Collective2: extreme-os


Science Trader has a downloadable spreadsheet from his blog http://scitra.blogspot.com/ that records the actual trades he took from the extreme-os system on Collective2.

This topic discusses his trades.
Science Trader: I just took a look at the "extreme os" spreadsheet that you uploaded and did a little =countif(...) exercise on the fills to see how your fills compared to the C2 fills:

Worse: 94
Better: 11
Same: 19

I assume that the cells that you filled in with NA are orders that the system generated that you either decided not to take or were unable to take?
Its nice to see that these data are helpful to others. extreme-os is well known for the fact that real-life fills are almost always worse than the hypothetical fills shown by C2. This is also confirmed by the low realism factor and the shape of the equity curve after slippage (as shown by C2). Even though I'll never be able to match the hypothetical C2 stats I still like the system, as it appears to be very consistent and has a 2 year track record. Trades shown as NA were not executed by my auto trading software for various reasons.
The return on that system is phenomenal and the smooth equity curve shows a low risk strategy which is confirmed by the mostly low and normal risk categories of each trade. I'd like to hear pt_emini's view on that system as I believe that is one of the things that he looks at when evaluating a trading system.

As you mentioned a low realism factor of 66.6.
quote:
The Collective2 Realism Factor is a mathematical algorithm that attempts to capture how likely you are to achieve the same trading results in real-life as shown in a hypothetical track record.

I'm guessing that the realism factor is an attempt at guessing the percentage fills that will be within an acceptable slippage range of the hypothetical fills. Your real world data shows that you managed to get 24% of your trades at the same or better prices and so I'm guessing that 42% of the worse-fills were within an acceptable slippage range. Would you agree with that?

I'm also looking at some of the comments that other traders have left about the system and they all seem to concur with your realism comment.
quote:
Getting stated fills is impossible to achieve. I agree with the low Realism Factor. Count on 10-30 cents slippage on sell side of each position, slightly less on the buy side, so you have to adjust the results accordingly to get an accurate picture. I am now using Autotrader and still experience 20-30 cents slippage on fills on BOTH sides.

quote:
I have been a subscriber for a short period and use the ATM to receive trade signals. Like other comments here I find the posted fills a fantasy. I am extremely attentive and respond immediately to all Instant Messages but cannot come close to achieving the posted fills.

quote:
After using this service for 2 months, I finally decide to unsubscribe it. Using above $80K for every trade is not something I can do all the time while the fills are usually hard to be completed...

That last comment indicates (as you would expect) that the larger the size you're trying to trade the more slippage you should experience.

So I have some questions for you. What autotrader do you use? What are the reasons that your autotrader would not have executed the trades?
Correct me if I'm wrong here but I've just been playing with your spreadsheet again and it shows that you have lost $24,105 while you've been using the system?

The hypothetical system made $13,615 on the same trades that you took (based on the number of shares you traded) and a further $31,165 on the trades that you didn't take for a total of $44,780.
The system hasn't been performing very well since the beginning of February. Your observations are correct, but note that the size you see in the file is the hypothetical C2 size, not the size in my account. The reasons trades were missed vary. In some cases the problems were of technical nature, in other cases a signal was issued as a limit order which did not get filled in my account. Ideally, one should monitor the auto trading software continuously (in which case all trades could have been executed), but I'm not able to do that. It was unfortunate that I missed a few very profitable trades, but I think this was just bad luck, it could have been other trades as well. The software I'm using for auto trading is TradeBullet. Also, I've been trading this system since October and earlier profits more than offset the recent losses. I would expect that the larger the size, the larger the slippage. It probably means that the system is not very scalable.
So do some of the signals get issued as market orders and others as limit orders?

By definition, if you're using limit orders, then you will always get filled on all the trades that get stopped out (because they will obviously go through your limit on the way to hitting your stop) but not on all the trades that are winners. For that reason, a hypothetical system that uses limit orders can only record a fill if the market trades 1 tick beyond the limit order. In real terms, this means that the actual system (i.e. real money) will perform better than the hypothetical system because it will get some fills that the hypothetical system will not get.
The way it works with extreme-os is that the entry order (always long) is issued as a limit order about 10 cents above the market. This is supposedly done to limit slippage on the entry to 10 cents at most. Because there can be a delay of several seconds between the time the signal is issued by the vendor and the time it gets to the auto trading software and is executed by the broker, in some cases the price has already slipped past the limit price. In these cases it has happened that some subscribers were filled and some were not. Usually these trades show up as filled at C2. Exit orders are always at market (to be sure trades are always closed), and this is reflected in the fact that slippage is generally larger for exit than for entry orders.

I recently found out that TradeBullet allows to change a limit order to a market order automatically in case it is reported as filled by C2 but has not yet been filled in your own account. By using this feature, I expect to have fewer missing trades in the future.
This also means that the extreme-os system has a "spontaneous" target and not a predetermined target at which you can enter a limit order.

From what you've said and what I've read and observed it seems that this system is very good but its implementation is where it falls down. It's being operated in a less than optimal environment. The current implementation is:
Signal -> Trader -> Execution
this needs to be changed to:
Signal -> Execution
...which means that the traders need to run the signalling software on their machines or the company running the extreme-os system has to execute (or issue) the orders directly to the broker's system/market on behalf of the subscribers.

Is that not an option for you?
I'm not sure if the slippage is due to the latency inherent in a setup with a chain like: signal -> C2 -> my PC -> broker.

I think the main problem is that each system is only scalable up to a certain level before it will start to move the market. The latter might possibly be happening with some C2 stock systems, where the total volume of shares from all subscribers together becomes just too large for the market to absorb instantaneously, resulting in slippage.

What I would expect is that systems that perform well will continue to attract new subscribers until their total volume will start to move the market and produce slippage (for a system that trades very liquid markets, it might take a very long time before this would happen). At that point some subscribers might become disappointed and stop trading the system, while new subscribers still come in. Perhaps this will lead to some equilibrium where the returns after slippage are still acceptable but not as spectacular as the C2 stats (before slippage) show. If this happens, vendors could try to issue several entries and exits to spread the volume over a larger time frame.
I've actually thought about this problem of subscriber base versus successful system in depth and believe that there is another solution based on an auction system.

Say the system can handle 20 subscribers before the volume starts affecting the market. (Let's assume all subscribes trade the same size.) Once the system is fully subscribed a new subscriber will need to pay a higher rate to become a subscriber and bump out the most recent subscriber at the lowest rate. Of course the most recent subscriber at the lowest rate would have the option to up the price their paying to the next tier and then the bump process would move on to the next subscriber etc.

This may sound somewhat unfair for subscribers but I don't believe that it is if all subscribers are aware of the rules and it allows the system to continue to function with the optimum number of subscribers and the system developer to receive just compensation. The laws of supply and demand working in a market economy.

Of course this will probably lead to a secondary "black" market in signals from this particular system.
There's an update on Science Trader's Blog called End of Portfolio where he states:
quote:
As you've probably noticed, my portfolio has not been doing well for almost the entire past 5 months. Despite all my efforts to analyze systems, I have not been able to pick a profitable set of systems. Whereas the S&P500 is almost exactly back to where it was when I started my portfolio, I am sitting on a loss of 13% (excluding the P/L from my various put options to hedge, the loss would be even larger).

The main problems I have encountered can be summarized as follows:
- Technology issues with auto trading (extreme-os)
- Vendor decided to terminate/change system after major losses (Trend Plays #1, Longstoch-ST)

In addition, I was close to signing up for Positive Forex, which completely collapsed shortly thereafter.

Time to move on!
Thanks for the updates Science Trader! Sorry that none of those systems ever worked out for you.