Wilmott Forums - what is high frequency trading?

Forum Navigation:

new topic

bookshop

FORUMS > Trading Forum

< refresh >

Topic Title: what is high frequency trading?
Created On Wed Apr 12, 06 09:20 AM

Topic View:


	Pages: [ 1 2 3 4 >> Next ]

needaclue
Senior Member

Posts: 319
Joined: Sep 2005

Wed Apr 12, 06 09:20 AM

I have several questions on this field if someone can answer:

1. What is high frequency trading?
2. Why is it a hot topic nowadays i.e. what are the new strategies it makes possible?
3. Which mathematical skills are in demand in this field?

would appreciate any input!

Quote

Top

Bottom

jomni
Senior Member

Posts: 989
Joined: Jan 2005

Wed Apr 12, 06 09:46 AM

In high frequency trading, you analyse trends in tick-by-tick data and make buy / sell decisions out of it.

It may sometimes lead to day-trading strategies...

-------------------------
visity my (dead) blog:

rmquant.blogspot.com

Edited: Wed Apr 12, 06 at 09:48 AM by jomni

Quote

Top

Bottom

DominicConnor
Senior Member

Posts: 10695
Joined: Jul 2002

Wed Apr 12, 06 12:14 PM

It's also sometimes used to include electronig trading, where you may have to respond in milliseconds.
Part of the reason for growth is that the links to the various exchanges & brokers can cope with it. This has taken longer than many poeple predicted.

Strategies vary, but they include market microstructure, volatility forecasting, liquidity modelling and a ragbag of thingsthat resemble technical analysis.

-------------------------
Discussion on the new regulations on bonuses here.

Quote

Top

Bottom

needaclue
Senior Member

Posts: 319
Joined: Sep 2005

Thu Apr 13, 06 04:52 AM

DCFC and jomni: is there any particular type of math that high freq trading shops would like to see on my resume? something apart from the usual stochastic calculus.

Edited: Thu Apr 13, 06 at 04:55 AM by needaclue

Posts: 83
Joined: Sep 2004

Thu Apr 13, 06 09:23 AM

High frequency trading means you measure your holding period in seconds (sometimes even in hundreds of milliseconds).
There are two very different techniques, passive and aggressive trading. In passive trading you're placing bids and offers and show volume waiting for someone to take you out. In aggressive trading you only take prices which are offered. Aggressive tradign is "easier" than passive trading which, to my limited knowlegde, very few teams have mastered.
There is interest in this area because it is perceived to be low-risk and the Sharpe Ratios you can achieve are double-digit.

Quote

Top

Bottom

DominicConnor
Senior Member

Posts: 10695
Joined: Jul 2002

Thu Apr 13, 06 01:06 PM

Aside from the usual is the ability to filter out "wrong" data. That requires the right sort of stats of course, but it's good if you have enough depth to make a good judgement on what is "wrong".
Time series analysis in general is good, as is signal processing.

Also there isn't a firm boundary between maths and programming her at all. Stuff works if it works, and you will need to be self sufficient in your IT in the sense of writing Matlab & C++ to prove your stuff works.
The smarter outfits also have specialist hardcore C++ guys to turn the prototypes into things that you can trust with real money. Given the speed of this stuff you can't run it on manual.

But as tk243 says, it should be relatively low risk, and a complementary set of skills is how you hedge this stuff in real time as well. It doesn't quite qualify as pure "risk free", but probalby has the best Sharpe ratio that'[s it's possible to sustain over time.

-------------------------
Discussion on the new regulations on bonuses here.

Quote

Top

Bottom

farmer
Senior Member

Posts: 7096
Joined: Dec 2002

Thu Apr 13, 06 02:40 PM

High frequency trading is probably characterized by

a) the limited amount of capital it can employ,
b) a high ratio of labor costs to trading profits, and
c) fundraising based on immediate empirical performance, rather than on theoretical expectation particular to an investment style.

-------------------------
Unemployment peaked at 9% two months after the October 1929 crash and then began drifting down over the next six months, falling to 6.3% by June 1930. In June 1930, against the advice of 1000 economists who took out newspaper ads warning against it, the US raised tariffs in order to save jobs by reducing imported goods.

Quote

Top

Bottom

needaclue
Senior Member

Posts: 319
Joined: Sep 2005

Fri Apr 14, 06 04:16 AM

thanks everyone for the replies.

if HF trading intervals are on the order of a second, then would that imply that
complex derivatives are not traded in this way? i don't believe they are so liquid.
even for most plain options it would not seem a possibility.

Quote

Top

Bottom

jomni
Senior Member

Posts: 989
Joined: Jan 2005

Mon Apr 17, 06 04:59 AM

Correct.
Derivatives (OTC) are not suitable for high-frequency trading.

I believe technical analysis (as DCFC said) is the key to high-frequency trading. It does not only mean that you should know a lot about TA techniques. All your competitors know that. To be competitive, you should also be creative and make your own TA techniques (and combos). TA are generally trading rules and metrics that result from data mining and some stat work.

Aside from data mining and stat, a technical trader should have intuition / good judgement. One should 'feel' the market to know when TA rules do not apply.

-------------------------
visity my (dead) blog:

rmquant.blogspot.com

Posts: 83
Joined: Sep 2004

Mon Apr 17, 06 08:57 AM

Technical analysis was the starting point for high frequency trading, but things have moved on since then. Refer to Olsen's book on high-frequency trading for pointers.

Re "Derivatives (OTC) are not suitable for high-ferquency trading"; well not exactly, you can trade these things high-frequency (though only a very small number of contracts).

Quote

Top

Bottom

farmer
Senior Member

Posts: 7096
Joined: Dec 2002

Mon Apr 17, 06 04:38 PM

Quote
Originally posted by: tk243
Refer to Olsen's book on high-frequency trading for pointers.

Refer also to Olsen's bizarre skewed returns distribution.

-------------------------
Unemployment peaked at 9% two months after the October 1929 crash and then began drifting down over the next six months, falling to 6.3% by June 1930. In June 1930, against the advice of 1000 economists who took out newspaper ads warning against it, the US raised tariffs in order to save jobs by reducing imported goods.

Quote

Top

Bottom

Lepperbe
Senior Member

Posts: 319
Joined: Nov 2002

Mon Apr 17, 06 06:23 PM

looks like he encountered a grey swan...

Quote

Top

Bottom

FrunkAdita
Junior Member

Posts: 16
Joined: Jul 2005

Mon Apr 17, 06 10:40 PM

When referring to Olsen's book, do you mean "An introduction to high frequency finance" ? Btw, has anyone a pdf version of chapter 11 "real time trading models" ? Each chapter looks like an article all its own, so it must have been published as an article someday.

Concerning high-frequency trading, if I understood what I've heard, the techniques currently used are:
* TA, using a combo of trend detectors (lets say), and agreeing that there's an up-trend when x% of the indicators agree. In this case, the indicators used are chosen by backtesting and checking which combination gives the best results (whatever that means).
* signal processing based trend indicators, ...

Does anyone have references or articles saying which (and possibly why) TA indicators are suitable for high-frequency data ?
And what are the signal processing based indicators used in the business ? Please, don't just tell me "wavelets" (for example). I'll be grateful for any details or references. Also, what is the state of the art in high frequency techniques ? Does anyone know or is it too much of a competitive field that no one knows what the other banks are doing ? Finaly, what are the best firms in high frequency trading ?

I know that's a lot of questions for one message, but I'm only a student.

Posts: 83
Joined: Sep 2004

Tue Apr 18, 06 11:54 AM

1) Olsen's book: yes.
2) Not sure about Olsen's track-record. No idea whay it's so erratic.
3) Intuitively, TA works "better" at high frequencies, becasue the actual economy/company doesn't change that much from one second to the next. Flows and trader psychology are much more important.
4) Refernces are far and few between, though there are some academics who pursue research in this area. Not much, that is to say almost zero is coming out of IBs or hedge funds.
5) The dominant IBs in this space are Goldman, BoA, Morgan Stanley, UBS (in no particular order and without any claims towards completeness). Not sure about hedge funds.

Quote

Top

Bottom

mit
Senior Member

Posts: 904
Joined: Feb 2005

Tue Apr 18, 06 10:57 PM

nice discussion.

any more info on the topic?

besides Olsen, any other reads?

Quote

Top

Bottom

mit
Senior Member

Posts: 904
Joined: Feb 2005

Wed Apr 19, 06 12:31 AM

http://www.behaviouralfinance.net/trend/Levi98.pdf

http://www.jamesgoulding.com/Research_II/Data%20Study/Intro_to_High_Frequency_Finance.pdf

Quote

Top

Bottom

FrunkAdita
Junior Member

Posts: 16
Joined: Jul 2005

Wed Apr 19, 06 01:14 AM

Thank you for your answers tk243. Out of curiosity, do you work in this area ?

Posts: 83
Joined: Sep 2004

Wed Apr 19, 06 10:10 AM

Yes I work in this area; fx, interest-rates and commodities.

Quote

Top

Bottom

QArbiTrader
Member

Posts: 37
Joined: Jan 2006

Wed Apr 19, 06 11:00 AM

Can anyboby give me a more or less comprehensive list of the type of
markets that are liquid enough for high frequency trading ?

Fx is the first choice I assume. Any other options ?

Is there any high frequency trading for derivatives ?

Alos, in FX markets, what is a realistic level that one can expect
for trading fees ?

Quote

Top

Bottom

FrunkAdita
Junior Member

Posts: 16
Joined: Jul 2005

Wed Apr 19, 06 09:42 PM

Suppose I have all the data for one day. Most of the time, it is rather easy to tell visually which periods are trend periods and which are not. But if you had to define a trend, how would you do it ? Are there TA definitions (for instance based on momentum) that are commonly accepted in a more quant sphere. Let's say I want to define a trend by the following conditions: over a time period delta_t prices vary by at least delta_price. Of course such a definition is not very stable. But how would you improve it ? Does anyone have references on the matter ?

Quote

Top

Bottom


	Pages: [ 1 2 3 4 >> Next ]

FORUMS > Trading Forum

< refresh >

Forum Navigation: