A Family of Financial Market Models

George Mason University CSS 739 under direction of Dr. Robert Axtell

Overall status:

We have a framework with several financial market models implementations (see paper with overview here). The framework allows for multiple assets and mixing of agents using different behavioral logics. You can download the jar file here and associated property files here. See RunGuide for further info.

The current version includes following graphs:

1. Bid and ask price history (up to last 10000 observations);
2. Return and absolute return history (up to last 10000 observations);
3. Trading volume history (up to last 10000 observations);
4. Graph of the autocorrelation of both returns and absolute returns (1 to 200 lags);
5. Trailing average volatility (absolute returns) (up to last 10000 observations, over 365 steps)
6. Frequency of returns;
7. Distribution of order book limit orders (if applicable);

Graphs are computationally expensive and are updated only when visible. Best practice is to run the model a while, then pause, 'Show' the graph you're interested in (from the Displays panel), and step the model once. Use 'Hide All' to dismiss them again -- closing the windows using the 'X' will render the graphs irretrievable.

Documentation

• How to run model: RunGuide;
• Details of OrderBook specification and implementation;
• Specification for GenericPlayer and LimitOrder;
• Document with Cont equivalent STATA graphs for SP500; Excel file with S&P 500 data: 1950-2008
• Data Summary Document with Cont equivalent STATA graphs for SP500 Daily Quotes, IBM Per Transaction Quotes, and Simulation (Farmer and Cont models) Quotes
• Data Analysis Slides: comparison between SP500, IBM and "Mason"
• All datasets have been uploaded FTP server to: wean.css.gmu.edu

 

• Data Analysis Team Update

  We have re-produced Rama Cont's 9 graphics in StataIC 10 using 14,643 observations on the S&P 500 index (1950-2008 data). We have also been doing some original analysis of historic S&P data and are trying to familiarize ourselves with "R" software tools since we're concerned that Stata won't be able to handle the large quantities of simulated data the financial market model will be producing. We're also trying to solve the problem of finding daily market information to boost the number of observations we're working with.

  We're not sure where to post a more detailed report on what we've done since our last class so I'll be forwarding a Word document to several individuals whose E-mail addresses I have on hand.

  Cheryl, Thomas, and Shengle Lin

  chansen on Mar 20, 2008 @ 07:51pm UTC

• Sensitivity

  Our model does appear to be obtaining roughly the same results as Cont. The only hitch is the results seem to be somewhat more sensitive to the balance of the parameters than he lets on. With his second example (Fig 4), there definitely seems to be periodic cyclicality in returns, but it's not noticeable with such a large volume of data for returns, or such a small domain for acf of absolute returns...

  jtbriggs on Mar 07, 2008 @ 08:10pm UTC

• added comments

  i wasn't sure how descriptive i should get with commenting, so i followed a buddist middle path: added a comment for almost every line of code, but didn't get very specific. i did not put any comments in gui files, since i don't understand the functionality.

  Maciek: could you please verify the validity of the comments, as i wasn't always sure of the exact functionality.

  also, when i run the model with gui, the first two plots with price and acf do not really show anything. not sure if it's my settings or if there is a problem with gui.

  sabdukad on Mar 05, 2008 @ 03:08am UTC

Copyright © 2006–2008 by Assembla, LLC / Phone 1.781.328.2241 / E-mail: info@assembla.com