Tid: 10 juni 2008 kl 11.15-12.00 . OBS! Dag och tid.

Plats : Seminarierummet 3733, Institutionen för matematik, KTH, Lindstedts väg 25, plan 7. Karta!

Föredragshållare: Daniel Rufelt

Titel: Fast trading: Stochastic Modeling and Simulation of Latency in Marketplace Systems

Sammanfattning: An increasing part of all the trading conducted at financial marketplaces (like stock exchanges) is executed by computers using different automated strategies. Whatever market one looks at, volume shares of over 30 % is not uncommon for this type of trading. Algorithmic trading has lead to increased volumes being traded (and thus better liquidity), as well as new opportunities to profit from small price changes and other information by quickly and automatically respond to the information by having a trading strategy that makes the right decision extremely fast. This has created a high demand for marketplaces with low latency (response time), and now the industry is talking about latency figures of milliseconds rather then seconds and the trend to lower figures is obvious.

In this master thesis the probability distribution of the latency in a computerized trading system (i.e. a marketplace) is analysed by making a simplified stochastic model of the system, with the goal to find bottlenecks, looking at scalability, and understanding how the latency distribution is affected when changing different parameters in the system. Special attention is given to the upper tail of the distribution, because the tail behaviour is crucial when specifying and agreeing on the latency demands of a marketplace.

The system is analysed using queue theory and stochastic discrete event simulation. The main conclusion is that stochastic simulation is probably the best approach when dealing with rather complex marketplace systems. This is valid in particular when it comes to systems with service stations having vacations and dependent arrival times which essentially differ from normal Poisson behaviour. The conclusion is mainly based on the fact that the framework of steady state queue theory is not general enough when it comes to modeling queue networks and relies too heavily on the assumptions of exponential distributions.

Till seminarielistan
To the list of seminars