Dear C2 customers trading via auto-trader, I am outraged as to how poorly C2 auto-trade system is, and how unwilling Matthew Klein is to spend money to hire good programmers who know what the are doing and have C2 auto-trader implemented the right way. If there really are 17K sytems here on C2, this alone would attribute to 1.5M in sales per month. Yet, company has only few employees, who clearly suck as programmers.
Today, I observed again something that should have not happened: my limit order to buy at 61.09 got filled at 62.35
BTO 140 DVN at limit 61.09 Filled: 62.35 at 12/15/11 9:30 ET
My price of 61.09 has not even been touched. Market low within 5 minute window was 62.06. I observe issues like this every day. Customers receive substantially worse price fills on my limit orders (which by definition should only be filled at limit price or better). Even observing the statistic for my system:
Keep after worst-case slippage 55.4%
shows how poor the fills are. And as one might assume, the statistic above should be in essence capturing market impact. However, this model with this amount of subscribers I have righ now has 0 market impact (trading Large Caps). So what exactly is the above number capturing ? It is capturing how poorly C2 is performing auto-trading on behalf of my trade signals. In other words, Mathew is reluctant to spend money to write the auto-trader properly, and as a result, You, The Customers, are the ones suffering substantially. I urge you to contact Mathew and express your dissatisfaction.
Thanks,
David Sepiashvili
Skip-n-Trade Daytrader
Hi, David:
I understand you’re agitated, and I’ve sent you a private email (probably a better way to have a conversation like this) explaining some of the issues we need to examine here. It does indeed seem that this conversion of your limit order into a market order was triggered prematurely, but we’re still investigating why the limit was triggered. The likely culprit is a bad bid/ask tick. However, as of this moment, I’m unable to say with certainty this is the cause, pending more investigation.
Matthew
Mathew, investigations as to this particular trade most definitely can be discussed via private messages. However, the issues that I rose are of concern to the whole community on C2. So I do believe this is a good time for everybody to participate and brainstorm on imporvements to the auto-trader. Since you mentioned possibly bad bid/ask quotes that can trigger trades prematurely, please let all of us know who is providing data feed to collective2. You are absolutely right that Fed Reg requires to report actual fill prices, and I have not disputed this. However, same Fed Reg requires that auto-trader acts based on NBBO bid/ask quotes. So this brings us to the question as to who is the data provider for Collective 2 ?
Thanks,
David A. Sepiashvili
David -
Our investigation revealed that we filled your limit erroneously because at the instant the market opened, we acted on the closing price of yesterday’s trading instead of the current last price. We will correct this software error. Thanks for bringing it to my attention.
After the fix has been made, you ought not see this (particular) issue again. (However the change is still not tested and committed.)
Hopefully later today – and at a minimum before tomorrow’s opening bell.
Matthew
A further update: We won’t be able to roll out fixes to this problem as quickly as I would have liked. We are still working on it. I will report here when problem is resolved.
Mathew, I am very glad to see you taking proactive steps. Clearly, so far you managed to identify one of the bugs in the auto-trader and working hard on resolving it. But let’s focus for a moment on more core question, which has been a concern of many C2 members already - on how auto-trader is based on pull based publisher-subscriber model instead of push based. Based on your API docs I learned that there is at least 10 second delay in refreshing data on the server side as to see if limit order has become marketable. In addition, I see that the server will instruct clients that are pulling for changes to use variable pulling interval that is recalculated dynamically based on current server load. All this applies to Gen1 API, and there are no docs for Gen3 API, however based on my observations, there still is substantial delay from the moment limit order becomes marketable to the moment it reaches the customer(s).
Since the whole concept of Collective2 is based on sharing of statistical information, and of members being able to see and analyze such statistics, I would like to see a statistic that will rank on how well C2 auto-trader is doing in delivering orders to the end clients. In particular, two statistics come to mind 1) average delay in seconds, which should be a sum of average processing time to go through the queue and identify marketable orders (note that it is not half of this amount because it is a poisson random process) plus average pulling time. 2) average slippage/gain attributable to C2’s auto-trader, which can be as well very easily computed. Such statisctics could be computed on per-model basis, or, even better, as whole and displayed on the main page.
Ranking individual models and not ranking C2’s ability to deliver these models promptly to the end clients is simply not fair. Let’s poll other members as to what they think about this, and if they would like to see as well a statstic that will rank C2.
Thanks,
David A. Sepiashvili
David - You are incorrect. Gen3 AutoTrade is push-based. - Matthew
Also, let me add that you are free to study the fill data, if you want. Gen3 fill data is posted publicly on the site. You can see all prices received by all AutoTraders in live accounts.
Matthew
