Showing posts from October, 2013

Computational Investing with Dr Tucker Balch

Just completed a 8 week track to build computational models for back testing trading strategies, event profiling, portfolio optimization, and simulation of trading strategies to analyze portfolio performance and risks. It was a great course and Dr Tucker Balch is awesome at mixing the theoretical and practical aspects of the subject matter.
The course introduced me to some new concepts and how to implement them and from here the fun starts as I look for other use cases and investment strategies where I can apply the knowledge from this course.

I look forward to Module 2 on Computational Investing which will be focused primarily on how to implement Machine Learning in Algorithmic Trading. Here are some of the useful links if you want to follow more on this subject and the instructor.…

Bollinger Bands event profiling continuation

A series of plots for different events based on Bollinger Band values. For this analysis, I am using all the stocks that were listed in S&P500 in the year 2012. The period of event study is Jan 1 2008 to Dec 31 2009. I am looking at 4 extreme events for individual stocks whereby the bollinger value of the stock fell below 2 Standard deviations whereas the overall market (SPX / SPY) Bollinger Value went above certain levels.

The basic reason for this back testing event analysis is to graphically visualize the Market relative return and hypothesize a trading strategy based on the behaviour.

 Event Study 1 :

 f_bollinger_symbol_yest >= -2 and f_bollinger_symbol_today < -2 and mkt_bollinger_market_today >= 0.5

Event Study 2:

f_bollinger_symbol_yest >= -2 and f_bollinger_symbol_today < -2 and mkt_bollinger_market_today >= 1.0

Event Study 3:

f_bollinger_symbol_yest >= -2 and f_bollinger_symbol_today < -2 and mkt_bollinger_market_today >= 1.5

Event Study 4:


Event Profiling based on Bollinger Bands

So my experiments with event profiling continue. This time, I look for event based on Bollinger Values.

Classified the event as the day when the standardized Bollinger value of a stock (Stock Price - 20 day running mean)/(SD of 20 day running price) fell below -2 ( 2 Standard Deviations) and the market (SPY) Bollinger Value at the day's close was > 1 ( 1 SD above the mean)

Observations :
From the time of event  till day 12, we see a market relative mean of 2.25% .

Over the next few weeks I will be experimenting with other event classifications based on technical indicators to see if we can beat the market.

Dirty Food in NYC !

Courtesy NYC Open data project, I was able to get the food safety violation data for restaurants/fast-food joints in NYC.
I filtered the data starting July 1 2012 onward and narrowed down on 10 zip codes that top the list in food safety violations and within those zip codes, I narrowed down on top 20 cuisines with violations.

Soon to follow : Geo-mapped image of  top 20 restaurants with violation code and date of violation.