Here are the bits of my homework -- no copying!! :)
Link to the course: https://www.coursera.org/course/compinvesting1
All of these depend on having the Quant Software ToolKit install (http://wiki.quantsoftware.org/index.php?title=QuantSoftware_ToolKit)
We are building a tool-chain that looks at the stock market for
- certain events in the price history of an equity
- operationalize the events by ouputting a list of trades
- simulate the outcomes of the trades by backtesting
- analyze the outcome of the event-trading strategy by comparing to $SPX performace
###HW1 HW1 is a brute-force portfolio optimizer. Given an arbitrary list of equities, it finds a set of allocations that maximizes Sharpe Ratio. It only considers long positions.
###HW2 HW2 performs event studies. It is currently looking for instances of equities crossing below the $5 at the close of trading. It then outputs a PDF sudy showing the average price change and standard deviation of the next few trading days
###HW3 HW3 comes in two parts, a market simulator/backtester, and a performance analyzer.
####Sim The simulator takes as input a CSV file specifying trades to be executed. It assumes executing the orders at adjusted closing price for a trading day. It outputs a CSV that lists the daily closing value of assets held in the portfolio.
####Analyzer The Analyzer takes the Sim output and compares it against a specified benchmark. It looks at Sharpe Ratio and risk (standard deviation of returns), Average daily returns, and total return.
###HW4 HW4 is derived from HW2. It performs the same event study analysis as HW2 but instead of creating the event study, it outputs a CSV file of trades that can be fed into the HW3 Sim/Analyzer chain.
###HW5 HW5 is an implementation of Bollinger Bands. Obviously there are smarter (http://www.ta-lib.org/) ways of using technical indicators. But this is a nice exercise in using Pandas. The real neat idea in the module is the thought process around normalizing your indicator ouput -- by defining a standarized range of output for any and all indicators you use, machine learning techniques are more easily applied.