Back to Community
Ability to specify offset of bars

If everyone trades wall clock minute cut off there will be burst of volume of trades either to buy or sell. This information might be able to exploited by games if the bars are not offset by some off set. The strategy should be able to provide offset from the wall clock on which bars are to be used.

This risk will be there when there are large number of Quantopians trading.

4 responses

Quantopian hasn't revealed the details of how live trading will work on a minute-by-minute basis with Interactive Brokers (IB) (or even if the trading will be based on on-the-minute tics of a wall clock). There is also the problem of reconciling their backtester to the various details of trading through an IB account. I've gathered that sorting all of this out is their present focus. My hunch is that IB has given them a bunch of interface requirements and that they are building their code according to them.

As a thought experiment, let's say that there are lots of quantopians, all initiating their orders every minute exactly. You know this and have lots of resources to throw into the ultimate "game the quantopians back to the stone age" strategy. What exactly would you do to maximize your profits from exploiting all of these poor quantopians? I would be very hard pressed to name anything actionable and I suspect you will too. "Predictable" and "exploitable" are two different things, and the first does not necessarily imply the second.

What is exploitable are algorithmic strategies with a clear "tell" that would tip onlookers off to a behavior that can be detected and then front run. Typically this would involve only larger orders, where you detect the presence of a large seller in the market and can front run them under the assumption that they will continue to sell and push prices down over some extended interval. Smaller orders tend not to have this problem because they don't impact the market as much, and because by the time you detect them, the order is probably already complete and it's too late to exploit. If you are trading larger orders, then you clearly want to avoid this type of predictability as you design your algos. Generally if you execute through a broker with decent electronic trading capabilities you will be fine. All the big brokers have specialized for years in anti-gaming logic so that even larger orders are difficult to detect. And you may want to stick to the top 500-1000 us stocks where the market velocity is fast enough to make algo detection impractical for would-be gamers. But this is getting off topic.

The other potential problem with 1-min increments is if it synchronizes similar strategies so that people are chasing exactly the same alpha opportunities at the same 1 minute increments. This would result in faster-than-expected alpha decay; in other words you tend to have trouble executing your orders before prices move away from you. This is a reasonable concern, but if it is real, it is more likely to affect people running shorter-timescale strategies, and event-based strategies. If you are running an algo with longer holding periods, it shouldn't make a big difference.

By the way, I suspect there are lots of algorithmic investment strategies that wake up at exactly the 1-min mark. It's not just a quantopian issue. But I also suspect that 95% of these are algos with very short-term holding periods. I also did some research last year on the percent of trading that occurs in the first second of every minute. If this 1-min phenomenon were a big issue, you'd expect the percentage of all trades in the first second of every minute to be high -- like 5-10% or even more. But it turned out to be quite low -- it was only about 1.5 - 2 x the share of the average second (1/60).

There will be some level of crowding in the strategies. Especially due to sharing. This means you can front run. This is one fear. I do not want others exploiting my strategy.

@Tomas, Is it possible to share your details. There is something of mutual benefit I want to discuss with you offline.