Along with the broker report, let's summarize a few points of the approach used so far in this illustration.
Instruments and approach
We have been focusing here mainly on
Futures and
ETFs (lately also considering their
options).
As to
futures, we have been focusing on energy, metals, indexes, and also tried out some agricultural (the "weakest" so far). They have been traded algorithmically by using multiple overlaid
long/short layers started on suitably spaced "corridors", and
superposed long/short players.
For
ETFs, we have been focusing mostly on leveraged instruments which, directly or through correlations, behave
"inversely" with respect to the mkt. This part has been carried out mostly through "enqueued orders" as entries (while closing can be done either by automation or through "enqueued orders", or manually, as preferred.)
Lately, we have also been adding to our games some futures
options to play in combination with the position taken algorithmically (we have seen possible uses of various possible option configurations, such as what we called "player anticipation", "covering", "protecting", etc.).
(Clearly, one could also integrate with stock related instruments (and their possible dividends), in which case I would mostly go based on sound companies with low price (undervalued security), and mostly long position, say similar to WB's approach.)
As to the algorithmic game played on each layer, we are using
long/short players which are simultaneously scalping and hedging. We continuously store
trading information in order to be able to recover the hedging orders ("stops") as well the rollover orders (to recover contango/backwardation).
To be "profitable" in the long term a layer will need
most of its players to be able to close, which requires, at player level, either that the player is trading in the "right" direction, or that at some point in the future, the price
reverts to the entry point until the player can close. While this is "guaranteed" (in an infinite time horizon) for most of the theoretical models which make sense to describe the mkt,
in practical terms there can be good reasons why the rate of player closing is not enough to "catch up" with the unrealized, within our lifetime. One of such reasons is for instance a prevalence of players open against some
structural drift or other forms of
decay (e.g., if you insist shorting the mkt indexes, it's quite likely you will not do well in the long period, or, if you insist in going long on leveraged inverse ETF, you may never come back from the "underwater" word, as conceptually the price may actually be "sliding" to -infinity, even tough the behavior is actually "masked" by periodic
reverse split events, etc.).
Trading information and long-term drifts
Central to our algorithmic treatment, we have been storing and using the
trading information, the concepts of
layer and
player. It is also crucial to recognize the main
long-term "forces" which drive our instruments, such as possible
"structural" drifts, contango, backwardation, decay. In fact, since they represent the main "determinants" over the long term, it is impossible to survive long time in the mkt, without taking them into account.
In more conceptual terms, within this approach, failure to include them would mean that, while we do
store trading information, we would not be able to
use most of it, because the players holding it, would in practice often remain permanently "out of reach" or scope, and therefore this is practically
equivalent to losing information, which is the ultimate reason, and actually (in my personal view) the
definition itself of a trading "loss".
Our edge has been essentially the creation of a
"statistical drift" within the PNL curve, based on the
accumulation and use of the trading information, the incessant scalping action and various mechanisms to recover hedging orders (stop) and rollover orders.
The positive PNL drift (essentially represented by the slope of the
G-L curve) is opposed by a
constant and powerful negative drift composed by various components such as, first of all the hedging orders (approximately corresponding to "stops" in a more naive sense), the mkt sliding up/down (through drifts, contango, backwardation, daily rebalancing, decays or a mix of these) in situations to practically cause some of our stored information to
remain permanently unused, the trading expenses, interests and other various fees.
As we have anticipated, the
positive drift of the PNL is fed by various components, mostly:
- Continued scalping action
- Hedging orders (stops) recovery mechanism ("player superposition")
- Contango and backwardation recovery mechanism
- "Decay" of the leveraged ETFs
- Time decay / volatility drop of possible short options
- Possible dividends
The
negative component is fueled mostly by:
- Hedging action of players
- Loss of information stored in "castaway" players which remain "stranded"
and practically excluded from the algorithmic game, due to possible instrument
drift or
decay (if not properly taken into account with the "layer contraints")
-
Contango/Backwardation,
rollover expenses, where not recovered
- Time decay / volatility increase of possible long options
- Commissions
- Short interests
- Possible negative dividends
Capital and diversification wrt to price
A main ingredient is also to gradually
distribute capital over all the exploitable trading range, and the capability to hold on while the
unrealized fluctuates and we carry out our scalping action, which also means that it is essential that
suitable risk capital is employed in the process.
While instrument diversification appears, in theory, a reasonable concept, in practice, it turns out pretty clear that
diversification through price range is actually
far more important than
instrument diversification (because various correlations, eg. mkt/crude oil recent "lockstep", often make instrument diversification work more
against us than in favor of a "smoother" PNL curve).
The gradual allocation and the fact that we need to be prepared for rather large adverse PNL fluctuations (but anyway "bounded" by practical limits or by structural drifts) also signifies that the concept of
compounding cannot be applied in a
deterministic sense, but rather
statistically over a long period.
There is, however, the continued scalping action which provides fuel to fight the adverse fluctuations and the negative components, an this
can vary greatly depending on the instrument microstructure (choppier and more volatile mkts, like energy and metals, would generally generate more fluctuations and scalps). Clearly, this contribution is "masked" (especially in the short term), in the PNL curve, by the
fluctuations of the unrealized component of the PNL, but it is always present. For instance, in our case in about
half year the scalping action has generated an amount equal to
about 1/3 of the capital, and most of the time we have been using only a fraction of the capital (of course, we needed the capital to hold on the peak DD however).
This is also a reason why good risk capital is needed in this context, along with a working understanding why some degree of DD is always necessary.
