Fully automated futures trading
- Thread starter globalarbtrader
- Start date
@globalarbtrader, Did you happen to test the historical performance of adding your intraday mean-reversion implementation to your current production system (i.e. without running the DO stage) to compare with that of your current system?
Last edited:
Same here: during the last 5~6 years that I have been using the API there were about two function calls which changed at a (forced) upgrade of the API. The updated function call had an extra argument.
But I also recall that I had to modify the code to start the reader thread when connecting to IB after the upgrade to API 9.72. At the time it caused some commotion among the IB API users, but can't recall the exact reason why everyone got upset.
Looking forward to it.
Excellent, thank you for the reply. I remain excited to read the details once my book pre-order arrives.
Since you tested it and your update above implied a decision against adding mean reversion if at the cost of dropping DO, to the extent possible to answer without stealing any book thunder, of course, I gather that the decision was driven purely by the expected performance trade-off (from adding MR while losing DO) versus an implementation constraint relating to intraday or another factor?
Might I run an idea by you guys;
It seems that flash crashes has become a part of today's algo-run markets. The fat fingered flash crash in the nordics a while a go, being the latest example. (it took the Swedish equity index OMX down 8%, and the move rippled out to all stock markets with an abaiting effect (Norwegian index was down 4,8%.)). The index recovered within minutes.
Since we already have servers running, and also have live data subscriptions, it wouldn't be much of an extra cost to have an algo looking for, and potentially exploiting, a flash crash.
A huge drop in a matter of minutes has only been caused by algos has it not? Black monday might be the exception (20% intraday drop).
I will admit that I added Michael A. Gayed to one of my twitter lists a short while ago. The guy is manic and I have no taste for his theory of "the end of the world is a bull case", as it is not laid out with any other other argument than - the bond market is in a state that cannot be sustained (or something to that extent - it's very vague).
I removed him from my list because his constant tweeting was basically spamming down myfeed, but the damage might already be done, because; he gave an anecdote about an experienced trader going all in during the bay of pigs debacle. His reasoning was if there is an nuclear money will not matter much anyway.
So I am left wondering; what other events could cause huge moves of 8% in a matter of 2 minutes, that was not really devastating in a way that my other investments haven't been severely impaired. I would not like to end up owing money, but loosing a bit more when you've already lost a metric ton doesn't matter that much does it.
It seems that flash crashes has become a part of today's algo-run markets. The fat fingered flash crash in the nordics a while a go, being the latest example. (it took the Swedish equity index OMX down 8%, and the move rippled out to all stock markets with an abaiting effect (Norwegian index was down 4,8%.)). The index recovered within minutes.
Since we already have servers running, and also have live data subscriptions, it wouldn't be much of an extra cost to have an algo looking for, and potentially exploiting, a flash crash.
A huge drop in a matter of minutes has only been caused by algos has it not? Black monday might be the exception (20% intraday drop).
I will admit that I added Michael A. Gayed to one of my twitter lists a short while ago. The guy is manic and I have no taste for his theory of "the end of the world is a bull case", as it is not laid out with any other other argument than - the bond market is in a state that cannot be sustained (or something to that extent - it's very vague).
I removed him from my list because his constant tweeting was basically spamming down myfeed, but the damage might already be done, because; he gave an anecdote about an experienced trader going all in during the bay of pigs debacle. His reasoning was if there is an nuclear money will not matter much anyway.
So I am left wondering; what other events could cause huge moves of 8% in a matter of 2 minutes, that was not really devastating in a way that my other investments haven't been severely impaired. I would not like to end up owing money, but loosing a bit more when you've already lost a metric ton doesn't matter that much does it.
Continuing very brief Twitter discussion from here:
One of the many things I don't understand about DO is how contract exposures can be quite different in the same DO basket.
I'm thinking of the value of the MBT contract (very small) compared to Micro Gold (reasonable size). I could have named a dozen others instead of Micro Gold.
Surely the different sized exposures can't all be put down to risk or volatility? Maybe costs? Hoping to see some explanation of this in an upcoming publication.
KH
One of the many things I don't understand about DO is how contract exposures can be quite different in the same DO basket.
I'm thinking of the value of the MBT contract (very small) compared to Micro Gold (reasonable size). I could have named a dozen others instead of Micro Gold.
Surely the different sized exposures can't all be put down to risk or volatility? Maybe costs? Hoping to see some explanation of this in an upcoming publication.
KH
Rob can probably also more comprehensively, but here's my take.
There are many differences between the systems of various people.
Signals: Each of use includes different strategies, models, lookbacks. As a result, each person will have a different signal strength for the same instrument.
Correlations: DO relies on a correlation matrix. Each person will calculate these correlations differently.
Instruments: Since DO looks at the correlations between instruments, having different instruments will result in different allocations to each.
Costs: How one accounts for costs could effect whether a person will trade a certain instrument or trade something that is somewhat correlated but cheaper to trade.
Capital and vol target: How much someone is willing to risk overall will effect whether an instrument is tradeable within his system.
Noise: Even if all of the above was the same, two people with identical systems will end up with some variation depending on when they trade, the bid-ask spread at the time, the execution price, etc. Over time, this variation may add up and the two systems may take different trades. In the end, though, the two systems will be highly correlated, but it may not appear that way when looking at individual trades.