Trading with automation (with IB) - II - spreads

We are still doing ok. The PNL touched today for the first time $402K and currently we have only a few layers (NQ and a CL puts) in more or less deep "red". I have defined 2 "hedgers" to take care of NQ while we patiently wait that the "load" gets absorbed by the scalping activity:

Folio13.png


In total, we have made so far 1,614 fills in 64 (solar) days and spent about $11.5K in commissions. 39 layers are currently getting tickdata, while 20 are actively autotrading.

Equity10.png


Most of the money extracted is coming from the main algorithmic scalping/hedging action. The options are contributing a very little bit, while the ETFs practically nothing.

So far we have being using a maximum of $1,299,273.62 as "FullMaintMarginReq" (maintenance margin requirement), while currently using: $937,806.17.
 
PNL is fluctuating around $435K after about 66 (solar) days and 1,680 orders (max FullMaintMarginReq so far: around $1.3M):

Equity12.png


It is probably too soon to talk, but it seems that this version of the trading games is doing better than previous ones. Also, the concept of "hedgers" is not bad (and serves well the purpose to provide some psychological comfort against the drawdown).

I have been improving on it, by adding the (optional) possibility to let the "protective players"open even on the hedgers (while previously only entries to hedge were allowed), and that seems to work ok. I am trying that on the 2 "hedging layers" for NQ:

Folio14.png


I think that the layers used as hedgers, when no more needed, can be "released", and let work according the algorithmic game, free to recover their open players too and to make their own games.
 
Last edited:
tradegee said:
Hi. What was your initial equity? Sorry if I missed that.
More...

Hello tradegee,

thank you for the question.

It was indicated on post #1. Let me copy and paste for you:

"Most of the time, we will be using a small fraction of the available funds ($6,172,520.74), so for performances purposes we might, for instance, use ratios which use (at denominator) the maximum decline of the NetLiquidation value, or similar stuff."

So far the maximum "FullMaintMarginReq" used has been:

FullMaintMarginReq: Max: 1,299,273.62

(the figure is transmitted by the broker.)

Recently, I have increased the packet sizes to increase the amount used (which will cause again large PNL swings.)
 
New week with some roller coaster moves. The combination of significant mkt moves and larger packets has caused a quite sharp drawdown (DD), which the application is now slowly converting into new profits.

However, most of the DD was not caused by the main algorithmic strategy, which is pretty good at hedging, but the surrounding (manual) strategy on options which is much more susceptible of powerful swings (and they hardly translate into new profits).

Currently PNL around $427K (in about 78 solar days), executed 2,084 orders, comms: $16,2K, max "FullMaintMarginReq": 1,446,645.19.

Equity13.png


If we consider a strategy performance ratio of the type: avg profit / max DD, we can order the various strategies played here as follows:

1. Algorithmic scalping/hedging (this is produces the smoother PNL curve, in relative terms, because the entries are just designed to create the maximum hedging action, while preserving the possibility of profit).

2. ETFs (decaying instruments): manual games on local peaks. This can produce quite large DD (on the other hand, the instruments are too "small" to be really harmful).

3. FOPs (futures options): mostly short selling "around the main game". This can produce the most powerful DD, due to the huge volatility of the instruments. If you play this, make sure to make entries very far away (like 300-500% on local peaks or more). This component can definitely spoil significantly the "smoothness" of the PNL curve.

Current state of folio (PNL individual instruments):


Folio15.png
 
Last edited:
Easter holidays passed, but not here, as we have an additional holiday day. In the meantime, we have been breaking our previous high-water mark, touching a peak profit of $730K.

Currently $723K profits, made in about 84 solar days, using max margin requirement ("FullMaintMarginReq") of $1,446,645.19. About 18K in commissions, for 2,258 orders.

Return ratio: profit/max margin = 0.5 = 50%, in our first 84 solar days. (Max DD has been pretty sharp, largely emphasized by the naked option position. The good side of it is that it lasted only a few days.)

Equity14.png


Situation, instrument by instrument:

Folio16.png


quite good, with NQ layers slowly catching up, turning the DD into new profits.
 
Our non-predictive approach is slowly paying off. Profits have just spiked to $1.2M, which is nice for 87 days in the mkt and using a max FullMaintMarginReq of about $1,4M. Commissions: about $20.5K for 2,487 fills.

Broker's report (html, zipped) attached.

Equity15.png


Folio instrument by instrument:

Folio17.png


Note that the IB document reports a slightly larger profit. But this depends on the method of PNL computation (they use a bit too "optimistic", imho, approach).

In this case, we just had a spike up in the PNL, just as previously we experienced some sharp DD (a "load up" phase). This also shows why, as we noted in post #3, the Sharpe ratio is not a too meaningful metric for pure algorithm performance. Obviously, variance can work both in your favor or against your interest. And, in addition and most importantly, most often the relationship between DD and profits is not linear, and basing optimization on Sharpe may lead to choose practically nonviable strategies.
 

Attachments

Last edited:
Currently, the most performing layer has been one trading ES (for this, I have increased the "packet size" traded to 5 contracts).

In the two pictures below the same ES layer is shown, in two different "views".

The first view shows the so called "players" (mostly "protective"), which could very roughly be interpreted as permanently sitting "stop loss orders", waiting to be recovered sometimes in the future.

The second view is the global "order cloud", comprising both the above players and and also the players which have already been closed. The closed players are no more considered from the algorithmic point of view, because since they resulted in profitable trades, they can be "forgotten" forever.

Open players only:

ES6.png


All orders:

ES7.png


In very intuitive terms, the recovery of what could be roughly interpreted as stop loss orders has the purposes to unbalance the 50/50 win loss ratio, typical of a purely random trading activity. On top of this, obviously the trading action is not purely random, but reacts instantaneously to price direction changes, pretty much as a "human" trader would do (with the difference that the human could quickly "lose track" of his trading activity, with the practical possibility of ending up with an unprofitable "order cloud", while the application instead is, at all times, checking the internal "consistency" of the order cloud generated, according to the programmed rules).

In this sense, the executions (all lmt orders) on full automated layers are completely determined by the flow of tickdata, and there nothing else intervening. No "prediction" whatsoever is used.

Clearly, what is very complex and delicate is the right balance between protective orders (hedging) and the scalping activity.

Just because, in intuitive terms, too much "stopping" would not allow the scalping profits to "catch up" with the hedging losses (which are continuously generated).

Another thing to watch out, at supervision level, is that instruments traded must not incorporate strong "structural" drift/decay (such has leveraged ETFs, or options), in which case the algorithmic games could be adjusted in order not to allow one of the two buy/sell fronts (most often the buy side) to grow too much, or else it would be impossible to recover the open players (mostly buy) due to the fast instrument decay. But it's quite difficult and time consuming to do.
 
You must log in or register to reply here.
Back
Top