A video on the consumption-based model - CCAPM

CCAPM is different from the traditional CAPM. Here is the 'root' formula of the consumption-based model of asset pricing:
                                                              
                               p = E(m x) = E(x* x) = <x*, x>

where
                p   - the asset price
                m  - stochastic discount factor
                x   - payoff
                x* - the unique discount factor in the payoff space X.

This video explains it in details.

Financial Data for Statistical Learning

There are plenty of financial and economic data which can be used in Machine Learning and Data Mining exercises. Here is list of mine:

FRED - Federal Research Economic Data.
CRSP - Center for Research in Security Prices
Ken-French - Fama-French 3-factor model data
John Cochrane - Data and Programs, Liquidity factor, Grumpy Economist.
Robet Shiller - Online data, and other research data.


Web Scraping is the new way to get the free "real-time" data.:)

Here is introduction given by Christopher Reeves.

Python (urllib, re, scrapy) and R(quantmod) are my favorite languages for FIN/ECON data scraping.

ffn

Intruction to ffn - Financial Functions for Python

ffn is a Python library for quantitative finance. It stands on the shoulders of giants (Pandas, Numpy, Scipy, etc.) and provides a vast array of utilities, from performance measurement and evaluation to graphing and common data transformations.

Its APIs support data retrieval, data manipulation, performance measurement, numerical routines and financial functions.

Statistical Significance vs. Economic Significance

Statistical Significance	Economic significance
Is it fitted well?	Is it an important factor?
[large] t-stat or [small] p-value	[large] bj values
Low t-stat => need more sample data?	Small bj values => Multicollinearity?

This is a sample regression output from fitlm() in Matlab:

Linear regression model:
    y ~ 1 + x1

Estimated Coefficients:
                   Estimate     SE           tStat     pValue    
    (Intercept)    0.0028283    0.0023652    1.1958    0.23514
    x1             0.91903      0.045009     20.419    2.5487e-34


Number of observations: 86, Error degrees of freedom: 84
Root Mean Squared Error: 0.0143
R-squared: 0.832,  Adjusted R-Squared 0.83
F-statistic vs. constant model: 417, p-value = 2.55e-34

Machine Learning vs Statistics (Statistical Learning)

In a sense, the difference between Machine Learning and Statistics (Statistical Learning) is what is the difference between a computer scientist and a statistician. In my humble opinion, they are becoming more similar as time goes by. On the other hand, Machine Learning is probably more empirical and works on problems by utilizing various optimization algorithms while Statistics (Statistical Learning) is more emphasizing on the assumptions and model validation -- more rigorous in mathematics and more often to talk about VC dimensions, KL divergence, conjugate prior, and etc. In practice, ML uses Matlab or Python while SL favors R. 

Rob Tibshirani, the author of "The Elements of Statistical Learning" gave a 'comparison of machine learning and statistics, here is a screenshot for your convenience:

In useR! 2004, Brian D. Ripley, another statistician, said: 'machine learning is statistics minus any checking of models and assumptions'.

A lecturer at Udacity gives a good comparison in one sentence, "Statistics focuses on analyzing the existing data and drawing valid conclusions, while Machine Learning focuses on making prediction and less worrying about the assumption as long as it makes good predictions".

Since 2001, the comparison become less  member the things have changed when William Cleveland introduced "data Science" as an independent discipline. As JOHN MOUNT said, "Machine learning and statistics may be the stars, but data science orchestrates the whole show."

In David Smith's blog, you can see CMU machine learning students "protest" at the G20 summit in Pittsburg, September 25 2009.

Network Analysis with NetworkX

NetworkX is a Python package for the creation, manipulation, and study of the structure, dynamics, and functions of complex networks.

NetworkX has been integrated with PyGraphviz, which is a Python interface to Graphviz -- the excellent open-source drawing tools initiated by AT&T Labs Research).

Here is the notes to build and integrate the Graphviz with NetworkX.

1, Install NetworkX

    I have WinPython 2.7, so installing NetworkX with WinPython's control panel is simple.

2, Install Graphviz

    Install Graphviz in "C:\graphviz"
    Add "C:\graphviz\bin" in PATH environment variable.
    Copy "C:\graphviz\include\graphviz" to "\python-2.7.x\include\"
    Copy cdt.lib and cgraph.lib in "C:\graphviz\lib\release\lib" to "\python-2.7.x\libs\"

3. Configure Visual C++ Compiler

    set VS90COMNTOOLS=C:\Program Files (x86)\Microsoft Visual Studio 11.0\Common7\Tools\

4. Build and Install PyGraphviz

    Open command prompt in WinPython and run
    pip install pygraphviz-1.3rc2.zip

Now import pygraphviz to test and start to use it with NetworkX.

Java Style Guide

Checkstyle is a tool to facilitate the use of coding standard. It allows us to define a coding style, and it can then check if our code follows the style automatically.

Not surprisingly, there exists an Eclipse plugin for Checkstyle. The URL of Checkstyle update site is:

     http://eclipse-cs.sf.net/update/

This is a sample checkstyle.xml of Google's Java Style Guide: google_checks.xml

Note that google_checks.xml is created for CheckStyle(CS) 5.8, not for CS 5.7. Since the Eclipse Plugin for CheckStyle is integrated with CS 5.7 (as of 10/12/2014), we may need to wait until the plugin is migrated to CS 5.8.