What are algos actually doing?
- Thread starter booked
- Start date
The misconception is that they're taking any positions and "participating in the market." HFT is in it solely to cross bids with asks, and nothing more. They do not trade systems that would ever lose money because all the algos do is cross trades and earn a spread. They do not try to make money in any other way than instantaneously, so that is probably where the public is most deluded. They aren't trading to make any larger a profit than what they can get immediately, and that is the main point I am making because it is that simple.
I like your thinking bowo, but it's a bit harder than that
For example:
When do you want to be quoting BBO, and when do you want to favour one side over the other, & back off a tick or so?
How likely is it you will be run over by a big dumb market order, consuming several price steps against you?
Then there's informational changes..
Also, some arb algorithms will cross spread(s).
If you guys look carefully at the Bid/Ask sizes you'll figure out how the HFT avoids staying on one side for too long. The length of time they keep a bid/ask is proportional to a statistically measured time (and a bit of a prediction) it takes for the market orders to get assymetrical to one side. When the market is moving really rapidly, they may hold one bid/ask for just a few seconds. For markets such as ES where ticks are big and move more slowly, they profit the most mid-day when the markets are quietest. They look at the frequency ratio of getting filled bid vs ask and estimate (with a little bit of noise) when this ratio is bound to get out of hand on one side. If it gets too assymetrical, ie 50 buys at ask, and 40 sells at bid, it pulls all of the asks, and hopes to get filled on 10 more bids. Usually it does, because conservative and well made HFTs know when to pull. Some of the more risky HFTs (or slower ones) even get to keep lack of symmetry to get filled two ticks difference and compensate for any previous mistakes, but this is rare, happens more on instruments with lower liquidity and larger spreads in terms of ticks. When the markets go wild in one direction, HFT usually don't put any bid/asks at all.
So there you have it, even HFT isn't a 0% risk game.
This also means that HFT is always first in queue, and the earliest posted Non-HFT limit order will get filled only when the price is about to move anyway - at least on liquid markets. That's why when backtesting, you should always count limit orders filled if the price breaks the limit. You can also play the HFT game and use their prediction on when price is about to move by looking at when the Bid/Ask size suddenly drops. Take a careful look at the order book and you'll see what i mean. If you get a fast and accurate data feed, you can predict with 90% probability where the price will move also, by looking at the ratio of fills bid vs ask before and after the HFT pulls it's bid/asks. You might have only a few seconds to act on it (on the SPY that's 0.05 seconds due to low tick spread), but it might give you 1 tick of an advantage if that's what you're after.
So there you have it, even HFT isn't a 0% risk game.
This also means that HFT is always first in queue, and the earliest posted Non-HFT limit order will get filled only when the price is about to move anyway - at least on liquid markets. That's why when backtesting, you should always count limit orders filled if the price breaks the limit. You can also play the HFT game and use their prediction on when price is about to move by looking at when the Bid/Ask size suddenly drops. Take a careful look at the order book and you'll see what i mean. If you get a fast and accurate data feed, you can predict with 90% probability where the price will move also, by looking at the ratio of fills bid vs ask before and after the HFT pulls it's bid/asks. You might have only a few seconds to act on it (on the SPY that's 0.05 seconds due to low tick spread), but it might give you 1 tick of an advantage if that's what you're after.
I hate HFT. They're killing the volatility and turning away market participants. You just don't get the same moves these days as you did just a couple years ago.
You guys are correct in their mechanical function, but forgetting one thing: they have access to front-running quotes. Get rid of those to level the playing field, and I think the market will be in a better place.
You guys are correct in their mechanical function, but forgetting one thing: they have access to front-running quotes. Get rid of those to level the playing field, and I think the market will be in a better place.
Care to comment on exactly how this is the case?
Sure. The ES e-mini is a good example here. The ticks are large and the spread is 1 tick wide virtually 100% of the time. Also, there is a message limit at the exchange so they can't post and pull too many bid/asks.
There are two cases where HFT will make itself first in line.
The speed case:
Say the current bid/ask is at a certain level. The buy orders at Ask deplete the ask size and the ask price moves 1 tick up. Now, where the ask once stood, there is room for a bid to be posted. One fraction of a nano-micro second later, the bid is already there, posted first by the HFT so they are the first in line at the bid.
The slow (order book) case:
Now that they got the bid, all they need is the Ask. Say 5 minutes ago, the price moved down. With the same method as above, they were the first to post the price at the new Ask and they kept it up until now (even as the price moved further down), so when the spread reaches their bid/ask, they're already first in line. However due to exchange limitations, they must sometimes economize with placing asks and bids so they don't keep too many active for too long. For this reason they will sometimes measure how long it might take the price to reach a certain level and post before everyone else. This is where different HFT strategies come into play and there are numerous models constructed around this, because remember, if they don't have the new Ask in place (are first in line), there's no point in speed-posting the bid either. Sometimes that's the edge, and some HFT shops might have multiple "accounts" to avoid exchange limitations, where they later consolidate their accounts. This is a shady practice, but i'm not sure how much it happens, nobody's talking.
How will this not get killed in a trend?