oxdk_day5b - Advanced Programming in Quantitative Economics...

Advanced Programming in Quantitative Economics
Introduction, structure, and advanced programming
techniques
17 – 21 August 2009, Aarhus, Denmark
Charles Bos
[email protected]
VU University Amsterdam
Tinbergen Institute
Advanced Programming in Quantitative Economics – p. 1

Day 5 - Afternoon
13.00L Efficiency and remaining topics
◦
Alternative algorithms
◦
Matrices
◦
Own code vs C-code
◦
Others
14.30P Finding the difference
◦
Concatenation
◦
Loop vs AR(p) vs ARFIMA package
15.30- Handing out exam, final remarks
Advanced Programming in Quantitative Economics – p. 2

Speed
•
Use matrices, avoid loops
•
Use the const argument qualifier
•
Use built-in functions
•
Optimise inner loop
•
Avoid using ‘hat’ matrices/outer products over large dimensions
•
Matrices are stored by row
•
Link in C or Fortran code
Advanced Programming in Quantitative Economics – p. 3

Speed: Loops vs matrices
Avoid loops like the plague.
Most of the time there is a matrix alternative, like for constructing
dummies:
speed loop2.ox
#include <oxstd.h>
#include <packages/oxutils/oxutils.h>
main()
{
decl iN, iR, vY, vDY, i, r;
iN= 10000;
iR= 1000;
vY= rann(iN, 1);
vDY= zeros(vY);
TrackTime("Loop");
for (r= 0; r < iR; ++r)
for (i= 0; i < iN; ++i)
if (vY[i] > 0)
vDY[i]= 1;
else
vDY[i]= -1;
TrackTime("Matrix");
for (r= 0; r < iR; ++r)
vDY= vY .> 0 .? 1 .: -1;
TrackTime(-1);
TrackReport();
}
Advanced Programming in Quantitative Economics – p. 4

Speed:
const
speed const.ox
SomeFuncConst(const mX)
{
// Do nothing
}
SomeFuncNotConst(mX)
{ // Do nothing
}
main()
{
decl iN, iK, iR, iRr, mX, r;
...
mX= rann(iN, iK);
for (r= 0; r < iR; ++r)
SomeFuncNotConst(mX);
for (r= 0; r < iR; ++r)
SomeFuncConst(mX);
}
Advanced Programming in Quantitative Economics – p. 5

Speed: Built-in functions
speed builtin.ox
#include <oxstd.h>
#include <packages/oxutils/oxutils.h>
MyOlsc(const vY, const mX, const avBeta)
{
avBeta[0]= invert(mX’mX)
*
mX’vY;
return !ismissing(avBeta[0]);
}
main()
{
decl iN, iK, iR, vY, mX, vBeta, r;
// Generate regression data
...
for (r= 0; r < iR; ++r)
MyOlsc(vY, mX, &vBeta);
for (r= 0; r < iR
*
iRr; ++r)
ols2c(vY, mX, &vBeta);
}
Advanced Programming in Quantitative Economics – p. 6

Speed: Inner/optimise
Target: Compute SSR= y − Xβ for a series of q different vectors of β ’s:
speed outer.ox
...
mBeta= vBeta + ranu(iK, iQ);
mE= vY - mX
*
mBeta;
TrackTime("Outer");
for (r= 0; r < iR; ++r)
vE2= diagonal(mE’mE);
TrackTime("Inner - matrix");
for (r= 0; r < iR; ++r)
vE2= sumc(mE .