fact it seems that the current distinction we have in "scalping"/"hedging" entries can be better articulated.

Displaying the Gain-Loss curve alone, with the PNL and unrealized seems much more effective and I am pretty happy with this change (I will also change the figures currently shown in the trading monitor, by replacing Gain and Loss with "G-L", that is their net difference.)

<img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=142657&stc=1&d=1391122902" />

 

[fullautotrading]

End of our first week. Still warming up.
Today we started with some DD, due to VXX pushing up. Then a few fluctuations of the mkt allowed to quickly recover some of it (through hedging buys).
I have added another copy of uco and tna, and still merely watching sco.

<img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=142679&stc=1&d=1391212198" />

 

[fullautotrading]

Today the mkt had the "worst drop since June":
http://www.reuters.com/article/2014...SBREA080LL20140203?feedType=nl&feedName=usdai

Our folio with a large component in volatility has experienced some DD.

These kind of occurrences do illustrate the scarce capability of the "short constrained" instruments to deal with large and violent unfavorable moves. Also, in this case, having put a (small) limit to the order size has worked against us, as the BUY hedging orders have been too "weak" to effectively contrast the adverse move.

This might also suggest some more tuning at game level, and I am currently thinking and actively working on these aspect. A possibility might also be to replace the "short only" constraint with a milder (and probably safer) condition at game level (eg., max difference between buy/sell players) which would allow some reversion (meaning the position to become temporarily long) in case of large moves like these, but also easily returning to being generally short.

<img src="http://www.elitetrader.com/vb/attachment.php?attachmentid=142760&stc=1&d=1391475275" />