Computer Science 286r
Homework 6
Due Monday, 28 Apr 2008, 11:59 pm

In this homework, you will consider the ideas of risk management,
multi-agent systems, and infrastructure discussed in the past few
classes.  Larger groups are expected to turn in more thorough 
responses; your entire submission should be on the order of one
single-spaced page of text per person in your group.

Part A. A Multi-Agent System for Risk Management

Pretend you are working for a hedge fund, and your pointy-haired boss
comes in and says, "I want you to build me a neural network that 
you can train on years of historical data to identify all of the 
risks that might lead to our portfolio losing more than 5% of
its value."

1. How would you respond?


Later, your pointy-haired boss's director walks in and says, "I
heard you know something about artificial markets.  I'd like you
to build us an artificial market -- a multi-agent system -- that
will test how our petroleum futures portfolio would hold up under
different market conditions."  She then mentions that you have
access to historical CFTC Commitment of Traders (COT) reports
(http://cftc.gov/marketreports/commitmentsoftraders/index.htm) and
United States DOE data (http://www.eia.doe.gov/).  She wants you
to model different kinds of energy traders in petroleum futures in
an artificial market to see how well hedged the portfolio is under
various scenarios, and use the historical data to inform your model.

2. Describe N different scenarios that might have a major impact on the 
   petroleum futures market that have not been seen in the last 5 years
   (where N is the number of people in your group.)

3. Think about an artificial market for risk analysis to test out these 
   scenarios by answering the following questions.  Be concise.
   a) Who would the simulated actors be?  
   b) Who among them will be trading?
   c) What data will the trading agents see?
   d) What kinds of rules will the trading agents follow based on these data?


Part B.  Understanding Computing Infrastructure

4. In class we discussed how much bandwidth it requires to get the 
   entire aggregated NASDAQ market feed.  You have as many of the 
   following servers as you want:
   - a 100 Mbps ethernet interface and 
   - two mirrored 750GB Seagate 7200 RPM SATA disks (total 750GB storage)
   - Intel 2.4 GHz Quad-Core Q6600 processor
   - 8 GB RAM

   a) How much bandwidth does the aggregated NASDAQ feed use?
   b) How many "servers" would be needed to accept such a feed and
      store it on local disk?
   c) How many days of data could you store before filling the disks?

5. Say you have found that by computing the implied volatility of 
   all of the real-time options prices of the S&P 500 equities and
   comparing them to the underlying prices and the futures prices,
   you can make a short-term prediction about the direction of the 
   e-Mini S&P futures traded on the CME.
   Assume the options and their underlying equities are quoted in 
   real-time on your feed in Manhattan, and your S&P futures are
   quoted in real-time on your feed in the Chicago Loop.

   a) What factors would you consider in determining whether you could
      take advantage of this prediction?
   b) Say you can predict the price movement in Chicago 10ms after 
      receiving your Manhattan data.  Is this useful?  What if the 
      prediction is 100ms?  1 second?
   c) Say you had to monitor all 500 S&P equities options chains,
      and the underlying equities, to make your prediction.  
      Describe the information you would require in order to
      design an appropriate infrastructure to do that.

6. Suggest N reasons why a bulge bracket firm would value reducing 
   execution delay by 10ms at $200 million USD in 2007, where N 
   is the number of people in your group.  Be concise.