org.akutan.drivers
Class DiversificationDriver

java.lang.Object
  org.akutan.drivers.DiversificationDriver

public class DiversificationDriver
extends java.lang.Object

This class is a driver program which reads a file of industry sector returns from the Fama/French data and generates a sequence of portfolios using 60 month rolling returns. These portfolios are then evalutated on the basis of ex post return.

Since:

11 September 2007

Field Summary

protected java.lang.String[]	_assets Array of the asset class names for use on output
protected cern.colt.matrix.DoubleMatrix2D	_mktCaps Market Cap in some units for each sector for each month
protected int[]	_months Index of months to match up with the data (both returns and market caps)
protected cern.colt.matrix.DoubleMatrix2D	_returns Annualized returns stored as a decimal

Constructor Summary

DiversificationDriver()
Constructs

Method Summary

double	computeOutOfSample(int ptr, double[] w) Computes the out of sample 1 month return
double	computeOutOfSample12(int ptr, double[] w) Computes the out of sample 12 month return for the next 12 month period starting with ptr.
void	dumpLastMktCap() For debugging dumps out the last line read from the file
void	dumpLastReturns() For debugging dumps out the last line read from the file
protected void	dumpPoints(java.util.List<SolvedPoint> points)
cern.colt.matrix.DoubleMatrix1D	evolve(int ptr, double[] aw) Evolves a portfolio one step into the future and computes new weights
int[]	getMonths() Getter for use in JUnit
void	loadData(java.lang.String fileName) Loads return data by asset and date
void	loadExcessFrontData(java.lang.String fileName) Loads return data by asset and date
void	loadFrontData(java.lang.String fileName) Loads return data by asset and date
void	loadMktCap(java.lang.String fileName) Loads market cap data
static void	main(java.lang.String[] args)
protected java.util.List<EfficientFrontier.SampleData>	prepareEstimates(int start, int length, cern.colt.matrix.DoubleMatrix1D mu, cern.colt.matrix.DoubleMatrix2D V) Called to compute sample returns and variances for the period specified.
protected void	processResults(int start, int length, SolvedPoint sp) Called to porocess a point solution and see how it will do in out of sample testing.
protected void	runTest1(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest10(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest11(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest12(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest13(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest14(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest15(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest16(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest2(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest3(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest4(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest5(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest6(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest7(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected void	runTest8(int start, int length) A Black-Litterman based momentum strategy ala the one Fabozzi uses in his book on Robust Portfolio managerment.
protected void	runTest9(int start, int length) For a given starting month, computes the mean and covariance for the window of size length.
protected SolvedPoint	sortResults(java.util.List<SolvedPoint> l) Sorts by sharpe ratio (note this didn't work so well in the past for some unknown reason.
protected SolvedPoint	sortResults10(java.util.List<SolvedPoint> l) Estimates the point with risk of 10% and returns the closest pt less than 10%

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

_assets

protected java.lang.String[] _assets

Array of the asset class names for use on output

_months

protected int[] _months

Index of months to match up with the data (both returns and market caps)

_returns

protected cern.colt.matrix.DoubleMatrix2D _returns

Annualized returns stored as a decimal

_mktCaps

protected cern.colt.matrix.DoubleMatrix2D _mktCaps

Market Cap in some units for each sector for each month

Constructor Detail

DiversificationDriver

public DiversificationDriver()

Constructs

Method Detail

getMonths

public int[] getMonths()

Getter for use in JUnit

Returns:

Array of month ids (yyyymm) for the data

loadData

public void loadData(java.lang.String fileName)

Loads return data by asset and date

Parameters:

fileName - The name of the input file. It is expected to be a text file contaning columnar data, with a single line header of asset titles and the date in the first column from the left. Returns are specified in percent.

loadFrontData

public void loadFrontData(java.lang.String fileName)
                   throws java.lang.Exception

Loads return data by asset and date

Parameters:

fileName - The name of the input file. It is expected to be a text file contaning columnar data, with a single line header of asset titles and the date in the first column from the left. Returns are specified in percent.

Throws:

java.lang.Exception

loadExcessFrontData

public void loadExcessFrontData(java.lang.String fileName)
                         throws java.lang.Exception

Loads return data by asset and date

Parameters:

fileName - The name of the input file. It is expected to be a text file contaning columnar data, with a single line header of asset titles and the date in the first column from the left. Returns are specified in percent.

Throws:

java.lang.Exception

loadMktCap

public void loadMktCap(java.lang.String fileName)

Loads market cap data

Parameters:

fileName - The name of the input file. It is expected to be a text file containing columnar data, with a single line header of asset titles and the date in the first column from the left. Returns are specified in percent.

runTest1

protected void runTest1(int start,
                        int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the return solver for the generation of an efficient frontier.

Parameters:

start - Starting index of window in instance data

length - Length of window

computeOutOfSample

public double computeOutOfSample(int ptr,
                                 double[] w)

Computes the out of sample 1 month return

Parameters:

ptr - Points to the out of sample month

w - The asset weights to use

Returns:

The out of sample 1 month return

computeOutOfSample12

public double computeOutOfSample12(int ptr,
                                   double[] w)

Computes the out of sample 12 month return for the next 12 month period starting with ptr.

Parameters:

ptr - Index in the instance data of the out of sample start month

w - Asset weights

Returns:

Returns the out of sample return

evolve

public cern.colt.matrix.DoubleMatrix1D evolve(int ptr,
                                              double[] aw)

Evolves a portfolio one step into the future and computes new weights

Parameters:

ptr - Points to the next time period returns

aw - Starting Asset weights

Returns:

Returns the ending asset weights for the specified step

prepareEstimates

protected java.util.List<EfficientFrontier.SampleData> prepareEstimates(int start,
                                                                        int length,
                                                                        cern.colt.matrix.DoubleMatrix1D mu,
                                                                        cern.colt.matrix.DoubleMatrix2D V)

Called to compute sample returns and variances for the period specified.

Parameters:

start - Starting period

length - Length in periods

mu - On return contains the arithmetic mean of the returns

V - On return contains the covariance matrix for the period

Returns:

List of period return data

runTest2

protected void runTest2(int start,
                        int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes the actual mean, but the state variance to the StateReturnSolver for solution and generation of an efficient frontier.

Parameters:

start -

length -

runTest3

protected void runTest3(int start,
                        int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the diversified solver for the generation of an efficient frontier.

Parameters:

start -

length -

runTest16

protected void runTest16(int start,
                         int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the diversified solver for the generation of an efficient frontier.

Parameters:

start -

length -

runTest11

protected void runTest11(int start,
                         int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the entropy solver for the generation of an efficient frontier.

Parameters:

start -

length -

runTest4

protected void runTest4(int start,
                        int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the state preference solver and converts back to mean variance space before it solves for maximum sharpe ratio portfolio.

Parameters:

start -

length -

runTest5

protected void runTest5(int start,
                        int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the state preference solver and solves for maximum sharpe ratio portfolio in state space. Note this case turns out to be not that interesting as the maximum sharpe ratio portfolio in state space is very close to the minimum variance portfolio.

Parameters:

start -

length -

runTest6

protected void runTest6(int start,
                        int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the spherical shrinkage solver.

Parameters:

start -

length -

runTest7

protected void runTest7(int start,
                        int length)

For a given starting month, computes the mean and covariance for the window of size length. Then we just use a cap weighted portfolio and see how it works going forwards. This should be the lowest turnover portfolio.

Parameters:

start -

length -

runTest8

protected void runTest8(int start,
                        int length)

A Black-Litterman based momentum strategy ala the one Fabozzi uses in his book on Robust Portfolio managerment. We use the summary statistics from the last 9 months as the forecast views and apply them to the equlibrium to determine our asset forecasts.

Parameters:

start -

length -

dumpPoints

protected void dumpPoints(java.util.List<SolvedPoint> points)

runTest9

protected void runTest9(int start,
                        int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes the actual mean, but the state variance to the StateReturnSolver for solution and generation of an efficient frontier. We use buckets of size 2, so we'll have 5 buckets of 2 rather than 10 buckets of 1 as we did in test 1.

Parameters:

start -

length -

runTest10

protected void runTest10(int start,
                         int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the Simulated Annealing solver for generating the efficient frontier subject to cardinality and minimum investment constraints

Parameters:

start -

length -

runTest13

protected void runTest13(int start,
                         int length)

For a given starting month, computes the mean and covariance for the window of size length. Uses a diffuse a prior.

Parameters:

start -

length -

runTest12

protected void runTest12(int start,
                         int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the Simulated Annealing solver for generating the efficient frontier subject to cardinality and minimum investment constraints

Parameters:

start -

length -

runTest14

protected void runTest14(int start,
                         int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the return solver for the generation of an efficient frontier.

Parameters:

start - Starting index of window in instance data

length - Length of window

runTest15

protected void runTest15(int start,
                         int length)

For a given starting month, computes the mean and covariance for the window of size length. Then passes it along to the return solver for the generation of an efficient frontier.

Parameters:

start - Starting index of window in instance data

length - Length of window

sortResults10

protected SolvedPoint sortResults10(java.util.List<SolvedPoint> l)

Estimates the point with risk of 10% and returns the closest pt less than 10%

Parameters:

l - List of points on the efficient frontier

Returns:

The best fit point

sortResults

protected SolvedPoint sortResults(java.util.List<SolvedPoint> l)

Sorts by sharpe ratio (note this didn't work so well in the past for some unknown reason.

Parameters:

l - List of points on the efficient frontier in order of return from least to greatest.

Returns:

List of points on the efficient frontier in sharpe ratio order.

processResults

protected void processResults(int start,
                              int length,
                              SolvedPoint sp)

Called to porocess a point solution and see how it will do in out of sample testing.

Parameters:

start - Starting index of the out of sample area for evaluation

length - Length of the window for the sample

sp - Starting point to evolve with the data

dumpLastReturns

public void dumpLastReturns()

For debugging dumps out the last line read from the file

dumpLastMktCap

public void dumpLastMktCap()

For debugging dumps out the last line read from the file

main

public static void main(java.lang.String[] args)
                 throws java.lang.Exception

Parameters:

args -

Throws:

java.lang.Exception