Can you give a summary? How big is the account? How much P/L%? In what timeframe? How much is the MDD% of acct? What are your expectations for the year?
Hi botpro, thanks for the question. The general initial information about the account is contained in the first posts of the thread. Anyway, we had a 1.5M of which we have been using only a portion, with increasing (and varying) margin requirements (for instance, current requirement is around 500K). As explained during the thread, the reason for this gradual money allocation is to be able to trade at the various price levels, in different "corridors" and with different layers, and thus be able to scalp in the whole instrument range, even when there is an accumulation of negative players (a "load phase" or "investment"). "Timeframe" does not really apply here. Or, if you prefer, we could also say we use all possible timeframes. In other words, a player remains active until it can make its profitable close (see initial posts for more details on players and approach), according to the programmed "game". Here, price rules, not time! The maximum DD was reached with the recent simultaneous drop of crude oil and mkt in "lockstep". This was strongly exacerbated by the manual "overload" of SCO, which was actually responsible of most of it (see pictures in previous posts). In this case, we peaked about 40% underwater. As explained, there was partly a mistake with the positions taken with SCO, but on the other hand the crude oil price was attractive, so taking some extra risk may have been appropriate, rather the problem was a miscalculation of SCO volatility and the too close initial "scheduled" entries, which are still causing some little problem at the moment. We have been running for 167 solar days so far, and as to expectations for a specific timeframe, it's usually not meaningful to make a prediction for a given precise date in terms of PNL, as, while we accumulate scalps, it is subject to swings due to the natural mkt volatility. So what happens "exactly" after one year depends on what kind of "phase" we are we the various instruments (we might fall in a period of heavy "load up" or we might be in a period where we are finally closing our trades). What really matters is being able to constantly grow the the G-L (gain loss) curve so that we establish a positive drift, and, after a while, the mkt price swings are less capable (statistically speaking) to "disturb" the PNL drift.
Hi recession2016, Yes, that's wise, and I did that when I was using real accounts kindly provided by investors. Since now I only show paper trading accounts in public threads, to avoid interference due to the intervention of some clueless idiot that punctually starts crying at the first DD (this happened in the past with just a 15% DD), thus possibly scaring the account owner and making impossible to pursue the desired goals, it is now not important to hide it. Actually, in this case, making no secret of it, enhances the transparency of the documentation (as I periodically also post the entire broker report), and there is no possibility of "selective reporting" (a typical plague on the web when trading is concerned). The general point on DD, is that it is actually a condition necessary (but not sufficient) to make long term profits, and I could make a rather long discourse to explain why this is so from a technical/scientific point of view. Unfortunately, the less experienced trader/investors is often mislead into believing that you can become easily rich with negligible DD (and associated existential pain), and even with small capital, which actually it is not the case. Anyway, I will stick to the practical illustration and avoid to start discussing that, as anyway anyone has to make his own journey.
Thanks fullautotrading, by "timeframe" I just meant in what time the P/L has been achieved, sorry for the confusion. Yes, I understand you very well about your saying "price rules, not time", as this is also reflected in my own sytem; it too doesn't use any fixed timeframe, the price action is the more important to watch for. Hmm., an MDD of 40% in 167 calendar days is really big. Is that figure based on the full account size, ie. 1.5M + P/L, ie. what was the peak account value before the MDD stroke? I would suggest to reduce the future MDD to less than 15% (or even <10%). It is IMO possible by using more diversification, ie. distributing the total risk over more titles by adding more titles into the basket.
Well, time is also to be factored in somehow, but I prefer to exploit it in form of price decay (eg. through the decaying ETFs, using "enqueued" orders), or by grabbing the option time decay (which we have not used so far in this illustration, so far). Yes, it was a bit large and actually exacerbated by the SCO position, which, as explained in the thread, was outside of automation and mostly my fault (manually "enqueued orders"). Another reason was the simultaneous load on mkt indexes (CL/ES strong correlation). SCO alone represented more than 34% of the whole DD. The larger PNL seen so far is +169.8K, while the current G-L (gain-loss, that can be interpreted somehow as max current "potential" PNL: what we have scalped away) figure is about 478K. The DD was anyway ok as the crude oil price was pretty low. (The only problem has been the "cluttering" of SCO positions in a too small price range, and we have now enforced a spacing "rule" about that. But we should recover that soon.)
Talking of grabbing some time value, today I have sold a few options on CL and ES just as "warm up", and easy small money since the expiry date was today. Actually, when the position becomes significant (and, in time, it usually does, due to the need of recovering the rollover orders), it may not be a bad idea to also use options within our games. In some cases they may be "covered" by the existing position. In other cases, shorted options can be interpreted as an "anticipation" of a future desired position at given price levels. If the options remain OTM we just grab the time value, if instead they expire ITM we get (by assignment) our desired position (well, if cash-settled obviously we just inject the corresponding position), which we can then transform into a player (or a "split sequence" of players) and leave it to the automated engine. This can make more sense when an instrument is close to its end range (like CL some weeks ago). CL is on the way up and several buy players could close. SCO instead has still way to go due to excessive cluttered load at low level: this mistake will be "erased" and "forgiven" only when there is enough reversion to close the short players which are "holding memory" of the trading actions. Most of the other load is now on natural gas NG, which has been falling down for several weeks. Actually, I should also roll it over and I was just waiting for some small spike up to proceed (as the contango is pretty large). DGAZ also starts to look interesting, and this time we will act more cautiously when spacing the "enqueued orders", noticing that its level of (avg) volatility is practically the same as SCO.
Another week is gone. We have started taking notice of the passage of time by writing some options. As said before, there are several ways to incorporate the use of options along with algorithmic trading. Some of these possible ways are the following: - "Anticipation of players" at a given price level (in his case we would short options with strike equal to the desired price). This can make sense for instruments that can be considered near the extreme of range. In case of instruments with drift, we would preferably use only PUTs (in case of positive drift) and preferably at low prices. - "Covered PUTs or CALLs", depending on the position taken by the algorithmic engine. With drifting instrument we would preferably use only option possibly resulting (on assignment) in a position along the drift. - "Protective options". In this case we would preferably create an option "collar", by selling ATM option and buying OTM farther options, in such a way to completely cover the decay of the long options. Clearly, more ideas are possible and with practice one can become pretty good at creating useful configurations. Our situation in the meantime is pretty fine. The only note still out of tune is SCO, but sooner or later those sell players will close too. CL has closed practically all buy players, which means that even if it goes down again to very low price, it won't be able to create any fatal damage to our account. NG is still going down and currently it represents (apart the anomaly of SCO) the maximum load we have currently open. We have been steadily increasing our G-L (gain - loss, currently around 480K) with a rate of about 7%, which is not bad considering that the simultaneous drop of crude oil and market indexes, aggravated by the SCO position has forced us to dampen drastically its rate of grow. Apart the disfunction with SCO, which, as we have noted, has caused an excessive peak, in abstract terms the PNL curve is pretty typical: We have in essence a drift which underlies the PNL curve and a number of fluctuations caused by the natural volatility of the mkt. Clearly, the extensions of these fluctuations also depend on the max position that we allow on our layers. In time, the statistical drift will pull the PNL curve higher and higher, so that, after a while, the PNL fluctuations should not be able to drag the PNL curve "underwater" anymore. It may also be interesting to note what are the components which essentially create the drift: - Continued scalping action - Hedging orders (stops) recover mechanism ("player superposition") - Contango and backwardation recovery mechanism - Decay of the leveraged ETFs - Time decay of possible options The PNL "drift" is represented by the slope of the G-L curve (dotted green in our picture), which can also be interpreted as our current maximum potential profit: PNL = (G-L) + Unrealized The "unrealized" is obviously the one causing the fluctuations in the PNL curve. This dynamic is pretty typical in real trading. There are contexts where you are also presented with PNL curves going "straight up", but these are normally outputs of curve-fitted backtests, where obviously the computer has adapted the entries to given known tickdata to create that specific shape.
Today I rolled over ES. Being essentially long, this time the backwardation worked in our favor, even though it's not much: Same we should do for NG, but I am waiting for some small spike up in the attempt to make easier some "recovery" of the unfavorable contango, which is pretty bad for this instrument. Anyway so far it did not show signs of slowing down the fall. (A portion of the decay will be recovered by gradually "loading up" on DGAZ.) In these cases, as explained, we close the existing position, on each layer, by a player and reopen using an "out-of-game" order, which will be later transformed into a player in case the player can successfully take profit (see scheme below). This way we can often "recover" the contango even through a long sequence of rollovers (which obviously can be pretty harmful). Here is a quick pictorial reference for rollovers:
I finally rolled NG. This means realizing quite a significant "loss" in terms of contango, but of course if it continues to fall we will recover that, through the mechanism we have seen in the previous posts (we close with a sell player, and if the sell player can close we transform the out-of-game order used to reopen into a buy player, thus completely recovering the contango "loss"): for the moment, the sell players used to close the previous contract of NG are causing, as it is right, a little drop in the G-L (gain-loss) curve: while the PNL has momentarily gone to 143.9K.