You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
HFT Myths
- Thread starter hft
- Start date
Without going into specifics on fees or products I think an individual can make about a grand a day with this setup fairly reasonably. With the experience I have, I'd probably aim for somewhere in the neighborhood of $5-10K/day on certain products...and most likely slip to about half that much...or into the red once I come across unexpected problems that I take for granted being resolved for me by other people now.Quote from jtrader33:
What would you estimate to be a feasible comm per share under such a set up? And what type of monthly pnl would you expect with it? Of course, I realize there are a ton of dependencies, but I'm just curious what sort of numbers would sound reasonable to you.
Quote from hft:
Absolutely. Well maybe 5K is pushing it but let's say 10K for the same server we use. You can be in the same ballpark as any HFT firm in terms of speed with that hardware, and the connection to the exchanges are identical, down to the cable length so it's even footing for everyone. In fact if I did exactly this, I have no doubt I'd be faster than my current setup because I could remove any iota of latency that is not crucial to my customized application (but is a part of the shared libraries that I use).
However you are severely limited in the number of markets/inputs you have so your PNL potential will be limited. Fees would be another factor depending on your clearing arrangement. Hence your profit margin and sharpe will be lower than mine, but given the right choice of market and product there is no financial reason you couldn't make consistent money with that basic setup.
If you include data and port fees, it quickly adds up. Would you say these rising costs are a barrier to entry? Also do you find it easier or harder to trade on exchanges with better technology e.g. Nasdaq vs a slower platform like CME?
Quote from chasinfla:
Are you saying that shorter time horizons are less random? I don't think you are.
Let's see if I can clarify. Two blips do not make a trend. A dozen or a hundred might. A trend is not random price action (though, yes, we observe what appear to be trends in random data).
If you watch a tick chart (the shortest time horizon), you will see a random succession of blips. On a 5 min chart, you might see a trend. If you are trading a tick chart, you are much more likely to encounter random results (excluding the spread); if you are trading a longer time horizon, you are more apt to encounter a trend.
Finally, back when I was testing trend-following systems, they demonstrated better profitability over longer time horizons.
I don't think I've quite said it as clearly as might like to, but hopefully shedding a bit more light on the point.
It does seem like time should be irrelevant, so I suppose what I'm describing is the number of trials, not necessarily the time horizon. Maybe someone could address that.
I think we're talking semantics here to some extent. I'd say it's easier to forecast the next 1-min interval than the next 1-year interval; the longer the forecasting period, the higher the likelihood for unanticipated events to happen. On the other hand, the shorter the time frame, the more noise - or more correctly, the lower the signal-to-noise ratio.
Quote from Drifter:
If you include data and port fees, it quickly adds up. Would you say these rising costs are a barrier to entry? Also do you find it easier or harder to trade on exchanges with better technology e.g. Nasdaq vs a slower platform like CME?
No, I don't think the hardware and intracolocation costs are a significant barrier to entry at all. They're paltry compared to the PNL potential of HFT.
Intercolocation costs, however, are certainly a large barrier to entry. Simply put, the more money you make, the more you can afford to pay for better network infrastructure. It's an example of the rich getting richer. Smaller firms already can't compete at those levels any longer, and the situation will only get more drastic as we get deeper into microwave world. The largest, most efficient firms will win out at the top tier here.
The other major barrier is simply the highly specialized knowledge and experience that is required to get an operation up and running. A lot of firms maintain a tight separation between departments and coming up with a team that can put together the pieces from end-to-end is becoming rarer. Add the fact that major firms already offer very competitive compensation to their best people and you're very hard-pressed to find any but the lowest performers leaving their employers.
I personally don't find a correlation between the quality of technology and ease of trading at a venue. In fact, when the technology is worse there can be more sources of inefficiency to capitalize on. My ease of trading is directly related to the PNL potential of an exchange since that's where we face the stiffest competition.
Quote from chasinfla:
Are you saying that shorter time horizons are less random? I don't think you are.
Let's see if I can clarify. Two blips do not make a trend. A dozen or a hundred might. A trend is not random price action (though, yes, we observe what appear to be trends in random data).
If you watch a tick chart (the shortest time horizon), you will see a random succession of blips. On a 5 min chart, you might see a trend. If you are trading a tick chart, you are much more likely to encounter random results (excluding the spread); if you are trading a longer time horizon, you are more apt to encounter a trend.
Finally, back when I was testing trend-following systems, they demonstrated better profitability over longer time horizons.
I don't think I've quite said it as clearly as might like to, but hopefully shedding a bit more light on the point.
It does seem like time should be irrelevant, so I suppose what I'm describing is the number of trials, not necessarily the time horizon. Maybe someone could address that.
When trading short term (trade duration in seconds and max minutes), there is info available that helps pin pointing trades accurately.
For example, I can predict next up/down tick with a high degree of accuracy, granted I can not monetize it, but this is just one of the examples, where short term "momentum" can be predicted accurately.
Is speed in the market making space more about sending orders quickly or more about cancelling quickly to limit adverse selection?
So is low latency for you, more important when you remove liquidity or send orders to get good queue placement, or is it more of a cancelling game to avoid getting hit when the market ticks against you?
So is low latency for you, more important when you remove liquidity or send orders to get good queue placement, or is it more of a cancelling game to avoid getting hit when the market ticks against you?
Both are equally as important.Quote from jb514:
Is speed in the market making space more about sending orders quickly or more about cancelling quickly to limit adverse selection?
So is low latency for you, more important when you remove liquidity or send orders to get good queue placement, or is it more of a cancelling game to avoid getting hit when the market ticks against you?